An automobile body surface pressure field prediction method, device, equipment and storage medium

Abstract

This application provides a method, apparatus, device, and storage medium for predicting the pressure field on the surface of an automobile body. The method includes: first, acquiring grid points on the surface of the automobile body; using each grid point as a term to construct a feature vector containing spatial coordinates and physical property parameters; calculating the radius of the physical influence domain corresponding to each grid point based on its local geometric features and physical field information; selecting grid points whose distances meet the constraints to construct a physical neighborhood set; generating an attention mask matrix from the neighborhood set; using this matrix to constrain the attention calculation range; and calculating the attention output for each grid point using a physical spatial bias function corresponding to the spatial distance; and finally integrating the calculation results of all grid points to complete the prediction output of the pressure field on the surface of the automobile body. This method effectively improves the orderliness of the overall prediction process while ensuring the actual fit of the pressure field prediction results.

Description

Technical Field

[0001] This application relates to the field of automotive aerodynamic performance simulation and prediction technology, and more specifically, to a method, device, equipment, and storage medium for predicting the pressure field on the surface of an automotive body. Background Technology

[0002] The pressure field on the vehicle body surface is a crucial analytical indicator in the development of automotive aerodynamics, directly reflecting the airflow distribution on the vehicle surface and providing data support for drag optimization and shape improvement. Currently, the acquisition of vehicle body pressure fields is mainly divided into two categories: wind tunnel testing and numerical simulation. Among them, computational fluid dynamics simulation offers high accuracy, but its computational process is cumbersome and the calculation cycle is long, which cannot meet the development needs of rapid vehicle iteration. Therefore, the industry has begun to adopt deep learning technology for intelligent prediction of vehicle body pressure fields to shorten computation time.

[0003] Existing deep learning techniques for predicting vehicle body pressure fields are early general-purpose fitting neural network models, which rely solely on a basic network structure for data fitting. Specifically, this technique uses vehicle body grid coordinates as input data and, relying on the network's own data fitting capabilities, establishes a mapping relationship between grid coordinates and pressure values. It then completes global pressure prediction of the entire vehicle body through global feature fitting, eliminating the need for complex fluid solvers and enabling rapid output of the pressure field.

[0004] This type of general neural network model relies solely on data fitting to complete predictions, resulting in significant limitations in its fitting ability. It lacks sensitivity to changes in the flow field in abrupt areas of the vehicle body, making it difficult to accurately capture the local pressure fluctuation characteristics of the vehicle body, leading to large prediction errors. Furthermore, the model has weak generalization ability and poor adaptability to vehicle body models of different sizes and shapes, failing to consistently meet the accuracy requirements of industrial vehicle development. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method, device, equipment and storage medium for predicting the pressure field on the surface of an automobile body, which can effectively improve the orderliness of the overall prediction process and ensure the actual fit of the pressure field prediction results.

[0006] In a first aspect, embodiments of this application provide a method for predicting the pressure field on the surface of an automobile body, the method comprising: Obtain the surface grid points on the car body surface, treat each surface grid point as a word, and the feature vector of each word contains the spatial coordinates and physical attribute parameters of that surface grid point; For each surface grid point, the radius of the physical influence domain of that surface grid point is calculated based on its local geometric features and physical field information. Based on the physical influence domain radius, a physical neighborhood set of the surface grid point is constructed. The physical neighborhood set includes other surface grid points whose spatial distance from the surface grid point satisfies the constraint of the physical influence domain radius. Generate an attention mask matrix based on the physical neighborhood set; The attention calculation range is limited to the physical neighborhood set by applying the attention mask matrix, and the attention output of the surface grid point is calculated by combining the physical spatial bias function based on spatial distance. The pressure field on the vehicle body surface is output based on the attention output of each surface grid point.

[0007] Optionally, calculating the physical influence radius of the surface grid point based on its local geometric features and physical field information includes: Local curvature is extracted from the feature vector of the word at the grid point on the surface as local geometric features, pressure field gradient is extracted as physical field information, and global geometric scale parameters of the vehicle body and incoming flow conditions are obtained. The local curvature, the pressure field gradient, the global geometric scale parameter, and the incoming flow condition are weighted and summed, and the summation result is mapped to a normalization coefficient between 0 and 1 through the Sigmoid function. The physical influence radius of the surface grid point is calculated by linear interpolation between the preset minimum and maximum neighborhood radii using the normalization coefficient.

[0008] Optionally, constructing the physical neighborhood set of the surface grid point based on the physical influence domain radius includes: Using the three-dimensional spatial coordinates of the surface grid point as the center of a sphere and the radius of the physical influence domain as the radius, other surface grid points whose spatial distance is not greater than the radius are searched among all grid points on the vehicle surface; The other surface grid points retrieved constitute the physical neighborhood set of this surface grid point.

[0009] Optionally, generating the attention mask matrix based on the physical neighborhood set includes: Define an attention mask matrix, whose rows and columns correspond to the surface grid points of the vehicle body, and are used to mask non-neighborhood grid points in the subsequent attention weight calculation; For any two surface grid points, if the latter belongs to the physical neighborhood set of the former, then the corresponding matrix element is set to 1; otherwise, it is set to 0.

[0010] Optionally, the application of the attention mask matrix to limit the attention calculation scope to the physical neighborhood set includes: For each surface grid point, obtain the initial attention score between that surface grid point and all other surface grid points; Obtain the matrix element values ​​of the attention mask matrix of the surface grid point that correspond to other surface grid points; If the value of the matrix element is 0, the initial attention score corresponding to the other surface grid point is replaced with an invalid score, which will cause the corresponding attention weight to be zero in the subsequent normalization process; If the value of the matrix element is 1, then the initial attention score corresponding to the other surface grid point is retained; After replacement, all attention scores are normalized to obtain the attention weights of each other surface grid point relative to the current surface grid point, where the attention weight corresponding to a matrix element value of 0 is zero.

[0011] Optionally, the step of calculating the attention output of the surface grid point by combining a physical spatial bias function based on spatial distance includes: The query vector of the surface grid point is multiplied by the key vectors of other surface grid points in the physical neighborhood set and then scaled to obtain the feature similarity score. The physical space offset score is obtained by multiplying the negative of the square of the spatial distance between the surface grid point and other surface grid points by the offset coefficient. The attention weights are obtained by summing the feature similarity scores, the physical space bias scores, and the logarithmic values ​​of the attention mask matrix and then applying a normalized exponential function. The attention output of the surface grid point is obtained by weighting and summing the value vectors of each grid point in the physical neighborhood set using the attention weights.

[0012] Optionally, the step of outputting the vehicle body surface pressure field based on the attention output of each surface grid point includes: The attention output of each surface grid point is input into the fully connected layer, and the pressure prediction value of that surface grid point is obtained by mapping. The predicted pressure values ​​of all surface grid points are organized according to their spatial coordinates to form a pressure distribution field on the vehicle surface, which is then used as the final output pressure field on the vehicle surface.

[0013] Secondly, embodiments of this application provide a device for predicting the pressure field on the surface of an automobile body, the device comprising: The surface grid point acquisition module is used to acquire surface grid points on the surface of the car body. Each surface grid point is treated as a word, and the feature vector of each word contains the spatial coordinates and physical attribute parameters of the surface grid point. The physical influence domain radius calculation module is used to calculate the physical influence domain radius of each surface grid point based on its local geometric features and physical field information. The physical neighborhood set construction module is used to construct a physical neighborhood set of the surface grid point based on the physical influence domain radius. The physical neighborhood set includes other surface grid points whose spatial distance from the surface grid point satisfies the constraint of the physical influence domain radius. An attention mask matrix generation module is used to generate an attention mask matrix based on the physical neighborhood set; The attention output calculation module is used to apply the attention mask matrix to limit the attention calculation range to the physical neighborhood set, and combine it with the physical spatial bias function based on spatial distance to calculate the attention output of the surface grid point. The vehicle body surface pressure field output module is used to output the vehicle body surface pressure field based on the attention output of each surface grid point.

