Method for improving three-dimensional high-precision flow field analysis efficiency and precision

Generating high-quality three-dimensional adaptive mesh through deep learning models and SOM neural network technology solves the problems of low efficiency and high computational cost of three-dimensional grid adaptive technology under complex conditions, and realizes high-precision flow field analysis while reducing it Calculate the cost.

CN118917232AActive Publication Date: 2024-11-08NANJING UNIV OF AERONAUTICS & ASTRONAUTICS

Patent Information

Application Number
CN202410919459.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-08
Estimated Expiration
2044-07-10

AI Technical Summary

Technical Problem

The existing three-dimensional grid adaptation technology is inefficient and has high computational cost under complex geometric and boundary conditions, making it difficult to reduce the computational cost and time of grid adaptation while ensuring high-precision three-dimensional flow field calculations.

Method used

Deep learning model combined with SOM neural network technology is used to generate a three-dimensional adaptive grid with node density distribution consistent with the flow field structure through flow field condition parameters, build a sample database and train a deep learning model to predict a high-quality network that matches the flow field to be analyzed. grid.

Benefits of technology

While ensuring high-precision three-dimensional flow field calculation, it significantly reduces the calculation cost and time of grid adaptability, improves the efficiency and accuracy of flow field analysis, and avoids the high computing burden in traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917232A_ABST
    Figure CN118917232A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a method for improving the analysis efficiency and precision of a three-dimensional high-precision flow field, and relates to the field of a three-dimensional grid adaptive technology and a deep learning technology oriented to computational fluid mechanics. According to the method, the high-quality three-dimensional adaptive grid of which the node density distribution accords with the flow field structure can be efficiently generated, so that the accuracy and efficiency of flow field numerical calculation are enhanced. The method comprises the following steps: a server receives initial three-dimensional grid data and flow field condition parameters sent by a client; generating initial three-dimensional flow field data based on the flow field condition parameters, and generating a corresponding three-dimensional adaptive grid by using the data; establishing a sample database according to the three-dimensional initial flow field data and the adaptive grid data; training a deep learning model by using the sample database; and inputting to-be-predicted flow field condition parameters into the trained deep learning model, predicting a three-dimensional adaptive grid, generating a high-quality three-dimensional grid conforming to a to-be-predicted flow field structure, and sending the high-quality three-dimensional grid to the client.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a three-dimensional grid adaptive technology for computational fluid dynamics, and in particular to a method for improving the efficiency and accuracy of three-dimensional high-precision flow field analysis. Background Art

[0002] In computational fluid dynamics (CFD), the accuracy of numerical calculations greatly depends on the quality of the mesh, especially when dealing with three-dimensional problems. The complexity of three-dimensional problems lies in the fact that the increase in spatial dimensions leads to a sharp expansion in the number of nodes and meshes and an increase in the difficulty of boundary processing, which increases the difficulty of calculation and resource requirements. In order to obtain high-precision results at the lowest possible computational cost, high-quality three-dimensional meshes should be dense in high-gradient areas of the flow field solution and relatively sparse in low-gradient areas. Therefore, efficient and stable three-dimensional mesh adaptive methods are essential for accurately simulating complex flow fields.

[0003] At present, the adaptive meshing technologies widely used in the CFD field mainly include local refinement method and moving mesh method. The local refinement method optimizes the mesh layout by adding mesh nodes in the area with large solution gradient. Although this method can improve the calculation accuracy of the local area, it also complicates the topological structure of the mesh and increases the calculation burden. On the other hand, the moving mesh method can improve the efficiency and accuracy of numerical calculations without significantly increasing the amount of calculations by moving nodes to the area where the solution changes dramatically without changing the total number of nodes and the mesh topological structure.

[0004] Although traditional methods have shown certain effects on two-dimensional problems, these adaptive methods are often accompanied by high costs and low efficiency when facing three-dimensional problems, especially under complex geometries and boundary conditions. In addition, most existing moving mesh methods also require the construction of auxiliary mesh partial differential equation models, which adds additional computational burden to CFD calculations. Any change in the target flow field requires the reconstruction of these auxiliary models, further increasing the demand for manpower and computing resources. Therefore, solving the technical challenges of three-dimensional mesh adaptation and optimizing its efficiency and cost are of great significance to supporting research and applications in the field of CFD. How to reduce the computational cost and time of mesh adaptation while ensuring high-precision three-dimensional flow field calculations has become a topic that needs further research and optimization. Summary of the invention

[0005] The embodiments of the present invention provide a method for improving the efficiency and accuracy of three-dimensional high-precision flow field analysis, which can reduce the computational cost and time of grid adaptation while ensuring high-precision three-dimensional flow field calculation.

