A method for reconstructing wing surface pressure, an electronic device, and a storage medium

Abstract

A method for reconstructing wing surface pressure, an electronic device and a storage medium, belonging to the technical field of wind tunnel test. To solve the problems of high efficiency and accuracy in wind tunnel pressure measurement tests. In the present invention, the original pressure data on the wing surface is collected through three-dimensional wing wind tunnel pressure measurement tests and three-dimensional wing model numerical simulations, and preprocessed to generate a deep neural network dataset for wing surface pressure reconstruction; a deep neural network model for wing surface pressure reconstruction is constructed, and datasets from two sources, namely wind tunnel tests and numerical simulations, are fused by modifying the loss function. The particle swarm optimization algorithm is used to optimize the model hyperparameters to obtain an optimized deep neural network model for training and testing; the method is applied to the model wind tunnel pressure measurement test to reconstruct the holographic pressure distribution on the wing surface, predict the aerodynamic load distribution data at non-measured points on the wing surface, and evaluate and verify the predicted holographic pressure distribution data. The present invention can be used for conventional pressure measurement tests of complex aircraft.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind tunnel tests, and particularly relates to a method for reconstructing the pressure on the wing surface, an electronic device, and a storage medium. Background Art

[0002] In all stages of aircraft design, the research on aerodynamics is crucial for estimating the aerodynamic characteristics of the aircraft. The pressure distribution on the surface of each component of the aircraft provides the original data of the aerodynamic load distribution for calculating the structural strength of the aircraft and each component, and at the same time provides a basis for studying the performance of the aircraft and each component and the flow characteristics around the model. It is a crucial part of aerodynamic research. Currently, the main ways to obtain the pressure distribution on the surface of each component of the aircraft are: wind tunnel pressure measurement tests and computational fluid dynamics numerical simulation and calculation methods (CFD, Computational Fluid Dynamics). However, both methods have defects and deficiencies.

[0003] The wind tunnel pressure measurement test technology is one of the conventional test technical capabilities of production wind tunnels. It is a landmark technology to measure whether the development of a wind tunnel test technology is mature, and it reflects the construction level of the test capabilities of the wind tunnel. The credibility of wind tunnel tests is relatively high, and the obtained aerodynamic force / load results are often used as the standard for evaluating the accuracy of numerical simulation methods. However, due to the long cycle of wind tunnel tests and the high dependence on the experience of test personnel, the rationality of the test plan directly affects the acquisition efficiency and effect of aerodynamic loads. Existing engineering practices generally believe that at least 50 to 100 pressure measurement holes need to be arranged on the airfoil surface, and the accuracy of the lift and pitching moment coefficients obtained through the integration of the pressure distribution is relatively credible. To obtain the complete flow field information on the airfoil surface, traditional methods usually arrange a sufficient number of pressure measurement holes on the airfoil surface for wind tunnel tests. Obtaining the pressure distribution on the entire airfoil surface through simple interpolation reconstruction requires a large number of pressure measurement holes. For complex aircraft, limited by spatial positions and test costs, the pressure measurement data obtained is insufficient, making the accuracy of traditional methods insufficient; moreover, the aerodynamic loads of complex aircraft in transonic wind tunnel tests are sensitive to parameters, and the measurement of refined aerodynamic forces / loads is more difficult and takes a longer time. Numerical simulation and calculation have the characteristics of simple implementation, convenience, and flexibility. However, due to the unclear physical model, the simulation of complex flows often has a large deviation from the real results and cannot achieve the same accuracy as wind tunnel tests. Summary of the Invention

[0004] The problem to be solved by the present invention is to solve the problems that the conventional pressure measurement test is limited by the spatial position and test cost, it is difficult to arrange a sufficient number of pressure measurement holes on the surface of a complex model to obtain complete surface pressure distribution information, and the accuracy of the lift and moment obtained by direct integration is insufficient; while the numerical simulation method, due to the unclear physical model, there are often certain discrepancies between the simulation of complex flows and the real results, and its laws cannot be simply applied to the wind tunnel test data. A method for reconstructing the surface pressure of an aircraft wing, an electronic device, and a storage medium are proposed.

