Implicit neural representation flow field reconstruction sampling method for aircraft simulation

Abstract

The application discloses an implicit neural representation flow field reconstruction sampling method for aircraft simulation, and relates to the technical field of fluid mechanics, and comprises the following steps: obtaining original flow field data of a target component in an aircraft simulation process, so as to extract physical descriptors representing flow characteristics from the original flow field data; performing feature space division on the physical descriptors, so as to generate a plurality of data clusters; calculating KL divergence values between probability distributions of the physical descriptors in two data clusters; constructing a directed graph structure representing the information transmission strength between the clusters based on the data clusters and all the KL divergence values; calculating the in-degree strength of each node in the directed graph structure, so as to generate sampling probability weights corresponding to the data clusters according to the in-degree strength; and extracting a sampling point set from each data cluster according to the sampling probability weights. The physical region in a core hub position in the entire flow field information network can be automatically identified, that is, the cluster corresponding to the node with high in-degree strength, and the preferential sampling of data points is realized.

Description

Technical Field

[0001] This invention relates to the field of fluid mechanics, and in particular to an implicit neural representation flow field reconstruction sampling method for aircraft simulation. Background Technology

[0002] Aircraft aerodynamic design and performance optimization heavily rely on the acquisition and analysis of high-fidelity turbulent flow field data. Flow field data obtained through computational fluid dynamics (CFD) simulations can reveal complex aerodynamic characteristics of aircraft, such as key flow phenomena like wing stall separation, shock wave boundary layer disturbances, and vortex evolution. However, traditional high-precision CFD simulations are difficult to use frequently in aircraft design iterations due to their enormous computational and storage costs, creating a prominent contradiction between simulation efficiency and accuracy.

[0003] In recent years, flow field reconstruction methods based on implicit neural representation (INR) have established a continuous mapping from spatial coordinates to field variables through neural networks, enabling the reconstruction of high-resolution flow fields from finite sampling points and reducing the dependence on the amount of original high-fidelity data. However, the reconstruction accuracy of INR is heavily dependent on the distribution strategy of sampling points during training. Aircraft flow fields have strong physical non-uniformity and multi-scale characteristics. Key aerodynamic phenomena, such as wingtip vortices, shock surfaces, and separation bubbles, are often limited to local areas with extremely large physical gradients. Existing INR methods mostly adopt uniform or random sampling strategies, which fail to fully consider this characteristic, leading to two problems: First, insufficient sampling in key areas: uniform sampling distributes too many sampling points in areas with gentle flow, while insufficient sampling in key flow areas with dense information and drastic gradient changes makes it difficult for INR models to accurately learn and reconstruct these fine flow structures that have a decisive impact on aircraft performance. Second, it is difficult to balance sampling efficiency and global coverage: Although some studies have attempted to adjust sampling weights based on local gradients or simple physical quantities, these methods lack quantitative analysis of the physical correlation and information transmission between flow regions. They are difficult to adaptively balance the need for global flow field coverage and fine analysis of local key regions under a limited sampling point budget, resulting in low sampling efficiency and high model training costs.

[0004] Therefore, there is an urgent need in this field for a method that can perform adaptive priority sampling during aircraft simulation to meet the needs of aircraft design for efficient and high-precision aerodynamic analysis. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide an implicit neural representation flow field reconstruction sampling method for aircraft simulation, which can overcome the blindness of random or uniform sampling strategies and efficiently capture the flow structure critical to reconstruction accuracy under a limited sampling point budget. The specific solution is as follows:

[0006] In a first aspect, this application discloses an implicit neural representation flow field reconstruction sampling method for aircraft simulation, including:

[0007] The raw flow field data of the target component in the aircraft simulation process is acquired in order to extract physical descriptors characterizing the flow features from the raw flow field data;

[0008] The physical descriptor is partitioned into a feature space to generate several data clusters; wherein each data cluster corresponds to a local region with similar physical states in the original flow field data;

[0009] Calculate the KL divergence between the probability distributions of the physical descriptors in the two data clusters;

[0010] A directed graph structure representing the intensity of information transmission between clusters is constructed based on each data cluster and all the KL divergence values; wherein, the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters.

[0011] Calculate the in-degree intensity of each node in the directed graph structure, so as to generate the sampling probability weights corresponding to each data cluster based on the in-degree intensity;

[0012] Sampling point sets are extracted from each data cluster according to the sampling probability weights; the sampling point sets are used to train implicit neural representation models to reconstruct the turbulent flow field of target components in the aircraft simulation process.

