Multi-parameter automobile body flow field prediction method and device and server

Abstract

The invention provides a multi-parameter automobile body flow field prediction method and device and a server, and relates to the technical field of vehicle fluid mechanics, and the method comprises the steps: carrying out the data preprocessing of a three-dimensional CAD model and a preset physical parameter set, and determining target input data; the method comprises the following steps: constructing a dynamic graph structure of a vehicle body flow field by performing dynamic graph edge convolution processing on target input data, and performing edge convolution processing on the dynamic graph structure to extract a local feature vector of the dynamic graph structure; and performing layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors, determining global feature vectors, and sending the global feature vectors to a preset full-connection network to predict various physical quantities in the automobile body flow field, and determining a target flow field parameter prediction result. According to the method, the expression ability of the model to a complex flow field can be enhanced based on dynamic graph edge convolution, so that the simulation efficiency and the prediction accuracy are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle fluid mechanics, and in particular to a method, device and server for predicting a multi-parameter automobile body flow field. Background Art

[0002] In automobile research and development, flow field prediction is an important part of evaluating the aerodynamic performance of the vehicle body. By accurately predicting the airflow distribution around the vehicle body (such as pressure field, velocity field and turbulent energy distribution, etc.), the vehicle body design can be effectively optimized, the drag coefficient can be reduced, and thus the vehicle performance and fuel economy can be improved. At present, relevant technologies have proposed that the vehicle body flow field can be predicted through traditional CFD simulation methods or based on neural networks. However, traditional CFD simulation methods have long calculation time and high resource consumption, and are not suitable for fast-iteration automobile design processes. At the same time, existing neural network-based prediction methods are insufficient in dealing with complex vehicle body geometry changes and nonlinear flow field characteristics, especially in the joint prediction of multi-parameter flow field characteristics, where the accuracy and generalization capabilities are limited. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a multi-parameter automobile body flow field prediction method, device and server, which can enhance the model's ability to express complex flow fields based on dynamic graph edge convolution, thereby significantly improving simulation efficiency and prediction accuracy.

[0004] In a first aspect, an embodiment of the present invention provides a multi-parameter automobile body flow field prediction method, the method comprising: obtaining a three-dimensional CAD model of a automobile body and a preset physical parameter set, and performing data preprocessing on the three-dimensional CAD model and the preset physical parameter set to determine target input data, wherein the preset physical parameter set includes: various physical parameters used in fluid dynamics calculations; constructing a dynamic graph structure of the vehicle body flow field by performing dynamic graph edge convolution processing on the target input data, and performing edge convolution processing on the dynamic graph structure to extract local feature vectors of the dynamic graph structure; performing layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors to determine the global feature vector, and sending the global feature vector to a preset fully connected network to predict various physical quantities in the automobile body flow field and determine the target flow field parameter prediction results.

[0005] In one embodiment, before the step of performing data preprocessing on the three-dimensional CAD model and the preset physical parameter set, it includes: performing data export processing on the three-dimensional CAD model to determine a point cloud model of the automobile body, as well as point cloud data in the point cloud model and additional features of the point cloud data, wherein the point cloud data includes: point cloud coordinates and surface normal vector features, and the additional features of the point cloud data are used to characterize local surface properties of the vehicle body surface.

[0006] In one embodiment, data preprocessing is performed on the three-dimensional CAD model and the preset physical parameter set to determine the target input data, including: normalizing and denoising the point cloud data, additional features and physical parameters, and completing the physical parameters through the Reynolds number formula to determine the target input data.

[0007] In one embodiment, the dynamic graph structure includes: a node set and an edge set. The steps of constructing the dynamic graph structure of the vehicle body flow field by performing dynamic graph edge convolution processing on the target input data include: determining the node set by capturing the point cloud coordinates, surface normal vector characteristics and physical parameters corresponding to each node in the target input data, and calculating the Euclidean distance between each node through the K nearest neighbor algorithm to dynamically select the K nearest neighbor nodes corresponding to each node to determine the edge set; determining the dynamic graph structure according to the node set and the edge set.

