Network structure and method for predicting aerodynamic parameters of airfoil profile surface
Through the improved network structure, the airfoil surface aerodynamic parameters are directly predicted using the airfoil shape coordinates and flow field parameters, which solves the problem of insufficient utilization of artificial errors and flow field information in the prior art, and achieves fast and accurate aerodynamic parameters prediction.
Patent Information
- Application Number
- CN202510790844.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-13
- Publication Date
- 2025-08-29
AI Technical Summary
The existing convolutional neural networks and traditional point cloud networks have artificial errors and failure to utilize flow field parameter information in the prediction of airfoil flow field, making it difficult to quickly and accurately predict the aerodynamic parameters of airfoils under different working conditions.
An improved network structure is designed, using the airfoil shape coordinates and flow field parameters as inputs, and the combination of multi-layer perceptron and maximum pooling layer avoids data pixelation, adds flow field parameter information, and realizes the prediction of airfoil surface aerodynamic parameters under any operating conditions.
It realizes fast and accurate prediction of aerodynamic parameters on the surface of the airfoil, avoids manual errors, retains the accuracy of CFD calculations, and can handle different flow field conditions.
Smart Images
Figure CN120562052A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the fields of aerodynamics and deep learning, and specifically relates to a network structure and method for predicting aerodynamic parameters of an airfoil surface. Background Art
[0002] In the process of airfoil optimization design, accurately predicting its surface aerodynamic parameters is of vital importance for airfoil design, optimization, safety assessment, and performance improvement. Traditional methods for obtaining airfoil aerodynamic parameters mainly include wind tunnel testing and computational fluid dynamics (CFD) simulation.
[0003] Wind tunnel testing is a classic method for obtaining aerodynamic parameters. It simulates the actual flight environment of the airfoil in a wind tunnel and directly measures the aerodynamic parameters of its surface. However, wind tunnel testing has many limitations. For example, the test cost is high, special wind tunnel equipment and a large number of test models are required, and the test cycle is long, making it difficult to quickly respond to design and optimization needs. In addition, the similarity law problem in wind tunnel testing also makes it difficult for the test results to fully and accurately reflect the aerodynamic performance of the airfoil under actual flight conditions. CFD simulation uses numerical methods to solve fluid mechanics equations to predict the flow field and surface aerodynamic parameters around the airfoil. Although CFD simulation has high theoretical accuracy, its calculation process is complex and time-consuming. For complex airfoil shapes and high Reynolds number flow conditions, a large amount of computing resources and time are required to obtain relatively accurate results.
[0004] With the rapid development of artificial intelligence (AI) technology, its application in aerodynamic parameter prediction has gradually attracted attention. Machine learning methods learn from large amounts of known data and establish a mapping relationship between input and output, thereby enabling rapid prediction of unknown data. Compared to traditional wind tunnel testing and CFD simulations, machine learning methods offer the advantages of fast computational speed and low cost, providing a new approach for the rapid prediction of airfoil aerodynamic parameters. Traditional convolutional neural networks (CNNs) typically convert input data into images, using image-based convolution operations to establish the mapping between input and output. This process of converting airfoil models into images inevitably introduces artifacts. Furthermore, the visualization process can also result in a loss of detail in the airfoil's shape. Point cloud data processing methods, however, offer unique advantages for predicting the aerodynamic parameters of complex airfoils. Point cloud data can directly and accurately describe the airfoil's shape, avoiding the drawbacks of visualization. As a typical point cloud processing network, the PointNet network boasts advantages such as direct processing of point cloud data and invariance to point cloud disorder and transformations, achieving promising results in tasks such as 3D shape classification and segmentation. However, the original PointNet network has some shortcomings when dealing with the problem of predicting airfoil aerodynamic parameters. For example, it fails to utilize additional information such as flow field parameters, which is crucial for the accurate prediction of aerodynamic parameters. For example, Reference 1 (KASHEFI A, REMPE D, GUIBAS L J. A point-cloud deep learning framework for prediction of fluid flow fields on irregular geometries[J / OL]. Physics of Fluids, 2021, 33(2): 027104. DOI:10.1063 / 5.0033376.) achieves accurate prediction of irregular geometry flow fields based on PointNet, but the prediction conditions are fixed and the flow field parameter information is not utilized. Reference 2 (Song Juzheng's A method for predicting aerodynamic parameters of airfoil surfaces based on point cloud networks) is a similar technology that has been made public. This document develops and applies PointNet to the prediction of aerodynamic parameters of airfoil surfaces. However, the point cloud network involved also fails to utilize flow field parameter information and can only predict aerodynamic parameters of airfoil surfaces under fixed working conditions. Summary of the Invention
[0005] In order to solve the problems of the above-mentioned convolutional neural networks and traditional point cloud networks in airfoil flow field prediction, and to address the deficiencies in the existing technology, this application proposes a network structure and method for predicting the aerodynamic parameters of the airfoil surface, and designs an improved network structure that can directly use the airfoil shape coordinates and flow field parameters as input, and use the surface aerodynamic parameters corresponding to the airfoil shape coordinates as output. Compared with convolutional neural networks, the improved point cloud network does not require imaging of the data, avoids the introduction of human errors, and simplifies the data preprocessing steps. At the same time, compared with traditional point cloud networks, the addition of flow field parameter information as input allows the network to predict the aerodynamic parameters of the airfoil surface under any working conditions.