[0013] Optionally, the physical descriptor includes any one or more of the following: vorticity, velocity gradient tensor, strain rate tensor, pressure gradient, turbulent kinetic energy, vorticity modulus, and Reynolds stress tensor.

[0014] Optionally, the step of partitioning the physical descriptor into a feature space to generate several data clusters includes:

[0015] Using a preset clustering algorithm, and based on the similarity metric between the physical descriptors, each physical descriptor is divided into several data clusters;

[0016] Based on the spatial position of the physical descriptor in the original flow field data, the mapping relationship between the data cluster where the physical descriptor is located and the corresponding spatial position is recorded.

[0017] Optionally, calculating the KL divergence value between the probability distributions of the physical descriptors in the two data clusters includes:

[0018] The probability distribution of the physical descriptors in each data cluster is statistically analyzed to determine the probability density function of each data cluster based on the probability distribution;

[0019] The relative entropy between the probability density functions corresponding to two data clusters is calculated based on the KL divergence equation to obtain the KL divergence value.

[0020] Optionally, the construction of a directed graph structure representing the intensity of inter-cluster information transfer based on each data cluster and all the KL divergence values ​​includes:

[0021] If the KL divergence value is greater than the preset divergence threshold, the corresponding data cluster pair is marked as a valid data cluster pair, and the valid data cluster pair and the corresponding KL divergence value are recorded in the divergence offset array.

[0022] Index management is performed based on the inter-cluster connection relationships and divergence weights in the divergence offset array, inter-cluster divergence index array, and divergence weight index array to construct a directed graph structure containing node, edge, and weight information; wherein, the inter-cluster divergence index array records the connection relationships between valid data cluster pairs, and the divergence weight index array records the divergence weights.

[0023] Optionally, calculating the in-degree intensity of each node in the directed graph structure, so as to generate sampling probability weights corresponding to each data cluster based on the in-degree intensity, includes:

[0024] The KL divergence weights of all in-degree edges of each node in the directed graph structure are accumulated to obtain the in-degree strength of each node in the network. The in-degree strength is then normalized according to the maximum entropy principle to generate sampling probability weights.

[0025] Optionally, the step of extracting a set of sampling points from each data cluster according to each sampling probability weight includes:

[0026] Sampling points are extracted from each data cluster according to the sampling probability weights to generate a sampling point set based on each sampling point.

[0027] Secondly, this application discloses an implicit neural representation flow field reconstruction sampling device for aircraft simulation, comprising:

[0028] The data extraction module is used to acquire the raw flow field data of the target component in the aircraft simulation process, so as to extract physical descriptors characterizing the flow features from the raw flow field data;

[0029] The data cluster partitioning module is used to partition the physical descriptor into a feature space to generate several data clusters; wherein each data cluster corresponds to a local region with similar physical states in the original flow field data;

[0030] The divergence calculation module is used to calculate the KL divergence value between the probability distributions of the physical descriptors in two data clusters;

[0031] The graph construction module is used to construct a directed graph structure representing the intensity of information transmission between clusters based on each data cluster and all the KL divergence values; wherein, the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters.

[0032] The sampling weight determination module is used to calculate the in-degree intensity of each node in the directed graph structure, so as to generate the sampling probability weight corresponding to each data cluster based on the in-degree intensity.

[0033] The sampling module is used to extract a set of sampling points from each data cluster according to the sampling probability weights; the set of sampling points is used to train an implicit neural representation model to reconstruct the turbulent flow field of the target component in the aircraft simulation process.

[0034] Thirdly, this application discloses an electronic device, comprising:

[0035] Memory, used to store computer programs;

[0036] A processor is configured to execute the computer program to implement the steps of the aforementioned disclosed implicit neural representation flow field reconstruction sampling method for aircraft simulation.

[0037] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed implicit neural representation flow field reconstruction sampling method for aircraft simulation.