[0006] To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

[0007] A method for improving the efficiency and accuracy of three-dimensional high-precision flow field analysis, comprising:

[0008] Step 1, the computing center accepts a business request, the business request includes: initial three-dimensional grid data and flow field condition parameters, wherein the flow field condition parameters at least include: geometric shape parameters and the Mach number Ma and angle of attack AOA of the flow field;

[0009] Step 2: calculating initial three-dimensional flow field data on the initial three-dimensional grid according to the flow field condition parameters, and then using the initial three-dimensional flow field data to generate a three-dimensional adaptive grid whose node density distribution is consistent with the flow field structure;

[0010] Step 3: constructing a sample database according to the flow field condition parameters and the three-dimensional adaptive grid;

[0011] Step 4: Using the sample database to train a deep learning model;

[0012] Step 5: Input the condition parameters of the flow field to be analyzed into the trained deep learning model to obtain a three-dimensional grid that conforms to the flow field structure of the flow field to be analyzed;

[0013] Step 6: Determine the client that sends the service request, and send the three-dimensional grid obtained in step 5 to the client and display it.

[0014] Specifically, in step 2, the initial three-dimensional flow field data is generated according to the flow field condition parameters, including: using the flow field condition parameters to perform CFD simulation calculations on the initial three-dimensional grid, wherein the initial three-dimensional flow field data obtained by the CFD simulation calculation is stored in a three-dimensional flow field library; Specifically, in this embodiment, the specific method of using the flow field condition parameters to perform CFD simulation calculations on the initial three-dimensional grid can adopt the currently developed general grid adaptive method for CFD, such as "A general grid adaptive method for CFD", which can calculate the flow field numerical calculation results that meet the accuracy requirements and return them to the client. It can effectively improve the problem that the node distribution of the initial CFD grid is not reasonable, so that the grid nodes are moved as much as possible to the area where the flow field solution changes violently and meets the high gradient distribution. The use of the initial three-dimensional flow field data to generate a three-dimensional adaptive grid whose node density distribution is consistent with the flow field structure includes: adaptively adjusting the initial three-dimensional grid by a three-dimensional SOM grid adaptive method, and storing the result of the adaptive adjustment as the three-dimensional adaptive grid that is consistent with the initial three-dimensional flow field structure in the three-dimensional grid library.

[0015] The method of adaptively adjusting the initial three-dimensional grid by the three-dimensional SOM grid adaptive method includes: step 101: collecting SOM neural network training data; step 102: constructing the SOM neural network structure according to the SOM neural network training data and the initial three-dimensional grid; step 103: training the constructed SOM neural network, and then updating the weight of the competition layer neurons; step 104: repeating the above steps 101 to 103 until the maximum number of iterations, and obtaining the result of the adaptive adjustment. The final adaptive optimization result is obtained and stored in the three-dimensional grid library.

[0016] Wherein, in step 101, it includes: taking out a single sample pair from the three-dimensional flow field library and the three-dimensional grid library to train a multilayer perceptron, and then using the trained multilayer perceptron to predict the spatial flow field solution in the high-resolution three-dimensional uniform background grid, and the prediction result is used as the SOM neural network training data, wherein the high resolution is a node density that is 5 times or more than 5 times of the initial three-dimensional grid; the three-dimensional uniform background grid is generated in the initial three-dimensional grid space, and the sampling point range [x min :x max ,y min :y max ,z min :z max ], where x min ,x max ,y min ,y max ,z min ,z max are the minimum and maximum values ​​of the initial three-dimensional grid in the x, y and z directions respectively; construct a three-dimensional high-resolution uniform background grid within the sampling point range, use the trained multi-layer perceptron to predict the flow field solutions on all nodes in the three-dimensional uniform background grid, and calculate the node gradients; I i is the gradient of the ith node in the three-dimensional uniform background grid, i and N are both positive integers, i<N, I s is the gradient of all nodes arranged from large to small, For the The gradient of nodes, threshold is the preset threshold.

[0017] In step 102, the construction of the SOM neural network structure includes: projecting the initial three-dimensional grid to the competition layer of the SOM neural network, the grid nodes of the SOM neural network are defined as competition layer neurons, the adjacency relationship between the neurons of the competition layer corresponds to the topological connection relationship between the initial three-dimensional grid nodes, and the topological distance between the neurons of the competition layer is: floor(·) represents the floor rounding function, Sij is the Euclidean distance from the i-th neuron to the j-th neuron, i and j are both positive integers; the input layer of the SOM neural network is defined as the constructed training data set sample library, the number of neurons in the input layer corresponds to the training data dimension, and for the three-dimensional grid adaptive optimization, the number of neurons in the input layer is 3; the neuron weights of the competition layer of the SOM neural network correspond to the three-dimensional grid node coordinates, and the output dimension of the competition layer is 3, where the initial three-dimensional grid node coordinates X 0 Used to initialize the neuron weights ξ of the competition layer 0 The initial three-dimensional flow field data is obtained through CFD calculation on the initial three-dimensional grid.