[0008] In one embodiment, before the step of determining the dynamic graph structure based on the node set and the edge set, it includes: determining the edge weights between the nodes by initializing the edge weights of the Euclidean distances and fluid characteristic differences between the nodes in the edge set; determining the dynamic graph structure based on the node set, the edge set and the edge weights, and performing dynamic graph update processing on the edge weights to make the dynamic graph structure adaptive to the dynamic changes of the vehicle body flow field.

[0009] In one embodiment, the step of performing edge convolution processing on the dynamic graph structure to extract the local feature vector of the dynamic graph structure includes: performing edge convolution processing on the dynamic graph structure to extract the local geometric relationship between each node and the interactive relationship between the physical characteristics of each parameter, and determining the local feature vector of the node.

[0010] In one embodiment, the step of performing layer-by-layer aggregation processing and global feature aggregation processing on local feature vectors to determine the global feature vector includes: performing layer-by-layer aggregation processing on local feature vectors to determine multiple layers of feature vectors, and aggregating the features of each node into a global feature vector through a maximum pooling operation.

[0011] In a second aspect, an embodiment of the present invention further provides a multi-parameter automobile body flow field prediction device, the device comprising: a data preprocessing module, which obtains a three-dimensional CAD model of a automobile body and a preset physical parameter set, and performs data preprocessing on the three-dimensional CAD model and the preset physical parameter set to determine target input data, wherein the preset physical parameter set includes: various physical parameters used in fluid dynamics calculations; a feature extraction module, which constructs a dynamic graph structure of the vehicle body flow field by performing dynamic graph edge convolution processing on the target input data, and performs edge convolution processing on the dynamic graph structure to extract local feature vectors of the dynamic graph structure; a parameter prediction module, which performs layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors to determine the global feature vector, and sends the global feature vector to a preset fully connected network to predict various physical quantities in the automobile body flow field and determine the target flow field parameter prediction results.

[0012] In a third aspect, an embodiment of the present invention further provides a server, comprising a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement any one of the methods provided in the first aspect.

[0013] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement any one of the methods provided in the first aspect.

[0014] The embodiments of the present invention bring the following beneficial effects: An embodiment of the present invention provides a multi-parameter automobile body flow field prediction method, device and server. After obtaining a three-dimensional CAD model of a car body and a preset physical parameter set, the method performs data preprocessing on the three-dimensional CAD model and the preset physical parameter set to determine the target input data, and then constructs a dynamic graph structure of the car body flow field by performing dynamic graph edge convolution processing on the target input data, and performs edge convolution processing on the dynamic graph structure to extract local feature vectors of the dynamic graph structure. Finally, layer-by-layer aggregation processing and global feature aggregation processing are performed on the local feature vectors to determine the global feature vector, and the global feature vector is sent to a preset fully connected network to predict various physical quantities in the automobile body flow field and determine the target flow field parameter prediction results. The embodiment of the present invention can dynamically capture the complex geometric structure of the automobile body and the local dependency of the flow field by introducing a dynamic graph edge convolution network, and at the same time support multi-parameter joint modeling and prediction, thereby significantly improving simulation efficiency and prediction accuracy, and providing strong technical support for automobile aerodynamic design.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A schematic flow chart of a multi-parameter automobile body flow field prediction method provided by an embodiment of the present invention; Figure 2 A schematic diagram of a multi-parameter automobile body flow field prediction method provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a multi-parameter automobile body flow field prediction device provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a server provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] At present, in automobile research and development, flow field prediction is an important part of evaluating the aerodynamic performance of the vehicle body. By accurately predicting the airflow distribution around the vehicle body (such as pressure field, velocity field and turbulent energy distribution, etc.), the vehicle body design can be effectively optimized, the drag coefficient can be reduced, and thus the vehicle performance and fuel economy can be improved. However, traditional CFD simulation calculations are usually time-consuming and require high computing resources, which is not conducive to rapid iteration of design. In recent years, with the development of deep learning technology, neural network-based flow field prediction methods have gradually been introduced to accelerate the calculation process and improve prediction efficiency. However, these methods usually rely on static graph structures or fixed input features, and are difficult to adapt to complex vehicle body geometry changes and nonlinear characteristics of the flow field, resulting in limited prediction accuracy and generalization ability. In addition, existing methods lack effective modeling methods when dealing with multi-parameter flow field characteristics (such as joint prediction of multiple aerodynamic indicators), which further limits their practicality.