[0005] To achieve the above object, the present invention is realized through the following technical solutions:

[0006] A method for reconstructing the surface pressure of an aircraft wing, comprising the following steps:

[0007] S1. Collect the original pressure data on the wing surface through three-dimensional wing wind tunnel pressure measurement tests and three-dimensional wing model numerical simulations. Preprocess the obtained original pressure data on the wing surface to construct a deep neural network dataset for wing surface pressure reconstruction;

[0008] S2. Construct a deep neural network model for wing surface pressure reconstruction. Through the modification of the loss function, the fusion of two types of data samples, namely experimental data and numerical simulation data, in the deep neural network dataset for wing surface pressure reconstruction obtained in step S1 is realized, and the constructed deep neural network model for wing surface pressure reconstruction is trained and tested;

[0009] S3. Optimize the hyperparameters of the deep neural network model for wing surface pressure reconstruction constructed in step S2 using the particle swarm optimization algorithm to obtain an optimized deep neural network model for wing surface pressure reconstruction;

[0010] S4. Use the optimized deep neural network model for wing surface pressure reconstruction obtained in step S3 and apply it to the wind tunnel pressure measurement test of a new aircraft model to reconstruct the holographic pressure distribution on the model wing surface, predict the aerodynamic load distribution data at non-measured points on the model wing surface, and evaluate and verify the predicted holographic pressure distribution data.

[0011] Further, the specific implementation method of step S1 includes the following steps:

[0012] S1.1. Collect the original pressure data on the wing surface through three-dimensional wing wind tunnel pressure measurement tests and three-dimensional wing model numerical simulations, including Mach number, angle of attack, sideslip angle, total pressure, Reynolds number, and surface pressure coefficient;

[0013] S1.2. Set the expected accuracy of the three-dimensional wing model such that the number of spanwise nodes of the three-dimensional wing is 57 - 81, and the number of chordwise nodes of the three-dimensional wing is 161 - 241. Perform mesh division according to the expected accuracy of the three-dimensional wing model to obtain the mesh nodes of the three-dimensional wing model, which are used as the holographic mesh nodes of the three-dimensional wing model;

[0014] S1.3. Arrange the holographic mesh nodes of the three-dimensional wing model obtained in step S1.2 in rows and columns according to the mesh nodes of the three-dimensional wing model to obtain the holographic mesh nodes of the unfolded three-dimensional wing model;

[0015] S1.4. Sample the original wing surface pressure data obtained in step S1.1 according to the coordinate information data of the holographic mesh nodes of the unfolded three-dimensional wing model obtained in step S1.3, and obtain the holographic mesh node pressure data of the three-dimensional wing model through interpolation; and obtain the two-dimensional unfolded coordinate information data of the corresponding holographic mesh nodes of the three-dimensional wing model according to the positions of the pressure measurement points in the wind tunnel pressure measurement experiment;

[0016] S1.5. Construct a deep neural network dataset for wing surface pressure reconstruction, including: the data of the deep neural network dataset for wing surface pressure reconstruction includes operating condition state parameters, expected accuracy grid point data, original surface pressure data from the pressure measurement experiment, position data of the holographic mesh nodes of the three-dimensional wing model corresponding to the original surface pressure data from the pressure measurement experiment, and numerical simulation surface pressure data of the holographic mesh nodes of the three-dimensional wing model corresponding to the original surface pressure data from the pressure measurement experiment.

[0017] Furthermore, the specific implementation method of step S2 includes the following steps:

[0018] S2.1. The wing surface pressure reconstruction deep neural network model adopts a two-dimensional U-shaped convolutional neural network structure, including a convolutional layer, a pooling layer, and a deconvolution layer. The convolutional kernel size of the convolutional layer is 3×3; the pooling layer is responsible for downsampling the spatial dimension of the input data, setting the receptive field of 2×2 as max pooling, and the sliding step size is 2; the convolutional kernel size of the deconvolution layer is 2×2 for upsampling;

[0019] S2.2. Set the learning rate, number of iterations, number of layers, number of neurons in each layer, batch size, activation function of the wing surface pressure reconstruction deep neural network model, and train it using the root mean square error as the loss function, and the root mean square error loss function The calculation formula is:

[0020] (1)

[0021] Among them, is the loss function, is the weight parameter, is the bias term parameter, is the number of training samples, and is the known sample data, is the model prediction value, is the model prediction function, is the input value is the function formula after forward propagation;

[0022] The training uses the backpropagation algorithm. Using the chain rule for differentiation, the partial derivatives of the loss function between the known sample data and the model prediction value with respect to each weight parameter or bias term are calculated, and then the weight parameters or bias term parameters are updated layer by layer in reverse. This is used to construct the deep neural network model for wing surface pressure reconstruction. The calculation formula for the gradient descent algorithm of the parameters in the deep neural network model for wing surface pressure reconstruction is:

[0023] (2)

[0024] (3)

[0025] Among them, is the l layer node to the l -1 layer node weight, is the layer node bias term, is the gradient operator, is the learning rate;

[0026] According to the chain rule for differentiation, the derivative of the loss function with respect to each weight or bias term is solved, and the calculation formula is:

[0027] (4)

[0028] (5)

[0029] Among them, 、 is the layer node output value, the layer node output value, is the layer node after calculation result, equals ;

[0030] Let the error of each neuron The calculation formula is:

[0031] (6)

[0032] Substituting formula (6) into formula (4) and formula (5), we get:

[0033] (7)

[0034] (8)

[0035] The calculation formula for the output layer is:

[0036] (9)

[0037] Substituting formula (9) into formula (6), the calculation formula for the error of each neuron is obtained as:

[0038] (10)

[0039] Substituting formula (10) into formula (7) and formula (8), the gradient update values of the weights and bias terms in the output layer can be calculated;

[0040] The calculation formula for the hidden layer is obtained according to the derivative formula of the composite function as:

[0041] (11)

[0042] Substituting formula (11) into formula (6), we get:

[0043] (12)

[0044] Substituting formula (12) into formula (7) and formula (8), the gradient update values of the weights and bias terms in the hidden layer are deduced; the model training is completed through iterative optimization;

[0045] S2.3. Calibrate the wing surface pressure reconstruction deep neural network model obtained in step S2.2: Modify the root mean square error loss function: By using the penalty coefficient for modification, the mean squared error loss function is obtained, and its calculation formula is:

[0046] (13)

[0047] Among them, is the number of numerical simulation sample data; is the number of wind tunnel test sample data; is the penalty coefficient of the numerical simulation sample data.

[0048] Furthermore, the specific implementation method of step S3 includes the following steps:

[0049] S3.1. First, improve the inertia factor in the particle swarm optimization algorithm by adopting a non-linear adjustment strategy. Perturb the global best particle by adjusting the inertia factor, and add stagnation detection. The calculation formula of the inertia factor is: by adjusting the inertia factor perturb the global best particle, and add stagnation detection. The inertia factor has the following calculation formula:

[0050] (14)

[0051] where represents the maximum inertia factor; represents the minimum inertia factor; represents the current iteration number; represents the maximum iteration number.

[0052] S3.2. Apply the improved particle swarm optimization algorithm in step S3.1 to the hyperparameter optimization of the wing surface pressure reconstruction deep neural network model constructed in step S2. Use the learning rate, iteration number, number of layers, number of neurons in each layer, batch size, and weights in the loss function as optimization parameters. Starting from a random solution, find the optimal hyperparameters through iteration;

[0053] S3.3. Use the optimal hyperparameters obtained in step S3.2 for the wing surface pressure reconstruction deep neural network model constructed in step S2 to obtain an optimized wing surface pressure reconstruction deep neural network model, and conduct training and testing on the optimized wing surface pressure reconstruction deep neural network model.