[0018] Specifically, the step 3 includes: using the initial three-dimensional flow field data and the three-dimensional adaptive grid generated in the step 2 to form an input-output data sample pair, and storing the input-output data sample pair in the sample database.

[0019] Specifically, in step 4, the training deep learning model includes: flow field feature extraction link, flow field grid feature interaction link and three-dimensional grid adaptation link; the flow field feature extraction link includes: twelve layers of grid convolution layer; the flow field grid feature interaction link includes: four layers of perception feature pooling; in the flow field grid feature interaction link, the four layers of features in the flow field feature extraction link are cascaded, and aligned with the input node features of the three-dimensional grid adaptation link through the perception feature pooling layer and then fused. The three-dimensional grid adaptation link includes: fifteen layers of grid convolution layer and eight layers of grid attention layer. In the three-dimensional grid adaptation link, the mapping relationship from flow field conditions to high-quality three-dimensional adaptive grids is learned.

[0020] The implicit output dimensions of the first two layers of grid convolution in the flow field feature extraction link are 64 and 128 respectively, the implicit output dimensions of the middle seven layers of grid convolution are 256, and the implicit output dimensions of the last three layers of grid convolution are 128, 64 and 1 respectively; the implicit output dimensions of the four layers of perception feature pooling layer in the flow field grid feature interaction link are 64, 128, 256 and 64 respectively; the implicit output dimensions of the first thirteen layers of graph convolution in the three-dimensional grid adaptation link are 256, the implicit output dimension of the fourteenth layer of graph convolution is 128, the implicit output dimension of the fifteenth layer of graph convolution is 3, and the implicit output dimensions of the eight layers of grid attention layer are 256.

[0021] The input node features of the three-dimensional grid adaptive link include the mean coordinates of the three-dimensional grid nodes F m and the local node density F d ; In the process of training the deep learning model, the loss function used is: L = L ada +λL rec , Lada represents the grid loss function, L rec Flow field reconstruction loss function, λ is L ada and L rec The balance coefficient.

[0022] The method for improving the efficiency and accuracy of three-dimensional high-precision flow field analysis provided by the embodiment of the present invention is that the client provides initial three-dimensional grid data and flow field condition parameters, and the server performs modeling analysis from flow field conditions to adaptive grids. The initial three-dimensional grid and initial flow field data are first obtained, and the SOM three-dimensional grid adaptive technology is used to generate a high-quality adaptive grid and build a sample database using the data, and the deep learning model is trained. According to the flow field condition parameters to be predicted and the initial three-dimensional grid provided by the client, the model predicts and generates a three-dimensional adaptive grid according to the flow field condition parameters and the initial grid provided by the client, and finally returns a high-quality grid with a node density distribution that conforms to the flow field structure to be predicted to the client and displays it. Thereby, while ensuring high-precision three-dimensional flow field calculation, the calculation cost and time of grid adaptation are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 A schematic diagram of a system architecture provided by an embodiment of the present invention;

[0025] Figure 2 A flow chart of a method provided by an embodiment of the present invention;

[0026] Figure 3 A flowchart of a SOM three-dimensional grid adaptive method provided by an embodiment of the present invention;

[0027] Figure 4 A schematic diagram of a deep learning model provided by an embodiment of the present invention;

[0028] Figure 5 A schematic diagram of a logic flow for three-dimensional adaptive grid prediction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention will be described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention. It can be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "one", "said" and "the" used herein may also include plural forms. It should be further understood that the term "including" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used here may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as generally understood by those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0030] The embodiment of the present invention provides a method for optimizing high-precision flow field analysis, such as Figure 1 As shown, including:

[0031] Step 1: The computing center accepts a business request, wherein the business request includes initial three-dimensional grid data and flow field condition parameters, wherein the flow field condition parameters at least include: geometric shape parameters and the Mach number Ma and angle of attack AOA of the flow field.

[0032] The computing center can receive business requests sent by clients. Servers, databases and other equipment can be deployed in the computing center. The computing center can also be called a data center and can be built using existing technologies. For example, it can be applied in Figure 1In the scenario architecture shown, the client provides initial grid data and flow field condition parameters; the server uses the information given by the client to complete the simulation calculation of the initial flow field data and the adaptive grid generation, thereby building a sample database; the sample database is used to build a deep learning network structure, and then the deep learning model is trained; finally, the server completes the simulation calculation of the flow field data to be predicted based on the initial grid and flow field condition parameters to be predicted provided by the client, and then uses the flow field condition parameters to be predicted, the initial three-dimensional grid and the trained deep learning model to predict high-quality three-dimensional adaptive grid data and return it to the client.