[0014] Optionally, calculating the physical influence radius of the surface grid point based on its local geometric features and physical field information includes: Local curvature is extracted from the lexical feature vector of the surface grid points as local geometric features, pressure field gradient is extracted as physical field information, and global geometric scale parameters of the vehicle body and incoming flow conditions are obtained. The local curvature, the pressure field gradient, the global geometric scale parameter, and the incoming flow condition are weighted and summed, and the summation result is mapped to a normalization coefficient between 0 and 1 through the Sigmoid function. The physical influence radius of the surface grid point is calculated by linear interpolation between the preset minimum and maximum neighborhood radii using the normalization coefficient.

[0015] Optionally, constructing the physical neighborhood set of the surface grid point based on the physical influence domain radius includes: Using the three-dimensional spatial coordinates of the surface grid point as the center of a sphere and the radius of the physical influence domain as the radius, other surface grid points whose spatial distance is not greater than the radius are searched among all grid points on the vehicle surface; The other surface grid points retrieved constitute the physical neighborhood set of this surface grid point.

[0016] Optionally, generating the attention mask matrix based on the physical neighborhood set includes: Define an attention mask matrix, whose rows and columns correspond to the surface grid points of the vehicle body, and are used to mask non-neighborhood grid points in the subsequent attention weight calculation; For any two surface grid points, if the latter belongs to the physical neighborhood set of the former, then the corresponding matrix element is set to 1; otherwise, it is set to 0.

[0017] Optionally, the application of the attention mask matrix to limit the attention calculation scope to the physical neighborhood set includes: For each surface grid point, obtain the initial attention score between that surface grid point and all other surface grid points; Obtain the matrix element values ​​of the attention mask matrix of the surface grid point that correspond to other surface grid points; If the value of the matrix element is 0, the initial attention score corresponding to the other surface grid point is replaced with an invalid score, which will cause the corresponding attention weight to be zero in the subsequent normalization process; If the value of the matrix element is 1, then the initial attention score corresponding to the other surface grid point is retained; After replacement, all attention scores are normalized to obtain the attention weights of each other surface grid point relative to the current surface grid point, where the attention weight corresponding to a matrix element value of 0 is zero.

[0018] Optionally, the step of calculating the attention output of the surface grid point by combining a physical spatial bias function based on spatial distance includes: The query vector of the surface grid point is multiplied by the key vectors of other surface grid points in the physical neighborhood set and then scaled to obtain the feature similarity score. The physical space offset score is obtained by multiplying the negative of the square of the spatial distance between the surface grid point and other surface grid points by the offset coefficient. The attention weights are obtained by summing the feature similarity scores, the physical space bias scores, and the logarithmic values ​​of the attention mask matrix and then applying a normalized exponential function. The attention output of the surface grid point is obtained by weighting and summing the value vectors of each grid point in the physical neighborhood set using the attention weights.

[0019] Optionally, the step of outputting the vehicle body surface pressure field based on the attention output of each surface grid point includes: The attention output of each surface grid point is input into the fully connected layer, and the pressure prediction value of that surface grid point is obtained by mapping. The predicted pressure values ​​of all surface grid points are organized according to their spatial coordinates to form a pressure distribution field on the vehicle surface, which is then used as the final output pressure field on the vehicle surface.

[0020] Thirdly, embodiments of this application provide a computer device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the automobile body surface pressure field prediction method described in any of the optional embodiments of the first aspect are performed.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the automobile body surface pressure field prediction method described in any of the optional embodiments of the first aspect.

[0022] The technical solution provided in this application includes, but is not limited to, the following beneficial effects: By uniformly setting each grid point on the surface of the car body as a word and constructing a feature vector containing spatial coordinates and physical attribute parameters, the data expression form of various grid points can be standardized, and the basic physical and spatial information corresponding to a single point can be fully collected, providing standard and complete basic data support for subsequent calculations.

[0023] By calculating the radius of the physical influence domain based on the local geometric features of the surface grid points and the corresponding physical field information, the appropriate range of influence can be determined according to the actual state of different positions on the vehicle body, so that the size of the divided influence domain fits the vehicle body structure and the actual flow situation in the field.

[0024] By constructing a corresponding physical neighborhood set based on the calculated physical influence domain radius, other grid points that are actually related to the target grid point can be accurately screened out, effectively reducing the range of data involved in subsequent calculations, reducing the participation of invalid data in calculations, and making the division of the relationship between points more in line with the actual situation.

[0025] An attention mask matrix is ​​generated based on the established physical neighborhood set. This matrix can clearly distinguish between effective and irrelevant points in a standardized form, thereby achieving a unified and regular definition of the relationships between grid points and providing a clear and executable basis for subsequent calculations.

[0026] By defining the effective range of attention computation through an attention mask matrix and combining it with a physical space bias function based on spatial distance to solve for the attention output corresponding to the grid points, we can not only strictly lock a reasonable computation range, but also reflect the degree of correlation between different grid points based on spatial distance differences, thereby further optimizing the actual effect of feature fusion.

[0027] The attention output results obtained by solving all surface grid points are used to complete the output of the vehicle body surface pressure field. It can rely on the effective feature information obtained by multi-level regularization and accurate calculation in the early stage to smoothly complete the output of the target physical field results, ensuring that the final output of the vehicle body surface pressure field results conforms to the actual use conditions.

[0028] In summary, this application completes the entire process of grid point feature construction, influence domain radius calculation, neighborhood set construction, mask matrix generation, attention output solution, and pressure field output in sequence. Each operation step is connected and cooperates with each other, which can steadily complete the complete prediction of the pressure field on the surface of the car body, effectively improve the orderliness of the overall prediction process, and at the same time ensure the actual fit of the pressure field prediction results.

[0029] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart of a method for predicting the pressure field on the surface of an automobile body, provided in Embodiment 1 of this application, is shown. Figure 2 A flowchart of a method for calculating the radius of a physical influence domain provided in Embodiment 1 of this application is shown; Figure 3 A flowchart of a method for constructing a physical neighborhood set provided in Embodiment 1 of this application is shown; Figure 4 A flowchart of an attention mask matrix generation method provided in Embodiment 1 of this application is shown; Figure 5 A flowchart of an attention calculation range limitation method provided in Embodiment 1 of this application is shown; Figure 6 A flowchart of an attention output calculation method provided in Embodiment 1 of this application is shown; Figure 7 A flowchart of a method for outputting a pressure field on a vehicle surface provided in Embodiment 1 of this application is shown; Figure 8 This paper shows a schematic diagram of the structure of a pressure field prediction device for the surface of an automobile body provided in Embodiment 2 of this application; Figure 9 A schematic diagram of the structure of a computer device provided in Embodiment 3 of this application is shown. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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 this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0033] Example 1 To facilitate understanding of this application, the following is combined with... Figure 1 The flowchart illustrating the method for predicting the pressure field on the surface of an automobile body provided in Embodiment 1 of this application will be used to describe Embodiment 1 of this application in detail.

[0034] See Figure 1 As shown, Figure 1 A flowchart of a method for predicting the pressure field on the surface of a car body, as provided in Embodiment 1 of this application, is shown, wherein the method includes steps S101 to S106: S101: Obtain the surface grid points on the car body surface, treat each surface grid point as a word, and the feature vector of each word contains the spatial coordinates and physical attribute parameters of the surface grid point.

