Pneumatic parameter prediction method, storage medium and computer program product

Abstract

The embodiment of the invention provides an aerodynamic parameter prediction method, a storage medium and a computer program product. The method comprises the steps of obtaining a to-be-predicted object; when the to-be-predicted object is a three-dimensional wing, an aerodynamic parameter prediction result corresponding to the to-be-predicted object is output through a wing prediction model, and the wing prediction model is obtained based on training of a plurality of section images and aerodynamic parameter labels of a plurality of three-dimensional wing samples; the aerodynamic parameter labels of the multiple three-dimensional wing samples are obtained through calculation based on an RANS method of a coupling transition model; when the to-be-predicted object is a three-dimensional blade, outputting an aerodynamic parameter prediction result corresponding to the to-be-predicted object through a blade prediction model, the blade prediction model being obtained by training based on a plurality of cross-section images and aerodynamic parameter labels of a plurality of three-dimensional blade samples; the aerodynamic parameter labels of the multiple three-dimensional blade samples are obtained through calculation based on an LES method. The embodiment of the invention aims to improve the speed of accurately predicting the aerodynamic parameters of the wing or the blade.

Description

Technical Field

[0001] Embodiments of the present application relate to the technical field of data processing, and in particular, to an aerodynamic parameter prediction method, storage medium, and computer program product. Background Art

[0002] In the early aerodynamic design of aircraft wings or other low-speed blades, wind tunnel experiments and theoretical calculations were mainly relied upon. However, the wind tunnel experiment method can only simulate relatively simple working conditions, and the high cost of wind tunnel experiments makes it impossible to apply them on a large scale.

[0003] With the development of computational fluid dynamics, numerical simulation algorithms can be used to more accurately solve the aerodynamic performance of most wings or blades. The flow field information and important aerodynamic parameters of aircraft or fan blades can be obtained through CFD simulation software. However, although the numerical simulation method reduces the experimental cost, it has the problem of long simulation time. Aerodynamic optimization requires repeated adjustment and verification of the geometric parameters of aircraft wings or fan blades. The long simulation time will seriously slow down this process.

[0004] Therefore, how to improve the speed of accurately predicting the aerodynamic parameters of a wing or blade is an urgent problem to be solved. Summary of the invention

[0005] Embodiments of the present application provide an aerodynamic parameter prediction method, storage medium, and computer program product, which are intended to increase the speed of accurately predicting the aerodynamic parameters of a wing or blade.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting aerodynamic parameters, the method comprising: Acquire an object to be predicted, wherein the object to be predicted includes a three-dimensional wing and a three-dimensional blade; When the object to be predicted is a three-dimensional wing, a plurality of cross-sectional images corresponding to the object to be predicted are input into a wing prediction model, and a prediction result of aerodynamic parameters corresponding to the object to be predicted is output through the wing prediction model, wherein the wing prediction model is obtained by training based on a plurality of cross-sectional images and aerodynamic parameter labels of a plurality of three-dimensional wing samples, and the aerodynamic parameter labels of the plurality of three-dimensional wing samples are calculated based on a RANS method of a coupled transition model; When the object to be predicted is a three-dimensional blade, multiple cross-sectional images corresponding to the object to be predicted are input into a blade prediction model, and the blade prediction model outputs aerodynamic parameter prediction results corresponding to the object to be predicted. The blade prediction model is trained based on multiple cross-sectional images and aerodynamic parameter labels of multiple three-dimensional blade samples, and the aerodynamic parameter labels of the multiple three-dimensional blade samples are calculated based on the LES method.

[0007] Optionally, the method further comprises: The geometric profiles of multiple three-dimensional wing samples are obtained, and the aerodynamic parameter labels corresponding to each three-dimensional wing sample are calculated based on the RANS method of the coupled transition model; Constructing a three-dimensional wing sample data set, wherein the three-dimensional wing sample data set includes a plurality of three-dimensional wing samples, and any three-dimensional wing sample includes a plurality of cross-sectional images and aerodynamic parameter labels corresponding to the three-dimensional wing; Constructing a first convolutional neural network, the first convolutional neural network comprising an input layer, a dimensionality reduction layer, a convolutional backbone network layer and an output layer, wherein the dimensionality reduction layer is used to extract common features in multiple cross-sectional images corresponding to any three-dimensional wing; The first convolutional neural network is iteratively trained based on the three-dimensional wing sample data set to obtain a wing prediction model.

[0008] Optionally, the geometric profiles of multiple three-dimensional wing samples are obtained, and based on the RANS method of the coupled transition model, the aerodynamic parameter labels corresponding to each three-dimensional wing sample are calculated respectively, including: For any three-dimensional wing sample, a geometric profile corresponding to the three-dimensional wing sample is obtained, and an SA model of a coupled transition model is input, wherein the SA model is a model based on a RANS method; The aerodynamic parameter labels corresponding to the three-dimensional wing sample are calculated through the SA model of the coupled transition model.

[0009] Optionally, the calculation formula of the SA model of the coupled transition model is:

[0010] in, By solving the convection transport equation, we can obtain is the dynamic viscosity coefficient, is the density, represents kinematic viscosity; are geometric coordinates; and are the generated term and dissipated term after coupling respectively; For additional generated items; is an empirical parameter; is the average speed.