[0033] Specifically, the flow field condition parameters include at least: three-dimensional geometric shape parameters, the Mach number Ma and the angle of attack α of the flow field; the flow field condition parameters also include the Reynolds number Re of the flow field; the initial grid data can be various three-dimensional grid data commonly understood by industry insiders, including structured grids, unstructured grids, tetrahedral grids and hexahedral grids, etc.

[0034] Step 2: generating initial three-dimensional flow field data according to the flow field condition parameters, and then using the initial flow field data to generate a three-dimensional adaptive grid whose node density distribution is consistent with the flow field structure;

[0035] The structure of the deep learning network is constructed according to the sample database to establish an end-to-end mapping from flow field condition parameters to a three-dimensional adaptive grid. The deep learning network structure refers to the specific components of the deep learning model.

[0036] Specifically, a three-dimensional adaptive grid that conforms to the flow field structure can be generated by a SOM three-dimensional grid adaptive method. The SOM three-dimensional grid adaptive method refers to a moving grid method based on a self-organizing competitive mapping neural network, which constructs SOM network training data according to the flow field solution, and updates the competition layer neuron weights according to the training data and the initial three-dimensional grid data to obtain a three-dimensional adaptive grid. By training the neural network, a high-quality three-dimensional adaptive grid with unchanged node number and grid topology can be obtained.

[0037] Step 3: construct a sample database based on the initial flow field data and the produced three-dimensional adaptive grid;

[0038] Step 4: Use the sample database to train a deep learning model.

[0039] The deep learning network structure is constructed according to the sample database to establish an end-to-end mapping from flow field condition parameters to a three-dimensional adaptive grid. The deep learning network structure refers to the specific components of the deep learning model.

[0040] Step 5: Input the parameters of the flow field condition to be predicted into the trained deep learning model, and simultaneously predict the three-dimensional flow field and the three-dimensional adaptive grid to obtain a high-quality three-dimensional grid that conforms to the structure of the flow field to be predicted.

[0041] Wherein, in said step 5, it includes: inputting the flow field condition parameters to be predicted into the trained deep learning model, the deep learning model outputs the predicted three-dimensional adaptive grid result, wherein the server simultaneously inputs the flow field condition parameters and the initial three-dimensional grid data into the deep learning model. Figure 5 The method for three-dimensional adaptive mesh prediction shown in the figure obtains a high-quality three-dimensional mesh that conforms to the flow field structure to be predicted and sends it to the client; wherein, the three-dimensional adaptive mesh prediction is performed based on the flow field condition parameters to be predicted, the initial three-dimensional mesh and the trained deep learning model, and then the high-quality three-dimensional adaptive mesh predicted by the deep learning model is returned to the client.

[0042] Step 6: Determine the client that sends the service request, and send the three-dimensional grid obtained in step 5 to the client and display it.

[0043] In this embodiment, in step 2, it includes: according to the flow field condition parameters and the initial three-dimensional grid data, using the flow field conditions to perform CFD simulation calculations on the initial three-dimensional grid, and storing the results in the three-dimensional flow field library; according to the results of the CFD simulation calculations, the initial three-dimensional grid is adaptively optimized by the SOM three-dimensional grid adaptive method and the optimization results are stored in the three-dimensional grid library.

[0044] In step 3, it includes: using the three-dimensional flow field data in the three-dimensional flow field library and the optimization results in the three-dimensional grid library to form input-output data sample pairs, and storing the input-output data sample pairs in the sample database.

[0045] The sample database also includes: the initial three-dimensional grid input by the client can be any specified type of three-dimensional grid data, including structured grid, unstructured grid, tetrahedral grid and hexahedral grid. The server performs CFD simulation calculation on the initial three-dimensional grid according to the flow field condition parameters input by the client to obtain the corresponding initial flow field data. Then, the initial flow field data and the initial three-dimensional grid input by the client are combined into the input training sample of the deep learning model. Then, using Figure 3 The SOM 3D mesh adaptation process shown in the figure, according to the initial flow field data, the server uses the SOM 3D mesh adaptation technology to perform mesh adaptive optimization on the initial 3D mesh, and the generated high-quality 3D mesh will be used as a label for supervised deep learning training.