[0035] Specifically, in predicting the pressure field on the surface of a car body, the three-dimensional curved surface of the car body needs to be discretized into a grid. Each grid point corresponds to a physical location on the surface of the car body and is directly associated with the pressure value to be predicted at that location.

[0036] To accurately characterize the physical state of each point on the vehicle body surface, each grid point on the vehicle body surface is... The mapping is to tokens, and its feature vector formula is: .

[0037] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface The lexical feature vector integrates all the physical information affecting surface pressure at that point; : Grid points on the vehicle body surface The three-dimensional spatial coordinates are expressed as follows: Locating the position on the vehicle body surface is the basis for determining the range of influence of local pressure; : Grid points on the vehicle body surface The normal vector of the surface reflects the local orientation of the vehicle body surface, determines the airflow impact angle, and directly affects the surface pressure at that point. : Grid points on the vehicle body surface The curvature of the surface area describes the degree of unevenness of the car body surface. The pressure changes drastically in high curvature areas such as the A-pillar, rearview mirror, and rear of the car, while the pressure is evenly distributed in flat areas. The incoming flow conditions for a car's movement, including the flow velocity, direction, and environmental pressure, are global physical parameters that determine the overall surface pressure level of the car body.

[0038] S102: For each surface grid point, calculate the radius of the physical influence domain of that surface grid point based on its local geometric features and physical field information.

[0039] Specifically, the pressure on the vehicle body surface has local propagation characteristics: the pressure at a certain grid point is only affected by a limited range of surrounding grid points. Areas with high curvature and large pressure gradients have a small influence range, while flat areas have a large influence range. It is necessary to dynamically determine the radius of the pressure influence domain of each grid point.

[0040] The formula for calculating the physical influence domain radius (i.e., the range of pressure influence at vehicle body mesh points) is as follows: .

[0041] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface The radius of the pressure influence domain characterizes the spatial extent in which the pressure at that point is affected by the surrounding grid points. The minimum radius of the pressure influence domain of the vehicle body grid points is preset to 0.05m, which corresponds to the minimum influence range of high curvature areas such as A-pillars and rearview mirrors. The maximum radius of the pressure influence area of ​​the vehicle body grid points is preset to 0.2m, which corresponds to the maximum influence range of flat areas such as the side and roof of the vehicle body. Sigmoid (S-type activation function) maps the fusion results of multiple physical parameters of the vehicle body to [0,1], achieving a smooth transition of the radius of the pressure influence domain; The weight of the influence range of the vehicle body surface curvature on the pressure is preset to 0.3. The greater the curvature, the stronger the weighting effect, thus reducing the influence range. : Grid points on the vehicle body surface The local curvature and local geometric features directly reflect the constraints of the vehicle body surface irregularities on the range of pressure influence. The weight of the pressure gradient on the vehicle body surface on the range of pressure influence is preset to 0.4. The larger the pressure gradient, the stronger the weighting effect, thus reducing the influence domain. : Grid points on the vehicle body surface The pressure field gradient, a physical field information, characterizes the degree of drastic pressure change near that point; the larger the gradient, the smaller the pressure influence range. The weight of the vehicle's global geometric dimensions on the range of pressure influence is preset to 0.15, adapting to different vehicle sizes. : Global geometric dimension parameters of the vehicle body, with a value of 4.8m (typical sedan length), relating to the matching relationship between the overall dimensions of the vehicle body and the range of pressure influence; The weight of the impact of oncoming traffic conditions on the pressure range is preset to 0.15, and the higher the vehicle speed, the stronger the weighting effect. The incoming flow conditions for vehicle operation are global physical parameters that reflect the regulatory effect of driving conditions on the range of vehicle body pressure.

[0042] The radius is adapted to the characteristics of different areas of the vehicle body: the radius is larger for flat areas such as the roof and doors to integrate pressure-related information over a wider range; the radius is smaller for high-curvature areas such as the A-pillar, rearview mirror, and rear of the vehicle to accurately capture local pressure changes.

[0043] S103: Based on the physical influence domain radius, construct a physical neighborhood set for the surface grid point. The physical neighborhood set includes other surface grid points whose spatial distance from the surface grid point satisfies the constraint of the physical influence domain radius.

[0044] Specifically, the surface pressures on the vehicle body are only correlated within a limited neighborhood, which is the set of vehicle body grid points. The "pressure-related mesh set" refers to the set where the physical state of the mesh points directly affects the physical state of the mesh points. Surface pressure at a point.

[0045] Formula for the pressure-related neighborhood set of vehicle body mesh points: .

[0046] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface The pressure-related neighborhood set contains all pairs Grid points affected by surface pressure; : Index of any grid point in the global grid on the vehicle body surface; : Global mesh computation domain on the vehicle body surface, covering all surface mesh points of the vehicle body to be predicted under all pressures; : Grid points on the vehicle body surface The three-dimensional spatial position vector; : Grid points on the vehicle body surface The three-dimensional spatial position vector; : Two-dimensional Euclidean distance, used to calculate the spatial straight-line distance between grid points on two surfaces of a vehicle body; : Grid points on the vehicle body surface The radius of the pressure influence domain constrains the spatial range of the neighborhood set.

[0047] The car body surface is a 200,000-level unstructured triangular mesh, using a KD-tree (K-Dimensional Tree) spatial index structure, which can quickly retrieve pressure-related mesh points on the car body surface that meet distance constraints, avoiding the high time consumption of global traversal and adapting to the scale of industrial-grade car body meshes.

[0048] The neighborhood set is dynamically adjusted according to the radius of the vehicle body pressure influence domain: the number of neighborhood grids in high curvature areas is small, which can accurately capture local pressure changes; the number of neighborhood grids in flat areas is large, which can ensure the continuity of pressure distribution on the vehicle body surface.

[0049] S104: Generate an attention mask matrix based on the physical neighborhood set.

[0050] Specifically, there is no cross-regional pressure correlation between grid points in different areas of the vehicle body. For example, the pressure of grid points at the rear of the vehicle will not affect grid points at the front of the vehicle. The mask matrix is ​​used to shield such invalid and non-physical pressure correlations, and only retain the valid pressure correlations within the neighborhood.

[0051] Transforming the pressure-related constraints of vehicle body mesh points into computational constraints, the attention mask matrix formula is as follows: .

[0052] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots (Target point) and The elements of the attention mask matrix (for related points) represent Is the point correct? Point surface pressure has an effective effect; : Index of target grid points on the vehicle body surface, corresponding to the grid points of the pressure to be predicted; : Index of grid points on the vehicle body surface whose pressure correlation needs to be determined; : Grid points on the vehicle body surface The pressure-related neighborhood set; Body mesh dots Not belonging to The point pressure is associated with the neighborhood, but there is no effective pressure association.

[0053] This mask matrix achieves precise constraints on the vehicle body pressure correlation range: when a matrix element is 1, it retains... Point to point The pressure correlation of the point is calculated; when it is 0, the invalid correlation is shielded to avoid non-physical pressure interference and conform to the local pressure propagation law of the vehicle body.

[0054] S105: Apply the attention mask matrix to limit the attention calculation range to the physical neighborhood set, and combine it with the physical spatial bias function based on spatial distance to calculate the attention output of the surface grid point.

[0055] Specifically, the pressure correlation strength on the vehicle body surface decreases with increasing spatial distance: the pressure correlation between adjacent grid points is strong, and the correlation weakens as the distance increases. This distance attenuation characteristic needs to be quantified by an offset function to conform to the physical law of vehicle body pressure propagation.