[0038] As can be seen, this application discloses a method for acquiring raw flow field data of a target component in an aircraft simulation process, extracting physical descriptors characterizing flow features from the raw flow field data, partitioning the physical descriptors into a feature space to generate several data clusters, wherein each data cluster corresponds to a local region with similar physical states in the raw flow field data, calculating the KL divergence value between the probability distributions of the physical descriptors in two data clusters, constructing a directed graph structure characterizing the information transmission intensity between clusters based on each data cluster and all the KL divergence values, wherein the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters, calculating the in-degree intensity of each node in the directed graph structure to generate sampling probability weights corresponding to each data cluster based on the in-degree intensity, extracting a set of sampling points from each data cluster based on each sampling probability weight, and using the set of sampling points to train an implicit neural representation model for reconstructing the turbulent flow field of the target component in the aircraft simulation process. Therefore, by extracting physical descriptors representing flow characteristics and partitioning the feature space, a neural network implicitly maps any continuous spatiotemporal coordinate (query point) in the physical space to the flow field attribute value (velocity, pressure, etc.) at that point. This achieves a continuous and analytical expression of the flow field distribution, shifting sampling from the geometric space to the physical feature space. This directly correlates the distribution of sampling points with the physical essence of the flow field (vortex structure, shear region), ensuring the physical meaning and relevance of the sampling strategy. Furthermore, calculating the KL divergence and constructing a directed graph representing the intensity of information transmission between clusters places data clusters in a global information interaction network. By calculating the in-degree strength of nodes to generate sampling probability weights, the system can automatically identify the physical regions that occupy a core position in the entire flow field information network, i.e., the clusters corresponding to nodes with high in-degree strength. This ensures that sampling resources are preferentially allocated to those regions with the richest information content and the greatest impact on the global flow. Finally, with the same total number of sampling points, the accuracy of the reconstruction model is significantly improved compared to traditional methods; and when achieving the same reconstruction accuracy, the amount of training data and training cost are significantly reduced. This provides an effective technical solution to the problem of high computational costs in high-fidelity simulation of aircraft. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Figure 1This application discloses a flowchart of a turbulent flow field reconstruction method for aerodynamic design in aircraft simulation.

[0041] Figure 2 This application discloses a specific method for reconstructing turbulent flow fields in aerodynamic design during aircraft simulation.

[0042] Figure 3 This is a schematic diagram of a turbulent flow field reconstruction device for aerodynamic design in aircraft simulation process disclosed in this application.

[0043] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0044] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0045] Aircraft aerodynamic design and performance optimization heavily rely on the acquisition and analysis of high-fidelity turbulent flow field data. Flow field data obtained through computational fluid dynamics (CFD) simulations can reveal complex aerodynamic characteristics of aircraft, such as key flow phenomena like wing stall separation, shock wave boundary layer disturbances, and vortex evolution. However, traditional high-precision CFD simulations are difficult to use frequently in aircraft design iterations due to their enormous computational and storage costs, creating a prominent contradiction between simulation efficiency and accuracy.

[0046] In recent years, flow field reconstruction methods based on implicit neural representation (INR) have established a continuous mapping from spatial coordinates to field variables through neural networks, enabling the reconstruction of high-resolution flow fields from finite sampling points and reducing the dependence on the amount of original high-fidelity data. However, the reconstruction accuracy of INR is heavily dependent on the distribution strategy of sampling points during training. Aircraft flow fields have strong physical non-uniformity and multi-scale characteristics. Key aerodynamic phenomena, such as wingtip vortices, shock surfaces, and separation bubbles, are often limited to local areas with extremely large physical gradients. Existing INR methods mostly adopt uniform or random sampling strategies, which fail to fully consider this characteristic, leading to two problems: First, insufficient sampling in key areas: uniform sampling distributes too many sampling points in areas with gentle flow, while insufficient sampling in key flow areas with dense information and drastic gradient changes makes it difficult for INR models to accurately learn and reconstruct these fine flow structures that have a decisive impact on aircraft performance. Second, it is difficult to balance sampling efficiency and global coverage: Although some studies have attempted to adjust sampling weights based on local gradients or simple physical quantities, these methods lack quantitative analysis of the physical correlation and information transmission between flow regions. They are difficult to adaptively balance the need for global flow field coverage and fine analysis of local key regions under a limited sampling point budget, resulting in low sampling efficiency and high model training costs.

[0047] To this end, the present invention provides an implicit neural representation flow field reconstruction sampling scheme for aircraft simulation, which can overcome the blindness of random or uniform sampling strategies and efficiently capture the flow structure that is critical to the reconstruction accuracy under a limited sampling point budget.

[0048] Reference Figure 1 As shown, this invention discloses an implicit neural representation flow field reconstruction sampling method for aircraft simulation, comprising:

[0049] Step S11: Obtain the raw flow field data of the target component in the aircraft simulation process, so as to extract physical descriptors characterizing the flow features from the raw flow field data.

