Aircraft flow field reconstruction method

Abstract

The invention discloses an aircraft flow field reconstruction method, and belongs to the technical field of aerodynamics. According to the method, the problems of unstable precision and poor reliability of a three-dimensional flow field reconstruction prediction result of an aircraft with a complex shape are solved. According to the method, geometric shapes of different aircrafts are generated through parameterization, and flow field calculation is carried out by adopting a CFD calculation method; carrying out geometric expression point sampling on the obtained different aircraft geometric shapes, and extracting geometric shape feature information by adopting a symbolic distance expression method; the aircraft appearance feature information and the incoming flow condition are combined into an input feature matrix, a three-direction velocity field and a pressure field obtained through calculation of a corresponding flow field serve as an output feature matrix, a flow field reconstruction data set is formed, and a training set, a verification set and a test set are divided; and importing the obtained data set into a network of a dual nested attention mechanism, and carrying out model training, so that flow field accurate prediction can be realized for aircrafts with different input appearance characteristics and different working conditions.

Description

Technical Field

[0001] The invention relates to a flow field reconstruction method and belongs to the technical field of aerodynamics. Background Art

[0002] In recent years, with the gradual application of artificial intelligence technology in the field of aerodynamics, flow field reconstruction technology based on deep learning has become a hot topic of continuous concern in the field of aerodynamics. This technology uses a convolutional neural network method to model and learn complex flow field data. This method can quickly and accurately capture multi-dimensional flow pattern information, thereby realizing rapid inversion and reconstruction of the flow field of aircraft with different shapes. After sufficient verification, the current research method can prove that the two-dimensional airfoil flow field reconstruction technology has gradually matured. However, in the process of transition from two-dimensional to three-dimensional, due to the substantial increase in the number of characteristic variables, it is more difficult to reconstruct the three-dimensional flow field with high precision. The current research progress at home and abroad still cannot completely solve the problems of unstable accuracy and poor reliability of the prediction results of the three-dimensional flow field reconstruction for aircraft with complex shapes. Therefore, further in-depth research on the three-dimensional flow field reconstruction algorithm of aircraft is still needed. Summary of the invention

[0003] In order to overcome the problem of unstable accuracy and poor reliability of the prediction results of three-dimensional flow field reconstruction for complex-shaped aircraft, the present invention provides an aircraft flow field reconstruction method. A brief overview of the present invention is given below to provide a basic understanding of certain aspects of the present invention. It should be understood that this overview is not an exhaustive overview of the present invention. It is not intended to identify the key or important parts of the present invention, nor is it intended to limit the scope of the present invention.

[0004] The technical solution of the present invention: A method for reconstructing an aircraft flow field comprises the following steps: S1. Generate different aircraft geometric shapes through parameterization, modify the setting parameters in batches, and use CFD calculation method to calculate the flow field; S2, sampling geometric expression points of different aircraft geometric shapes obtained in step S1, and extracting geometric shape feature information using a signed distance expression method; S3, combining the aircraft shape feature information obtained in step S2 with the incoming flow conditions into an input feature matrix, using the three-directional velocity field and pressure field calculated from the corresponding flow field as the output feature matrix, forming a flow field reconstruction data set, and dividing it into a training set, a validation set, and a test set; S4. Import the data set obtained in step S3 into the Unet++ network with a double nested attention mechanism for model training, so as to achieve accurate flow field prediction for aircraft with different shape features and different working conditions.

[0005] Furthermore, step S1 generates the aircraft shape parameters by parameterization, and automatically generates meshes for aircraft with different shapes by using Fun3D; at the same time, batch processing modifies the setting parameters, and uses the RANS equation to solve the flow field of the aircraft under different working conditions.

[0006] Furthermore, the method for extracting geometric features of the aircraft shape in step S2 is specifically as follows: S2.1. Sampling geometric expression points of the aircraft shape by marking and recording the geometric positions on the grid points and forming a geometric position matrix; S2.2, using an implicit expression method based on geometric signed distance function, by solving the pixel points of each geometric position matrix in the image space of the aircraft shape To region boundary The shortest distance , and the sign of the distance is defined differently, and its expression is: ; in, is a symbolic function used to distinguish the definition; when the point is inside the region boundary, ; When the point is outside the region boundary, ; When the point is on the boundary, Based on the above expression, by extracting the grid points calculated by CFD, the signed distance function is used to calculate the distance from each grid point to the aircraft surface.