[0011] Optionally, the method further comprises: The geometric profiles of multiple three-dimensional blade samples are obtained, and based on the LES method, the aerodynamic parameter labels corresponding to each three-dimensional blade sample are calculated respectively; Constructing a three-dimensional blade sample data set, wherein the three-dimensional blade sample data set includes a plurality of three-dimensional blade samples, and any three-dimensional blade sample includes a plurality of cross-sectional images and aerodynamic parameter labels corresponding to the three-dimensional blade; Constructing a second convolutional neural network, the second convolutional neural network comprising an input layer, a dimensionality reduction layer, a convolutional backbone network layer and an output layer, wherein the dimensionality reduction layer is used to extract common features in multiple cross-sectional images corresponding to any three-dimensional leaf; The second convolutional neural network is iteratively trained based on the three-dimensional blade sample data set to obtain a blade prediction model.

[0012] Optionally, after outputting the aerodynamic parameter prediction result corresponding to the object to be predicted through the wing prediction model, the method further includes: In response to the target operation, obtaining a target aerodynamic parameter corresponding to the object to be predicted, wherein the target aerodynamic parameter includes at least one parameter type of a target drag coefficient, a target pressure coefficient, and a maximum lift-to-drag ratio; According to the target aerodynamic parameters corresponding to the object to be predicted and the aerodynamic parameter prediction result, the geometric parameters of the object to be predicted are iteratively updated until the object to be predicted after iterative update meets the target aerodynamic parameters.

[0013] Optionally, after outputting the aerodynamic parameter prediction result corresponding to the object to be predicted through the wing prediction model, the method further includes: In response to the target operation, obtaining a target aerodynamic parameter corresponding to the object to be predicted, wherein the target aerodynamic parameter includes at least one parameter type of a target drag coefficient, a target pressure coefficient, and a maximum lift-to-drag ratio; Iteratively updating the geometric parameters of the object to be predicted according to the target aerodynamic parameters corresponding to the object to be predicted and the aerodynamic parameter prediction result until the iteratively updated object to be predicted meets the target aerodynamic parameters, including: Calculating a target difference according to the target aerodynamic parameter corresponding to the object to be predicted and the aerodynamic parameter prediction result; When the target difference is greater than a preset target threshold, the geometric parameters of the object to be predicted are adjusted, the object to be predicted after the geometric parameters are adjusted is used as the object to be predicted, and a prediction result of aerodynamic parameters corresponding to the current object to be predicted is calculated based on the wing prediction model or the blade prediction model; When the target difference is less than the target threshold, the iteration is stopped, and the prediction results of the geometric parameters and aerodynamic parameters corresponding to the current object to be predicted are output.

[0014] Optionally, the target difference is calculated as follows:

[0015] Wherein, MSE is the target difference; n is the number of parameter types in the target aerodynamic parameter; is the i-th aerodynamic parameter in the aerodynamic parameter prediction result; is the i-th aerodynamic parameter among the target aerodynamic parameters.

[0016] In a second aspect, an embodiment of the present application provides a computer device, comprising: at least one processor, and a memory, wherein the memory stores a computer program that can be run on the processor, wherein when the processor executes the computer program, the aerodynamic parameter prediction method described in the first aspect of the embodiment is executed.

[0017] In a third aspect, an embodiment of the present application provides a non-volatile readable storage medium, wherein the non-volatile readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the aerodynamic parameter prediction method described in the first aspect of the embodiment is performed.

[0018] In a fourth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the aerodynamic parameter prediction method described in the first aspect of the embodiment.

[0019] Beneficial effects: In this method, when predicting the aerodynamic parameters of a high-speed three-dimensional wing, a wing prediction model is used for prediction. The wing prediction model is trained based on multiple cross-sectional images and aerodynamic parameter labels of multiple three-dimensional wing samples, and the aerodynamic parameter label of each three-dimensional wing sample is calculated based on the RANS method of the coupled transition model. The RANS method of the coupled transition model can obtain more accurate aerodynamic parameters of the three-dimensional wing, and the wing prediction model learns the relationship between multiple cross-sectional images and aerodynamic parameter labels of the three-dimensional wing samples. In the subsequent prediction process, the wing prediction model can quickly and accurately generate aerodynamic parameter prediction results for high-speed three-dimensional wings.

[0020] When predicting the aerodynamic parameters of low-speed three-dimensional blades, a blade prediction model is used for prediction. The blade prediction model is trained based on multiple cross-sectional images and aerodynamic parameter labels of multiple three-dimensional blade samples. The aerodynamic parameter labels of the three-dimensional blade samples are calculated based on the LES method. The blade prediction model learns the relationship between the multiple cross-sectional images and aerodynamic parameter labels of the three-dimensional blade samples. In subsequent predictions, the blade prediction model can quickly generate aerodynamic parameter prediction results for the three-dimensional blades.

[0021] Compared with the existing aerodynamic parameter calculation methods that all adopt the RANS method, this method first uses different methods to calculate the aerodynamic parameter labels of high-speed three-dimensional wings and low-speed three-dimensional blades, which can improve the accuracy of aerodynamic parameter calculation of three-dimensional wings and three-dimensional blades. Based on the different methods used to calculate the aerodynamic parameter labels to obtain the wing prediction model and the blade prediction model through classification training, the aerodynamic parameter prediction results of the three-dimensional wing and the three-dimensional blade can be quickly predicted, thereby improving the speed of accurately predicting the aerodynamic parameters of the wing or the blade. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art.

