A wheel impact stress prediction method, device, equipment and storage medium

Abstract

The application provides a wheel impact stress prediction method, device, equipment and storage medium, comprising: acquiring a plurality of different wheel three-dimensional geometric information corresponding to each wheel, working condition parameters and corresponding stress field distribution results; converting each wheel three-dimensional geometric information into a geometric representation in a unified space; converting each working condition parameter into a condition feature vector; training a proxy model based on a geometric information neural operator with the geometric representation and the condition feature vector as input and the corresponding stress field distribution as output to obtain a stress prediction model; and predicting the target stress field distribution result of the target wheel according to the target wheel three-dimensional geometric information and the target working condition parameter of the target wheel using the stress prediction model. The above method can realize efficient and accurate prediction of the wheel impact stress field, effectively simplify the stress analysis process, improve the prediction efficiency, and adapt to the actual engineering needs in the wheel design stage.

Description

Technical Field

[0001] This application relates to the field of automotive structural analysis technology, and more specifically, to a method, apparatus, device, and storage medium for predicting wheel impact stress. Background Technology

[0002] As a core load-bearing and force-transmitting component of a vehicle, the wheel must withstand complex mechanical forces such as road impacts and dynamic loads during driving. Its structural strength directly determines the overall vehicle's driving safety performance. Wheel impact testing is a crucial test for verifying the wheel's impact resistance reliability and identifying failure risks such as spoke cracking and flange detachment. Stress distribution analysis under impact conditions is a core step in wheel structure design, strength verification, and safety validation. Accurately and efficiently obtaining stress field data is essential for shortening the R&D cycle and ensuring product reliability.

[0003] Currently, wheel impact stress analysis in the industry generally relies on traditional finite element simulation technology. This technology is based on the finite element numerical method, which requires first completing a three-dimensional fine model of the wheel, then meshing the structure, and then defining material parameters, impact boundary conditions, contact relationships, etc. By iteratively solving the mechanical control equations, the final output of the stress field distribution results of the entire wheel is obtained. It is the mainstream and mature stress analysis method in the industry and has been used in the entire wheel R&D process for a long time.

[0004] Traditional finite element simulation technology has a cumbersome and time-consuming analysis process. A single complete impact simulation usually takes tens of minutes to several hours. When the wheel structure is fine-tuned or multiple schemes are compared in parallel, the entire process of modeling, subdivision, and solving must be repeated. The iteration cycle is long and the computing power consumption is high, which seriously slows down the iteration rhythm of wheel design and cannot meet the actual engineering needs of rapid screening and comparison of a large number of candidate schemes in the early stage of design. 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 wheel impact stress, which can realize efficient and accurate prediction of wheel impact stress field, effectively simplify the stress analysis process, improve prediction efficiency and adapt to the actual engineering needs in the wheel design stage.

[0006] In a first aspect, embodiments of this application provide a method for predicting wheel impact stress, the method comprising: Obtain the three-dimensional geometric information, working parameters, and corresponding stress field distribution results of various different wheels; The three-dimensional geometric information of each wheel is converted into a geometric representation in a unified space; Each of the aforementioned operating condition parameters is converted into a conditional feature vector; Using the geometric representation and the conditional feature vector as inputs and the corresponding stress field distribution as outputs, a surrogate model based on geometric information neural operators is trained to obtain a stress prediction model. Based on the three-dimensional geometric information of the target wheel and the target working condition parameters, the target stress field distribution of the target wheel is predicted using the stress prediction model.

[0007] Optionally, converting the three-dimensional geometric information of each wheel into a geometric representation in a unified space includes: The three-dimensional geometry of each wheel is globally normalized so that the geometry of different wheels has a consistent spatial reference datum under a unified coordinate system. Construct a cubic latent space of a predetermined size, and embed the normalized wheel geometry into the cubic latent space; Multiple sampling points are set within the cubic hidden space, and the signed distance function value from each sampling point to the geometric surface of the wheel is calculated; The spatial coordinates of each sampling point and its signed distance function value together constitute the geometric feature vector of that sampling point; The geometric representation is formed by arranging the geometric feature vectors of all sampling points in the spatial order of the sampling points in the latent space of the cube.

[0008] Optionally, calculating the signed distance function value from each sampling point to the wheel geometry includes: A surface mesh model for obtaining the three-dimensional geometric information of the wheel; For each sampling point, calculate the shortest Euclidean distance from that sampling point to the surface mesh model; Depending on whether the sampling point is located inside or outside the surface mesh model, the shortest Euclidean distance is assigned a positive or negative sign to obtain the sign distance function value.

[0009] Optionally, converting each of the operating condition parameters into a conditional feature vector includes: Extract the impact mass and impact height from the aforementioned operating parameters; The impact mass and the impact height are respectively normalized numerically. The normalized values ​​are combined into a multidimensional vector, which serves as the conditional feature vector.

[0010] Optionally, the step of training a surrogate model based on a geometric information neural operator, using the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain a stress prediction model, includes: An improved geometric information neural operator model is constructed as a surrogate model; The geometric representation and the conditional feature vector are input into the surrogate model to obtain the predicted stress field; Calculate the error between the predicted stress field and the corresponding stress field distribution; The parameters of the proxy model are updated based on the error; Repeat the above input, calculation, and update steps until the preset training stopping condition is met; The trained surrogate model is used as the stress prediction model.