[0007] Furthermore, in step S3, a data set is constructed for the flow field reconstruction method of different aircraft shapes. By changing the wing geometric shape parameters, incoming flow state, and flight parameters, CFD calculation is used to solve the velocity, pressure, and density distribution of the compressible flow around the wing, thereby generating a 7-channel data set. The first three channels, the wing geometric feature expressions SDF, Mach number, and angle of attack, are used as neural network inputs; the last four channels, U velocity field, V velocity field, W velocity field, and P pressure field, are used as neural network outputs.

[0008] Furthermore, in step S3, the data is divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0009] Furthermore, in step S4, the data set obtained in step S3 is imported into the Unet++ network of the double nested attention mechanism for model training. The flow field calculation results of the three-dimensional wing are modeled by a neural network. The model continuously iterates and optimizes the weight coefficients between neurons in the neural network through the gradient descent method, so as to find the optimal fitting solution between the input variables and the output variables, and establish a direct mapping from the three-dimensional wing geometric features, the incoming flow characteristics to the speed and pressure, thereby realizing the construction of a flow field reconstruction prediction model.

[0010] Furthermore, the Unet++ network of the dual nested attention mechanism in step S4 is composed of a backbone Unet++ network part and a dual parallel attention mechanism part, wherein the backbone Unet++ network is used to obtain multi-scale information of wing geometric shape features, and the backbone Unet++ network has the ability to deeply mine feature information; at the same time, the dual parallel attention mechanism part is used to focus on key feature information and capture the correlation between features.

[0011] Furthermore, Unet++ adopts multiple Unet structures as a whole.

[0012] Furthermore, the dual parallel attention mechanism part extracts features from the information in the input variables by combining the channel attention mechanism and the spatial attention mechanism through the feature importance extraction method based on the channel domain and the spatial domain; the channel attention mechanism compresses the current feature dimension by performing global pooling and average pooling on the input flow field image. The final pooling result is learned in the multi-layer perceptron MLP and outputted through the Sigmoid function. Its expression is: ; in, and Represents average pooling and maximum pooling, represents the activation function, and represents the weight coefficient of MLP, ; and Respectively represent the average pooling and maximum pooling results under the channel attention mechanism; The spatial attention mechanism performs global pooling and average pooling on the flow field features, concatenates the pooled results, and finally processes them through convolution and activation functions. Its expression is: ; in, and Represents average pooling and maximum pooling, represents the activation function, Expressed as The convolution operation, and Respectively represent the average pooling and maximum pooling results under the spatial attention mechanism.

[0013] Furthermore, the dual parallel attention mechanism part adopts the method of parallelizing the channel and space dual attention mechanisms to realize the parallel design structure. The expression of the attention mechanism structure is: ; ; ; Among them, × represents the fusion operator, + represents the addition operator, Represents the result after fusion with the spatial attention mechanism, Represents the result after fusion with the channel attention mechanism, Represented as the final result of the attention mechanism.

[0014] The beneficial effects of the present invention are embodied in: Compared with the prior art, the present invention is a Unet++ network with a dual nested attention mechanism, which combines the parallel design structure of the attention mechanism with the nested network architecture, and reconstructs the three-directional velocity field and pressure field of aircraft with different geometric parameter shape characteristics and different working conditions through the signed distance expression method. It has the advantages of stable accuracy and high reliability of the three-dimensional flow field reconstruction prediction results for aircraft with complex shapes. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flow chart of an aircraft flow field reconstruction method described in Example 1; Figure 2 A schematic diagram of the architecture of the Unet++ network model of the dual nested attention mechanism described in the present invention; Figure 3 The present invention is a flowchart for describing in detail a method for reconstructing an aircraft flow field. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is described below by the specific embodiments shown in the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In addition, in the following description, the description of known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present invention. It should also be understood that, unless otherwise specified or indicated, the terms "first", "second", "third" and the like in the specification are only used to distinguish between the various components, elements, steps, etc. in the specification, rather than to represent the logical relationship or sequential relationship between the various components, elements, steps, etc.