[0021] Based on this, the multi-parameter automobile body flow field prediction method, device and server provided by the implementation of the present invention can dynamically capture the local dependency between the complex geometry of the automobile body and the flow field characteristics through the dynamic graph convolution mechanism, and realize efficient prediction of various physical quantities in the flow field (such as pressure field, velocity field, etc.) through multi-parameter joint modeling. Compared with traditional methods, the present invention can significantly improve the efficiency and accuracy of flow field prediction, and at the same time provide an automated and efficient aerodynamic performance evaluation method for complex body design, which helps the rapid innovation and optimization of automobile research and development.

[0022] To facilitate understanding of this embodiment, a multi-parameter automobile body flow field prediction method disclosed in an embodiment of the present invention is first described in detail. The method is applied to a multi-parameter automobile body flow field prediction model. Training the multi-parameter automobile body flow field prediction model specifically includes the following steps (1) to (5): (1) Data division: The training data is generated by CFD simulation and wind tunnel experiments, covering thousands of point cloud data of different car body shapes and their corresponding flow field characteristic parameters, including pressure distribution, drag coefficient, lift coefficient, etc. In order to improve the generalization ability of the model, the data set is divided into training set, validation set and test set in a ratio of 7:2:1.

[0023] (2) Loss function design: In order to measure the difference between the model prediction value and the actual flow field parameters, the present invention uses the mean square error (MSE) as the loss function:

[0024] in, is the real flow field parameter, is the model prediction value, N is the number of samples, and this loss function can effectively reduce the impact of large deviation samples on the training process.

[0025] (3) Regularization and stability improvement: Introduce the Dropout technology in the fully connected layer (Dropout is a commonly used method to prevent overfitting), set the drop probability to 0.5, and randomly block some neurons to prevent the model from overfitting the training data.

[0026] In one implementation, batch normalization techniques are applied after each convolution and fully connected operation to improve training stability and accelerate convergence.

[0027] (4) Model training: During the training process, the model inputs data in batch form (batch size = 64). Each iteration (epoch) contains multiple batches of training samples. After each iteration, the model performance is evaluated in real time using the validation set to ensure that the loss value gradually converges. When the validation set loss does not decrease significantly within five consecutive epochs, the training is terminated early to prevent overfitting.

[0028] (5) Performance evaluation: On the test set, the mean square error (MSE) was used to evaluate the model performance. The experimental results show that the proposed model can predict the flow field parameters with high accuracy, and the errors of the predicted pressure distribution, drag coefficient and lift coefficient are all controlled within 5%.

[0029] See also Figure 1 The flowchart of a multi-parameter automobile body flow field prediction method shown in FIG. 1 mainly includes the following steps S102 to S106: Step S102, obtain the 3D CAD model of the car body and the preset physical parameter set, and perform data preprocessing on the 3D CAD model and the preset physical parameter set to determine the target input data, wherein the preset physical parameter set includes: various physical parameters used in fluid dynamics calculations. In one embodiment, when the 3D CAD model of the car body and the multi-physical parameters related to fluid dynamics are used as the core input, the input information includes the following aspects: 1. Geometric information: The point cloud model of the car body is derived through 3D scanning or CAD data, and each point is represented by a 3D coordinate (x, y, z) and a surface normal vector n. The point cloud data provides a basis for subsequent graph structure construction and edge convolution operations; 2. Local surface properties: the initial pressure, temperature distribution and local roughness of the car body surface, etc., as additional features of the point cloud data, in addition, it also includes the initial fluid environment setting, that is, setting the air density, aerodynamic viscosity and fluid velocity of the current environment.