[0006] The technical solution of the present invention is In a first aspect, the present invention provides a network structure for predicting aerodynamic parameters of an airfoil surface, the network structure sequentially comprising: N×2 input layer 1, 64k×64k multilayer perceptron 1, 64k×64k multilayer perceptron 2, 64k×128k multilayer perceptron 3, maximum pooling layer 1, 1×m input layer 2, multilayer perceptron 2, merging layer of maximum pooling layer 1 and 1×m input layer 2, 128k×128k multilayer perceptron 4, 128k×n multilayer perceptron 5, N×n output layer; where N is the number of coordinate points on the airfoil surface, n is the number of aerodynamic parameters on the airfoil surface, m is the number of airfoil flow field condition parameters, and k is the network structure scaling factor, which is 1, 1 / 2, or 2.
[0007] Furthermore, the merging layer of the multilayer perceptron 2, the maximum pooling layer 1, and the 1×m input layer 2 is specifically: The 1×128k features output by the maximum pooling layer 1 are concatenated with the 1×m features of the input layer 2 to form 1×(128k+m) features, which are replicated N times to form N×(128k+m) features. This is then concatenated with the N×64k features output by the multilayer perceptron 2 to finally form N×(64k+128k+m) features.
[0008] Furthermore, several 64k×64k multilayer perceptrons are added between the 64k×64k multilayer perceptron 1 and the 64k×64k multilayer perceptron 2. Add several 64k×64k multilayer perceptrons between the 64k×64k multilayer perceptron 2 and the 64k×128k multilayer perceptron 3; Several 128k×128k multilayer perceptrons are added between the 128k×128k multilayer perceptron 4 and the 128k×n multilayer perceptron 5.
[0009] In a second aspect, the present invention further provides a method for predicting aerodynamic parameters of an airfoil surface, which is implemented using the network structure described in the first aspect, and the method comprises: A network structure for predicting the aerodynamic parameters of the airfoil surface was established, and then 80% of the data set was randomly selected as the training data set, and the remaining 20% of the data set was used as the test data set. The network structure for predicting the aerodynamic parameters of the airfoil surface was trained using the training data set, and the prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface was tested and verified using the test data set.
[0010] Furthermore, a test data set is used to test and verify the prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface. Specifically, an airfoil and flow field conditions are selected from the test data set, the coordinates of the airfoil surface coordinate points and the flow field conditions are used as input, the trained network structure for predicting the aerodynamic parameters of the airfoil surface is input, and the output value of the network structure is used as the predicted value of the aerodynamic parameters of the airfoil surface to complete the prediction of the aerodynamic parameters of the airfoil surface.
[0011] Furthermore, the method further comprises: Step 1: Select a reference airfoil and establish a two-dimensional coordinate system Oxy on the reference airfoil, where the coordinate origin O is the leading edge of the airfoil, the Ox axis passes through the trailing edge of the airfoil and points to the trailing edge of the airfoil, and the Oy axis is perpendicular to the Ox axis and points upward; parameterize the reference airfoil using the airfoil parameterization method to obtain the geometric parameters of the reference airfoil; Step 2: Generate N groups of random geometric parameter groups using a random sampling method, and generate N random airfoils according to the airfoil parameterization method to form an airfoil library; Step 3: Set the flow field conditions of the airfoil and obtain the surface aerodynamic parameters of each airfoil in the random airfoil library under the set flow field conditions based on CFD simulation calculations; Step 4: Use the coordinates of the airfoil surface coordinate points in the random airfoil library and the flow field conditions of the airfoil as input data, and the surface aerodynamic parameters corresponding to the airfoil surface coordinate points as output data to establish a data set.