[0023] Figure 1 A flowchart showing the steps of the aerodynamic parameter prediction method provided in an embodiment of the present application is shown; Figure 2 A schematic diagram of the span direction provided by an embodiment of the present application is shown; Figure 3 A flowchart showing the steps of the training process of the wing prediction model provided by an embodiment of the present application is shown; Figure 4 A schematic diagram of the structure of a first convolutional neural network provided in an embodiment of the present application is shown; Figure 5 A flowchart showing the steps of the training process of the blade prediction model provided in an embodiment of the present application is shown; Figure 6 A schematic diagram of a genetic algorithm optimization process provided by an embodiment of the present application is shown; Figure 7 A schematic diagram of a non-volatile readable storage medium provided in an embodiment of the present application is shown; Figure 8 A schematic diagram of a computer program product provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0025] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, each embodiment of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in each embodiment of the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not constitute any limitation, and the various embodiments can be combined with each other and referenced to each other without contradiction.

[0026] PCA: principal components analysis, which aims to transform multi-index data into a few comprehensive indicators by using the idea of ​​dimensionality reduction.

[0027] RANS: Reynolds equation, a calculation method in computational fluid dynamics, which averages the physical quantities in the flow field over time and brings the averaged physical quantities into the traditional NS equations in computational fluid dynamics for solution.

[0028] LES: Large Eddy Simulation, is also a calculation method in computational fluid dynamics. It divides the vortices in the flow field into large vortices and small vortices at a certain scale. The large vortex directly solves the NS equations, while the small vortex constructs a new sub-grid stress model to solve it.

[0029] Transition: The process by which the model flow changes from a laminar state to a turbulent state.

[0030] In the complex process of designing and developing large aircraft such as airplanes, the aerodynamic optimization design of wing blades is the core link, which has a very important impact on the aerodynamic performance of the entire aircraft. With the continuous improvement of the requirements for the comprehensive performance of aircraft, improving the aircraft's lift-to-drag ratio and lift-to-drag coefficient and other aerodynamic parameters through the aerodynamic optimization design of aircraft wings has become an important and urgent task. In addition, for small aircraft such as low-altitude drones and low-speed blades such as fan blades, aerodynamic optimization is still needed to improve their aerodynamic performance.

[0031] In the early aerodynamic design of aircraft wings or other low-speed blades, wind tunnel experiments and theoretical calculations were mainly relied upon. However, the wind tunnel experiment method can only simulate relatively simple working conditions, and the high cost of wind tunnel experiments makes it impossible to apply them on a large scale.

[0032] With the development of computational fluid dynamics, the use of numerical simulation algorithms can more accurately solve the aerodynamic performance of most wings or blades. The flow field information and important aerodynamic parameters of aircraft or fan blades can be obtained through CFD simulation software. Aerodynamic design is divided into two directions: inverse design and optimization design. Inverse design is to obtain the geometric shape of the wing or blade that meets the above distribution through a given wing or blade surface pressure distribution or velocity distribution; while optimization design is to use the specified aerodynamic performance parameters (such as maximum lift-to-drag ratio) as the optimization target, and finally adjust the wing or blade shape in combination with the optimization algorithm; however, although the numerical simulation method has reduced the experimental cost, it has the problem of long simulation time. Aerodynamic optimization requires repeated adjustment and verification of the geometric parameters of the aircraft wing or fan blade. The long simulation time will seriously slow down this process.

[0033] Therefore, how to improve the speed of accurately predicting the aerodynamic parameters of a wing or a blade is an urgent problem to be solved; in order to improve the speed of accurately predicting the aerodynamic parameters of a wing or a blade, an embodiment of the present application provides an aerodynamic parameter prediction method.

[0034] Reference Figure 1 , shows a flowchart of a method for predicting aerodynamic parameters provided by an embodiment of the present application, the method may include the following steps: S101: Acquire an object to be predicted, where the object to be predicted includes a three-dimensional wing and a three-dimensional blade.

[0035] In the actual implementation process, the object to be predicted is first obtained, and the object to be predicted includes a three-dimensional wing and a three-dimensional blade. For example, the user can provide the three-dimensional shape of the three-dimensional wing or the three-dimensional blade, and the existing image recognition technology can be used to identify that the object to be predicted belongs to the three-dimensional wing or the three-dimensional blade. The prediction options of the three-dimensional wing and the prediction options of the three-dimensional blade can also be set separately. The user can upload multiple cross-sectional images of the three-dimensional wing to be predicted through the prediction option of the three-dimensional wing, and can also upload multiple cross-sectional images of the three-dimensional blade to be predicted through the prediction option of the three-dimensional blade.

[0036] Reference Figure 2 , shows a spanwise schematic diagram provided in an embodiment of the present application. The cross-sectional image refers to a contour image containing the geometric features of a three-dimensional wing or a three-dimensional blade. For any three-dimensional wing or three-dimensional blade, there are corresponding cross-sectional images in different spanwise directions. The spanwise direction refers to the direction extending in a certain direction. Different cross-sectional images can be obtained by cutting off the three-dimensional wing or three-dimensional blade along different spanwise positions.