[0046] In this embodiment, the adaptive optimization of the initial three-dimensional grid by the SOM three-dimensional grid adaptive method includes:

[0047] Step 101: Collect SOM neural network training data;

[0048] Specifically, step 101 includes:

[0049] A single sample pair is taken from the three-dimensional flow field library and the three-dimensional grid library to train a multi-layer perceptron, and then the trained multi-layer perceptron is used to predict the spatial flow field solution in a high-resolution three-dimensional uniform background grid.

[0050] The multi-layer perceptron uses training data consisting of single three-dimensional flow field data and three-dimensional grid node coordinates to train an artificial neural network consisting of multiple hidden layers, thereby fitting the complex flow field rules in the three-dimensional grid space.

[0051] The three-dimensional uniform background grid is generated in the initial three-dimensional grid space, and the sampling point range [x min :x max ,y min :y max ,z min :z max ], where x min ,x max ,y min ,y max ,z min ,z max are the minimum and maximum values ​​of the initial three-dimensional grid in the x, y and z directions respectively. A three-dimensional high-resolution uniform background grid is constructed within this range, and the trained multi-layer perceptron is used to predict the flow field solution at all nodes in the high-resolution background grid and calculate the node gradient. Set the threshold The training data obtained by sampling and distribution conforming to the high gradient characteristics of the flow field is input into the sample library: Among them I i is the gradient of the i-th node in the background grid, I s is the gradient of all nodes arranged from large to small, For the I s I s can be understood as a set that includes the gradients of all nodes, and For the The gradient of the nodes is For set I s The The gradient of the node

[0052] Step 102: constructing a SOM neural network structure according to the training data and the initial three-dimensional grid;

[0053] Specifically in step 102, the SOM neural network structure construction includes:

[0054] The initial three-dimensional grid is projected onto the competitive layer of the SOM neural network. The grid nodes are defined as competitive layer neurons. The adjacency relationship between neurons is defined by the topological connection relationship between the initial three-dimensional grid nodes. The topological distance between neurons is defined as follows:

[0055]

[0056] Where floor(·) represents the floor function, S ij is the Euclidean distance from the i-th neuron to the j-th neuron;

[0057] The SOM neural network input layer is defined as the constructed training data set sample library, and the number of input layer neurons is determined by the dimension of the training data. For three-dimensional grid adaptive optimization, the number of input layer neurons is 3; wherein, the sample content in the "training data sample library" can be the same as the sample content in the "sample database" in step 3, so the SOM neural network input layer is defined as the constructed training data set sample library, which can be understood as follows: the SOM neural network input layer is defined as the initial three-dimensional grid and the initial three-dimensional flow field data.

[0058] The neuron weights of the competition layer of the SOM neural network correspond to the three-dimensional grid node coordinates, and the initial three-dimensional grid node coordinates X 0 Used to initialize the neuron weights ξ of the competition layer 0 , the output dimension of the competition layer is 3.

[0059] Step 103: Train the SOM neural network and update the weights of the competition layer neurons;

[0060] Specifically in step 102, the training of the SOM neural network includes:

[0061] A sample is randomly collected from the training data sample library as the input of SOM, and the spatial position x of the input point is used t Select the winning neuron from the current competitive layer Where t is the current iteration step, and the winning neighborhood is calculated based on the topological distance between the winning neuron and all neurons in the current competitive layer. Then calculate the learning rate δ of the current competitive layer neuron t , and then according to the center position of all cells in the current three-dimensional grid Add a grid smoothness constraint to the neuron weight update, and finally update the neuron weights of the current competition layer:

[0062]

[0063] Where λ is the smoothing rate, and the coordinates of the three-dimensional grid nodes in the next iteration step are obtained: t+1 =ξ t+1 ,ξ t Indicates the three-dimensional grid node coordinates of the current iteration step, x t Represents the input point coordinates of the current iteration step.

[0064] Step 104: Repeat the above steps until the maximum number of iterations is reached, and obtain the final optimization result of the adaptive optimization and store it in the three-dimensional mesh library.

[0065] In this embodiment, the training deep learning model includes: a flow field feature extraction link, a flow field grid feature interaction link and a three-dimensional grid adaptation link; the flow field feature extraction link is used to extract the global structural features of the three-dimensional flow field; the flow field feature extraction link includes: twelve layers of grid convolution layers, the flow field grid feature interaction link includes four layers of perception feature pooling layers, and the three-dimensional grid adaptation link includes fifteen layers of grid convolution; the flow field grid feature interaction link includes: fusing the extracted flow field features with the grid node features; the three-dimensional grid adaptation link includes: learning the mapping relationship from flow field condition parameters to high-quality three-dimensional adaptive grids. In this embodiment, the flow field feature extraction link adopts the Unet architecture, and the flow field features output by the first, second, ninth and eleventh layers of grid convolution in the encoder are sequentially cascaded with the flow field features output by each layer of convolution of the generator through skip connections. Therefore, the flow field feature extraction module can fully integrate shallow features containing more spatial structure information and deep features containing more semantic features, improve the accuracy of flow field reconstruction, and provide global structural features that retain more fluid physical structure information for subsequent three-dimensional grid adaptation.