[0056] The physical space bias function formula for quantifying the pressure correlation strength between mesh points on the vehicle body is: .

[0057] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots and The pressure correlation distance offset value represents the attenuation effect of spatial distance on the pressure correlation strength; : Grid points on the vehicle body surface The three-dimensional spatial position vector; : Grid points on the vehicle body surface The three-dimensional spatial position vector; Pressure correlation distance attenuation coefficient, with a value of 0.5~1.5, controls how quickly the distance between grid points on the vehicle body affects the attenuation of pressure correlation strength; The square of the spatial distance between two grid points on the vehicle body. The larger the distance, the smaller the offset value and the weaker the pressure correlation strength.

[0058] The bias function fits the physical characteristics of vehicle body pressure: the smaller the distance between adjacent grid points on the vehicle body, the larger the bias value and the higher the pressure correlation weight; the farther the distance, the smaller the bias value and the lower the weight, which conforms to the local dominance of vehicle body pressure propagation law.

[0059] By combining an attention mask matrix, invalid cross-regional pressure correlations of the vehicle body can be masked, transforming global attention calculation into local pressure correlation calculation, thus reducing computational complexity from... Down to ( The number of neighborhoods associated with the average pressure at each grid point on the vehicle body is much smaller than the total number of grids. ), adapted for efficient calculation of vehicle body mesh at the 200,000 level.

[0060] S106: Output the pressure field on the vehicle surface based on the attention output of each surface grid point.

[0061] Specifically, the pressure field on the vehicle body surface is a spatial distribution set of pressure values ​​at all surface grid points. It is necessary to integrate the features of each grid point with local pressure correlation information and map them to the actual surface pressure value to finally restore the complete pressure distribution of the vehicle body.

[0062] By integrating the pressure prediction values ​​of all grid points on the vehicle body, it can replace the traditional CFD (Computational Fluid Dynamics) steady-state solution process and directly output the pressure distribution on the vehicle body surface. It is suitable for engineering scenarios such as automotive aerodynamic design, drag coefficient optimization, and rapid aerodynamic performance evaluation.

[0063] The output pressure field on the vehicle body surface is continuous and smooth, without non-physical phenomena such as sudden changes in flow velocity or pressure discontinuity. The surface Cp (Pressure Coefficient) error is ≤10%, and the error in high curvature areas such as A-pillars and rearview mirrors is ≤3%, meeting the engineering precision requirements of automotive aerodynamic design.

[0064] Modeling is based solely on the vehicle body surface mesh, eliminating the need for volume mesh in the calculation. This reduces the 3D PDE (Partial Differential Equation) problem to a 2.5D surface modeling problem, reducing memory usage by 1-2 orders of magnitude and significantly improving the efficiency of vehicle body pressure prediction.

[0065] In an optional implementation, see Figure 2 As shown, Figure 2The flowchart illustrates a method for calculating the physical influence domain radius provided in Embodiment 1 of this application. The method for calculating the physical influence domain radius of a surface grid point based on its local geometric features and physical field information includes steps S201-S203: S201: Extract local curvature as local geometric features from the lexical feature vectors of the surface grid points, extract pressure field gradient as physical field information, and obtain global geometric scale parameters of the vehicle body and incoming flow conditions.

[0066] Specifically, the feature vector of the vehicle body grid points contains all physical quantities that affect the pressure influence domain. Core parameters need to be extracted for dynamically calculating the radius of the pressure influence domain to ensure that the radius fits the actual pressure distribution characteristics of the vehicle body.

[0067] The formula for the word feature vector of the vehicle body grid points is as follows: , where local curvature , inflow conditions The pressure field gradient is an inherent physical characteristic of the lexical unit. As a core indicator of pressure change on the vehicle body surface, global geometric parameters The vehicle body length is a necessary parameter for calculating the radius of the pressure influence zone.

[0068] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface lexical feature vectors; : Grid points on the vehicle body surface The local curvature and local geometric features reflect the degree of unevenness on the vehicle body surface; : Grid points on the vehicle body surface The pressure field gradient, physical field information, characterizes the degree of drastic change in local pressure. : Global geometric parameters of the vehicle body, which are related to the overall dimensions of the vehicle body; The incoming flow conditions for vehicle movement are global physical parameters that reflect the driving conditions.

[0069] S202: The local curvature, the pressure field gradient, the global geometric scale parameter, and the incoming flow condition are weighted and summed, and the summation result is mapped to a normalization coefficient between 0 and 1 through the Sigmoid function.

[0070] Specifically, the size of the pressure influence domain is determined by the curvature of the vehicle body, the pressure gradient, the vehicle body size, and the driving conditions. These physical quantities need to be weighted and fused, normalized, and then used for radius interpolation to adapt to different areas of the vehicle body.

[0071] The weighted summation formula for multiple physical parameters of the vehicle body is: The normalized mapping formula is: .

[0072] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body curvature weight parameter, preset 0.3; : Grid points on the vehicle body surface Local curvature; Vehicle body pressure gradient weight parameter, preset 0.4; : Grid points on the vehicle body surface The pressure field gradient; The global scale weight parameter for the vehicle body is preset to 0.15. Global geometric parameters of the vehicle body; The weighting parameter for the oncoming traffic conditions is preset to 0.15. : Traffic flow conditions for vehicle movement; The Sigmoid function outputs a normalization coefficient in the range [0,1]. The larger the value, the larger the radius of the pressure influence field in the smooth area of ​​the vehicle body.

[0073] The weighting parameters can be fine-tuned according to different vehicle models and driving conditions to ensure that the fusion result conforms to the actual pressure distribution pattern of the vehicle body.

[0074] S203: The physical influence radius of the surface grid point is calculated by linear interpolation between the preset minimum and maximum neighborhood radii using the normalization coefficient.

[0075] Specifically, by interpolating with normalization coefficients, the fusion results of the vehicle body's physical parameters are transformed into the actual pressure influence domain radius, enabling the radius to be adaptively adjusted according to the local characteristics of the vehicle body.

[0076] The linear interpolation formula for the radius of influence of vehicle body pressure is: .

[0077] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface The radius of the pressure influence zone; Minimum radius of the vehicle body pressure influence zone, preset 0.05m; Maximum radius of the vehicle body pressure influence zone, preset 0.2m; The normalized coefficients output by the Sigmoid function; The definitions of the remaining characters are the same as in formula S202.

[0078] The interpolation logic adapts to different areas of the vehicle body: flat areas such as the roof and doors have a large normalization coefficient and a large radius; high curvature areas such as the A-pillar and rearview mirrors have a small normalization coefficient and a small radius, accurately matching the local pressure change characteristics of the vehicle body.

[0079] In an optional implementation, see Figure 3 As shown, Figure 3 The flowchart illustrates a method for constructing a physical neighborhood set according to Embodiment 1 of this application, wherein the step of constructing the physical neighborhood set of the surface grid points based on the physical influence domain radius includes steps S301-S302: S301: Using the three-dimensional spatial coordinates of the surface grid point as the center of the sphere and the radius of the physical influence domain as the radius, search for other surface grid points whose spatial distance is not greater than the radius among all grid points on the vehicle surface.

[0080] Specifically, the vehicle body pressure correlation neighborhood is centered on the target grid point and the range is the radius of the pressure influence domain. It is searched in the grid on the vehicle body surface to ensure that all grid points in the neighborhood are valid correlation points that have an impact on the pressure of the target point.

[0081] Using the three-dimensional spatial coordinates of the vehicle body grid points Radius of the pressure influence zone, centered on the sphere. To determine the retrieval radius, a KD-tree spatial index structure is used to traverse the global grid points on the vehicle surface.