[0030] Step S104, constructing a dynamic graph structure of the vehicle body flow field by performing dynamic graph edge convolution processing on the target input data, and performing edge convolution processing on the dynamic graph structure to extract local feature vectors of the dynamic graph structure. The dynamic graph structure includes: a node set and an edge set. In one embodiment, the dynamic graph structure can be determined based on the node set and the edge set. Specifically, the node set V: each node corresponds to a point cloud The multi-dimensional node features include the three-dimensional coordinates of the point, the normal vector and other physical parameters related to the fluid; the edge set E: by calculating the Euclidean distance in the feature space , dynamically select the k-nearest neighbor nodes of each node, where the edge set is defined as follows:

[0031] Furthermore, in the process of point cloud processing, since the connection relationship of the points is not fixed, dynamic graph edge convolution can dynamically adjust the selection of neighbor points according to the geometric position or features of the points, so as to better describe the structural information. Specifically, dynamic graph edge convolution not only considers the node features, but also assigns weights to each edge. This edge weight can be calculated by feature similarity, geometric information, etc., so that the model can focus on edges that are more relevant to the target. Since dynamic graph edge convolution usually combines local feature extraction with global feature update for information integration, it has stronger adaptability to data in complex geometric shapes or dynamic environments.

[0032] Step S106, performing layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors, determining the global feature vector, and sending the global feature vector to the preset fully connected network to predict various physical quantities in the flow field of the automobile body, and determine the prediction results of the target flow field parameters. In one embodiment, after completing the multi-layer feature extraction, the features of all nodes can be aggregated into a global feature vector through a maximum pooling operation:

[0033] Among them, N(i) is based on the distance , for all nodes Sorting, select the k nodes with the closest distance as the domain N(i). The global feature vector integrates the overall information of all local structures and their physical properties in the point cloud.

[0034] Then, the global features are input into the fully connected network (Fully Connected Layers) to predict the target flow field parameters using the following formula:

[0035] Among them, y is the predicted value of the target flow field parameters (such as pressure distribution on the vehicle body surface, drag coefficient, etc.). A large number of experiments have shown that the error between the predicted value and the actual value is controlled within 5%.

[0036] The above-mentioned multi-parameter automobile body flow field prediction method provided by the embodiment of the present invention can enhance the model's ability to express complex flow fields based on dynamic graph edge convolution, thereby significantly improving simulation efficiency and prediction accuracy.

[0037] See also Figure 2 The schematic diagram of a multi-parameter automobile body flow field prediction method is shown. The embodiment of the present invention also provides an implementation method of the vehicle body flow field prediction. For details, see (1) to (3) below: (1) The 3D CAD model is exported to determine the point cloud model of the automobile body, as well as the point cloud data and additional features of the point cloud data in the point cloud model. The point cloud data, additional features and physical parameters are then normalized and denoised, and the physical parameters are supplemented by the Reynolds number formula to determine the target input data. The point cloud data includes: point cloud coordinates and surface normal vector features. The additional features of the point cloud data are used to characterize the local surface properties of the vehicle body surface. In a real-time manner, in the data preprocessing stage, the point cloud data must first be normalized to normalize the geometric coordinates and physical parameters to a uniform range to prevent the influence of different feature scales on model training. Secondly, noise reduction technology is used to remove measurement errors or redundant points to ensure the quality and sparse characteristics of the input data. In addition, some physical parameters are supplemented by the Reynolds number formula to better characterize the fluid environment:

[0038] Where Re is the Reynolds number, which is used to determine the type of fluid flow and is dimensionless; is the fluid density, the mass of fluid per unit volume, in kg / m3; is the fluid velocity, relative to the flow velocity of the object, in m / s; L is the characteristic length, in m; μ: the fluid dynamic viscosity, in Pascal seconds (Pa s).

[0039] (2) The node set is determined by capturing the point cloud coordinates, surface normal vector features and physical parameters corresponding to each node in the target input data, and the Euclidean distance between each node is calculated through the K nearest neighbor algorithm to dynamically select the K nearest neighbor nodes corresponding to each node to determine the edge set. The edge weights between each node are then determined by initializing the edge weights based on the Euclidean distances and fluid feature differences between the nodes in the edge set. Finally, the dynamic graph structure is determined based on the node set, edge set and edge weights, and the edge weights are dynamically updated to make the dynamic graph structure adaptive to the dynamic changes of the vehicle body flow field.