[0037] S102: When the object to be predicted is a three-dimensional wing, multiple cross-sectional images corresponding to the object to be predicted are input into a wing prediction model, and the wing prediction model outputs aerodynamic parameter prediction results corresponding to the object to be predicted, wherein the wing prediction model is trained based on multiple cross-sectional images and aerodynamic parameter labels of multiple three-dimensional wing samples, and the aerodynamic parameter labels of the multiple three-dimensional wing samples are calculated based on the RANS method of the coupled transition model.

[0038] The aerodynamic parameters included in this embodiment include: pressure coefficient, drag coefficient, lift coefficient, etc.

[0039] This method uses a RANS method based on a coupled transition model to calculate aerodynamic parameter labels for multiple three-dimensional wing samples, and then trains a wing prediction model based on multiple cross-sectional images and aerodynamic parameter labels for the multiple three-dimensional wing samples. The wing prediction model predicts the aerodynamic parameter prediction results of the three-dimensional wing, and there is no need to use a numerical simulation algorithm to calculate the aerodynamic parameters of the three-dimensional wing every time, thereby reducing the calculation cost of the prediction and speeding up the efficiency of predicting the aerodynamic parameters.

[0040] Moreover, for high-speed aircraft wings, when the speed is fast, if the LES method is used, the amount of grids required is very large. Therefore, for high-speed three-dimensional wings, this embodiment uses the RANS method of the coupled transition model to calculate the aerodynamic parameters of the three-dimensional wing as training labels, which can obtain more accurate aerodynamic parameter calculation results for the three-dimensional wing. Then, the wing prediction model learns the relationship between multiple cross-sectional images of the three-dimensional wing samples and the aerodynamic parameter labels calculated by the RANS method of the coupled transition model, which can quickly obtain more accurate aerodynamic parameter prediction results for the three-dimensional wing.

[0041] Reference Figure 3 , shows a flowchart of the steps of the training process of the wing prediction model provided in an embodiment of the present application. In a feasible implementation, the training process of the wing prediction model includes: A1: Get the geometric profiles of multiple 3D wing samples, and calculate the aerodynamic parameter labels corresponding to each 3D wing sample based on the RANS method of the coupled transition model.

[0042] First, multiple three-dimensional wing samples are obtained, and the three-dimensional wings with more applications can be obtained as the three-dimensional wing samples. The network of the geometric contours of the three-dimensional wing samples is used as input, and the velocity distribution around each three-dimensional wing is simulated through open source CFD simulation software. The aerodynamic parameter labels of each three-dimensional wing sample can be calculated through the velocity distribution.

[0043] In this method, a RANS method based on a coupled transition model is used to calculate the aerodynamic parameter labels of each three-dimensional wing sample in an open source CFD simulation software.

[0044] Specifically, the core equation in the RANS method, the momentum equation, has the following form:

[0045] In the formula, is the density, For time, For pressure, are geometric coordinates, is the average velocity, the subscripts i, j and k represent the components of the ergodic velocity in three directions in three-dimensional space, and the short horizontal line above represents the time average, which is represented by get, and are the total speed and pulsating speed respectively; is the Kronecker function, Characterizes centrifugal force and Coriolis force, and is applicable to problems such as fan blades that require a rotating coordinate system.

[0046] In the formula, is the viscous stress tensor, and its expression is: ,in, represents kinematic viscosity; In the formula, is the Reynolds stress tensor, which is expressed as: ,in, is the turbulent eddy viscosity.

[0047] The RANS method is mainly obtained by substituting some physical quantities into the traditional NS equation after time averaging, which makes the traditional NS equation non-closed. The reason for the non-closed is that The solution.

[0048] Therefore, it is necessary to further model it. There are many forms of modeling. In this embodiment, for any three-dimensional wing sample, the geometric contour corresponding to the three-dimensional wing sample is obtained, and the SA model of the coupled transition model is input, and the SA model is a model based on the RANS method; through the SA model of the coupled transition model, the aerodynamic parameter label corresponding to the three-dimensional wing sample is calculated.

[0049] Specifically, in the SA model, is represented as ,in is the empirical parameter term, It is solved by the convection transport equation, which is in the form of:

[0050] in, is the dynamic viscosity coefficient, and are the generated term and dissipated term of the equation, which are also obtained from the empirical formula. is an empirical parameter.

[0051] The current RANS method does not have the ability to predict the transition process, so it is necessary to add an additional transition equation to solve the intermittent factor. .

[0052] The transition equation is of the form:

[0053] in, and are the generated term and dissipated term of the transition equation respectively.

[0054] After coupling the transition equation with the traditional SA model equation, the equation of the SA model coupled with the transition model becomes:

[0055] in, and are the generation term and dissipation term after coupling, is the transformed intermittent factor, It is an additional generated item used to ensure the accuracy of the calculation results under certain special conditions.

[0056] In the equation of the SA model of the above coupled transition model, based on the intermittent factor The different values ​​of can realize the conversion between laminar and turbulent states, so that more accurate aerodynamic parameter labels can be calculated for high-speed three-dimensional wing samples.

[0057] A2: Construct a three-dimensional wing sample data set, wherein the three-dimensional wing sample data set includes multiple three-dimensional wing samples, and any three-dimensional wing sample includes multiple cross-sectional images and aerodynamic parameter labels corresponding to the three-dimensional wing.

[0058] After generating corresponding aerodynamic parameter labels for each three-dimensional wing sample, a three-dimensional wing sample dataset is constructed. Each three-dimensional wing sample includes multiple cross-sectional images and aerodynamic parameter labels calculated by the RANS method based on the coupled transition model.

