Reconstruction Method and Device for Wing Holographic Pressure Coefficient in Wind Tunnel Experiment

Abstract

An embodiment of the present invention discloses a method and device for reconstructing the holographic pressure coefficient of an airfoil for wind tunnel experiments, which relates to the sensor layout technology for wind tunnel experiments and can reconstruct reasonable holographic data based on the real data of wind tunnel experiments, reducing the amount of calculation and improving the utilization rate of wind tunnel experiment data. The present invention includes: receiving the original wind tunnel experiment data, operating condition parameter, and fineness expectation information sent by a client; constructing a training sample database using the original wind tunnel experiment data; training a deep learning model using the sample database; constructing a holographic grid node layout corresponding to the fineness expectation; using the trained deep learning model and the holographic grid node layout to perform reconstruction processing on the holographic pressure coefficient of the airfoil, and sending the processing result to the client.

Description

Technical Field

[0001] The present invention relates to the technology of sensor layout in wind tunnel experiments, and particularly to a method and device for reconstructing the holographic pressure coefficient of an airfoil for wind tunnel experiments. Background Art

[0002] The distribution of the pressure coefficient on the airfoil surface is an important characterization of the aerodynamic performance of the airfoil and plays a crucial role in the design of aircraft. Computational Fluid Dynamics (CFD) assisted by wind tunnel experiments is the mainstream method for obtaining the holographic pressure coefficient. First, a limited number of sensors are placed at specific positions on the airfoil surface to collect the corresponding physical quantities of aerodynamic performance in wind tunnel experiments, and these physical quantities are usually regarded as true values; then, CFD-based analysis and calculation are carried out under the same working condition conditions to obtain holographic data; finally, the true values collected in wind tunnel experiments are used to correct the accuracy of the results of CFD analysis and calculation to ensure the validity of the simulation results. This operation mode has been widely applied in the laboratory research and product verification processes of major scientific research institutions and technology enterprises.

[0003] However, high-quality CFD analysis and calculation require a large amount of computing resources and high requirements for computing personnel, which have seriously dragged down the experimental cycle and the product verification cycle. Therefore, how to reduce the dependence on CFD simulation calculation has become a problem to be further studied and solved. Summary of the Invention

[0004] Embodiments of the present invention provide a method and device for reconstructing the holographic pressure coefficient of an airfoil for wind tunnel experiments, which can reconstruct reasonable holographic data relying on the true data of wind tunnel experiments, reduce the amount of calculation and improve the utilization rate of wind tunnel experiment data.

[0005] To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

[0006] In the first aspect, the method provided by the embodiments of the present invention includes:

[0007] S1. Receiving the original data of the wind tunnel experiment, the working condition parameter, and the fineness expectation information sent by the client;

[0008] S2. Using the original data of the wind tunnel experiment to construct a training sample database;

[0009] S3. Using the sample database to train a deep learning model;

[0010] S4. Constructing a holographic grid node arrangement corresponding to the fineness expectation;

[0011] S5. Using the trained deep learning model and the holographic grid node arrangement to perform reconstruction processing on the holographic pressure coefficient of the airfoil, and sending the processing result to the client.

[0012] In a second aspect, the device provided by an embodiment of the present invention includes:

[0013] A data receiving module, configured to receive the original wind tunnel experiment data, operating condition parameters, and fineness expectation information sent by a client;

[0014] A sample processing module, configured to construct a training sample database by using the original wind tunnel experiment data;

[0015] A training module, configured to train a deep learning model by using the sample database;

[0016] A grid adjustment module, configured to construct a holographic grid node arrangement corresponding to the fineness expectation;

[0017] A data processing module, configured to reconstruct the wing holographic pressure coefficient by using the trained deep learning model and the holographic grid node arrangement;

[0018] A data sending module, configured to send the processing result to the client.