[0082] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface The three-dimensional spatial coordinates (searching for the center of the sphere); : Grid points on the vehicle body surface X-axis coordinate; : Grid points on the vehicle body surface Y-axis coordinate; : Grid points on the vehicle body surface Z-axis coordinate; : Grid points on the vehicle body surface The radius of the pressure influence domain (retrieval radius).

[0083] KD-tree adapts to unstructured surface meshes of vehicle bodies at the 200,000 level, quickly filtering vehicle body surfaces that meet distance constraints. The grid points efficiently locate pressure-related neighborhoods, avoiding the high time consumption of global traversal.

[0084] S302: Construct the physical neighborhood set of the other retrieved surface grid points.

[0085] Specifically, the retrieved vehicle body mesh points are all associated points that have an effective impact on the pressure of the target point. After integration, they form a pressure-associated neighborhood set, providing effective pressure association information for subsequent attention calculations.

[0086] Formula for the neighborhood set of vehicle body pressure correlation: The retrieved vehicle body grid points Form a set .

[0087] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: : Grid points on the vehicle body surface The pressure-related neighborhood set; : Index of retrieved vehicle body pressure-related grid points; : Global mesh computation domain of the vehicle body surface; : Grid points on the vehicle body surface The three-dimensional spatial position vector; : Retrieved vehicle body grid points The three-dimensional spatial position vector; Two-dimensional Euclidean distance; : Grid points on the vehicle body surface The radius of the pressure influence zone.

[0088] The neighborhood set dynamically adapts to the vehicle body pressure distribution: the neighborhood size is small in high curvature areas to focus on local pressure abrupt changes; the neighborhood size is large in flat areas to ensure the continuity of the pressure distribution on the vehicle body surface.

[0089] In an optional implementation, see Figure 4 As shown, Figure 4 The flowchart illustrates an attention mask matrix generation method provided in Embodiment 1 of this application, wherein the step of generating the attention mask matrix based on the physical neighborhood set includes steps S401-S402: S401: Define an attention mask matrix. The rows and columns of this attention mask matrix correspond to the surface grid points of the vehicle body. It is used to mask non-neighborhood grid points in the subsequent attention weight calculation.

[0090] Specifically, the dimensions of the mask matrix are matched with the total number of grids on the vehicle body surface. Each row corresponds to a target pressure prediction grid point, and each column corresponds to a grid point whose pressure correlation needs to be judged, thereby achieving accurate screening of global pressure correlations on the vehicle body.

[0091] The attention mask matrix has a dimension of ,in The total number of grid points on the vehicle body surface, with the row index corresponding to the target grid point on the vehicle body. The column index corresponds to the grid point on the vehicle body that needs to be judged. .

[0092] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: Total number of grid points on the vehicle body surface; : Mask matrix row index (vehicle target pressure prediction grid point); : Mask matrix column index (grid points on the vehicle body to be judged related to pressure).

[0093] The core function of the mask matrix is ​​to filter out cross-regional and non-physical pressure correlations on the vehicle body, strictly limit attention calculations to the neighborhood of pressure correlations, reduce invalid calculations, and conform to the local propagation law of vehicle body pressure.

[0094] S402: For any two surface grid points, if the latter belongs to the physical neighborhood set of the former, then set the corresponding matrix element to 1; otherwise, set it to 0.

[0095] Specifically, the binary value rule directly maps the pressure correlation between vehicle body grid points, with 1 representing a valid correlation and 0 representing an invalid correlation, thus concisely constraining the range of pressure correlation.

[0096] Attention mask matrix formula: .

[0097] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots (line) and (column) mask matrix elements; : Target pressure prediction grid points on vehicle body; : Grid points on the vehicle body to be determined regarding pressure correlation; Body mesh dots The pressure-related neighborhood set; 1: Body mesh dots belong The pressure-related neighborhood is preserved, and pressure-related calculations are retained. 0: Body mesh points Not belonging to Pressure-related neighborhood, shielding pressure-related calculations; Body mesh dots Not belonging to All cases in the pressure-related neighborhood.

[0098] The binary value selection rule provides clear constraints for subsequent attention calculations, ensuring that only grid points in the neighborhood of the vehicle body pressure are involved in the weight calculation, thus eliminating interference from non-physical pressure.

[0099] In an optional implementation, see Figure 5 As shown, Figure 5 The flowchart illustrates a method for limiting the attention calculation range according to Embodiment 1 of this application, wherein applying the attention mask matrix to limit the attention calculation range to the physical neighborhood set includes steps S501-S505: S501: For each surface grid point, obtain the initial attention score between that surface grid point and all other surface grid points.

[0100] Specifically, the initial attention score reflects the similarity of the lexical features of two grid points on the vehicle body. The lexical features include pressure-related physical quantities such as coordinates, normals, curvature, and incoming flow. The higher the similarity, the higher the feature matching degree of the pressure association between the two grid points.

[0101] The initial attention score is obtained by scaling the dot product of the Query and Key points of the vehicle body mesh, as shown in the formula: .

[0102] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots The query vector is composed of lexical features. Linear projection is used to obtain the characterization Point pressure-related characteristics; Body mesh dots The key vector is composed of lexical features. Linear projection is used to obtain the characterization Point pressure-related characteristics; : Matrix transpose symbol; : Query the feature dimensions of the key vector and the query vector to uniformly represent the dimensions of vehicle body pressure-related features; The square root of the feature dimension is used to scale the dot product result, avoiding excessive scores due to an increase in the vehicle body pressure feature dimension.

[0103] The initial score is a quantification of the matching degree of pressure association features of the vehicle body grid points, providing a basic score for subsequent calculation of pressure association weights.

[0104] S502: Obtain the matrix element values ​​of the attention mask matrix of the surface grid point that correspond to other surface grid points.

[0105] Specifically, the vehicle body mesh points are quickly obtained using an index mask matrix. Does it belong to The pressure correlation neighborhood of a point provides a basis for subsequent determination of whether to retain the pressure correlation calculation of that point.

[0106] Through the index mask matrix Obtain vehicle body mesh points Corresponding rows, grid points The element value (0 or 1) of the corresponding column.

[0107] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: Body mesh dots and mask matrix elements; : Grid points for predicting the current target pressure on the vehicle body; Other grid points on the vehicle body whose pressure correlation needs to be determined; 0 / 1: Binary values ​​of the mask matrix, representing Does the point belong to Point pressure is related to the neighborhood.

[0108] Indexing operations efficiently locate the pressure correlation between vehicle body grid points and quickly filter out valid pressure-correlated grid points.

[0109] S503: If the value of the matrix element is 0, the initial attention score corresponding to the other surface grid point is replaced with an invalid score. This invalid score will cause the corresponding attention weight to be zero in the subsequent normalization process.

[0110] Specifically, a mask matrix element of 0 represents a vehicle body mesh point. and If there is no effective pressure correlation, set its score to an invalid value to ensure that the point does not participate in the pressure correlation weight calculation and to shield non-physical interference.

[0111] Invalid scores are set to negative infinity. The normalization exponent (normalization exponent, normalization exponent function) becomes 0 after normalization to negative infinity, completely shielding the pressure correlation influence of non-neighbor grid points on the vehicle body.

[0112] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: 0: Mask matrix element value, representing the vehicle body mesh points. Not belonging to Point pressure is related to the neighborhood; Invalid scores, with values ​​approaching negative infinity, indicate no valid stress correlation; Normalized index: The normalized index function maps the score to stress-related weights in the interval [0,1].