[0059] A3: Construct a first convolutional neural network, which includes an input layer, a dimensionality reduction layer, a convolutional backbone network layer and an output layer, wherein the dimensionality reduction layer is used to extract common features in multiple cross-sectional images corresponding to any three-dimensional wing.

[0060] Reference Figure 4 , showing a schematic diagram of the structure of the first convolutional neural network provided in an embodiment of the present application, the input object of the input layer of the first convolutional neural network is a plurality of cross-sectional images of any three-dimensional wing, for example, the cross-sectional image includes a contour image of the cross-section of the three-dimensional wing in any span direction; a dimensionality reduction layer is also included after the input layer, and the dimensionality reduction layer uses the PCA method to capture the common features between the cross-sectional images of any three-dimensional wing at different span directions. Since different cross-sectional images of the same three-dimensional wing are continuous, the dimensionality reduction layer can extract common feature data, thereby reducing the amount of data by reducing the dimensionality, thereby reducing the data processing cost of the first convolutional neural network and accelerating the data processing speed.

[0061] The dimensionality reduction layer is followed by the convolution backbone network layer, which includes three parts: convolution layer, pooling layer and fully connected layer. The convolution layer acts on the input data in sequence through a convolution kernel of a certain size to extract different features of the input data. The convolution operation will generate a large amount of data. The pooling layer is used to compress redundant data, which can prevent overfitting while reducing the amount of calculation. The final fully connected layer integrates the local features of the previous convolution, pooling, nonlinear activation and other operations to obtain complete feature information.

[0062] The fully connected layer is connected to the output layer, which outputs the aerodynamic parameters of any three-dimensional wing.

[0063] A4: Iteratively train the first convolutional neural network based on the three-dimensional wing sample data set to obtain a wing prediction model.

[0064] Specifically, a first convolutional neural network is trained according to a three-dimensional wing sample data set, multiple cross-sectional images of each of the multiple three-dimensional wing samples are input into the first convolutional neural network, and the first convolutional neural network is iteratively trained based on the deviation between the aerodynamic parameter prediction results of each of the multiple three-dimensional wing samples output by the first convolutional neural network and the aerodynamic parameter labels of each of the multiple three-dimensional wing samples.

[0065] In the actual implementation process, the first convolutional neural network can be iteratively trained based on an adjusted optimizer, such as Adam, ReLU or Leaky ReLU, or iteratively trained based on a mean square error loss function with an L2 regularization term added.

[0066] You can also use a custom loss function to improve the accuracy of the model's prediction results, but this implementation does not impose any restrictions.

[0067] During the network construction process, different hyperparameters of the first convolutional neural network can be adjusted based on the deviation between the aerodynamic parameter prediction results of the first convolutional neural network and the aerodynamic parameter labels, as well as the convergence during training. If the prediction results are overfitting, you can try to increase the number of pooling layers; if the prediction results converge poorly or the prediction results are less accurate, you can try to increase the number of hidden layers in the fully connected neural network or the number of neural units in each hidden layer.

[0068] When the iterative training of the first convolutional neural network reaches the target state, such as when the number of iterations reaches the target number or the value of the error function is less than the calibration value, the training can be stopped to obtain a trained wing prediction model.

[0069] In the application process of the prediction model, for any three-dimensional wing to be predicted, multiple cross-sectional images of the three-dimensional wing are input into the wing prediction model, and the aerodynamic parameter prediction results of the three-dimensional wing can be quickly predicted through the wing prediction model.

[0070] S103: When the object to be predicted is a three-dimensional blade, multiple cross-sectional images corresponding to the object to be predicted are input into a blade prediction model, and the blade prediction model outputs a prediction result of aerodynamic parameters corresponding to the object to be predicted. The blade prediction model is trained based on multiple cross-sectional images and aerodynamic parameter labels of multiple three-dimensional blade samples, and the aerodynamic parameter labels of the multiple three-dimensional blade samples are calculated based on the LES method.

[0071] This method is based on the aerodynamic parameter labels of multiple three-dimensional blade samples calculated by the LES method, and then trains a blade prediction model based on multiple cross-sectional images and aerodynamic parameter labels of the multiple three-dimensional blade samples. The blade prediction model predicts the aerodynamic parameter prediction results of the three-dimensional blade, so there is no need to use a numerical simulation algorithm to calculate the aerodynamic parameters of the three-dimensional blade every time, thereby reducing the calculation cost of the prediction and speeding up the efficiency of predicting the aerodynamic parameters.

[0072] Reference Figure 5 , shows a flowchart of the steps of the training process of the blade prediction model provided in an embodiment of the present application. In a feasible implementation manner, the training process of the blade prediction model includes: B1: Obtain the geometric profiles of multiple 3D blade samples, and calculate the aerodynamic parameter labels corresponding to each 3D blade sample based on the LES method.

[0073] First, multiple three-dimensional blade samples are obtained. Three-dimensional blades with more applications can be obtained as three-dimensional blade samples. The network of the geometric contours of the three-dimensional blade samples is used as input. The velocity distribution around each three-dimensional blade is simulated through open source CFD simulation software. The aerodynamic parameter label of each three-dimensional blade sample can be calculated through the velocity distribution.