[0017] Embodiment 1, combination Figure 1-Figure 3 This embodiment describes a method for reconstructing an aircraft flow field, including the following steps: A method for reconstructing an aircraft flow field comprises the following steps: S1. Generate different aircraft geometric shapes through parameterization, modify the setting parameters in batches, and use CFD calculation method to calculate the flow field; Step S1 generates the aircraft shape parameters by parameterization, and automatically generates meshes for aircraft with different shapes by using Fun3D; at the same time, batch modify the setting parameters, and use the RANS equation to solve the flow field of the aircraft under different working conditions.

[0018] S2, sampling geometric expression points of different aircraft geometric shapes obtained in step S1, and extracting geometric shape feature information using a signed distance expression method; The method for extracting the geometric features of the aircraft shape in step S2 is specifically as follows: S2.1. Sampling geometric expression points of the aircraft shape by marking and recording the geometric positions on the grid points and forming a geometric position matrix; S2.2, using an implicit expression method based on geometric signed distance function (SDF), by solving the pixel points of each geometric position matrix in the image space of the aircraft shape To region boundary The shortest distance , and the sign of the distance is defined differently, and its expression is: ; in, is a symbolic function used to distinguish the definition; when the point is inside the region boundary, ; When the point is outside the region boundary, ; When the point is on the boundary, Based on the above expression, by extracting the grid points calculated by CFD, the signed distance function is used to calculate the distance from each grid point to the aircraft surface, so as to achieve accurate expression of the aircraft's geometric shape and provide effective geometric shape input features for the subsequent construction of a deep neural network.

[0019] S3, combining the aircraft shape feature information obtained in step S2 with the incoming flow conditions into an input feature matrix, using the three-directional velocity field and pressure field calculated from the corresponding flow field as the output feature matrix, forming a flow field reconstruction data set, and dividing it into a training set, a validation set, and a test set in a ratio of 8:1:1; In step S3, a data set is constructed for the flow field reconstruction method of different aircraft shapes. By changing the wing geometric shape parameters, incoming flow state, and flight parameters, CFD calculation is used to solve the velocity, pressure, and density distribution of the compressible flow around the wing, thereby generating a 7-channel data set. The first three channels, the wing geometric feature expression SDF, Mach number, and angle of attack, are used as neural network inputs; the last four channels, U velocity field, V velocity field, W velocity field, and P pressure field, are used as neural network outputs.

[0020] S4, importing the data set obtained in step S3 into the Unet++ network of the double nested attention mechanism to perform model training, so as to achieve accurate flow field prediction for aircraft with different shape features and different working conditions; The data set obtained in step S3 is imported into the Unet++ network of the double nested attention mechanism for model training. The neural network modeling is performed on the flow field calculation results of the three-dimensional wing. The model continuously iterates and optimizes the weight coefficients between neurons in the neural network through the gradient descent method, so as to find the optimal fitting solution between the input variables and the output variables, and establish a direct mapping from the three-dimensional wing geometric characteristics, the incoming flow characteristics to the flow parameters such as speed and pressure, so as to realize the construction of the flow field reconstruction prediction model.

[0021] Specifically, the Unet++ network of the dual nested attention mechanism in step S4 is composed of a backbone Unet++ network part and a dual parallel attention mechanism part, wherein the backbone Unet++ network is used to obtain multi-scale information of the wing geometric shape features, and the backbone Unet++ network has the ability to deeply mine feature information; at the same time, the dual parallel attention mechanism part is used to focus on key feature information and capture the correlation between features.

[0022] Unet++ adopts multiple Unet structures as a whole, and has multi-layer connection structure characteristics in the network. Specifically, each upsampling block will not only be connected to its corresponding downsampling block, but also to other downsampling blocks with the same depth; this makes the downsampling and skipping process of Unet++ more intensive. This multi-layer connection structure helps to pass the underlying features to the upper layer, improve the perception of the model's feature extraction, and reduce the loss of eigenvalues ​​during upsampling and skip connections. The narrowing of the subnetwork between the encoder and decoder can enhance the ability to identify flow characteristics in the flow field and improve prediction accuracy and fit.