[0113] This processing logic ensures that only grid points within the effective pressure-related neighborhood of the vehicle body participate in weight allocation, eliminating cross-regional non-physical pressure interference.

[0114] S504: If the value of this matrix element is 1, then retain the initial attention score corresponding to the other surface grid point.

[0115] Specifically, an element of 1 in the mask matrix represents a mesh point on the vehicle body. and If there is a valid stress correlation, retain its initial score for subsequent calculation of stress correlation weights.

[0116] A value of 1 in the mask matrix represents a mesh point on the vehicle body. belong The pressure-related neighborhood set of a point, its initial attention score Keep it as is, without making any changes.

[0117] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: 1: Mask matrix element values, representing vehicle body mesh points. belong Point pressure is related to the neighborhood; : The initial attention score is retained, and the definition of each character is the same as in the S501 formula.

[0118] Only the scores of grid points in the neighborhood related to the effective vehicle body pressure are retained, and the attention calculation is moved from the global scope. Complexity reduced to local Complexity is reduced, and the system is adapted for efficient calculation of vehicle body meshes up to 200,000 levels.

[0119] S505: Normalize all attention scores after replacement processing to obtain the attention weights of each other surface grid point relative to the surface grid point, where the attention weight corresponding to a matrix element value of 0 is zero.

[0120] Specifically, the normalized weights represent the proportion of each grid point in the vehicle's neighborhood to the pressure at the target point. The larger the weight, the stronger the influence of that point on the pressure at the target point, which is consistent with the local contribution law of vehicle pressure.

[0121] Score normalization is performed using a normalized exponential function, and the attention weight formula is as follows: .

[0122] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots right Stress-related attention weights; : Normalization function, outputs the pressure correlation weights in the interval [0,1]; Initial attention score, representing the degree of matching of stress features; : Physical space bias function, characterizing the attenuation of the pressure correlation by distance; Logarithmic value of the mask matrix At that time (Blocking invalid associations) When the value is 0 (valid associations are retained); The definitions of the remaining characters are the same as those in the formula above.

[0123] After normalization, the sum of the weights is 1, which accurately quantifies the contribution of each grid point in the neighborhood of the vehicle body to the pressure of the target point. The weight directly reflects the strength of the local pressure influence.

[0124] In an optional implementation, see Figure 6 As shown, Figure 6 The flowchart of an attention output calculation method provided in Embodiment 1 of this application is shown, wherein the step of calculating the attention output of the surface grid point by combining a physical spatial bias function based on spatial distance includes steps S601 to S604: S601: Perform a dot product between the query vector of the surface grid point and the key vector of other surface grid points in the physical neighborhood set, and then scale the result to obtain the feature similarity score.

[0125] Specifically, only the feature similarity of grid points in the neighborhood of vehicle body pressure is calculated to avoid global redundant calculation. The higher the similarity, the higher the matching degree of pressure-related features between the two grid points and the stronger the pressure correlation.

[0126] Feature similarity score formula: Only calculate the neighborhood set associated with vehicle body pressure. Inner grid points The score.

[0127] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots The query vector; Vehicle body pressure is related to the grid points in the neighborhood. The key vector; : Matrix transpose; : Stress-related feature dimensions; : The square root of the feature dimension; Body mesh dots The pressure-related neighborhood set.

[0128] This score focuses on the effective pressure correlation range of the vehicle body, reduces invalid calculations, and is suitable for high-resolution mesh scenarios of industrial-grade vehicle bodies.

[0129] S602: Calculate the negative of the square of the spatial distance between this surface grid point and other surface grid points, multiply it by the offset coefficient, and obtain the physical space offset score.

[0130] Specifically, the bias score quantifies the attenuation of the pressure correlation strength due to the spatial distance between grid points on the vehicle body. The closer the distance, the higher the score and the stronger the pressure correlation, which is consistent with the physical law of local pressure propagation on the vehicle body.

[0131] Physical space bias score formula: .

[0132] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots and Pressure-related distance bias score; Pressure-related distance attenuation coefficient, with a value ranging from 0.5 to 1.5; Body mesh dots The three-dimensional spatial position vector; Vehicle body pressure is related to the grid points in the neighborhood. The three-dimensional spatial position vector; : The square of the spatial distance between two grid points on the vehicle body.

[0133] The bias score decreases as the distance between grid points on the vehicle body increases, which strengthens the pressure correlation weight of grid points with close proximity, consistent with the local dominance of vehicle body pressure propagation characteristics.

[0134] S603: The feature similarity score, the physical space bias score, and the logarithm of the attention mask matrix are added together, and the attention weights are obtained by normalizing the exponential function.

[0135] Specifically, by integrating three factors—pressure feature matching degree, distance attenuation, and association effectiveness—the pressure association weights between vehicle body grid points are calculated to ensure that the weights simultaneously conform to feature matching and physical laws.

[0136] Attention weight formula: .

[0137] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots right Stress-related attention weights; : Normalization function; Pressure feature similarity score; Pressure-related distance bias score; Logarithmic value of the mask matrix; Attention mask matrix elements; The definitions of the remaining characters are the same as those in the formula above.

[0138] hour for The corresponding weight is 0, which masks the invalid pressure association of the vehicle body; The time weight is determined by both feature similarity and distance bias, taking into account both feature matching and physical constraints.

[0139] S604: Use the attention weights to perform a weighted summation of the value vectors of each grid point in the physical neighborhood set to obtain the attention output of the surface grid point.

[0140] Specifically, the pressure-related features of all grid points in the vehicle body pressure correlation neighborhood are weighted and summed to output a feature vector that integrates local pressure correlation information, providing accurate features for subsequent mapping to actual pressure values.

[0141] Attention output formula: .

[0142] The characters in the above formula correspond to the physical quantities used for predicting vehicle body pressure, and are defined as follows: Body mesh dots Attention output after fusing local pressure correlation information; : Summation symbol, integrates the pressure correlation characteristics of all grid points in the neighborhood; The summation range is the vehicle body mesh points. The pressure-related neighborhood set; Body mesh dots right Stress-related attention weights; Vehicle body pressure is related to the grid points in the neighborhood. The value vector is composed of lexical features. Linear projection is used to obtain the characterization Point pressure-related characteristics.

[0143] The weighted summation result integrates local pressure correlation information of the vehicle body, which not only preserves the similarity of pressure features but also conforms to the spatial distance attenuation law, providing high-dimensional feature support for accurate prediction of vehicle body surface pressure.

[0144] In an optional implementation, see Figure 7 As shown, Figure 7 The flowchart illustrates a method for outputting a vehicle surface pressure field according to Embodiment 1 of this application, wherein the step of outputting the vehicle surface pressure field based on the attention output of each surface grid point includes steps S701-S702: S701: Input the attention output of each surface grid point into the fully connected layer to obtain the pressure prediction value of that surface grid point.

[0145] Specifically, the fully connected layer integrates high-dimensional features of local pressure correlation information and maps them to the actual pressure values ​​of grid points on the vehicle surface, directly outputting the physical quantity to be predicted, which meets the final requirement of vehicle pressure prediction.

[0146] Attention output of vehicle body mesh points is achieved through a fully connected layer. The high-dimensional feature mapping is converted into a one-dimensional pressure value, which is suitable for the task of predicting pressure on the surface of a car body.

[0147] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: Body mesh dots Attention output, integrating local pressure-related information; Fully connected layer: A neural network layer that maps high-dimensional stress features to one-dimensional actual stress values; Predicted pressure values: vehicle body grid points The corresponding actual value of surface pressure.

[0148] The prediction accuracy meets the requirements of automotive aerodynamic design engineering: surface Cp error ≤10%, and error in high curvature areas such as A-pillar and rearview mirror ≤3%, which is better than traditional algorithms.