[0012] Furthermore, the method further comprises: In step 2, the specific method for generating N random airfoils is to randomly generate at least multiple groups of geometric parameters in a preset airfoil geometric parameter space, and then deform the reference airfoil based on the multiple groups of randomly generated geometric parameters according to the airfoil parameterization method. The deformed airfoil is used as a random airfoil, and one group of geometric parameters corresponds to one group of random airfoils.
[0013] Furthermore, in step 3, the flow field conditions of the airfoil are used as flow field parameters input into the CFD simulation calculation, and the flow field conditions of the airfoil are the incoming flow Mach number, the incoming flow angle of attack or the incoming flow temperature; the flow field conditions are generated in a preset flow field parameter space by a random sampling method; the surface aerodynamic parameters of the airfoil are the airfoil surface aerodynamic pressure coefficient, the airfoil surface tangential friction coefficient, the airfoil surface normal friction coefficient, or the airfoil surface aerodynamic parameters obtained by CFD simulation calculation.
[0014] The present invention provides a network structure and method for predicting aerodynamic parameters of an airfoil surface. The network structure predicts the aerodynamic parameters of the airfoil surface based on the airfoil coordinates and flow field conditions. Compared with the traditional CFD solution method, rapid prediction of aerodynamic parameters of the airfoil surface is achieved. Compared with the traditional convolutional neural network, this method does not involve data pixelation or data dimensionality reduction, retains the accuracy of the CFD calculation results, and avoids the introduction of artificial errors. Compared with the traditional point cloud network, the network can additionally input airfoil flow field information to predict the aerodynamic parameters of the airfoil surface under different flow field conditions. The method of the present application provides a new prediction model and prediction method for the rapid prediction of aerodynamic parameters of the airfoil surface under different flow field conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 Figure 2 is a schematic diagram of the RAE2822 reference airfoil and the random airfoil library generated based on the reference airfoil. The solid line is the reference airfoil, denoted as RAE2822, and the shaded area is the coordinate range of the generated random airfoil library, denoted as airfoil library, where c is the chord length of the airfoil. Figure 2 Figure 1 is a schematic diagram of the improved point cloud network structure, where MLP refers to multi-layer perceptron. The numerical label (64,128) after MLP indicates that the size of the first fully connected layer in the multi-layer perceptron is 64, and the size of the second fully connected layer is 128. The meanings of other numerical labels are similar. Figure 3 Comparison of the predicted values of the airfoil surface pressure coefficient distribution under different flow conditions using the trained point cloud model with the CFD-calculated values. The solid dots in the figure represent the CFD-calculated values, denoted as "Truth," while the hollow dots represent the neural network-predicted values, denoted as "Prediction." Three samples were selected for this comparison: sample-1, sample-2, and sample-3, corresponding to Mach numbers of 0.42, 0.45, and 0.44, respectively. DETAILED DESCRIPTION
[0016] The present application is described in further detail below with reference to the accompanying drawings of the embodiments.
[0017] The embodiment of the present invention provides a network structure for predicting aerodynamic parameters of airfoil surface, such as Figure 2As shown, the network structure includes: N×2 input layer 1, 64k×64k multilayer perceptron 1, 64k×64k multilayer perceptron 2, 64k×128k multilayer perceptron 3, maximum pooling layer 1, 1×m input layer 2, multilayer perceptron 2, merging layer of maximum pooling layer 1 and 1×m input layer 2, 128k×128k multilayer perceptron 4, 128k×n multilayer perceptron 5, N×n output layer; where N is the number of coordinate points on the airfoil surface, n is the number of aerodynamic parameters on the airfoil surface, m is the number of airfoil flow field condition parameters, and k is the network structure scaling factor, which is 1, 1 / 2, or 2.