[0050] In this embodiment, the raw flow field data of the target component in the aircraft simulation process is first acquired. The target component is a core aerodynamic component or region in the aircraft aerodynamic simulation whose flow field contains complex turbulent structures and has a critical impact on the overall aerodynamic performance of the aircraft. Specifically, it includes, but is not limited to, main lifting surfaces and control surfaces, fuselage and blended wing-body structures, etc. Main lifting surfaces and control surfaces include wings and tail fins; fuselage and blended wing-body structures include the fuselage, cockpit, and blended wing-body structures. Then, the raw flow field data of the target component is acquired to extract physical descriptors. The raw flow field data refers to discrete or continuous datasets that characterize the flow details of the target component under operating conditions, obtained through high-fidelity numerical simulations, such as Large Eddy Simulation (LES), Direct Numerical Simulation (DNS), or high-resolution experimental measurements. Typical data variables include, but are not limited to, three-dimensional velocity fields, pressure fields, temperature fields, and density fields. One or more state variable fields are selected from the original flow field data as the basis for analysis. In this embodiment, the velocity vector field and scalar pressure field are selected. For each local location (grid point) in the flow field, spatial differentiation operations are performed on the state variables in its neighborhood to calculate physical quantities that can characterize the local flow structure and dynamic characteristics. These operations include gradient calculation, vorticity calculation, and divergence calculation, etc., and higher-order derivative operations are performed based on the above operations to obtain physical descriptors. The physical descriptors transform the original state variable field into one or more physical descriptor tensor fields. These descriptors constitute a high-dimensional feature space, in which the coordinates of each point are defined by a set of physical quantity values. The physical descriptors include any one or more of the following: vorticity, velocity gradient tensor, strain rate tensor, pressure gradient, turbulent kinetic energy, vorticity modulus, and Reynolds stress tensor.

[0051] Specifically, after preprocessing the data and extracting high-dimensional physical descriptors that can characterize local and global flow behavior, including but not limited to vorticity, velocity gradient tensor, strain rate tensor, pressure gradient, turbulent kinetic energy, vorticity modulus, Reynolds stress tensor, and other second- or higher-order physical quantities that have a significant impact on the flow state, for any selected local flow field sub-region R[i,j], the descriptor tensor field is calculated by performing local spatial partial derivative, gradient, divergence, or curl operations on the main flow state variables (velocity vector, pressure field, or temperature field).

[0052] Step S12: Divide the physical descriptor into a feature space to generate several data clusters; wherein each data cluster corresponds to a local region with similar physical states in the original flow field data.

[0053] In this embodiment, a preset clustering algorithm is used to divide each physical descriptor into several data clusters based on the similarity metric between the physical descriptors. Based on the spatial location of the physical descriptors in the original flow field data, the mapping relationship between the data cluster containing the physical descriptor and its corresponding spatial location is recorded. It can be understood that the clustering algorithm is used to partition the extracted weighted physical descriptors into feature spaces, generating several physically meaningful data clusters that characterize the similarity and structural consistency of local physical states within the flow field. The clustering algorithm can be K-means, density-based clustering (DBSCAN), spectral clustering, or hierarchical clustering suitable for high-dimensional flow field data, etc., and the similarity between physical descriptors can be determined based on the cosine similarity, Euclidean distance, or Mahalanobis distance of the feature vectors. If a physical descriptor has a similarity higher than a preset threshold with its neighboring physical descriptors, it is classified into the same cluster. The cluster offset array (cluster_ptr), the local descriptor index array (local_desc_pos), and the intra-cluster descriptor index array (cluster_desc_pos) form the index basis for subsequent calculation of inter-cluster relative entropy and derivation of information relationships.

[0054] Step S13: Calculate the KL divergence between the probability distributions of the physical descriptors in the two data clusters.

[0055] In this embodiment, the probability distribution of the physical descriptors in each data cluster is statistically analyzed to determine the probability density function of each data cluster. The relative entropy between the probability density functions of two data clusters is calculated based on the KL divergence equation to obtain the KL divergence value. It can be understood that for any two generated data clusters C[i] and C[j], the probability distribution characteristics of their intra-cluster descriptors are statistically analyzed, and their respective probability density functions are determined. The relative entropy between the probability distributions of the two clusters is then calculated based on the Kullback-Leibler divergence formula and the two probability density functions to measure the inter-cluster information transmission strength and physical similarity.