[0149] S702: Organize the pressure prediction values ​​of all surface grid points into a pressure distribution field on the vehicle surface according to their spatial coordinates, which is the final output pressure field on the vehicle surface.

[0150] Specifically, it integrates the pressure prediction values ​​of all grid points on the vehicle body surface, restores the continuous pressure distribution of the three-dimensional curved surface of the vehicle body, and intuitively presents the high and low pressure areas on the vehicle body surface, thus meeting the needs of aerodynamic design optimization.

[0151] Integrate all vehicle body grid point pressure prediction values, based on spatial coordinates The two-dimensional pressure distribution field on the vehicle body surface can be restored without the need for volume mesh, reducing the 3D PDE problem to a 2.5D surface modeling problem.

[0152] The above parameters correspond to the physical quantities used in predicting vehicle body pressure, and are defined as follows: Body mesh dots Three-dimensional spatial coordinates; Body mesh dots X-axis coordinate; Body mesh dots Y-axis coordinate; Body mesh dots Z-axis coordinate; Pressure distribution field: A continuous and smooth data set consisting of pressure values ​​at all grid points on the vehicle body surface.

[0153] The output pressure field on the vehicle body surface has no non-physical phenomena such as pressure discontinuity or abrupt changes in flow velocity, and can be directly used for automotive aerodynamic shape optimization and wind resistance reduction design, significantly shortening the R&D cycle and reducing testing costs.

[0154] For a clear view of the implementation path of the above effects, please refer to [link / reference]. Figure 8 As shown, Figure 8This document illustrates a flowchart of a vehicle body surface physics field prediction method based on adaptive physical neighborhood constraints, as provided in Embodiment 1 of this application. The flowchart fully presents the entire technical process from basic input to physics field output. The process begins with the "Input Data and Physical Conditions" module, which provides the basic information required for prediction. This module includes two core input types: "Input Conditions: Geometry and Direction," clarifying the vehicle body geometry model and the incoming flow direction, among other working condition information. Simultaneously, the input conditions are connected to the "Physical Marker Construction" module, which transforms discrete grid points on the vehicle body surface into physical marks that the model can process. Each mark's features include "Marker Features: 3D Coordinates (x, y, z), Normal Vector (n), Local Curvature (K), Incoming Flow Condition (v)," comprehensively characterizing the spatial location, local geometric characteristics, and global flow condition information of the grid points.

[0155] The completed physical tags are then fed into the "Adaptive Physical Neighborhood Search and Influence Domain Calculation" module, which is designed with "..." The core computational logic is based on four types of physical constraints: local curvature, physical field gradient, global scale, and flow conditions. It dynamically calculates the physical influence domain radius of each grid point, ultimately generating a "local domain set". ", that is, the effective pressure of each grid point is associated with the set of grid points, which provides physical boundary constraints for subsequent calculations.

[0156] The generated local domain set is connected to the "Neighborhood Search and Influence Domain Calculation Spatial Index" module. This module constructs an "Attention Mask Matrix" based on the neighborhood set, which uses a binary matrix to shield invalid associations of non-neighborhood grid points. On the other hand, it clarifies the association rule of "distance and weight: higher for nearby and lower for distant", that is, the closer the grid point is in space, the higher the weight of its physical influence on the target point, which is in line with the physical law of local propagation of vehicle body pressure.

[0157] Information constrained by spatial indexing enters the "Multi-layer Transformer Operator Learning" module. This module uses "Attention Mask" as the physical constraint and learns through multi-layer Transformer operators (labeled as...). , to Iterative learning is performed to achieve "local and global physical modeling", which not only captures the local correlation features of grid points in the neighborhood, but also integrates the global flow law of the vehicle body to complete the deep modeling of physical field features.

[0158] The final modeling results are connected to the "Output Target Object Physical Field Prediction" module, which outputs the prediction results of the vehicle body surface. The results include two types of outputs: "Output Global Physical Field (such as pressure cloud map) and Global Integral Quantity (such as drag coefficient)". The global physical field is presented in the form of a cloud map to intuitively show the pressure distribution on the vehicle body surface, while the global integral quantity provides aerodynamic performance indicators such as drag coefficient, which directly serve the optimization of automotive aerodynamic design.

[0159] Example 2 See Figure 8 As shown, Figure 8 The diagram shows a structural schematic of a pressure field prediction device for an automobile body surface provided in Embodiment 2 of this application, wherein the device includes: The surface grid point acquisition module 801 is used to acquire surface grid points on the surface of the car body. Each surface grid point is treated as a word, and the feature vector of each word contains the spatial coordinates and physical attribute parameters of the surface grid point. The physical influence domain radius calculation module 802 is used to calculate the physical influence domain radius of each surface grid point based on its local geometric features and physical field information. The physical neighborhood set construction module 803 is used to construct a physical neighborhood set of the surface grid point based on the physical influence domain radius. The physical neighborhood set includes other surface grid points whose spatial distance from the surface grid point satisfies the constraint of the physical influence domain radius. Attention mask matrix generation module 804 is used to generate an attention mask matrix based on the physical neighborhood set; Attention output calculation module 805 is used to apply the attention mask matrix to limit the attention calculation range to the physical neighborhood set, and combine it with the physical space bias function based on spatial distance to calculate the attention output of the surface grid point. The vehicle body surface pressure field output module 806 is used to output the vehicle body surface pressure field based on the attention output of each surface grid point.

[0160] In an optional implementation, calculating the physical influence radius of the surface grid point based on its local geometric features and physical field information includes: Local curvature is extracted from the lexical feature vector of the surface grid points as local geometric features, pressure field gradient is extracted as physical field information, and global geometric scale parameters of the vehicle body and incoming flow conditions are obtained. The local curvature, the pressure field gradient, the global geometric scale parameter, and the incoming flow condition are weighted and summed, and the summation result is mapped to a normalization coefficient between 0 and 1 through the Sigmoid function. The physical influence radius of the surface grid point is calculated by linear interpolation between the preset minimum and maximum neighborhood radii using the normalization coefficient.

[0161] In an optional implementation, constructing the physical neighborhood set of the surface grid points based on the physical influence domain radius includes: Using the three-dimensional spatial coordinates of the surface grid point as the center of a sphere and the radius of the physical influence domain as the radius, other surface grid points whose spatial distance is not greater than the radius are searched among all grid points on the vehicle surface; The other surface grid points retrieved constitute the physical neighborhood set of this surface grid point.

[0162] In an optional implementation, generating the attention mask matrix based on the physical neighborhood set includes: Define an attention mask matrix, whose rows and columns correspond to the surface grid points of the vehicle body, and are used to mask non-neighborhood grid points in the subsequent attention weight calculation; For any two surface grid points, if the latter belongs to the physical neighborhood set of the former, then the corresponding matrix element is set to 1; otherwise, it is set to 0.

[0163] In an optional implementation, applying the attention mask matrix to limit the attention computation to the physical neighborhood set includes: For each surface grid point, obtain the initial attention score between that surface grid point and all other surface grid points; Obtain the matrix element values ​​of the attention mask matrix of the surface grid point that correspond to other surface grid points; If the value of the matrix element is 0, the initial attention score corresponding to the other surface grid point is replaced with an invalid score, which will cause the corresponding attention weight to be zero in the subsequent normalization process; If the value of the matrix element is 1, then the initial attention score corresponding to the other surface grid point is retained; After replacement, all attention scores are normalized to obtain the attention weights of each other surface grid point relative to the current surface grid point, where the attention weight corresponding to a matrix element value of 0 is zero.