[0074] In this embodiment, the aerodynamic parameter labels of each three-dimensional blade sample are calculated in the open source CFD simulation software based on the LES method.

[0075] Specifically, the momentum equation of the LES method is in the form of:

[0076] in, , is the velocity of the large-scale vortex, and the tilde above the physical quantity represents the filtering operation. and are the total velocity and the velocity of the small-scale vortex respectively; For the sub-grid stress tensor, additional model equations need to be added to solve it.

[0077] B2: constructing a three-dimensional blade sample data set, wherein the three-dimensional blade sample data set includes a plurality of three-dimensional blade samples, and any three-dimensional blade sample includes a plurality of cross-sectional images and aerodynamic parameter labels corresponding to the three-dimensional blade.

[0078] After generating corresponding aerodynamic parameter labels for each 3D blade sample, a 3D blade sample dataset is constructed. Each 3D blade sample includes multiple cross-sectional images and aerodynamic parameter labels calculated based on the LES method.

[0079] B3: Construct a second convolutional neural network, which includes an input layer, a dimensionality reduction layer, a convolutional backbone network layer and an output layer, wherein the dimensionality reduction layer is used to extract common features in multiple cross-sectional images corresponding to any three-dimensional leaf.

[0080] The network structure of the second convolutional neural network is consistent with that of the first convolutional neural network, which will not be repeated here. In the second convolutional neural network, a dimension reduction layer based on the PCA method is also included after the input layer. The dimension reduction layer captures the common features between the cross-sectional images of any three-dimensional blade at different spanwise positions. Since there is continuity between different cross-sectional images of the same three-dimensional blade, the dimension reduction layer can extract common feature data, thereby reducing the amount of data by reducing the dimension, which can reduce the data processing cost of the second convolutional neural network and speed up the data processing.

[0081] B4: Iteratively train the second convolutional neural network based on the three-dimensional blade sample data set to obtain a blade prediction model.

[0082] The process of training the second convolutional neural network based on the three-dimensional leaf sample data set is consistent with the process of training the first convolutional neural network in step A4, and will not be repeated here.

[0083] When the iterative training of the second convolutional neural network reaches the target state, such as when the number of iterations reaches the target number or the value of the error function is less than the calibration value, the training can be stopped to obtain a trained blade prediction model.

[0084] In the application process of the prediction model, for any three-dimensional blade to be predicted, multiple cross-sectional images of the three-dimensional blade are input into the blade prediction model, and the aerodynamic parameter prediction results of the three-dimensional blade can be quickly predicted through the blade prediction model.

[0085] This method first obtains the aerodynamic parameter labels of the three-dimensional wing and the low-speed three-dimensional blade based on different methods. Specifically, the aerodynamic parameter label of the three-dimensional wing is obtained by the RANS method of the coupled transition model, and the aerodynamic parameter label of the three-dimensional blade is obtained by the LES method. Compared with the existing aerodynamic parameter calculation methods that all use the RANS method, a more suitable aerodynamic parameter calculation method can be selected according to the speed difference of the three-dimensional wing and the three-dimensional blade, thereby obtaining more accurate aerodynamic parameter labels of the three-dimensional wing and the low-speed three-dimensional blade.

[0086] Then, a wing prediction model is obtained by training with more accurate aerodynamic parameter labels of the three-dimensional wing and multiple cross-sectional images of the three-dimensional wing, so that the wing prediction model learns the relationship between the multiple cross-sectional images and aerodynamic parameter labels of the three-dimensional wing samples. In the subsequent prediction process, the wing prediction model can quickly and accurately generate aerodynamic parameter prediction results for high-speed three-dimensional wings. It is no longer necessary to calculate the aerodynamic parameters of each three-dimensional wing in turn based on the numerical simulation method, thereby speeding up the prediction of the aerodynamic parameters of the three-dimensional wing.

[0087] The blade prediction model is obtained by training with more accurate aerodynamic parameter labels of three-dimensional blades and multiple cross-sectional images of three-dimensional blades, so that the blade prediction model learns the relationship between multiple cross-sectional images and aerodynamic parameter labels of three-dimensional blade samples. In the subsequent prediction process, the blade prediction model can quickly and accurately generate aerodynamic parameter prediction results for low-speed three-dimensional blades. It is no longer necessary to calculate the aerodynamic parameters of each three-dimensional blade in turn based on numerical simulation methods, thereby speeding up the prediction of the aerodynamic parameters of the three-dimensional blades.

[0088] In a feasible implementation manner, after the wing prediction model outputs the aerodynamic parameter prediction result corresponding to the object to be predicted, the method further includes: C1: In response to a target operation, a target aerodynamic parameter corresponding to the object to be predicted is obtained, where the target aerodynamic parameter includes at least one parameter type of a target drag coefficient, a target pressure coefficient, and a maximum lift-to-drag ratio.

[0089] After the geometric shape of any three-dimensional wing or three-dimensional blade is determined, the corresponding aerodynamic parameter prediction results are predicted by the wing prediction model and the blade prediction model, and then the target aerodynamic parameters are set as constraints. The geometric shape of the three-dimensional wing or three-dimensional blade can be adjusted so that the aerodynamic parameters of the adjusted three-dimensional wing or three-dimensional blade are close to or equal to the target aerodynamic parameters.

[0090] C2: iteratively updating the geometric parameters of the object to be predicted according to the target aerodynamic parameters corresponding to the object to be predicted and the aerodynamic parameter prediction result, until the object to be predicted after iterative update meets the target aerodynamic parameters.