[0019] The method and device for reconstructing the wing holographic pressure coefficient for wind tunnel experiments provided by the embodiments of the present invention use the original wind tunnel experiment data provided by the client and the fineness expectation value for reconstructing the holographic pressure coefficient, perform modeling analysis on the wing holographic pressure coefficient on the server side, first obtain the original data of the wind tunnel experiment, construct a sample database by using the original data according to the grid arrangement suitable for deep learning training, train a deep learning model, construct a grid node arrangement suitable for holographic pressure coefficient reconstruction according to the fineness expectation value of the client, reconstruct the wing holographic pressure coefficient by using the trained model, and finally return the wing holographic pressure coefficient meeting the fineness requirement to the client for selection. This solution does not require using CFD simulation that consumes computing resources to obtain holographic data, and can reconstruct reasonable holographic data only relying on the real data of wind tunnel experiments, reducing the amount of calculation and improving the utilization rate of wind tunnel experiment data. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0021] Figure 1 It is a schematic diagram of the hardware environment provided by an embodiment of the present invention;

[0022] Figure 2 It is a flowchart of the method provided by an embodiment of the present invention;

[0023] Figure 3 Schematic diagram of the original grid of the wind tunnel experiment;

[0024] Figure 4 Schematic diagram of the deep learning model structure;

[0025] Figure 5 Schematic diagram of the method framework for wing holographic pressure coefficient reconstruction;

[0026] Figure 6 Schematic diagram of the device structure provided by the embodiment of the present invention. Detailed implementation manners

[0027] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners. The implementation manners of the present invention will be described in detail below. Examples of the implementation manners are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions from beginning to end. The implementation manners described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as limiting the present invention. Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The phrase "and / or" used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art of the present technology can understand that unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless defined as here.

[0028] The embodiment of the present invention provides a method for reconstructing the wing holographic pressure coefficient for a wind tunnel experiment, and this method can be applied to a hardware system as shown in Figure 1 as shown below, and this method is as shown in Figure 2 below and includes:

[0029] S1. Receive the original data of the wind tunnel experiment, the working condition parameter, and the expected fineness information sent by the client.

[0030] S2. Use the original data of the wind tunnel experiment to construct a training sample database. Specifically, organize the original wind tunnel data according to the grid node arrangement method applicable to the training of the deep learning model, and then construct the training sample database.

[0031] S3. Use the sample database to train the deep learning model.

[0032] S4. Construct a holographic grid node arrangement corresponding to the expected fineness. Specifically, construct a holographic grid node arrangement applicable to the reconstruction of the holographic pressure coefficient according to the expected fineness value.

[0033] S5. Use the trained deep learning model and the holographic grid node arrangement to reconstruct the holographic pressure coefficient of the wing, and send the processing result to the client.

[0034] In this embodiment, it further includes: the client collects the pressure coefficient on the surface of the wing model through sensors arranged at specified positions on the surface of the wing model. The specified positions are the traditional sensor arrangement positions in the wind tunnel experiment, as Figure 3 shown. And the client obtains the expected fineness information, which includes the number of holographic grid nodes for the reconstruction of the holographic pressure coefficient.

[0035] Specifically, the original data of the wind tunnel experiment specifically refers to: in the wind tunnel blowing experiment under a certain working condition, the pressure coefficients are collected by sensors arranged at multiple positions on the surface of the wing model. For each sensor, its original attribute data form is (x, y, z, M, α, β, C p ), where x, y, z represent the three-dimensional coordinates in space, M represents the Mach number, α represents the angle of attack, β represents the sideslip angle, and C p represents the pressure coefficient. The working condition parameters include the Mach number, the angle of attack, and the sideslip angle, which together with the three-dimensional space coordinates of the sensors form the input data of the present invention. Therefore, the original data corresponding to one wind tunnel experiment is multiple (x, y, z, M, α, β, C p ) vectors. For example, Figure 3 as shown, the original data under this sensor arrangement is a 5×25×7 matrix, where 5 is the number of cross-sections, 25 is the number of sensors on each cross-section, and 7 is the vector (x, y, z, M, α, α, C p) dimension. Due to the limited number of sensor arrangements in the wind tunnel experiment, the pressure coefficient data is highly sparse in space. The present invention can use the original data under multiple working conditions to construct a training sample database. Therefore, the form of the wind tunnel original data provided by the client is a matrix of N×S×G×7, where N is the number of groups of working conditions, S is the number of cross-sections, G is the number of sensors on each cross-section, and 7 is the vector (x, y, z, M, β, β, C p ) dimension.