[0054] Furthermore, the specific implementation method of step S4 includes the following steps:

[0055] S4.1. Use the optimized wing surface pressure reconstruction deep neural network model obtained in step S3 to retrain and test, reconstruct the holographic pressure distribution on the wing surface of the model, and perform an inverse operation on the holographic pressure distribution result on the wing surface of the model to realize the visualization of the holographic pressure distribution on the wing surface of the model;

[0056] S4.2. Compare, verify and evaluate the prediction effects of the wing surface pressure reconstruction deep neural network models before and after optimization;

[0057] S4.3. Apply the obtained optimized wing surface pressure reconstruction deep neural network model to the model wind tunnel pressure measurement test, and conduct holographic surface pressure reconstruction modeling and verification evaluation on the wing component with limited pressure measurement points.

[0058] An electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the wing surface pressure reconstruction method are implemented.

[0059] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the wing surface pressure reconstruction method is implemented.

[0060] Advantages of the present invention:

[0061] For the wing surface pressure reconstruction method of the present invention, by fusing experimental data and CFD simulation data, the characteristics of different data sources are correlated with each other, the aerodynamic data consistency is improved, the internal relationship of the aerodynamic data is effectively explored, and the internal deviation of multi-source data is reduced.

[0062] The wing surface pressure reconstruction method of the present invention has commonality, certain universality and popularization, and can be used for the conventional pressure measurement test of complex aircraft. The three-dimensional model is unfolded into a two-dimensional model in the data preprocessing stage, which reduces the difficulty of model construction and improves the model training efficiency; and through the hyperparameter optimization algorithm, the problem of complex and inefficient process of manually setting and adjusting hyperparameters is solved. By fusing numerical simulation and wind tunnel test data, both the measured data of the wind tunnel test are utilized and the advantages of CFD technology can be exerted. The complete pressure distribution is reconstructed by using sparse experimental pressure measurement data, the spatial resolution of the wind tunnel test data is improved, and higher-accuracy aerodynamic characteristic data are provided for the aerodynamic load design of non-measurement points, solving the problem of refined reconstruction of distributed loads under the condition of limited space and sparse observation. Description of the Drawings

[0063] Figure 1 It is a flowchart of the wing surface pressure reconstruction method of the present invention;

[0064] Figure 2 It is a schematic diagram of holographic grid nodes of the three-dimensional wing model of the present invention;

[0065] Figure 3 It is a spatial schematic diagram of the three-dimensional to two-dimensional conversion of the three-dimensional wing model of the present invention. Detailed Embodiments

[0066] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only a part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention usually described and shown in the accompanying drawings here can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0067] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents the selected specific embodiments of the present invention. All other specific embodiments obtained by those skilled in the art based on the specific embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0068] To further understand the content, features and effects of the present invention, the following specific embodiments are exemplified and are accompanied by Figure 1 - Appendix Figure 3 The detailed description is as follows: Specific Embodiment 1:

[0070] A method for reconstructing the surface pressure of a wing, comprising the following steps:

[0071] S1. Collect the original pressure data on the wing surface through three-dimensional wing wind tunnel pressure measurement tests and three-dimensional wing model numerical simulations, preprocess the obtained original pressure data on the wing surface, and construct a deep neural network dataset for wing surface pressure reconstruction;

[0072] Further, the specific implementation method of step S1 includes the following steps:

[0073] S1.1. Collect the original pressure data on the wing surface through three-dimensional wing wind tunnel pressure measurement tests and three-dimensional wing model numerical simulations, including Mach number, angle of attack, sideslip angle, total pressure, Reynolds number, and surface pressure coefficient;

[0074] S1.2. Set the expected accuracy of the three-dimensional wing model to have the number of spanwise nodes of the three-dimensional wing being 57 - 81 and the number of chordwise nodes of the three-dimensional wing being 161 - 241. Perform mesh division according to the expected accuracy of the three-dimensional wing model to obtain the mesh nodes of the three-dimensional wing model, which are used as the holographic mesh nodes of the three-dimensional wing model;

[0075] S1.3. Arrange the holographic mesh nodes of the three-dimensional wing model obtained in step S1.2 in rows and columns according to the mesh nodes of the three-dimensional wing model to obtain the expanded holographic mesh nodes of the three-dimensional wing model;

[0076] S1.4. Sample the original wing surface pressure data obtained in step S1.1 according to the holographic grid node coordinate information data of the unfolded three-dimensional wing model obtained in step S1.3, and obtain the holographic grid node pressure data of the three-dimensional wing model through interpolation; and obtain the two-dimensional unfolded coordinate information data of the corresponding holographic grid nodes of the three-dimensional wing model according to the pressure measurement point positions of the wind tunnel pressure measurement test.