[0066] The implicit output dimensions of the four-layer perception feature pooling layer in the flow field grid feature interaction link are 64, 128, 256 and 64 respectively; the implicit output dimensions of the first thirteen layers of graph convolution in the three-dimensional grid adaptation link are all 256, the implicit output dimension of the fourteenth layer of graph convolution is 128, the implicit output dimension of the fifteenth layer of graph convolution is 3, and the implicit output dimensions of the eight-layer grid attention layer are all 256.

[0067] The grid convolution layer weight function update function is:

[0068]

[0069] in, is the activation function, K is the number of neurons in the lth layer, N i is the number of first-order neighbors of the ith node, b[n] l is the bias of the nth neuron in the lth layer.l,n,k is the weight connecting the kth neuron in the lth layer and the nth neuron in the l+1th layer, is the weight of the kth neuron in the lth layer connecting the jth neighboring point of the ith node and the nth neuron in the l+1th layer of the ith node, where l is a positive integer. represents the features of the i-th node in the l-th layer and its j-th neighboring node on the k-th neuron, Represents the feature of the i-th node in the l-th layer on the k-th neuron.

[0070] Specific as Figure 4 As shown, the flow field feature extraction link inputs flow field condition parameters, outputs flow field features and three-dimensional reconstructed flow field; the flow field grid feature interaction link inputs flow field features and initial grid node features, and outputs fused grid node features; the three-dimensional grid adaptive link inputs grid node features, and outputs adaptively optimized grid node coordinates. Among them, the flow field feature extraction link can be divided into an encoder and a generator. The encoder takes the initial flow field condition parameters as input and the flow field features as output. The generator takes the flow field features as input and the reconstructed three-dimensional flow field as output. In this embodiment, the initial flow field and the reconstructed flow field are distinguished. When the two are more similar, the flow field features extracted by the encoder retain more global structural information of the flow field. The flow field grid feature interaction link inputs flow field features and initial grid node features, and uses a fully connected network to perform feature alignment and fusion through the perception feature pooling layer. The three-dimensional grid adaptive link takes the fused grid node features and the initial three-dimensional grid as input, and outputs the predicted high-quality three-dimensional adaptive grid through the aggregation and dimensionality reduction learning of fifteen layers of grid convolution. Among them, the deep learning model is implemented by computer language, and after the training is completed, the results are saved in the server for subsequent use.

[0071] In this embodiment, during the training of the deep learning model, the initial grid node features of the input flow field grid feature interaction link include:

[0072] The mean coordinates of the three-dimensional grid nodes F m , where the mean coordinate of the i-th node is mvc i for:

[0073]

[0074] Among them, w k represents the weight between the i-th node and the k-th boundary point, n is the number of three-dimensional grid boundary points, ξ i is the spatial coordinate of the ith node, ξ k is the spatial coordinate of the kth boundary point, ||ξ k -ξ i || is the Euclidean distance from the i-th node to the k-th boundary point, αk is the angle ∠ξ k ξ i ξ k+1 The value of α k-1 is the angle ∠ξ k-1 ξ i ξ k The three-dimensional grid node mean coordinates provide the deep learning model with generalization to different geometric shapes and far-field boundaries.

[0075] The local mesh density is F d , specifically the local density F of the i-th node i d for:

[0076]

[0077] Among them, N a is the number of adjacent cells of the ith node, |E j | is the volume of the jth adjacent cell of the ith node. The local grid density enhances the model's ability to distinguish different regions in the physical domain.

[0078] In this embodiment, during the training of the deep learning model, the loss function used is:

[0079] L=L ada +λL rec , where L ada represents the grid loss function, L rec The flow field reconstruction loss, λ, is used to balance the two loss functions. The grid loss function is used to supervise the training process from the initial three-dimensional grid to the three-dimensional adaptive grid; the flow field reconstruction loss is used to supervise the training process from the flow field condition parameters to the reconstructed three-dimensional flow field.

[0080] Specifically, the grid loss function is:

[0081]

[0082] Where N is the number of initial 3D grid nodes and ξ is the predicted 3D grid of the deep learning model. The node coordinates of is the label grid optimized by the SOM 3D grid adaptive method The node coordinates of .

[0083] The flow field reconstruction loss is:

[0084]

[0085] Among them, F i pis the predicted flow field data of the deep learning model at the i-th node, is the initial flow field data at the i-th node, that is, the CFD calculation simulation value on the initial three-dimensional grid.