[0018] The merging layer of the multilayer perceptron 2, the maximum pooling layer 1, and the 1×m input layer 2 is: The 1×128k features output by the maximum pooling layer 1 are concatenated with the 1×m features of the input layer 2 to form 1×(128k+m) features, which are replicated N times to form N×(128k+m) features. This is then concatenated with the N×64k features output by the multilayer perceptron 2 to finally form N×(64k+128k+m) features.
[0019] Add several 64k×64k multilayer perceptrons between the 64k×64k multilayer perceptron 1 and the 64k×64k multilayer perceptron 2; Add several 64k×64k multilayer perceptrons between the 64k×64k multilayer perceptron 2 and the 64k×128k multilayer perceptron 3; Several 128k×128k multilayer perceptrons are added between the 128k×128k multilayer perceptron 4 and the 128k×n multilayer perceptron 5.
[0020] An embodiment of the present invention further provides a method for predicting aerodynamic parameters of an airfoil surface, which is implemented using the aforementioned network structure. The method includes: A network structure for predicting the aerodynamic parameters of the airfoil surface was established, and then 80% of the data set was randomly selected as the training data set, and the remaining 20% of the data set was used as the test data set. The network structure for predicting the aerodynamic parameters of the airfoil surface was trained using the training data set, and the prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface was tested and verified using the test data set.
[0021] Specifically, a test data set is used to test and verify the prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface. Specifically, an airfoil and flow field conditions are selected from the test data set, the coordinates of the airfoil surface coordinate points and the flow field conditions are used as input, the trained network structure for predicting the aerodynamic parameters of the airfoil surface is input, and the output value of the network structure is used as the predicted value of the aerodynamic parameters of the airfoil surface to complete the prediction of the aerodynamic parameters of the airfoil surface.
[0022] Specifically, the method further includes: Step 1: Select a reference airfoil and establish a two-dimensional coordinate system Oxy on the reference airfoil, where the coordinate origin O is the leading edge of the airfoil, the Ox axis passes through the trailing edge of the airfoil and points to the trailing edge of the airfoil, and the Oy axis is perpendicular to the Ox axis and points upward; parameterize the reference airfoil using the airfoil parameterization method to obtain the geometric parameters of the reference airfoil; Step 2: Generate N groups of random geometric parameter groups using a random sampling method, and generate N random airfoils according to the airfoil parameterization method to form an airfoil library; Step 3: Set the flow field conditions of the airfoil and obtain the surface aerodynamic parameters of each airfoil in the random airfoil library under the set flow field conditions based on CFD simulation calculations; Step 4: Use the coordinates of the airfoil surface coordinate points in the random airfoil library and the flow field conditions of the airfoil as input data, and the surface aerodynamic parameters corresponding to the airfoil surface coordinate points as output data to establish a data set.
[0023] Specifically, the method further includes: In step 2, the specific method for generating N random airfoils is to randomly generate at least multiple groups of geometric parameters in a preset airfoil geometric parameter space, and then deform the reference airfoil based on the multiple groups of randomly generated geometric parameters according to the airfoil parameterization method. The deformed airfoil is used as a random airfoil, and one group of geometric parameters corresponds to one group of random airfoils.
[0024] Specifically, in step 3, the flow field conditions of the airfoil are used as flow field parameters input into the CFD simulation calculation, and the flow field conditions of the airfoil are the incoming flow Mach number, the incoming flow angle of attack or the incoming flow temperature; the flow field conditions are generated in a preset flow field parameter space by a random sampling method; the surface aerodynamic parameters of the airfoil are the airfoil surface aerodynamic pressure coefficient, the airfoil surface tangential friction coefficient, the airfoil surface normal friction coefficient, or the airfoil surface aerodynamic parameters obtained by CFD simulation calculation.
[0025] Example A method for predicting aerodynamic parameters of an airfoil surface provided by the present invention includes: Step 1: Select the supercritical RAE2822 airfoil as the reference airfoil. Establish a two-dimensional coordinate system, Oxy, on this reference airfoil, where the origin O is the leading edge of the airfoil, the Ox-axis passes through and points toward the trailing edge, and the Oy-axis is perpendicular to the Ox-axis and points upward. Parameterize the reference airfoil using the free-form deformation method.