[0091] Specifically, during each optimization iteration, a target difference is calculated according to the target aerodynamic parameter corresponding to the object to be predicted and the aerodynamic parameter prediction result.

[0092] The target difference is calculated as follows:

[0093] Wherein, MSE is the target difference; n is the number of parameter types in the target aerodynamic parameter; is the i-th aerodynamic parameter in the aerodynamic parameter prediction result; is the i-th aerodynamic parameter among the target aerodynamic parameters.

[0094] When the target difference is greater than a preset target threshold, the geometric parameters of the object to be predicted are adjusted, the object to be predicted after the geometric parameters are adjusted is used as the object to be predicted, and a prediction result of aerodynamic parameters corresponding to the current object to be predicted is calculated based on the wing prediction model or the blade prediction model; When the target difference is less than the target threshold, the iteration is stopped, and the prediction results of the geometric parameters and aerodynamic parameters corresponding to the current object to be predicted are output.

[0095] In the process of iteratively updating the geometric parameters of the object to be predicted, the existing common gradient method, genetic algorithm and particle swarm algorithm can be used, and this embodiment does not limit this.

[0096] Taking genetic algorithm as an example, the optimization process of genetic algorithm is as follows: (1) Initialize and generate the first generation population; (2) Calculate the fitness of each individual in the population; (3) Determine whether the fitness of the current population meets the requirements. If so, output the optimization result; if not, continue with the subsequent steps; (4) Select the optimal individual; (5) Generate the next generation population through crossover and mutation, and repeat the cycle until the fitness meets the requirements.

[0097] Reference Figure 6 , shows a schematic diagram of the genetic algorithm optimization process provided by an embodiment of the present application. By way of example, in this embodiment, multiple three-dimensional wings can be used as the first-generation population, and the aerodynamic parameter prediction results of each three-dimensional wing are predicted by the wing prediction model respectively. Then, the target difference between the aerodynamic parameter prediction results of each three-dimensional wing and the target aerodynamic parameters is used as the fitness. If there is any three-dimensional wing whose fitness, that is, the target difference, is less than or equal to the target threshold, the three-dimensional wing with the smallest fitness is used as the optimized three-dimensional wing.

[0098] If the fitness of any three-dimensional wing, that is, the target difference, is greater than the target threshold, multiple three-dimensional wings can be crossed and mutated to produce the next generation population, and the wing prediction model is continued to predict the aerodynamic parameter prediction results of each three-dimensional wing in the current population. The cycle continues until there is a three-dimensional wing whose fitness, that is, the target difference, is less than or equal to the target threshold. The three-dimensional wing with the smallest fitness is used as the optimized three-dimensional wing.

[0099] The method provided in this embodiment has at least the following beneficial effects: 1. For high-speed three-dimensional wings, the RANS method of coupled transition model can obtain more accurate aerodynamic parameter labels of three-dimensional wings, and the trained wing prediction model can generate more accurate aerodynamic parameter prediction results of three-dimensional wings; for low-speed three-dimensional blades, the LES method is used to obtain more accurate aerodynamic parameter labels of three-dimensional wings, and the trained blade prediction model can generate more accurate aerodynamic parameter prediction results of three-dimensional blades; 2. Predicting the aerodynamic parameter prediction results of a three-dimensional wing based on the wing prediction model and predicting the aerodynamic parameter prediction results of a three-dimensional blade based on the blade prediction model can reduce the calculation cost and improve the speed of accurately predicting the aerodynamic parameters of the wing or blade compared to calculating the aerodynamic parameters of the three-dimensional wing or three-dimensional blade one by one using the numerical simulation method; 3. Integrate the PCA method into the framework of the convolutional neural network and use the PCA method to reduce the dimension of the input data so that the trained wing prediction model and blade prediction model can be applied to three-dimensional problems; 4. Combining the wing prediction model and the blade prediction model with the optimization algorithm can increase the speed of optimizing the geometric shapes of the three-dimensional wing and the three-dimensional blade.

[0100] An embodiment of the present application further provides a computer device, comprising: at least one processor, and a memory, wherein the memory stores a computer program that can be run on the processor, wherein the processor executes the aerodynamic parameter prediction method described in the embodiment when executing the computer program.

[0101] Reference Figure 7 , showing a schematic diagram of a non-volatile readable storage medium provided in an embodiment of the present application, wherein the non-volatile readable storage medium 700 stores a computer program 701, wherein the computer program 701, when executed by a processor, executes the aerodynamic parameter prediction method described in the embodiment.

[0102] Reference Figure 8 , showing a schematic diagram of a computer program product provided in an embodiment of the present application, the computer program product 800 includes a computer program / instruction 801, and when the computer program / instruction 801 is executed by a processor, the aerodynamic parameter prediction method described in the embodiment is implemented.