[0036] Specifically, the expected fineness value refers to the number of holographic grid nodes applicable to the reconstruction of the holographic pressure coefficient. The holographic grid nodes are different from the physical grids used in CFD calculations. The present invention uses the arrangement of sensors in the original wind tunnel experiment as the grid. As Figure 3 shown, this grid has a total of 5 cross-sections, with 25 points on each cross-section. Then the corresponding expected fineness information is 5×25. Obviously, the expected fineness of 125 cannot present the holographic information of the wing. Therefore, a holographic grid needs to be constructed based on a larger expected fineness. The larger expected fineness can be achieved from three different perspectives: First, increase the number of cross-sections. For example, increase 195 cross-sections. At this time, the expected fineness information is 200×25. Second, increase the number of nodes on the cross-section. For example, increase 75 points on each cross-section. At this time, the expected fineness information is 5×100. Third, increase both the number of cross-sections and the number of nodes on the cross-section. For example, first increase 75 points on each cross-section, and then increase 195 cross-sections. At this time, the expected fineness information is 200×100. Therefore, the format of the expected fineness information input by the client is (number of cross-sections, number of nodes on each cross-section). According to this index, the server can output holographic pressure coefficients with different fineness levels.

[0037] Furthermore, the method of arranging sensors at specified positions on the surface of the wing model includes:

[0038] Select cross-sections at intervals from the end near the fuselage to the wing tip, and arrange sensors at intervals on the selected cross-sections respectively. Among them, the number of selected cross-sections is not less than 5. And in each cross-section, the density of sensors at the leading edge and the trailing edge is greater than that at other positions in the cross-section. In this embodiment, it is required that the selection of cross-sections covers the entire wing as much as possible, such as Figure 3 the 5 cross-sections in, which are more evenly distributed throughout the wing. For the arrangement of sensors on a single cross-section, the leading edge and the trailing edge should be denser than other positions.

[0039] In a preferred embodiment, 25 sensors are arranged in each cross-section. Among them, sensors numbered 1 to 13 are located on the lower wing surface, and sensors numbered 14 to 25 are located on the upper wing surface. In practical applications, the grid node arrangement method applicable to deep learning model training specifically refers to: in order to increase the universality of the present invention, the grid node arrangement method applicable to deep learning model training directly uses the original arrangement of sensors in the wind tunnel experiment. Figure 3 The original arrangement of sensors in the wind tunnel experiment is shown. Five cross-sections are relatively evenly distributed on the entire wing. 25 sensors are arranged in an orderly manner on each cross-section. Sensors numbered 1 to 13 are located on the lower wing surface, and sensors numbered 14 to 25 are located on the upper wing surface.

[0040] Among them, in the process of constructing a training sample database using the original wind tunnel experiment data, it includes: obtaining the spatial coordinates, pressure coefficient values, and operating condition parameters of each sensor arranged on the wing model surface as the original sensor data. The sensors on each cross-section are divided into ordered subsequences in the grid. Among them, starting from each node, at least 4 nodes are taken backward to form a subsequence with a length of at least 5. Specifically, in this embodiment, constructing a training sample database specifically means: in the wind tunnel experiment, there is an interaction between adjacent points on the wing cross-section, and this interaction is called the spatial dependence between nodes. The present invention learns the distribution law of the pressure coefficient on the wing surface in the wind tunnel experiment by modeling the spatial dependence between adjacent nodes, and further realizes the purpose of reconstructing a reasonable holographic pressure coefficient from sparse pressure coefficients. The modeling method is the deep learning model designed by the present invention, and its core module is the LSTM that is good at capturing the implicit physical laws in sequence data. The LSTM requires the input data to be in the form of a sequence, and the sequence needs to have a fixed sequence length. Therefore, it is necessary to make an adaptive adjustment to the grid nodes of the original wind tunnel experiment data. The present invention divides the original sequence on each cross-section in the grid node arrangement into multiple ordered subsequences. The original sequence refers to Figure 3 the long sequence numbered 1 to 25 along the cross-section direction in. The division method is: starting from each node, 4 nodes are taken backward to form a subsequence with a length of 5. For example, numbers 1 to 5 form the first subsequence, numbers 2 to 6 form the second subsequence, and so on. A total of 25 subsequences are obtained. Starting from the node numbered 22, the part exceeding number 25 is supplemented starting from number 1. For example, (23, 24, 25, 1, 2) is the subsequence at number 23. Finally, for a single cross-section, taking Figure 3 as an example, a total of 25 subsequences with a length of 5 are obtained. The data form of the five cross-sections is 5×25×5×7, where 7 is the attribute vector of each node (x, y, z, M, α, β, C p) dimension. For customers with wind tunnel experiment data under multiple sets of working conditions, a training sample database for multiple sets of working conditions can be constructed, thereby reducing the model training time consumption when using the present invention subsequently. The data form of the training sample database for multiple sets of working conditions is a matrix of N×5×25×5×7, where N is the number of sets of working conditions.