[0026] Step 2: Generate 4000 sets of control point displacement values of the free deformation method based on the Latin hypercube sampling method, deform the reference airfoil, and the set of 4000 sets of new airfoils after deformation is used as the random airfoil library. The schematic diagram of the reference RAE2822 airfoil and the random airfoil library is shown in Figure 1 shown.
[0027] Step 3: Select the incoming Mach number as the airfoil flow field condition. Using the Latin Hypercube sampling method, generate 4000 sets of Mach numbers between 0.4 and 0.6 as the incoming Mach number for each airfoil in the airfoil library. Based on CFD simulations, obtain the surface pressure coefficient for each airfoil in the random airfoil library at the set incoming Mach number.
[0028] Step 4: Use the horizontal and vertical coordinates of the airfoil surface coordinate points in the random airfoil library as input data 1, the incoming flow Mach number of the airfoil flow field as input data 2, and the surface pressure coefficient corresponding to the airfoil surface coordinate points as output data to establish a data set. Randomly select 80% of the data set samples as the training data set, and the remaining 20% of the data set samples as the test data set.
[0029] Step 5: Establish a network structure for predicting the aerodynamic parameters of the airfoil surface, then randomly select 80% of the data set as the training data set and the remaining 20% of the data set as the test data set. Use the training data set to train the network structure for predicting the aerodynamic parameters of the airfoil surface, and use the test data set to test and verify the prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface.
[0030] As attached Figure 2 As shown in the figure, the network structure for predicting airfoil surface aerodynamic parameters is as follows: N×2 input layer 1, 64×64 multilayer perceptron 1, 64×64 multilayer perceptron 2, 64×128 multilayer perceptron 3, max pooling layer 1, multilayer perceptron 2, a merging layer of max pooling layer 1 and 1×m input layer 2, 128×128 multilayer perceptron 4, 128×n multilayer perceptron 5, and an N×n output layer. N is the number of coordinate points on the airfoil surface, N=192. n is the number of aerodynamic parameters on the airfoil surface, n=1. m is the number of airfoil flow field condition parameters, m=1. The network structure for predicting airfoil surface aerodynamic parameters is trained using a training dataset, and the network's prediction performance is verified using a test dataset.
[0031] Step 5: Select three airfoils from the test set as the airfoils to be predicted, use the horizontal and vertical coordinates of the coordinate points on the airfoil surface to be predicted as input 1, use the incoming Mach number of the airfoil to be predicted as input 2, input the trained network structure for predicting the aerodynamic parameters of the airfoil surface, use the output value of the network structure as the predicted value of the aerodynamic parameters of the airfoil surface, and complete the prediction of the aerodynamic parameters of the airfoil surface. The comparison between the predicted values of the pressure coefficient distribution of the three airfoil surfaces and the CFD calculated values is shown in the attached figure. Figure 3 shown.
[0032] The present invention provides a network structure and method for predicting aerodynamic parameters of an airfoil surface. The network structure predicts the aerodynamic parameters of the airfoil surface based on the airfoil coordinates and flow field conditions. Compared with the traditional CFD solution method, rapid prediction of aerodynamic parameters of the airfoil surface is achieved. Compared with the traditional convolutional neural network, this method does not involve data pixelation or data dimensionality reduction, retains the accuracy of the CFD calculation results, and avoids the introduction of artificial errors. Compared with the traditional point cloud network, the network can additionally input airfoil flow field information to predict the aerodynamic parameters of the airfoil surface under different flow field conditions. The method of the present application provides a new prediction model and prediction method for the rapid prediction of aerodynamic parameters of the airfoil surface under different flow field conditions.
Claims
1. A network structure for predicting aerodynamic parameters of airfoil surfaces, characterized in that: The network structure includes: N×2 input layer 1, 64k×64k multilayer perceptron 1, 64k×64k multilayer perceptron 2, 64k×128k multilayer perceptron 3, maximum pooling layer 1, 1×m input layer 2, multilayer perceptron 2, a merged layer of maximum pooling layer 1 and 1×m input layer 2, 128k×128k multilayer perceptron 4, 128k×n multilayer perceptron 5, and an N×n output layer; where N is the number of coordinate points on the airfoil surface, n is the number of aerodynamic parameters on the airfoil surface, m is the number of airfoil flow field condition parameters, and k is the network structure scaling factor, which is 1, 1 / 2, or 2.