[0040] In one embodiment, when establishing a graph structure, in order to better characterize the relationship between nodes, it is necessary to initialize a weight for each edge. This weight is not only based on the geometric distance, but also combines the difference in fluid characteristics to reflect the similarity between the two. The weight of the initialized edge is , which is defined by combining geometric distance and physical parameter similarity, taking into account the node and The geometric distance and fluid characteristic difference Δ :

[0041] in, and is an adjustment factor used to control the influence of geometric distance and fluid pressure difference on edge weight. In practical applications, they can be respectively =0.5 and =0.2.

[0042] After each layer of the network, the neighborhood and edge weights are dynamically recalculated according to the feature changes of the points, ensuring that the graph structure can adapt to the dynamic characteristics of the complex flow field.

[0043] (3) Performing edge convolution processing on the dynamic graph structure to extract the local geometric relationship between each node and the interactive relationship between the physical features of each parameter, and determining the local feature vector of the node; performing layer-by-layer aggregation processing on the local feature vector to determine the multi-layer feature vector, and aggregating the features of each node into a global feature vector through a maximum pooling operation. In one embodiment, based on the dynamic graph structure, edge convolution (EdgeConv) is used to extract the interaction between the local geometric relationship and the multi-parameter physical features. Specifically, for the node The feature update formula is:

[0044] in, , Representation Node The feature vector of MLP is a shared multi-layer perceptron that learns the interaction relationship between local features. For input to the node The concatenation vector of the feature and the feature difference of adjacent nodes is used to capture the local geometric relationship.

[0045] In the edge convolution operation, not only the translation invariance of the local geometric structure is preserved, but also the impact of physical characteristics on the changes in the local flow field is captured, thereby enhancing the model's ability to express complex flow fields. Through multi-layer EdgeConv operations, local features are aggregated layer by layer, and the ReLU activation function and batch normalization technology are used to improve the stability of feature extraction and nonlinear expression capabilities.

[0046] In addition, compared with the existing neural network model for predicting the flow field of the vehicle body, the advantages of dynamic graph edge convolution are: 1. Ability to process irregular data: The wind resistance prediction model usually needs to extract geometric features from the point cloud data of the vehicle body. Point cloud data is inherently sparse and irregular, while traditional convolutional networks are more friendly to regular grid data. In dynamic graph edge convolution, the network can directly operate on point cloud data, avoiding the risk of information loss when converting point clouds into voxels or projected images; 2. Sensitivity to local geometric deformation: Wind resistance is significantly affected by the local geometric shape of the vehicle body surface. Dynamic graph edge convolution can capture subtle shape changes in the point cloud by adaptively constructing neighborhood relationships, thereby more accurately predicting wind resistance; 3. Dynamic relationship modeling: The relationship between wind resistance and the overall structure of the vehicle body is usually dynamic. Dynamic graph edge convolution can dynamically adjust edge weights and adjacency relationships to ensure that the most relevant features between different parts of the vehicle body are captured during training; 4. Improved computational efficiency: The sparsity of point cloud data will lead to a waste of computational resources in traditional convolution operations. The dynamic graph edge convolution network constructs a sparse adjacency graph and only propagates information between related points, which significantly reduces the computational overhead while retaining high-quality feature representation; 5. Better generalization ability: When processing car body data of different shapes, dynamic graph edge convolution can dynamically adjust according to the geometric characteristics of the input data, while traditional convolutional networks need to design special structures or preprocessing methods for different car body shapes. Therefore, the former performs better when generalizing to different data.