[0041] In this embodiment, training the deep learning model using the sample database includes: using the spatial three-dimensional coordinates of the grid nodes and the working condition parameters as the input of the deep learning model, and using the measured wind tunnel experiment pressure coefficient corresponding to the grid nodes as the label to obtain a data matrix for model training. Among them, the structure of the deep learning model includes: three layers of transposed convolution, three layers of bidirectional LSTM, and four layers of fully connected layers. Specifically, training the deep learning model specifically means: using the spatial three-dimensional coordinates of the grid nodes and the three-dimensional working condition parameters together as the input of the model, that is, (x, y, z, M, α, β), and the measured wind tunnel experiment pressure coefficient corresponding to this grid node as the label, that is, C p . Finally, the data form for model training is a matrix of N×5×25×5×6, and the data form of the corresponding label is a matrix of N×5×25×5×1.

[0042] The loss function for supervised training is: Among them, N represents the last node in the subsequence, and y (i) represents the true value of the i-th node in the subsequence, that is, the measured wind tunnel experiment pressure coefficient, represents the predicted value of the i-th node in the subsequence. Among them, the deep learning model is implemented by a computer language, and the result is saved in the server for subsequent use after training.

[0043] In the preferred solution, the kernel size of each layer of transposed convolution is 3, and the stride is 2. The hidden output dimension of the three layers of bidirectional LSTM is 713. The hidden output dimensions of the four layers of fully connected layers are 1426, 713, 356, and 1 respectively. Specifically, the specific structure of the deep learning model is: three layers of transposed convolution, three layers of bidirectional LSTM, and four layers of fully connected layers. The kernel size of the transposed convolution is 3, and the stride is 2. The hidden output dimension of the three layers of bidirectional LSTM is 713. The hidden output dimensions of the four layers of fully connected layers are 1426, 713, 356, and 1 respectively. The data flow of the deep learning model structure is specifically: the input data is first upsampled to a high-dimensional space through three layers of transposed convolution, then the three layers of LSTM model the spatial dependence relationship of the data in the high-dimensional space and give a coarse-grained output, and finally the four layers of fully connected layers further fit the coarse-grained output to obtain the final model output. For example Figure 4As shown, it is a schematic diagram of the deep learning model structure. The deep learning model is specifically composed of three sub-modules: three layers of transposed convolution, three layers of bidirectional LSTM, and four layers of fully connected layers. The sub-modules are connected in series. The kernel size of the transposed convolution is 3, and the stride is 2. The hidden output dimensions of the three layers of bidirectional LSTM are all 713. The hidden output dimensions of the four layers of fully connected layers are 1426, 713, 356, and 1 respectively. The data flow of the deep learning model structure is specifically as follows: The input data is first upsampled to a high-dimensional space through three layers of transposed convolution, then the three layers of LSTM model the spatial dependence relationship of the data in the high-dimensional space and give a coarse-grained output, and finally the four layers of fully connected layers further fit the coarse-grained output to obtain the final model output.

[0044] The construction of the holographic grid node arrangement corresponding to the expected fineness includes: on the basis of the data grid arrangement of the training samples, the number of nodes on the grid arrangement is increased by using the method of inserting nodes at equal intervals. Among them, what is increased in the data grid arrangement of the training samples includes: increasing the number of cross-sections and the number of nodes on a single cross-section. Specifically, in practical applications of this embodiment, it is necessary to construct a holographic grid node arrangement suitable for holographic pressure coefficient reconstruction according to the expected fineness value. Among them, according to the expected fineness value, on the basis of the training sample data grid arrangement, the method of inserting nodes at equal intervals can be used to increase the number of nodes on the grid arrangement between a single airfoil cross-section and cross-sections to meet the requirements of the expected fineness value. The holographic grid nodes are different from the physical grids used in CFD calculations. In the present invention, the arrangement of sensors in the original wind tunnel experiment is used as the grid, as Figure 3 shown. This grid has a total of 5 cross-sections, with 25 points on each cross-section. Then the corresponding expected fineness information is 5×25. Obviously, the expected fineness of 125 cannot present the holographic information of the wing. Therefore, it is necessary to construct a holographic grid based on a larger expected fineness. The larger expected fineness can be achieved from three different perspectives:

[0045] First, increase the number of cross-sections. For example, increase 195 cross-sections. At this time, the expected fineness information is 200×25.

[0046] Second, increase the number of nodes on the cross-section. For example, increase 75 points on each cross-section. At this time, the expected fineness information is 5×100.

[0047] Third, increase the number of cross-sections and the number of nodes on the cross-section at the same time. For example, first increase 75 points on each cross-section, and then increase 195 cross-sections. At this time, the expected fineness information is 200×100. Therefore, the format of the expected fineness information input by the client is (number of cross-sections, number of nodes on each cross-section). Specifically, increase in the third way Figure 3Taking the expected information on the fineness of the medium grid arrangement as an example, 75 points are added to each cross-section, and at the same time, 100 cross-sections are added. Adding 75 points to a single cross-section is equivalent to evenly inserting 3 points between every two of the original 25 nodes. This way of inserting points does not change the positional relationship of the original 25 nodes and retains the spatial dependence relationship in the original data. Similarly, adding 100 cross-sections is equivalent to evenly adding 25 cross-sections between every two of the original 5 cross-sections.

[0048] After that, the trained model and the holographic grid node arrangement are used to reconstruct the wing holographic pressure coefficient. Specifically: as Figure 5 shown, the server loads the trained deep learning model, takes the coordinates of the holographic grid nodes and the working condition parameters as inputs, and outputs the pressure coefficient values of all nodes. Taking the expected fineness information of 100×200 as an example, the present invention first automatically generates holographic grid nodes according to this expected fineness information. Then, the three-dimensional spatial coordinates of each grid node and the working condition parameters input by the client are combined into a 6-dimensional vector, namely (x, y, z, M, α, β). Next, the server loads the trained deep learning model from the local, makes predictions on the input of the holographic grid nodes, and obtains the holographic pressure coefficient, in the form of a matrix of 100×200×1.

[0049] This embodiment also provides a device for reconstructing the wing holographic pressure coefficient in a wind tunnel experiment. Specifically, the reconstruction device can run on the server in the laboratory, as Figure 6 shown, the reconstruction device includes:

[0050] A data receiving module, configured to receive the original data of the wind tunnel experiment, the working condition parameters, and the expected fineness information sent by the client.

[0051] A sample processing module, configured to construct a training sample database by using the original data of the wind tunnel experiment.

[0052] A training module, configured to train a deep learning model by using the sample database.

[0053] A grid adjustment module, configured to construct a holographic grid node arrangement corresponding to the expected fineness.

[0054] A data processing module, configured to reconstruct and process the wing holographic pressure coefficient by using the trained deep learning model and the holographic grid node arrangement.

[0055] A data sending module, configured to send the processing result to the client.

[0056] The advantages of the implementation of the present invention are as follows: It is not necessary to use CFD simulations that consume computing resources to obtain holographic data. Reasonable holographic data can be reconstructed only relying on the real data of wind tunnel experiments, reducing the dependence on simulation calculations and improving the utilization rate of wind tunnel experiment data. At the same time, the deep learning sub-module involved in the present invention is easy to implement, and once the model is trained, the efficiency of reconstructing holographic data far exceeds that of CFD simulation calculations. For example, under the condition of keeping the computing device consistent, a comparative experiment on obtaining holographic data for the M6 wing is carried out. The expected information of the fineness of the holographic data is 80×240, that is, the arrangement of the holographic grid nodes is: 8 cross-sections, with 240 nodes in each cross-section. The results show that the time consumption for CFD to calculate holographic data is 15 hours, and it only takes 30 seconds to predict holographic data using the trained model in the present invention.