2. A network structure for predicting aerodynamic parameters of an airfoil surface according to claim 1, characterized in that: The merging layer of the multilayer perceptron 2, the maximum pooling layer 1, and the 1×m input layer 2 is: The 1×128k features output by the maximum pooling layer 1 are concatenated with the 1×m features of the input layer 2 to form 1×(128k+m) features, which are replicated N times to form N×(128k+m) features. This is then concatenated with the N×64k features output by the multilayer perceptron 2 to finally form N×(64k+128k+m) features.
3. The network structure for predicting aerodynamic parameters of an airfoil surface according to claim 1, characterized in that: Add several 64k×64k multilayer perceptrons between the 64k×64k multilayer perceptron 1 and the 64k×64k multilayer perceptron 2; Add several 64k×64k multilayer perceptrons between the 64k×64k multilayer perceptron 2 and the 64k×128k multilayer perceptron 3; Several 128k×128k multilayer perceptrons are added between the 128k×128k multilayer perceptron 4 and the 128k×n multilayer perceptron 5.
4. A method for predicting aerodynamic parameters of an airfoil surface, characterized in that: The method is implemented using the network structure according to any one of claims 1 to 3, comprising: A network structure for predicting the aerodynamic parameters of the airfoil surface was established, and then 80% of the data set was randomly selected as the training data set, and the remaining 20% of the data set was used as the test data set. The network structure for predicting the aerodynamic parameters of the airfoil surface was trained using the training data set, and the prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface was tested and verified using the test data set.
5. The method for predicting aerodynamic parameters of an airfoil surface according to claim 4, characterized in that: The prediction performance of the network structure for predicting the aerodynamic parameters of the airfoil surface is tested and verified using a test data set. Specifically, an airfoil and flow field conditions are selected from the test data set, the coordinates of the airfoil surface coordinate points and the flow field conditions are used as input, the trained network structure for predicting the aerodynamic parameters of the airfoil surface is input, and the output value of the network structure is used as the predicted value of the aerodynamic parameters of the airfoil surface to complete the prediction of the aerodynamic parameters of the airfoil surface.
6. The method for predicting aerodynamic parameters of an airfoil surface according to claim 5, characterized in that: The method further comprises: Step 1: Select a reference airfoil and establish a two-dimensional coordinate system Oxy on the reference airfoil, where the coordinate origin O is the leading edge of the airfoil, the Ox axis passes through the trailing edge of the airfoil and points to the trailing edge of the airfoil, and the Oy axis is perpendicular to the Ox axis and points upward; parameterize the reference airfoil using the airfoil parameterization method to obtain the geometric parameters of the reference airfoil; Step 2: Generate N groups of random geometric parameter groups using a random sampling method, and generate N random airfoils according to the airfoil parameterization method to form an airfoil library; Step 3: Set the flow field conditions of the airfoil and obtain the surface aerodynamic parameters of each airfoil in the random airfoil library under the set flow field conditions based on CFD simulation calculations; Step 4: Use the coordinates of the airfoil surface coordinate points in the random airfoil library and the flow field conditions of the airfoil as input data, and the surface aerodynamic parameters corresponding to the airfoil surface coordinate points as output data to establish a data set.
7. The method for predicting aerodynamic parameters of an airfoil surface according to claim 6, characterized in that: The method further comprises: In step 2, the specific method for generating N random airfoils is to randomly generate at least multiple groups of geometric parameters in a preset airfoil geometric parameter space, and then deform the reference airfoil based on the multiple groups of randomly generated geometric parameters according to the airfoil parameterization method. The deformed airfoil is used as a random airfoil, and one group of geometric parameters corresponds to one group of random airfoils.
8. The method for predicting aerodynamic parameters of an airfoil surface according to claim 6, characterized in that: In step 3, the flow field conditions of the airfoil are used as flow field parameters input for CFD simulation calculation, and the flow field conditions of the airfoil are the incoming flow Mach number, the incoming flow angle of attack or the incoming flow temperature; the flow field conditions are generated in a preset flow field parameter space by a random sampling method; the surface aerodynamic parameters of the airfoil are the airfoil surface aerodynamic pressure coefficient, the airfoil surface tangential friction coefficient, the airfoil surface normal friction coefficient, or the airfoil surface aerodynamic parameters obtained by CFD simulation calculation.