[0047] In practical applications, if an automobile manufacturing company plans to design a high-performance electric sports car, in order to achieve the best aerodynamic performance while meeting the appearance design requirements, the goal is to reduce the drag coefficient to improve fuel efficiency and high-speed stability. However, traditional wind tunnel experiments are expensive and have a long iteration cycle. Therefore, the above-mentioned dynamic graph modeling technology in the present invention can be used to improve design efficiency by combining numerical simulation with machine learning methods, including the following (A) to (C): (A) Data acquisition and preprocessing include input data and data preprocessing, where input data includes: Geometric information acquisition: Use a 3D scanner to scan the car body to generate point cloud data, and save the data in .ply format. Each point is represented by a 3D coordinate (x, y, z) and a surface normal vector n; set local surface properties: initial surface pressure distribution, temperature distribution, material roughness; complete flow field parameters through the Reynolds number formula; initial fluid environment setting: set the air density, aerodynamic viscosity, and fluid velocity of the current environment.

[0048] In addition, data preprocessing includes: normalizing the point cloud coordinates and physical parameters to eliminate the influence of feature scale differences; using Gaussian filtering technology to remove scanning noise to ensure the sparsity and structural integrity of the point cloud.

[0049] (B) Dynamic graph modeling and updating include: graph structure construction, edge weight initialization and dynamic graph update. Specifically, graph structure construction includes: based on point cloud data P={p1,p2,...,pN}, constructing a graph structure G=(V,E), where the node set V contains the geometric coordinates, normal vector and physical parameters of each point; the edge set E is constructed by the k-nearest neighbor algorithm (k=8), and the Euclidean distance between nodes is calculated. To define the connection relationship:

[0050] When initializing edge weights, based on edge weights Comprehensive consideration of geometric distance and physical parameter difference Δ :

[0051] Among them, the adjustment factors are set as σd=0.5 and σp=0.2.

[0052] Dynamic graph updating includes: after each layer of neural network calculation, the neighborhood and edge weights are recalculated according to the feature changes of the points, so that the graph structure can adapt to the dynamic changes of the flow field.

[0053] (C) Feature extraction and global aggregation include: edge convolution operation, global feature aggregation and output prediction. Specifically, the edge convolution operation includes: extracting local characteristics and geometric relationships between nodes through edge convolution, where the node feature update formula is:

[0054] in, , Representation Node The characteristic vector of is the input vector, MLP is a shared multi-layer perceptron that learns the interaction relationship between local features.

[0055] When performing global feature aggregation, the maximum pooling operation is used to aggregate local features into a global feature vector:

[0056] When making output predictions, the global feature vector Input into the fully connected network FC to get the final prediction result:

[0057] The prediction result outputs the drag coefficient of the vehicle body, and the actual experimental results show that the predicted drag coefficient of the vehicle is less than the actual value and is controlled within 5%. The present invention improves the model's ability to express complex flow fields, avoids the high calculation cost of traditional CFD methods, significantly improves simulation efficiency and ensures prediction accuracy.

[0058] In summary, the present invention can combine the three-dimensional point cloud geometric information of the automobile body with the multi-physical parameters of fluid dynamics, construct a dynamic graph structure to dynamically capture the interactive characteristics of complex surface geometry and fluid parameters, use edge convolution to extract local features and aggregate them layer by layer into global feature vectors, and then map the global features into flow field target parameters through a fully connected network, thereby achieving high-precision prediction of the pressure distribution and flow velocity field on the vehicle body surface; this method completes the physical parameters through the Reynolds number formula, optimizes the edge weights in combination with the geometric distance and fluid property differences, improves the model's ability to express complex flow fields, avoids the high computational cost of traditional CFD methods, and significantly improves simulation efficiency and prediction accuracy.

[0059] For the multi-parameter automobile body flow field prediction method provided in the above embodiment, the embodiment of the present invention provides a multi-parameter automobile body flow field prediction device, see Figure 3 The schematic diagram of the structure of a multi-parameter automobile body flow field prediction device is shown, and the device includes the following parts: The data preprocessing module 302 obtains a three-dimensional CAD model of the automobile body and a preset physical parameter set, and performs data preprocessing on the three-dimensional CAD model and the preset physical parameter set to determine target input data, wherein the preset physical parameter set includes: various physical parameters used in fluid dynamics calculation; The feature extraction module 304 constructs a dynamic graph structure of the vehicle body flow field by performing a dynamic graph edge convolution process on the target input data, and performs an edge convolution process on the dynamic graph structure to extract a local feature vector of the dynamic graph structure; The parameter prediction module 306 performs layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors, determines the global feature vectors, and sends the global feature vectors to a preset fully connected network to predict various physical quantities in the flow field of the automobile body and determine the target flow field parameter prediction results.