[0056] Specifically, firstly, taking data cluster C[i] as an example, the distribution of all members within the data cluster (i.e., the physical descriptor vectors belonging to this cluster) across various feature dimensions is statistically analyzed to obtain the probability distribution characteristics of the descriptors within the cluster. Then, based on the statistical results of the probability distribution characteristics, a specific probability density function is derived for the data cluster to approximate its intrinsic distribution. This embodiment assumes that the descriptors within the data cluster follow a multivariate Gaussian distribution to determine the probability density function of the data cluster. Then, based on the process steps of determining the probability density function of data cluster C[i], the probability density function of data cluster C[j] is also determined. Finally, the relative entropy between the probability density functions of the two data clusters is calculated using the Kullback-Leibler divergence formula to obtain the KL divergence value, which characterizes the information cost required to understand cluster C[j] from cluster C[i], i.e., the intensity of information transmission. This reflects the similarity in physical state of the local flow field regions represented by the two clusters. The smaller the KL divergence value, the more similar the flow structures of the two regions are; the larger the KL divergence value, the greater the difference in physical state, or the existence of unidirectional physical influence, such as the transmission of shear layer disturbances to the downstream vortex structure.

[0057] Step S14: Construct a directed graph structure representing the intensity of information transmission between clusters based on each data cluster and all the KL divergence values; wherein, the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters.

[0058] In this embodiment, if the KL divergence value is greater than a preset divergence threshold, the corresponding data cluster pair is marked as a valid data cluster pair, and the valid data cluster pair and its corresponding KL divergence value are recorded in the divergence offset array. Based on the inter-cluster connection relationships and divergence weights in the divergence offset array, inter-cluster divergence index array, and divergence weight index array, index management is performed to construct a directed graph structure containing node, edge, and weight information. The inter-cluster divergence index array records the connection relationships between valid data cluster pairs, and the divergence weight index array records the divergence weights. It can be understood that if the KL divergence value of a cluster pair is greater than a set threshold, it indicates that there is significant information interaction between the cluster pairs, and it should be considered an important part of the information transmission chain. The effective cluster pairs and their divergence values ​​are recorded by the divergence offset array (kl_ptr), and the cluster pair relationships and their divergence weights are indexed and managed by the inter-cluster divergence index array (kl_edge_pos) and the divergence weight index array (kl_weight_pos), forming a directed graph containing node, edge and weight information, which provides the network structure foundation for subsequent sampling weight allocation.

[0059] Specifically, based on the above calculations, a weighted directed graph G=(V,E,W) is constructed to formally express the global inter-cluster information relationship. V represents the set of nodes, E represents the set of edges between nodes, and W represents the set of edge weights. The node set V of the directed graph consists of all data clusters {C[1], C[2], ..., C[N]}. Directed edges and weights: For any two different nodes C[i] and C[j], the KL divergence value between them is calculated. The divergence value calculation process is as described in the aforementioned disclosed embodiment and will not be repeated here. If the value is greater than a preset divergence threshold ϵ, it is considered that there is a significant information flow from cluster C[i] to cluster C[j], thereby creating a directed edge from node C[i] to node C[j]. The weight parameter of this directed edge is determined by the KL divergence value between the two nodes. The resulting directed graph structure characterizes the network topology of information interaction between different physical regions in the flow field, providing a crucial data foundation for the next step of adaptive sampling based on network centrality (in-degree strength).

[0060] Step S15: Calculate the in-degree intensity of each node in the directed graph structure so as to generate the sampling probability weights corresponding to each data cluster based on the in-degree intensity.

[0061] In this embodiment, the KL divergence weights of all in-degree edges of each node in the directed graph structure are accumulated to obtain the in-degree strength of each node in the network. The in-degree strength is then normalized according to the maximum entropy principle to generate sampling probability weights. It can be understood that for any node V[i] (i.e., a data cluster) in the constructed directed graph, its in-degree count and the KL divergence weights of its in-degree edges are counted, and the in-degree weights are accumulated to obtain the information transmission strength index of the node in the information network. Combining the node's local information entropy value and the in-degree information strength, the in-degree count of the node is recorded through a node offset array (node_ptr), and the node index array (node_pos) and node weight index array (node_weight_pos) identify the position index of the node and its in-degree weight in the directed graph. Based on the maximum entropy principle, the sampling probability weights are generated after normalizing the node information strength, ensuring that high-information-content regions have higher sampling density, while redundant regions have lower sampling density, thereby achieving information-driven adaptive and efficient subsampling.

[0062] Step S16: Extract a set of sampling points from each data cluster according to the sampling probability weights; the set of sampling points is used to train an implicit neural representation model to reconstruct the turbulent flow field of the target component in the aircraft simulation process.