[0077] Furthermore, considering the spatial dependence of the pressure data of adjacent points on the model surface, sample the original surface pressure data of the numerical simulation in order according to the unfolded grid nodes; at the same time, consider the pressure measurement hole position information of the wind tunnel test data, correspond it to the holographic grid node positions, and obtain their corresponding position information; preprocess all the data according to the above requirements.

[0078] S1.5. Construct a deep neural network dataset for wing surface pressure reconstruction, including: the data of the deep neural network dataset for wing surface pressure reconstruction includes working condition state parameters, expected accuracy grid point data, original surface pressure data of the pressure measurement test, position data of the holographic grid nodes corresponding to the original surface pressure data of the pressure measurement test of the three-dimensional wing model, and numerical simulation surface pressure data of the holographic grid nodes corresponding to the original surface pressure data of the pressure measurement test of the three-dimensional wing model.

[0079] S2. Construct a deep neural network model for wing surface pressure reconstruction, realize the fusion of the two source data samples of the test data and the numerical simulation data in the deep neural network dataset for wing surface pressure reconstruction obtained in step S1 through the modification of the loss function, and train and test the constructed deep neural network model for wing surface pressure reconstruction.

[0080] Furthermore, the specific implementation method of step S2 includes the following steps:

[0081] S2.1. The deep neural network model for wing surface pressure reconstruction adopts a two-dimensional U-shaped convolutional neural network structure, including a convolutional layer, a pooling layer, and a deconvolutional layer, where the convolutional kernel size of the convolutional layer is 3×3; the pooling layer is responsible for downsampling the spatial dimension of the input data, sets the receptive field of 2×2 as the maximum pooling, and the sliding step size is 2; the convolutional kernel size of the deconvolutional layer is 2×2 for upsampling.

[0082] S2.2. Set the learning rate, number of iterations, number of layers, number of neurons in each layer, batch size, activation function of the deep neural network model for wing surface pressure reconstruction, and train with the root mean square error as the loss function, and the root mean square error loss function The calculation formula is:

[0083] (1)

[0084] Among them, is the loss function, is the weight parameter, is the bias term parameter, is the number of training samples, and are the known sample data, is the model prediction value, is the model prediction function, is the input value The function formula after forward propagation;

[0085] For the convolutional neural network model, the smaller the error between the predicted value and the actual value, the better the model. Therefore, the training process of the network structure is the process of minimizing the loss function, that is, solving the extreme point of the loss function;

[0086] The training uses the backpropagation algorithm. Using the chain rule of differentiation, calculate the partial derivative of the loss function between the known sample data and the model prediction value with respect to each weight parameter or bias term, and then update the weight parameter or bias term parameter layer by layer in reverse. This is used to construct the deep neural network model for wing surface pressure reconstruction. The calculation formula of the gradient descent algorithm for the parameters in the deep neural network model for wing surface pressure reconstruction is:

[0087] (2)

[0088] (3)

[0089] Among them, is the l layer node to the l -1 layer node weight, is the layer node bias term, is the gradient operator, is the learning rate;

[0090] According to the chain rule of differentiation, solve the derivative of the loss function with respect to each weight or bias term. The calculation formula is:

[0091] (4)

[0092] (5)

[0093] Among them, 、 is the layer node output value, the The output value of the layer node is the result calculated by the layer node after calculation, and is equal to ;

[0094] Let the error of each neuron be calculated by the formula:

[0095] (6)

[0096] Substituting formula (6) into formula (4) and formula (5) gives:

[0097] (7)

[0098] (8)

[0099] The calculation formula for the output layer is:

[0100] (9)

[0101] Substituting formula (9) into formula (6), the calculation formula for the error of each neuron is obtained as:

[0102] (10)

[0103] Substituting formula (10) into formula (7) and formula (8) can calculate the gradient update values of the weights and bias terms in the output layer;