[0023] Specifically, the dual parallel attention mechanism part extracts features from the information in the input variables by combining the channel attention mechanism and the spatial attention mechanism through the feature importance extraction method based on the channel domain and the spatial domain; the channel attention mechanism is an attention mechanism for selecting and weighting input features in the channel dimension. This mechanism compresses the current feature dimension by performing global pooling and average pooling on the input flow field image. The final pooling result is learned in a multi-layer perceptron (MLP) and outputted through the Sigmoid function. Its expression is: ; in, and Represents average pooling and maximum pooling, represents the activation function, and Represents the weight coefficient of MLP ( ); and They represent the average pooling and maximum pooling results under the channel attention mechanism respectively. Based on the above expressions, the attention weight of each channel is calculated through MLP, which enables the model to focus on the channels associated with the current task objectives when processing multi-channel input features. Finally, the weights are applied to the channel dimensions of the input features to form weighted channel feature results. The spatial attention mechanism is an attention mechanism that focuses on a specified position in image feature extraction. This mechanism performs global pooling and average pooling on flow field features, concatenates the pooled results, and finally processes them through convolution and activation functions. Its expression is: ; in, and Represents average pooling and maximum pooling, represents the activation function, Expressed as The convolution operation, and They respectively represent the average pooling and maximum pooling results under the spatial attention mechanism; based on the above expressions, the importance of features in the image is weighted, so that more attention can be paid to important areas in the image and the representation ability of the features can be improved.

[0024] Specifically, the dual parallel attention mechanism part adopts the method of parallelizing the channel and space dual attention mechanisms to realize the parallel design structure. The expression of the attention mechanism structure is: ; ; ; Among them, × represents the fusion operator, + represents the addition operator, Represents the result after fusion with the spatial attention mechanism, Represents the result after fusion with the channel attention mechanism, Represented as the final result of the attention mechanism; Based on the above expression, an attention mechanism is added after each output module in the Unet++ network to enhance the ability to extract effective information while ensuring that the number of input channels does not change.

[0025] Embodiment 2: An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of an aircraft flow field reconstruction method described in Example 1 when executing the computer program.

[0026] The computer device of the present invention may be a device including a processor and a memory, such as a single chip microcomputer including a central processing unit. Furthermore, the processor is used to implement the steps of the above-mentioned method for predicting the drag reduction of fuselage turbulent micro-ribs based on the Bayesian probability model when executing the computer program stored in the memory.

[0027] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0028] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0029] Embodiment 3: A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of an aircraft flow field reconstruction method described in Example 1.

[0030] The computer-readable storage medium of the present invention can be any form of storage medium that can be read by the processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. A computer program is stored on the computer-readable storage medium. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned aircraft flow field reconstruction method can be implemented.

[0031] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer readable media do not include electric carrier signals and telecommunication signals.

[0032] The embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for reconstructing an aircraft flow field, characterized in that: The following steps are involved: S1. Generate different aircraft geometric shapes through parameterization, modify the setting parameters in batches, and use CFD calculation method to calculate the flow field; S2, sampling geometric expression points of different aircraft geometric shapes obtained in step S1, and extracting geometric shape feature information using a signed distance expression method; S3, combining the aircraft shape feature information obtained in step S2 with the incoming flow conditions into an input feature matrix, using the three-directional velocity field and pressure field calculated from the corresponding flow field as the output feature matrix, forming a flow field reconstruction data set, and dividing it into a training set, a validation set, and a test set; S4. Import the data set obtained in step S3 into the Unet++ network with a double nested attention mechanism for model training, so as to achieve accurate flow field prediction for aircraft with different shape features and different working conditions.

2. The method for reconstructing an aircraft flow field according to claim 1, characterized in that: Step S1 generates the aircraft shape parameters by parameterization, and automatically generates meshes for aircraft with different shapes by using Fun3D; at the same time, batch modify the setting parameters, and use the RANS equation to solve the flow field of the aircraft under different working conditions.