[0057] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A reconstruction method for the wing holographic pressure coefficient in wind tunnel experiments, characterized in that Including: S1. Receive the original data of the wind tunnel experiment, the working condition parameter, and the fineness expectation information sent by the client; S2. Use the original data of the wind tunnel experiment to construct a training sample database; S3. Use the sample database to train a deep learning model. Among them, take the three-dimensional spatial coordinates of the grid nodes and the working condition parameters as the input of the deep learning model, and take the measured wind tunnel pressure coefficient value corresponding to the grid node as the label to obtain a data matrix for model training; S4. Construct a holographic grid node arrangement corresponding to the fineness expectation. Among them, on the basis of the data grid arrangement of the training sample, use the method of inserting nodes at equal intervals to increase the number of nodes on the grid arrangement. The increase in the data grid arrangement of the training sample includes: increasing the number of cross-sections and the number of nodes on a single cross-section; S5. Use the trained deep learning model and the holographic grid node arrangement to reconstruct the holographic pressure coefficient of the wing, and send the processing result to the client.

2. The method according to claim 1, characterized in that Also including: The client collects the pressure coefficient on the surface of the wing model through sensors arranged at specified positions on the surface of the wing model; And, the client obtains the fineness expectation information, and the fineness expectation information includes: the number of holographic grid nodes for holographic pressure coefficient reconstruction.

3. The method according to claim 2, wherein The method of arranging sensors at specified positions on the surface of the wing model includes: Select cross-sections at intervals from the end near the fuselage to the wing tip, and arrange sensors at intervals on the selected cross-sections respectively, where the number of selected cross-sections is not less than 5; And in each cross-section, the density of sensors at the leading edge and the trailing edge is greater than other positions in the cross-section.

4. The method according to claim 3, characterized in that, 25 sensors are arranged in each cross-section. Among them, sensors numbered 1 to 13 are located on the lower wing surface, and sensors numbered 14 to 25 are located on the upper wing surface.

5. The method according to any one of claims 2-4, characterized in that, In the process of using the original data of the wind tunnel experiment to construct a training sample database, it includes: Obtain the spatial coordinates, pressure coefficient values, and working condition parameters of each sensor arranged on the surface of the wing model, and use them as the original sensor data; Divide the sensors on each cross-section into ordered subsequences in the grid. Among them, starting from each node, take at least 4 nodes backward to form a subsequence with a length of at least 5.

6. The method according to claim 2, wherein During the process of training the deep learning model using the sample database, the loss function for supervised training is as follows: where N represents the last node in the subsequence, and y (i) represents the true value of the i-th node in the subsequence, that is, the measured value of the wind tunnel experiment pressure coefficient, represents the predicted value of the i-th node in the subsequence.

7. The method according to claim 1 or 6, characterized in that, The structure of the deep learning model includes: three layers of transposed convolution, three layers of bidirectional LSTM, and four layers of fully connected layers.

8. The method according to claim 7, wherein The kernel size of each layer of transposed convolution is 3, and the stride is 2; The hidden output dimension of the three layers of bidirectional LSTM is 713; The hidden output dimensions of the four layers of fully connected layers are 1426, 713, 356, and 1 respectively.

9. A reconstruction device for the wing holographic pressure coefficient in a wind tunnel experiment, characterized in that, Including: A data receiving module, used to receive the original data of the wind tunnel experiment, the working condition parameter, and the fineness expectation information sent by the client; A sample processing module, used to use the original data of the wind tunnel experiment to construct a training sample database; A training module for training a deep learning model using the sample database, wherein the three-dimensional spatial coordinates of the grid nodes and the working condition parameters are used as the input of the deep learning model, and the measured wind tunnel experiment pressure coefficient values corresponding to the grid nodes are used as labels to obtain a data matrix for model training; A grid adjustment module for constructing a holographic grid node arrangement corresponding to the expected fineness. Based on the data grid arrangement of the training samples, the number of nodes on the grid arrangement is increased by using the method of equally spaced inserting nodes. The increase in the data grid arrangement of the training samples includes: increasing the number of cross-sections and the number of nodes on a single cross-section; A data processing module for reconstructing the holographic pressure coefficient of the wing by using the trained deep learning model and the holographic grid node arrangement; A data sending module for sending the processing result to the client.

Images