[0104] The calculation formula for the hidden layer, obtained according to the composite function derivative formula, is:

[0105] (11)

[0106] Substituting formula (11) into formula (6) gives:

[0107] (12)

[0108] Substituting formula (12) into formula (7) and formula (8) to deduce the gradient update values of the weights and bias terms in the hidden layer; complete the model training through iterative optimization;

[0109] S2.3. Calibrate the wing surface pressure reconstruction deep neural network model obtained in step S2.2: Correct the root mean square error loss function: By using the penalty coefficient for correction, the mean squared error loss function is obtained​ The calculation formula is as follows:

[0110] (13)

[0111] Wherein, is the number of numerical simulation sample data; is the number of wind tunnel test sample data; is the penalty coefficient of numerical simulation sample data;

[0112] S3. Optimize the hyperparameters of the wing surface pressure reconstruction deep neural network model constructed in step S2 by using the particle swarm optimization algorithm to obtain the optimized wing surface pressure reconstruction deep neural network model;

[0113] Furthermore, the specific implementation method of step S3 includes the following steps:

[0114] S3.1. First, use a non-linear adjustment strategy to improve the inertia factor in the particle swarm optimization algorithm. Perturb the global best particle by adjusting the inertia factor . Add stagnation detection. If after continuous N iterations, the change of the global optimal particle is less than a certain threshold, it indicates that the population may have fallen into a local optimum and a stagnation phenomenon occurs. Then perturb the current particle to change its position, and the remaining particles are still updated using the original method; the inertia weight decreases in the case of algorithm stagnation and remains unchanged when the optimization is stable value; during this improvement process, the calculation expression of the inertia weight is:

[0115] (14)

[0116] Wherein, represents the maximum inertia factor; represents the minimum inertia factor; represents the current iteration number; represents the maximum iteration number.

[0117] S3.2. Apply the improved particle swarm optimization algorithm in step S3.1 to the hyperparameter optimization of the wing surface pressure reconstruction deep neural network model constructed in step S2. Use the learning rate, iteration number, number of layers, number of neurons in each layer, batch size, and weight value in the loss function as optimization parameters. Starting from a random solution, find the optimal hyperparameters through iteration;

[0118] S3.3. Use the optimal hyperparameters obtained in step S3.2 for the wing surface pressure reconstruction deep neural network model constructed in step S2 to obtain an optimized wing surface pressure reconstruction deep neural network model, and perform training and testing on the optimized wing surface pressure reconstruction deep neural network model;

[0119] S4. Apply the optimized wing surface pressure reconstruction deep neural network model obtained in step S3 to the wind tunnel pressure measurement test of the new aircraft model, reconstruct the holographic pressure distribution on the model wing surface, predict the aerodynamic load distribution data at the non-measured points on the model wing surface, and evaluate and verify the predicted holographic pressure distribution data;

[0120] Further, the specific implementation method of step S4 includes the following steps:

[0121] S4.1. Use the optimized wing surface pressure reconstruction deep neural network model obtained in step S3 to retrain and test, reconstruct the holographic pressure distribution on the model wing surface, and perform an inverse operation on the holographic pressure distribution result on the model wing surface to realize the visualization of the holographic pressure distribution on the model wing surface;

[0122] S4.2. Compare, verify and evaluate the prediction effects of the wing surface pressure reconstruction deep neural network models before and after optimization;

[0123] S4.3. Apply the obtained optimized wing surface pressure reconstruction deep neural network model to the wind tunnel pressure measurement test of the aircraft model, and perform holographic surface pressure reconstruction modeling and verification evaluation on the wing components with limited pressure measurement points.

[0124] From Figure 3 It can be seen that the mutual conversion operation of the three-dimensional wing model from three-dimensional to two-dimensional mainly relies on the different expressions of the surface grid node position information data. The three-dimensional position information data of the same node can be represented in the form of coordinates, and the two-dimensional can be displayed in the form of numbers in different directions, with a one-to-one correspondence. The mutual conversion operation of the three-dimensional wing model from three-dimensional to two-dimensional is realized through the mutual conversion of different position information data representation methods. Specific Embodiment 2:

[0126] An electronic device includes a memory and a processor. The memory stores a computer program, and the processor implements the steps of the wing surface pressure reconstruction method when executing the computer program.