3. The method for reconstructing an aircraft flow field according to claim 1, characterized in that: The method for extracting the geometric features of the aircraft shape in step S2 is specifically as follows: S2.

1. Sampling geometric expression points of the aircraft shape by marking and recording the geometric positions on the grid points and forming a geometric position matrix; S2.2, using an implicit expression method based on geometric signed distance function, by solving the pixel points of each geometric position matrix in the image space of the aircraft shape To region boundary The shortest distance , and the sign of the distance is defined differently, and its expression is: ； in, is a symbolic function used to distinguish the definition; when the point is inside the region boundary, ; When the point is outside the region boundary, ; When the point is on the boundary, Based on the above expression, by extracting the grid points calculated by CFD, the signed distance function is used to calculate the distance from each grid point to the aircraft surface.

4. The method for reconstructing an aircraft flow field according to claim 1, 2 or 3, characterized in that: In step S3, a data set is constructed for the flow field reconstruction method of different aircraft shapes. By changing the wing geometric shape parameters, incoming flow state, and flight parameters, CFD calculation is used to solve the velocity, pressure, and density distribution of the compressible flow around the wing, thereby generating a 7-channel data set. The first three channels, the wing geometric feature expression SDF, Mach number, and angle of attack, are used as neural network inputs; the last four channels, U velocity field, V velocity field, W velocity field, and P pressure field, are used as neural network outputs.

5. The method for reconstructing an aircraft flow field according to claim 4, characterized in that: In step S3, the data is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1.

6. The method for reconstructing an aircraft flow field according to claim 5, characterized in that: In step S4, the data set obtained in step S3 is imported into the Unet++ network of the double nested attention mechanism for model training. The flow field calculation results of the three-dimensional wing are used for neural network modeling. The model continuously iterates and optimizes the weight coefficients between neurons in the neural network through the gradient descent method, so as to find the optimal fitting solution between the input variables and the output variables, and establish a direct mapping from the three-dimensional wing geometric features, the incoming flow characteristics to the speed and pressure, thereby realizing the construction of the flow field reconstruction prediction model.

7. The method for reconstructing an aircraft flow field according to claim 6, characterized in that: The Unet++ network of the dual nested attention mechanism in step S4 is composed of a backbone Unet++ network part and a dual parallel attention mechanism part, wherein the backbone Unet++ network is used to obtain multi-scale information of the wing geometric shape features, and the backbone Unet++ network has the ability to deeply mine feature information; at the same time, the dual parallel attention mechanism part is used to focus on key feature information and capture the correlation between features; Unet++ as a whole adopts multiple Unet structures.

8. The method for reconstructing an aircraft flow field according to claim 7, characterized in that: The dual parallel attention mechanism extracts features from the input variables by combining the channel attention mechanism with the spatial attention mechanism through the feature importance extraction method based on the channel domain and the spatial domain. The channel attention mechanism compresses the current feature dimension by performing global pooling and average pooling on the input flow field image. The final pooling result is learned in the multi-layer perceptron MLP and outputted through the Sigmoid function. Its expression is: ； in, and Represents average pooling and maximum pooling, represents the activation function, and represents the weight coefficient of MLP, ; and Respectively represent the average pooling and maximum pooling results under the channel attention mechanism; The spatial attention mechanism performs global pooling and average pooling on the flow field features, concatenates the pooled results, and finally processes them through convolution and activation functions. Its expression is: ； in, and Represents average pooling and maximum pooling, represents the activation function, Expressed as The convolution operation, and Respectively represent the average pooling and maximum pooling results under the spatial attention mechanism; The dual parallel attention mechanism part adopts the method of parallelizing the channel and space dual attention mechanisms to realize the parallel design structure. The expression of the attention mechanism structure is: ； ； ； Among them, × represents the fusion operator, + represents the addition operator, Represents the result after fusion with the spatial attention mechanism, Represents the result after fusion with the channel attention mechanism, Represented as the final result of the attention mechanism.

9. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of an aircraft flow field reconstruction method as claimed in any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of an aircraft flow field reconstruction method according to any one of claims 1 to 8 are implemented.

Images