[0063] In this embodiment, sampling points are extracted from each data cluster according to the sampling probability weights to generate a sampling point set based on each sampling point. Specifically, representative sub-sampling points S[i] are extracted from each data cluster according to the adaptive sampling probability weights. The sampling points S[i] preferentially cover regions with dense information, significant local disturbances, and drastic flow changes to enhance the surrogate model's ability to analyze local details and boundary features. The sampling points are then used as input query points and fed into the Implicit Neural Representation (INR) architecture to perform mapping and function approximation of field variables in continuous space, achieving high-fidelity continuous reconstruction and efficient generalizable representation of the flow field.

[0064] As can be seen, this application discloses a method for acquiring raw flow field data of a target component in an aircraft simulation process, extracting physical descriptors characterizing flow features from the raw flow field data, partitioning the physical descriptors into a feature space to generate several data clusters, wherein each data cluster corresponds to a local region with similar physical states in the raw flow field data, calculating the KL divergence value between the probability distributions of the physical descriptors in two data clusters, constructing a directed graph structure characterizing the information transmission intensity between clusters based on each data cluster and all the KL divergence values, wherein the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters, calculating the in-degree intensity of each node in the directed graph structure to generate sampling probability weights corresponding to each data cluster based on the in-degree intensity, extracting a set of sampling points from each data cluster based on each sampling probability weight, and using the set of sampling points to train an implicit neural representation model for reconstructing the turbulent flow field of the target component in the aircraft simulation process. Therefore, by extracting physical descriptors representing flow characteristics and partitioning the feature space, a neural network implicitly maps any continuous spatiotemporal coordinate (query point) in the physical space to the flow field attribute value (velocity, pressure, etc.) at that point. This achieves a continuous and analytical expression of the flow field distribution, shifting sampling from the geometric space to the physical feature space. This directly correlates the distribution of sampling points with the physical essence of the flow field (vortex structure, shear region), ensuring the physical meaning and relevance of the sampling strategy. Furthermore, calculating the KL divergence and constructing a directed graph representing the intensity of information transmission between clusters places data clusters in a global information interaction network. By calculating the in-degree strength of nodes to generate sampling probability weights, the system can automatically identify the physical regions that occupy a core position in the entire flow field information network, i.e., the clusters corresponding to nodes with high in-degree strength. This ensures that sampling resources are preferentially allocated to those regions with the richest information content and the greatest impact on the global flow. Finally, with the same total number of sampling points, the accuracy of the reconstruction model is significantly improved compared to traditional methods; and when achieving the same reconstruction accuracy, the amount of training data and training cost are significantly reduced. This provides an effective technical solution to the problem of high computational costs in high-fidelity simulation of aircraft.

[0065] like Figure 2 As shown, this invention also discloses a specific implicit neural representation flow field reconstruction sampling method for aircraft simulation, including:

[0066] 1. Extract flow field attribute descriptors.

[0067] After acquiring high-fidelity raw flow field data of the target components in the aircraft simulation process, data preprocessing is performed first. By performing local spatial differentiation operations (including gradient, curl, divergence, etc.) on the main state variables such as velocity field and pressure field, high-dimensional physical descriptors that can characterize the essential features of the flow are extracted. These descriptors include, but are not limited to, vorticity, velocity gradient tensor, strain rate tensor, pressure gradient, and turbulent kinetic energy, forming the characteristic basis for subsequent analysis.

[0068] 2. Divide the feature space to generate data clusters.

[0069] Clustering algorithms are used to partition the feature space of the extracted physical descriptors. By calculating similarity metrics between descriptors (such as Euclidean distance and cosine similarity), descriptors that are geographically close and have similar physical properties are grouped into the same set, forming multiple data clusters with clear physical meanings. Each data cluster corresponds to a local region in the flow field with similar flow characteristics, and its spatial location information is recorded through an index structure.

[0070] 3. Calculate the relative entropy of the inter-cluster probability distribution.

[0071] For each generated data cluster, the probability distribution characteristics of its internal descriptors are statistically analyzed, and the corresponding probability density functions are derived. Based on the Kullback-Leibler divergence formula, the relative entropy between the probability distributions of any two data clusters is systematically calculated. This relative entropy value quantifies the information transfer intensity and physical similarity between different flow regions.

[0072] 4. Construct a directed graph that represents the intensity of information transmission.

[0073] Using data clusters as nodes and the relative entropy between clusters as edge weights, a weighted directed graph representing the information transmission relationships within the flow field is constructed. When the KL divergence between a pair of clusters exceeds a set threshold, a directed connection is established between the corresponding nodes. The resulting directed graph structure comprehensively depicts the information interaction network between different characteristic regions in the flow field, enabling the determination of the propagation paths and influence ranges of key flow structures in the turbulent flow field.