[0127] The computer device of the present invention can be a device including a processor and a memory, such as a single-chip microcomputer including a central processing unit. And the processor is used to implement the steps of the above wing surface pressure reconstruction method when executing the computer program stored in the memory.

[0128] The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0129] The memory may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices. Specific Embodiment Three:

[0131] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the described wing surface pressure reconstruction method is implemented.

[0132] The computer-readable storage medium of the present invention may be any form of storage medium readable by the processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. A computer program is stored on the computer-readable storage medium. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the described wing surface pressure reconstruction method can be implemented.

[0133] The computer program includes computer program code, which may be in the form of source code, object code, executable files, or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, external hard drives, magnetic disks, optical discs, computer memories, read-only memories (ROMs), random access memories (RAMs), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0134] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0135] Although the present application has been described above with reference to specific embodiments, various improvements can be made thereto and components therein can be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application can be combined with each other in any way. The reason for not exhaustively describing the situations of these combinations in this specification is only to save space and resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for reconstructing wing surface pressure, characterized in that, It includes the following steps: S1. Collect the original pressure data on the wing surface through three-dimensional wing wind tunnel pressure measurement tests and numerical simulations of three-dimensional wing models. Preprocess the obtained original pressure data on the wing surface to construct a deep neural network dataset for wing surface pressure reconstruction; S2. Construct a deep neural network model for wing surface pressure reconstruction. Through the modification of the loss function, fuse the two-source data samples, namely experimental data and numerical simulation data, in the deep neural network dataset for wing surface pressure reconstruction obtained in step S1, and conduct training and testing on the constructed deep neural network model for wing surface pressure reconstruction; The specific implementation method of step S2 includes the following steps: S2.

1. The deep neural network model for wing surface pressure reconstruction adopts a two-dimensional U-shaped convolutional neural network structure, including a convolutional layer, a pooling layer, and a deconvolution layer. The convolutional kernel size of the convolutional layer is 3×3; the pooling layer is responsible for downsampling the spatial dimension of the input data. The receptive field of 2×2 is set as max pooling, and the sliding step is 2; the convolutional kernel size of the deconvolution layer is 2×2 for upsampling; S2.

2. Set the learning rate, number of iterations, number of layers, number of neurons in each layer, batch size, and activation function of the deep neural network model for wing surface pressure reconstruction, and train it using the root mean square error as the loss function. The root mean square error loss function Loss MSE is calculated as follows: Among them, E(W, b) is the loss function, W is the weight parameter, b is the bias term parameter, n is the number of training samples, x j and y j are known sample data, is the model prediction value, g[h W,b (x j )] is the model prediction function, h W,b (x j ) is the function formula of the input value x j after forward propagation; The training uses the backpropagation algorithm. Use the chain rule of differentiation to calculate the partial derivatives of the loss function between the known sample data and the model prediction values with respect to each weight parameter or bias term, and then update the weight parameters or bias term parameters layer by layer in reverse to construct a deep neural network model for wing surface pressure reconstruction; S2.

3. Calibrate the wing surface pressure reconstruction deep neural network model obtained in step S2.2: Modify the root mean square error loss function: Modify it by using the penalty coefficient ρ to obtain the mean squared error loss function Loss MSE1 The calculation formula of Among them, p is the number of numerical simulation sample data; q is the number of wind tunnel test sample data; ρ is the penalty coefficient; S3. Optimize the hyperparameters of the deep neural network model for wing surface pressure reconstruction constructed in step S2 using the particle swarm optimization algorithm to obtain an optimized deep neural network model for wing surface pressure reconstruction; S4. Use the optimized deep neural network model for wing surface pressure reconstruction obtained in step S3 and apply it to the wind tunnel pressure measurement test of the new aircraft model to reconstruct the holographic pressure distribution on the model wing surface, predict the aerodynamic load distribution data at the non-measured points on the model wing surface, and evaluate and verify the predicted holographic pressure distribution data.