[0164] In an optional implementation, calculating the attention output for the surface grid point by combining a physical spatial bias function based on spatial distance includes: The query vector of the surface grid point is multiplied by the key vectors of other surface grid points in the physical neighborhood set and then scaled to obtain the feature similarity score. The physical space offset score is obtained by multiplying the negative of the square of the spatial distance between the surface grid point and other surface grid points by the offset coefficient. The attention weights are obtained by summing the feature similarity scores, the physical space bias scores, and the logarithmic values ​​of the attention mask matrix and then applying a normalized exponential function. The attention output of the surface grid point is obtained by weighting and summing the value vectors of each grid point in the physical neighborhood set using the attention weights.

[0165] In an optional implementation, outputting the vehicle body surface pressure field based on the attention output of each surface grid point includes: The attention output of each surface grid point is input into the fully connected layer, and the pressure prediction value of that surface grid point is obtained by mapping. The predicted pressure values ​​of all surface grid points are organized according to their spatial coordinates to form a pressure distribution field on the vehicle surface, which is then used as the final output pressure field on the vehicle surface.

[0166] Example 3 Based on the same application concept, see [link / reference] Figure 9 As shown, Figure 9 This illustration shows a structural schematic diagram of a computer device provided in Embodiment 3 of this application, wherein, as shown... Figure 9 As shown, the computer device 900 provided in Embodiment 3 of this application includes: The computer device 900 includes a processor 901, a memory 902, and a bus 903. The memory 902 stores machine-readable instructions that can be executed by the processor 901. When the computer device 900 is running, the processor 901 communicates with the memory 902 via the bus 903. When the machine-readable instructions are executed by the processor 901, the steps of the automobile body surface pressure field prediction method shown in Embodiment 1 are performed.

[0167] Example 4 Based on the same concept, this application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the automobile body surface pressure field prediction method described in any of the above embodiments.

[0168] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0169] The computer program product for predicting the pressure field on the surface of a car body provided in this application includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0170] The vehicle body surface pressure field prediction device provided in this application embodiment can be specific hardware on the device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in this application embodiment are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0171] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0172] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0173] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0174] If the aforementioned functions are implemented as 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 this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0175] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0176] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for predicting the pressure field on the surface of a car body, characterized in that, The method includes: Obtain the surface grid points on the car body surface, treat each surface grid point as a word, and the feature vector of each word contains the spatial coordinates and physical attribute parameters of that surface grid point; For each surface grid point, the radius of the physical influence domain of that surface grid point is calculated based on its local geometric features and physical field information. Based on the physical influence domain radius, a physical neighborhood set of the surface grid point is constructed. The physical neighborhood set includes other surface grid points whose spatial distance from the surface grid point satisfies the constraint of the physical influence domain radius. Generate an attention mask matrix based on the physical neighborhood set; The attention calculation range is limited to the physical neighborhood set by applying the attention mask matrix, and the attention output of the surface grid point is calculated by combining the physical spatial bias function based on spatial distance. The pressure field on the vehicle body surface is output based on the attention output of each surface grid point.

2. The method according to claim 1, characterized in that, The calculation of the physical influence domain radius of the surface grid point based on its local geometric features and physical field information includes: Local curvature is extracted from the feature vector of the word at the grid point on the surface as local geometric features, pressure field gradient is extracted as physical field information, and global geometric scale parameters of the vehicle body and incoming flow conditions are obtained. The local curvature, the pressure field gradient, the global geometric scale parameter, and the incoming flow condition are weighted and summed, and the summation result is mapped to a normalization coefficient between 0 and 1 through the Sigmoid function. The physical influence radius of the surface grid point is calculated by linear interpolation between the preset minimum and maximum neighborhood radii using the normalization coefficient.

3. The method according to claim 1, characterized in that, The construction of the physical neighborhood set of the surface grid point based on the physical influence domain radius includes: Using the three-dimensional spatial coordinates of the surface grid point as the center of a sphere and the radius of the physical influence domain as the radius, other surface grid points whose spatial distance is not greater than the radius are searched among all grid points on the vehicle surface; The other surface grid points retrieved constitute the physical neighborhood set of this surface grid point.

4. The method according to claim 1, characterized in that, The step of generating an attention mask matrix based on the physical neighborhood set includes: Define an attention mask matrix, whose rows and columns correspond to the surface grid points of the vehicle body, and are used to mask non-neighborhood grid points in the subsequent attention weight calculation; For any two surface grid points, if the latter belongs to the physical neighborhood set of the former, then the corresponding matrix element is set to 1; otherwise, it is set to 0.

5. The method according to claim 1, characterized in that, The application of the attention mask matrix to limit the attention calculation scope to the physical neighborhood set includes: For each surface grid point, obtain the initial attention score between that surface grid point and all other surface grid points; Obtain the matrix element values ​​of the attention mask matrix of the surface grid point that correspond to other surface grid points; If the value of the matrix element is 0, the initial attention score corresponding to the other surface grid point is replaced with an invalid score, which will cause the corresponding attention weight to be zero in the subsequent normalization process; If the value of the matrix element is 1, then the initial attention score corresponding to the other surface grid point is retained; After replacement, all attention scores are normalized to obtain the attention weights of each other surface grid point relative to the current surface grid point, where the attention weight corresponding to a matrix element value of 0 is zero.

6. The method according to claim 1, characterized in that, The calculation of the attention output for the surface grid point, using a physical spatial bias function based on spatial distance, includes: The query vector of the surface grid point is multiplied by the key vectors of other surface grid points in the physical neighborhood set and then scaled to obtain the feature similarity score. The physical space offset score is obtained by multiplying the negative of the square of the spatial distance between the surface grid point and other surface grid points by the offset coefficient. The attention weights are obtained by summing the feature similarity scores, the physical space bias scores, and the logarithmic values ​​of the attention mask matrix and then applying a normalized exponential function. The attention output of the surface grid point is obtained by weighting and summing the value vectors of each grid point in the physical neighborhood set using the attention weights.

7. The method according to claim 1, characterized in that, The step of outputting the vehicle body surface pressure field based on the attention output of each surface grid point includes: The attention output of each surface grid point is input into the fully connected layer, and the pressure prediction value of that surface grid point is obtained by mapping. The predicted pressure values ​​of all surface grid points are organized according to their spatial coordinates to form a pressure distribution field on the vehicle surface, which is then used as the final output pressure field on the vehicle surface.

8. A device for predicting the pressure field on the surface of an automobile body, characterized in that, The device includes: The surface grid point acquisition module is used to acquire surface grid points on the surface of the car body. Each surface grid point is treated as a word, and the feature vector of each word contains the spatial coordinates and physical attribute parameters of the surface grid point. The physical influence domain radius calculation module is used to calculate the physical influence domain radius of each surface grid point based on its local geometric features and physical field information. The physical neighborhood set construction module is used to construct a physical neighborhood set of the surface grid point based on the physical influence domain radius. The physical neighborhood set includes other surface grid points whose spatial distance from the surface grid point satisfies the constraint of the physical influence domain radius. An attention mask matrix generation module is used to generate an attention mask matrix based on the physical neighborhood set; The attention output calculation module is used to apply the attention mask matrix to limit the attention calculation range to the physical neighborhood set, and combine it with the physical spatial bias function based on spatial distance to calculate the attention output of the surface grid point. The vehicle body surface pressure field output module is used to output the vehicle body surface pressure field based on the attention output of each surface grid point.

9. A computer device, characterized in that, include: The computer device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the vehicle body surface pressure field prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method for predicting the pressure field on the surface of a vehicle body as described in any one of claims 1 to 7.

Images