[0074] 5. Generate sampling weights based on node in-degree and extract samples.

[0075] The in-degree strength of each node in the directed graph is analyzed, which is the sum of the weights of all edges pointing to that node. Based on the maximum entropy principle, the in-degree strength is normalized into sampling probability weights to ensure that regions with high information convergence receive higher sampling priority. According to the generated weight distribution, representative sample points are adaptively extracted from each data cluster to form an optimized subsampled dataset.

[0076] 6. Use subsampled data to train the INR flow field reconstruction model.

[0077] The adaptively extracted sample points are input into the implicit neural representation architecture as query points, and the neural network learns a continuous mapping function from spatial coordinates to flow field properties. The trained INR model can achieve high-fidelity reconstruction of the turbulent flow field of the target component, accurately reproducing the detailed features of key flow structures such as vortex cores and shear layers while reducing training costs.

[0078] Reference Figure 3 As shown, the present invention also provides an implicit neural representation flow field reconstruction sampling device for aircraft simulation, comprising:

[0079] Data extraction module 11 is used to acquire the original flow field data of the target component in the aircraft simulation process, so as to extract physical descriptors characterizing the flow features from the original flow field data;

[0080] The data cluster partitioning module 12 is used to partition the physical descriptor into a feature space to generate several data clusters; wherein each data cluster corresponds to a local region with similar physical states in the original flow field data;

[0081] The divergence calculation module 13 is used to calculate the KL divergence value between the probability distributions of the physical descriptors in two data clusters;

[0082] Graph construction module 14 is used to construct a directed graph structure representing the intensity of information transmission between clusters based on each data cluster and all the KL divergence values; wherein, the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters.

[0083] The sampling weight determination module 15 is used to calculate the in-degree intensity of each node in the directed graph structure, so as to generate the sampling probability weight corresponding to each data cluster based on the in-degree intensity.

[0084] The sampling module 16 is used to extract a set of sampling points from each data cluster according to the sampling probability weights; the set of sampling points is used to train an implicit neural representation model to reconstruct the turbulent flow field of the target component in the aircraft simulation process.

[0085] It can be seen that by combining the principle of information entropy, cluster analysis and dynamic error feedback mechanism, the sampling device can mine the local information content of the physical quantities of the flow field and dynamically generate the optimal set of query points, thereby improving the INR surrogate model's ability to express complex flow field details and its training convergence speed. It is especially suitable for high-fidelity continuous reconstruction and time evolution prediction under typical working conditions such as large-scale unstructured flow fields, turbulence and magnetohydrodynamics.

[0086] Furthermore, embodiments of this application also disclose an electronic device, Figure 4This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0087] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the implicit neural representation flow field reconstruction sampling method for aircraft simulation disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0088] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0089] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0090] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0091] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. It can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the implicit neural representation flow field reconstruction sampling method for aircraft simulation executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0092] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned implicit neural representation flow field reconstruction sampling method for aircraft simulation. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0094] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0095] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0096] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for sampling implicit neural representation flow field reconstruction for aircraft simulation, characterized in that, include: The raw flow field data of the target component in the aircraft simulation process is acquired in order to extract physical descriptors characterizing the flow features from the raw flow field data; The physical descriptor is partitioned into a feature space to generate several data clusters; wherein each data cluster corresponds to a local region with similar physical states in the original flow field data; Calculate the KL divergence between the probability distributions of the physical descriptors in the two data clusters; A directed graph structure representing the intensity of information transmission between clusters is constructed based on each data cluster and all the KL divergence values; wherein, the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters. Calculate the in-degree intensity of each node in the directed graph structure, so as to generate the sampling probability weights corresponding to each data cluster based on the in-degree intensity; Sampling point sets are extracted from each data cluster according to the sampling probability weights described above; the sampling point sets are used to train the implicit neural representation model to reconstruct the turbulent flow field of the target component in the aircraft simulation process; The calculation of the KL divergence value between the probability distributions of the physical descriptors in the two data clusters includes: The probability distribution of the physical descriptors in each data cluster is statistically analyzed to determine the probability density function of each data cluster based on the probability distribution; The relative entropy between the probability density functions corresponding to two data clusters is calculated based on the KL divergence equation to obtain the KL divergence value. The construction of a directed graph structure representing the intensity of information transfer between clusters based on each data cluster and all the KL divergence values ​​includes: If the KL divergence value is greater than the preset divergence threshold, the corresponding data cluster pair is marked as a valid data cluster pair, and the valid data cluster pair and the corresponding KL divergence value are recorded in the divergence offset array. Index management is performed based on the inter-cluster connection relationships and divergence weights in the divergence offset array, inter-cluster divergence index array, and divergence weight index array to construct a directed graph structure containing node, edge, and weight information; wherein, the inter-cluster divergence index array records the connection relationships between valid data cluster pairs, and the divergence weight index array records the divergence weights. The calculation of the in-degree intensity of each node in the directed graph structure, in order to generate sampling probability weights corresponding to each data cluster based on the in-degree intensity, includes: The KL divergence weights of all in-degree edges of each node in the directed graph structure are accumulated to obtain the in-degree strength of each node in the network. The in-degree strength is then normalized according to the maximum entropy principle to generate sampling probability weights.