2. A method for reconstructing the surface pressure of a wing according to claim 1, characterized in that, The specific implementation method of step S1 includes the following steps: S1.

1. Collect the original pressure data on the wing surface through three-dimensional wing wind tunnel pressure measurement tests and numerical simulations of three-dimensional wing models, including Mach number, angle of attack, sideslip angle, total pressure, Reynolds number, and surface pressure coefficient; S1.

2. Set the expected accuracy of the three-dimensional wing model. The number of spanwise nodes of the three-dimensional wing is 57-81, and the number of chordwise nodes of the three-dimensional wing is 161-241. Perform grid division according to the expected accuracy of the three-dimensional wing model to obtain the grid nodes of the three-dimensional wing model, which are used as the holographic grid nodes of the three-dimensional wing model; S1.

3. Arrange the holographic grid nodes of the three-dimensional wing model obtained in step S1.2 in rows and columns according to the grid nodes of the three-dimensional wing model to obtain the expanded holographic grid nodes of the three-dimensional wing model; S1.

4. Sample the original wing surface pressure data obtained in step S1.1 according to the holographic grid node coordinate information data of the unfolded three-dimensional wing model obtained in step S1.3, and obtain the holographic grid node pressure data of the three-dimensional wing model through interpolation; and obtain the two-dimensional unfolded coordinate information data of the corresponding holographic grid nodes of the three-dimensional wing model according to the positions of the pressure measurement points in the wind tunnel pressure measurement test. S1.

5. Construct a deep neural network dataset for wing surface pressure reconstruction, including: the data of the deep neural network dataset for wing surface pressure reconstruction includes working condition state parameters, expected accuracy grid point data, original surface pressure data of the pressure measurement test, position data of the holographic grid nodes of the three-dimensional wing model corresponding to the original surface pressure data of the pressure measurement test, and numerical simulation surface pressure data of the holographic grid nodes of the three-dimensional wing model corresponding to the original surface pressure data of the pressure measurement test.

3. A method for reconstructing the wing surface pressure according to claim 2, characterized in that The specific implementation method of step S3 includes the following steps: S3.

1. First, improve the inertia factor ω in the particle swarm optimization algorithm by using a non-linear adjustment strategy, perturb the global best particle by adjusting the inertia factor ω, and increase the stagnation detection. The calculation formula of the inertia factor ω is: Among them, ω max represents the maximum inertia factor, ω min represents the minimum inertia factor, i represents the current iteration number, and Maxinter represents the maximum number of iterations; S3.

2. Apply the improved particle swarm optimization algorithm in step S3.1 to the hyperparameter optimization of the wing surface pressure reconstruction deep neural network model constructed in step S2. Use the learning rate, number of iterations, number of layers, number of neurons in each layer, batch size, and penalty coefficient in the loss function as optimization parameters, and start from a random solution to find the optimal hyperparameters through iteration. S3.

3. Apply the optimal hyperparameters obtained in step S3.2 to the wing surface pressure reconstruction deep neural network model constructed in step S2 to obtain an optimized wing surface pressure reconstruction deep neural network model, and perform training and testing on the optimized wing surface pressure reconstruction deep neural network model.

4. A method for wing surface pressure reconstruction according to claim 3, characterized in that The specific implementation method of step S4 includes the following steps: S4.

1. Use the optimized wing surface pressure reconstruction deep neural network model obtained in step S3 to retrain and test, reconstruct the holographic pressure distribution on the model wing surface, and perform an inverse operation on the holographic pressure distribution result on the model wing surface to realize the visualization of the holographic pressure distribution on the model wing surface. S4.

2. Compare, verify, and evaluate the prediction effects of the wing surface pressure reconstruction deep neural network models before and after optimization. S4.

3. Apply the obtained optimized wing surface pressure reconstruction deep neural network model to the model wind tunnel pressure measurement test to perform holographic surface pressure reconstruction modeling and verification and evaluation on the wing components with limited pressure measurement points.

5. An electronic device, characterized in that, It includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for wing surface pressure reconstruction according to any one of claims 1-4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a method for wing surface pressure reconstruction according to any one of claims 1-4.

Images