[0086] In actual applications, this embodiment uses the data provided by the client to perform high-quality three-dimensional adaptive mesh prediction on the server side. First, the initial three-dimensional mesh and initial flow field data of the simulation calculation are obtained, and a training sample database suitable for deep learning is constructed. A deep learning network structure suitable for three-dimensional adaptive mesh prediction is constructed based on the training sample database, and the deep learning model is trained. The corresponding three-dimensional adaptive mesh is generated according to the flow field condition parameters to be predicted and the initial three-dimensional mesh provided by the client user. Finally, a high-quality three-dimensional mesh with a node density distribution consistent with the flow field structure is sent to the client for use.

[0087] The advantages of this implementation are: it successfully realizes the direct generation of high-quality three-dimensional adaptive grids from flow field condition parameters, while avoiding the high computational cost and time consumption of traditional methods when dealing with three-dimensional problems. The deep learning model involved in the present invention efficiently processes three-dimensional grid data and flow field data, is easy to implement and fully utilizes the complex structural information of the flow field. Various problems that may arise, such as grid entanglement and uneven density distribution, are fully avoided during the model training process. Once the model is trained, it can efficiently and accurately capture the complex flow characteristics of high-dimensional flow field data and generate a high-quality three-dimensional adaptive grid corresponding to the initial flow field condition parameters.

[0088] The method for improving the efficiency and accuracy of three-dimensional high-precision flow field analysis provided by the embodiment of the present invention is that the client provides initial three-dimensional grid data and flow field condition parameters, and the server performs modeling analysis from flow field conditions to adaptive grids. The initial three-dimensional grid and initial flow field data are first obtained, and the SOM three-dimensional grid adaptive technology is used to generate high-quality adaptive grids and build a sample database using the data, and the deep learning model is trained. According to the flow field condition parameters to be predicted and the initial three-dimensional grid provided by the client, the model predicts and generates a three-dimensional adaptive grid according to the flow field condition parameters and the initial grid provided by the client, and finally returns the high-quality grid whose node density distribution conforms to the flow field structure to be predicted to the client and displays it. The sub-links of the deep learning model involved in this scheme are easy to implement, and the collaborative training of these sub-links effectively utilizes the flow field data information of the sample library, thereby not only providing a reliable optimization direction for training, but also improving the accuracy of three-dimensional grid prediction, and also realizing the rapid prediction of flow field condition parameters to three-dimensional adaptive grids, effectively improving the accuracy of CFD calculations and the speed of three-dimensional adaptive grid optimization. Thereby, while ensuring high-precision three-dimensional flow field calculations, the computational cost and time of grid adaptation are reduced.

[0089] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The above is only a specific implementation method of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for improving the efficiency and accuracy of three-dimensional high-precision flow field analysis, characterized in that: include: Step 1, the computing center accepts a business request, the business request includes: initial three-dimensional grid data and flow field condition parameters, wherein the flow field condition parameters at least include: geometric shape parameters and the Mach number Ma and angle of attack AOA of the flow field; Step 2: calculating initial three-dimensional flow field data on the initial three-dimensional grid according to the flow field condition parameters, and then using the initial three-dimensional flow field data to generate a three-dimensional adaptive grid whose node density distribution is consistent with the flow field structure; Step 3: constructing a sample database according to the flow field condition parameters and the three-dimensional adaptive grid; Step 4: Using the sample database to train a deep learning model; Step 5: Input the condition parameters of the flow field to be analyzed into the trained deep learning model to obtain a three-dimensional grid that conforms to the flow field structure of the flow field to be analyzed; Step 6: Determine the client that sends the service request, and send the three-dimensional grid obtained in step 5 to the client.

2. The method according to claim 1, characterized in that In step 2, generating initial three-dimensional flow field data according to the flow field condition parameters includes: Performing CFD simulation calculation on the initial three-dimensional grid using the flow field condition parameters, wherein the initial three-dimensional flow field data obtained by the CFD simulation calculation is stored in a three-dimensional flow field library; The initial three-dimensional flow field data is used to generate a three-dimensional adaptive grid whose node density distribution is consistent with the flow field structure, including: adaptively adjusting the initial three-dimensional grid through a three-dimensional SOM grid adaptive method, and storing the result of the adaptive adjustment as the three-dimensional adaptive grid that is consistent with the initial three-dimensional flow field structure in a three-dimensional grid library.