[0103] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0104] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0106] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0108] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0109] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0110] Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for predicting aerodynamic parameters, characterized in that: The method comprises: Acquire an object to be predicted, wherein the object to be predicted includes a three-dimensional wing and a three-dimensional blade; When the object to be predicted is a three-dimensional wing, a plurality of cross-sectional images corresponding to the object to be predicted are input into a wing prediction model, and a prediction result of aerodynamic parameters corresponding to the object to be predicted is output through the wing prediction model, wherein the wing prediction model is obtained by training based on a plurality of cross-sectional images and aerodynamic parameter labels of a plurality of three-dimensional wing samples, and the aerodynamic parameter labels of the plurality of three-dimensional wing samples are calculated based on a RANS method of a coupled transition model; When the object to be predicted is a three-dimensional blade, multiple cross-sectional images corresponding to the object to be predicted are input into a blade prediction model, and the blade prediction model outputs aerodynamic parameter prediction results corresponding to the object to be predicted. The blade prediction model is trained based on multiple cross-sectional images and aerodynamic parameter labels of multiple three-dimensional blade samples, and the aerodynamic parameter labels of the multiple three-dimensional blade samples are calculated based on the LES method.

2. The method according to claim 1, characterized in that The method further comprises: The geometric profiles of multiple three-dimensional wing samples are obtained, and the aerodynamic parameter labels corresponding to each three-dimensional wing sample are calculated based on the RANS method of the coupled transition model; Constructing a three-dimensional wing sample data set, wherein the three-dimensional wing sample data set includes a plurality of three-dimensional wing samples, and any three-dimensional wing sample includes a plurality of cross-sectional images and aerodynamic parameter labels corresponding to the three-dimensional wing; Constructing a first convolutional neural network, the first convolutional neural network comprising an input layer, a dimensionality reduction layer, a convolutional backbone network layer and an output layer, wherein the dimensionality reduction layer is used to extract common features in multiple cross-sectional images corresponding to any three-dimensional wing; The first convolutional neural network is iteratively trained based on the three-dimensional wing sample data set to obtain a wing prediction model.

3. The method according to claim 2, characterized in that The geometric profiles of multiple 3D wing samples are obtained, and the aerodynamic parameter labels corresponding to each 3D wing sample are calculated based on the RANS method of the coupled transition model, including: For any three-dimensional wing sample, a geometric profile corresponding to the three-dimensional wing sample is obtained, and an SA model of a coupled transition model is input, wherein the SA model is a model based on a RANS method; The aerodynamic parameter labels corresponding to the three-dimensional wing sample are calculated through the SA model of the coupled transition model.

4. The method according to claim 3, characterized in that: The calculation formula of the SA model of the coupled transition model is: in, By solving the convection transport equation, we can obtain is the dynamic viscosity coefficient, is the density, represents kinematic viscosity; are geometric coordinates; and are the generated term and dissipated term after coupling respectively; For additional generated items; is an empirical parameter; is the average speed.

5. The method according to claim 1, characterized in that The geometric profiles of multiple three-dimensional blade samples are obtained, and based on the LES method, the aerodynamic parameter labels corresponding to each three-dimensional blade sample are calculated respectively; Constructing a three-dimensional blade sample data set, wherein the three-dimensional blade sample data set includes a plurality of three-dimensional blade samples, and any three-dimensional blade sample includes a plurality of cross-sectional images and aerodynamic parameter labels corresponding to the three-dimensional blade; Constructing a second convolutional neural network, the second convolutional neural network comprising an input layer, a dimensionality reduction layer, a convolutional backbone network layer and an output layer, wherein the dimensionality reduction layer is used to extract common features in multiple cross-sectional images corresponding to any three-dimensional leaf; The second convolutional neural network is iteratively trained based on the three-dimensional blade sample data set to obtain a blade prediction model.

6. The method according to claim 1, characterized in that After the wing prediction model outputs the aerodynamic parameter prediction result corresponding to the object to be predicted, the method further includes: In response to the target operation, obtaining a target aerodynamic parameter corresponding to the object to be predicted, wherein the target aerodynamic parameter includes at least one parameter type of a target drag coefficient, a target pressure coefficient, and a maximum lift-to-drag ratio; Iteratively updating the geometric parameters of the object to be predicted according to the target aerodynamic parameters corresponding to the object to be predicted and the aerodynamic parameter prediction result until the iteratively updated object to be predicted meets the target aerodynamic parameters, including: Calculating a target difference according to the target aerodynamic parameter corresponding to the object to be predicted and the aerodynamic parameter prediction result; When the target difference is greater than a preset target threshold, the geometric parameters of the object to be predicted are adjusted, the object to be predicted after the geometric parameters are adjusted is used as the object to be predicted, and a prediction result of aerodynamic parameters corresponding to the current object to be predicted is calculated based on the wing prediction model or the blade prediction model; When the target difference is less than the target threshold, the iteration is stopped, and the prediction results of the geometric parameters and aerodynamic parameters corresponding to the current object to be predicted are output.

7. The method according to claim 6, characterized in that The target difference is calculated as follows: Wherein, MSE is the target difference; n is the number of parameter types in the target aerodynamic parameter; is the i-th aerodynamic parameter in the aerodynamic parameter prediction result; is the i-th aerodynamic parameter among the target aerodynamic parameters.

8. A computer device, characterized in that: include: At least one processor, and a memory, wherein the memory stores a computer program that can be run on the processor, wherein the processor executes the aerodynamic parameter prediction method according to any one of claims 1 to 7 when executing the computer program.

9. A non-volatile readable storage medium, characterized in that: The non-volatile readable storage medium stores a computer program, wherein the computer program, when executed by a processor, executes the aerodynamic parameter prediction method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the aerodynamic parameter prediction method according to any one of claims 1 to 7 is implemented.

Images