2. The implicit neural representation flow field reconstruction sampling method for aircraft simulation according to claim 1, characterized in that, The physical descriptors include any one or more of the following: vorticity, velocity gradient tensor, strain rate tensor, pressure gradient, turbulent kinetic energy, vorticity modulus, and Reynolds stress tensor.

3. The implicit neural representation flow field reconstruction sampling method for aircraft simulation according to claim 1, characterized in that, The step of partitioning the physical descriptor into a feature space to generate several data clusters includes: Using a preset clustering algorithm, and based on the similarity metric between the physical descriptors, each physical descriptor is divided into several data clusters; Based on the spatial position of the physical descriptor in the original flow field data, the mapping relationship between the data cluster where the physical descriptor is located and the corresponding spatial position is recorded.

4. The implicit neural representation flow field reconstruction sampling method for aircraft simulation according to any one of claims 1 to 3, characterized in that, The step of extracting a set of sampling points from each data cluster according to each sampling probability weight includes: Sampling points are extracted from each data cluster according to the sampling probability weights to generate a sampling point set based on each sampling point.

5. A sampling device for implicit neural representation flow field reconstruction for aircraft simulation, characterized in that, include: The data extraction module is used to acquire the raw flow field data of the target component in the aircraft simulation process, so as to extract physical descriptors characterizing the flow features from the raw flow field data; The data cluster partitioning module is used to partition the physical descriptor into a feature space to generate several data clusters; wherein each data cluster corresponds to a local region with similar physical states in the original flow field data; The divergence calculation module is used to calculate the KL divergence value between the probability distributions of the physical descriptors in two data clusters; The graph construction module is used to construct a directed graph structure representing the intensity of information transmission between clusters based on each data cluster and all the KL divergence values; wherein, the nodes of the directed graph structure are data clusters, and the weight parameters of the directed edges between nodes are determined by the KL divergence values ​​between the corresponding data clusters. The sampling weight determination module is used to calculate the in-degree intensity of each node in the directed graph structure, so as to generate the sampling probability weight corresponding to each data cluster based on the in-degree intensity. The sampling module is used to extract a set of sampling points from each data cluster according to the sampling probability weights; the set of sampling points is used to train an implicit neural representation model to reconstruct the turbulent flow field of the target component in the aircraft simulation process. The divergence calculation module is specifically used to statistically analyze the probability distribution of the physical descriptors in each data cluster, so as to determine the probability density function of each data cluster based on the probability distribution; and to calculate the relative entropy between the probability density functions corresponding to two data clusters based on the KL divergence equation, so as to obtain the KL divergence value. The graph construction module is specifically used to mark the corresponding data cluster pair as a valid data cluster pair if the KL divergence value is greater than a preset divergence threshold, and record the valid data cluster pair and the corresponding KL divergence value in the divergence offset array; based on the inter-cluster connection relationship and divergence weight in the divergence offset array, inter-cluster divergence index array, and divergence weight index array, index management is performed to construct a directed graph structure containing node, edge, and weight information; wherein, the inter-cluster divergence index array records the connection relationship between valid data cluster pairs, and the divergence weight index array records the divergence weight; The sampling weight determination module is specifically used to accumulate the KL divergence weights of all in-degree edges of each node in the directed graph structure to obtain the in-degree intensity of each node in the network, and then normalize the in-degree intensity according to the maximum entropy principle to generate sampling probability weights.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the implicit neural representation flow field reconstruction sampling method for aircraft simulation as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the implicit neural representation flow field reconstruction sampling method for aircraft simulation as described in any one of claims 1 to 4.

Images