3. The method according to claim 2, characterized in that The method of adaptively adjusting the initial three-dimensional grid by using the three-dimensional SOM grid adaptive method includes: Step 101: Collect SOM neural network training data; Step 102: constructing a SOM neural network structure according to the SOM neural network training data and the initial three-dimensional grid; Step 103: training the constructed SOM neural network, and then updating the weights of the competition layer neurons; Step 104: Repeat the above steps 101 to 103 until the maximum number of iterations is reached to obtain the result of the adaptive adjustment.

4. The method according to claim 3, characterized in that In step 101, it includes: A single sample pair is taken out from the three-dimensional flow field library and the three-dimensional grid library to train a multilayer perceptron, and then the trained multilayer perceptron is used to predict the spatial flow field solution in a high-resolution three-dimensional uniform background grid, and the prediction result is used as the SOM neural network training data, wherein the high resolution is a node density that is 5 times or more than 5 times that of the initial three-dimensional grid.

5. The method according to claim 4, characterized in that The three-dimensional uniform background grid is generated in the initial three-dimensional grid space, and the sampling point range [x min :x max ,y min :y max ,z min :z max ], where x min ,x max ,y min ,y max ,z min ,z max are the minimum and maximum values ​​of the initial three-dimensional grid in the x, y and z directions respectively; Constructing a three-dimensional high-resolution uniform background grid within the sampling point range, using a trained multi-layer perceptron to predict the flow field solutions at all nodes in the three-dimensional uniform background grid, and calculating the node gradients; in, I i is the gradient of the i-th node in the three-dimensional uniform background grid, i and N are both positive integers and i<N, I s is the gradient of all nodes arranged from large to small, For the The gradient of nodes, threshold is the preset threshold.

6. The method according to claim 4, characterized in that In step 102, constructing a SOM neural network structure includes: The initial three-dimensional grid is projected onto the competition layer of the SOM neural network. The adjacency relationship between the neurons in the competition layer corresponds to the topological connection relationship between the nodes in the initial three-dimensional grid. The topological distance between the neurons in the competition layer is: floor(·) represents the floor rounding function, S ij is the Euclidean distance from the i-th neuron to the j-th neuron, i and j are both positive integers; The input layer of the SOM neural network is defined as the constructed training data set sample library, and the number of neurons in the input layer is 3; The neuron weights of the competition layer of the SOM neural network correspond to the three-dimensional grid node coordinates, and the output dimension of the competition layer is 3.

7. The method according to claim 1, characterized in that The step 3 comprises: The initial three-dimensional flow field data and the three-dimensional adaptive grid generated in step 2 are used to form input-output data sample pairs, and the input-output data sample pairs are stored in the sample database.

8. The method according to claim 1, characterized in that In step 4, the training deep learning model includes: a flow field feature extraction link, a flow field grid feature interaction link and a three-dimensional grid adaptation link; The flow field feature extraction link includes: twelve layers of grid convolution layers; The flow field grid feature interaction link includes: four-layer perception feature pooling; The three-dimensional grid adaptation link includes: fifteen layers of grid convolution layers and eight layers of grid attention layers.

9. The method according to claim 1 or 7, characterized in that: The implicit output dimensions of the first two layers of grid convolution in the flow field feature extraction link are 64 and 128 respectively, the implicit output dimensions of the middle seven layers of grid convolution are all 256, and the implicit output dimensions of the last three layers of grid convolution are 128, 64 and 1 respectively; The implicit output dimensions of the four-layer perception feature pooling layer in the flow field grid feature interaction link are 64, 128, 256 and 64 respectively; The implicit output dimensions of the first thirteen layers of graph convolution in the three-dimensional grid adaptation link are all 256, the implicit output dimension of the fourteenth layer of graph convolution is 128, the implicit output dimension of the fifteenth layer of graph convolution is 3, and the implicit output dimensions of the eight-layer grid attention layer are all 256.

10. The method according to claim 1, characterized in that The input node features of the three-dimensional grid adaptive link include the mean coordinates of the three-dimensional grid nodes F m and the local node density F d ; In the process of training deep learning models, the loss function used is: L = L ada +λL rec , L ada represents the grid loss function, L rec Flow field reconstruction loss function, λ is L ada and L rec The balance coefficient.

Citation Information

Patent Citations

  • General grid adaptive method for CFD

    CN113283186A

  • Analysis method for predicting high-precision airfoil flow field

    CN115438575A

  • Method for optimizing high-precision flow field analysis

    CN117010292A

  • Aircraft grid optimization method and device, computer equipment and storage medium

    CN117933146A

Cited By

  • Three-dimensional flow field simulation method and system based on clean air conditioner

    CN121189236A

  • A three-dimensional flow field simulation method and system based on clean air conditioning

    CN121189236B

  • Unstructured grid full-automatic generation method

    CN122065446A

  • A fully automatic method for generating unstructured grids

    CN122065446B