[0060] The above-mentioned multi-parameter automobile body flow field prediction device provided in the embodiment of the present application can enhance the model's ability to express complex flow fields based on dynamic graph edge convolution, thereby significantly improving simulation efficiency and prediction accuracy.

[0061] In one embodiment, before performing the step of data preprocessing the three-dimensional CAD model and the preset physical parameter set, the data preprocessing module 302 is also used to: perform data export processing on the three-dimensional CAD model to determine the point cloud model of the automobile body, as well as the point cloud data in the point cloud model and additional features of the point cloud data, wherein the point cloud data includes: point cloud coordinates and surface normal vector features, and the additional features of the point cloud data are used to characterize the local surface properties of the vehicle body surface.

[0062] In one embodiment, when performing data preprocessing on the three-dimensional CAD model and the preset physical parameter set to determine the target input data, the above-mentioned data preprocessing module 302 is also used to: normalize and denoise the point cloud data, additional features and physical parameters, and complete the physical parameters through the Reynolds number formula to determine the target input data.

[0063] In one embodiment, the dynamic graph structure includes: a node set and an edge set. When performing the step of constructing the dynamic graph structure of the vehicle body flow field by performing dynamic graph edge convolution processing on the target input data, the above-mentioned feature extraction module 304 is also used to: determine the node set by capturing the point cloud coordinates, surface normal vector features and physical parameters corresponding to each node in the target input data, and calculate the Euclidean distance between each node through the K nearest neighbor algorithm to dynamically select the K nearest neighbor nodes corresponding to each node to determine the edge set; determine the dynamic graph structure according to the node set and the edge set.

[0064] In one embodiment, when performing the step of determining the dynamic graph structure based on the node set and the edge set, the feature extraction module 304 is also used to: determine the edge weights between the nodes by initializing the edge weights of the Euclidean distances and fluid feature differences between the nodes in the edge set; determine the dynamic graph structure based on the node set, the edge set and the edge weights, and perform dynamic graph update processing on the edge weights to make the dynamic graph structure adapt to the dynamic changes of the vehicle body flow field.

[0065] In one embodiment, when performing edge convolution processing on the dynamic graph structure to extract the local feature vector of the dynamic graph structure, the feature extraction module 304 is also used to: perform edge convolution processing on the dynamic graph structure to extract the local geometric relationship between each node and the interactive relationship between the physical characteristics of each parameter, and determine the local feature vector of the node.

[0066] In one embodiment, when performing layer-by-layer aggregation processing on local feature vectors and global feature aggregation processing to determine the global feature vector, the above-mentioned parameter prediction module 306 is also used to: perform layer-by-layer aggregation processing on local feature vectors, determine multiple layers of feature vectors, and aggregate the features of each node into a global feature vector through a maximum pooling operation.

[0067] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0068] An embodiment of the present invention provides a server. Specifically, the server includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above-mentioned embodiments.

[0069] Figure 4 A structural diagram of a server provided in an embodiment of the present invention, the server 100 includes: a processor 40, a memory 41, a bus 42 and a communication interface 43, wherein the processor 40, the communication interface 43 and the memory 41 are connected via the bus 42; the processor 40 is used to execute an executable module stored in the memory 41, such as a computer program.

[0070] The memory 41 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 43 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0071] The bus 42 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0072] Among them, the memory 41 is used to store programs, and the processor 40 executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0073] The processor 40 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 40. The above processor 40 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 41, and the processor 40 reads the information in the memory 41 and completes the steps of the above method in combination with its hardware.

[0074] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be referred to the previous method embodiments, which will not be repeated here.

[0075] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0076] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A multi-parameter automobile body flow field prediction method, characterized in that: The method comprises: Acquire a three-dimensional CAD model of a vehicle body and a set of preset physical parameters, and perform data preprocessing on the three-dimensional CAD model and the set of preset physical parameters to determine target input data, wherein the set of preset physical parameters includes: various physical parameters used in fluid dynamics calculations; By performing a dynamic graph edge convolution process on the target input data, a dynamic graph structure of the vehicle body flow field is constructed, and an edge convolution process is performed on the dynamic graph structure to extract a local feature vector of the dynamic graph structure; The local feature vectors are subjected to layer-by-layer aggregation processing and global feature aggregation processing to determine a global feature vector, and the global feature vector is sent to a preset fully connected network to predict various physical quantities in the flow field of the automobile body and determine the prediction results of the target flow field parameters.

2. The multi-parameter automobile body flow field prediction method according to claim 1 is characterized in that: Before the step of performing data preprocessing on the three-dimensional CAD model and the preset physical parameter set, the method includes: The three-dimensional CAD model is subjected to data export processing to determine a point cloud model of a vehicle body, point cloud data in the point cloud model, and additional features of the point cloud data, wherein the point cloud data includes: point cloud coordinates and surface normal vector features, and the additional features of the point cloud data are used to characterize local surface properties of the vehicle body surface.

3. The multi-parameter automobile body flow field prediction method according to claim 1 is characterized in that: The step of performing data preprocessing on the three-dimensional CAD model and the preset physical parameter set to determine the target input data includes: The point cloud data, additional features and physical parameters are normalized and denoised, and the physical parameters are completed using a Reynolds number formula to determine the target input data.

4. The multi-parameter automobile body flow field prediction method according to claim 1, characterized in that: The dynamic graph structure includes: a node set and an edge set. The step of constructing the dynamic graph structure of the vehicle body flow field by performing dynamic graph edge convolution processing on the target input data includes: The node set is determined by capturing the point cloud coordinates, surface normal vector features and physical parameters corresponding to each node in the target input data, and the Euclidean distance between each node is calculated by a K-nearest neighbor algorithm to dynamically select the K-nearest neighbor nodes corresponding to each node to determine the edge set; The dynamic graph structure is determined according to the node set and the edge set.

5. The multi-parameter automobile body flow field prediction method according to claim 4 is characterized in that: Before the step of determining the dynamic graph structure according to the node set and the edge set, the method further comprises: Determine the edge weights between the nodes by performing edge weight initialization processing on the Euclidean distances and fluid characteristic differences between the nodes in the edge set; The dynamic graph structure is determined according to the node set, the edge set and the edge weights, and the edge weights are subjected to dynamic graph update processing, so that the dynamic graph structure can adapt to the dynamic changes of the vehicle body flow field.

6. The multi-parameter automobile body flow field prediction method according to claim 1, characterized in that: The step of performing edge convolution processing on the dynamic graph structure to extract a local feature vector of the dynamic graph structure includes: The dynamic graph structure is subjected to edge convolution processing to extract the local geometric relationship between nodes and the interactive relationship between the physical characteristics of various parameters, and determine the local feature vector of the node.

7. The multi-parameter automobile body flow field prediction method according to claim 1, characterized in that: The step of performing layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors to determine the global feature vector comprises: The local feature vectors are aggregated layer by layer to determine multiple layers of feature vectors, and the features of each node are aggregated into the global feature vector through a maximum pooling operation.

8. A multi-parameter automobile body flow field prediction device, characterized in that: The device comprises: A data preprocessing module, which obtains a three-dimensional CAD model of a vehicle body and a set of preset physical parameters, and performs data preprocessing on the three-dimensional CAD model and the set of preset physical parameters to determine target input data, wherein the set of preset physical parameters includes: various physical parameters used in fluid dynamics calculations; A feature extraction module, which constructs a dynamic graph structure of the vehicle body flow field by performing a dynamic graph edge convolution process on the target input data, and performs an edge convolution process on the dynamic graph structure to extract a local feature vector of the dynamic graph structure; The parameter prediction module performs layer-by-layer aggregation processing and global feature aggregation processing on the local feature vector, determines the global feature vector, and sends the global feature vector to a preset fully connected network to predict various physical quantities in the flow field of the automobile body and determine the prediction results of the target flow field parameters.

9. A server, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images