[0011] Optionally, the step of predicting the target stress field distribution of the target wheel using the stress prediction model based on the target wheel's three-dimensional geometric information and target working parameters includes: The three-dimensional geometric information of the target wheel is converted into a target geometric representation; Convert the target operating condition parameters into a target condition feature vector; The target geometric representation and the target condition feature vector are input into the stress prediction model; The three-dimensional continuous stress field distribution result output by the stress prediction model is obtained as the target stress field distribution result.

[0012] Optionally, the method further includes: The high-stress regions or stress concentration areas in the target wheel are identified using the target stress field distribution results; and / or, The stress field distribution results of the target stress field are used to compare and rank the stress fields of multiple candidate wheel designs; and / or, The target stress field distribution results are used to evaluate the response stability of the same wheel under multiple different operating conditions; and / or, The optimization effect can be judged by comparing the stress field changes before and after the local structural modification using the target stress field distribution results; and / or, The target stress field distribution results are used as a preliminary screening basis for subsequent high-precision finite element analysis.

[0013] Secondly, embodiments of this application provide a wheel impact stress prediction device, the device comprising: The wheel data acquisition module is used to acquire the three-dimensional geometric information, working parameters and corresponding stress field distribution results of various different wheels. A wheel 3D geometric information conversion module is used to convert the 3D geometric information of each wheel into a geometric representation in a unified space; The operating condition parameter conversion module is used to convert each of the operating condition parameters into a condition feature vector; The stress prediction model determination module is used to train a surrogate model based on geometric information neural operators, taking the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain the stress prediction model. The stress field distribution prediction module is used to predict the target stress field distribution of the target wheel based on the target wheel's three-dimensional geometric information and target working condition parameters using the stress prediction model.

[0014] Optionally, converting the three-dimensional geometric information of each wheel into a geometric representation in a unified space includes: The three-dimensional geometry of each wheel is globally normalized so that the geometry of different wheels has a consistent spatial reference datum under a unified coordinate system. Construct a cubic latent space of a predetermined size, and embed the normalized wheel geometry into the cubic latent space; Multiple sampling points are set within the cubic hidden space, and the signed distance function value from each sampling point to the geometric surface of the wheel is calculated; The spatial coordinates of each sampling point and its signed distance function value together constitute the geometric feature vector of that sampling point; The geometric representation is formed by arranging the geometric feature vectors of all sampling points in the spatial order of the sampling points in the latent space of the cube.

[0015] Optionally, calculating the signed distance function value from each sampling point to the wheel geometry includes: A surface mesh model for obtaining the three-dimensional geometric information of the wheel; For each sampling point, calculate the shortest Euclidean distance from that sampling point to the surface mesh model; Depending on whether the sampling point is located inside or outside the surface mesh model, the shortest Euclidean distance is assigned a positive or negative sign to obtain the sign distance function value.

[0016] Optionally, converting each of the operating condition parameters into a conditional feature vector includes: Extract the impact mass and impact height from the aforementioned operating parameters; The impact mass and the impact height are respectively normalized numerically. The normalized values ​​are combined into a multidimensional vector, which serves as the conditional feature vector.

[0017] Optionally, the step of training a surrogate model based on a geometric information neural operator, using the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain a stress prediction model, includes: An improved geometric information neural operator model is constructed as a surrogate model; The geometric representation and the conditional feature vector are input into the surrogate model to obtain the predicted stress field; Calculate the error between the predicted stress field and the corresponding stress field distribution; The parameters of the proxy model are updated based on the error; Repeat the above input, calculation, and update steps until the preset training stopping condition is met; The trained surrogate model is used as the stress prediction model.

[0018] Optionally, the step of predicting the target stress field distribution of the target wheel using the stress prediction model based on the target wheel's three-dimensional geometric information and target working parameters includes: The three-dimensional geometric information of the target wheel is converted into a target geometric representation; Convert the target operating condition parameters into a target condition feature vector; The target geometric representation and the target condition feature vector are input into the stress prediction model; The three-dimensional continuous stress field distribution result output by the stress prediction model is obtained as the target stress field distribution result.

[0019] Optionally, the apparatus further includes a predicted structure post-processing module for: The high-stress regions or stress concentration areas in the target wheel are identified using the target stress field distribution results; and / or, The stress field distribution results of the target stress field are used to compare and rank the stress fields of multiple candidate wheel designs; and / or, The target stress field distribution results are used to evaluate the response stability of the same wheel under multiple different operating conditions; and / or, The optimization effect can be judged by comparing the stress field changes before and after the local structural modification using the target stress field distribution results; and / or, The target stress field distribution results are used as a preliminary screening basis for subsequent high-precision finite element analysis.

[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 wheel impact stress 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 that, when executed by a processor, performs the steps of the wheel impact stress 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 acquiring the three-dimensional geometric information, working parameters, and corresponding stress field distribution results of various different wheels, we can accumulate sample data covering different wheel structures and impact conditions, ensuring the diversity and comprehensiveness of training samples. This provides a sufficient and reliable data foundation for the subsequent surrogate model to learn the mapping law between wheel structure, working conditions, and stress field.

[0023] Converting the three-dimensional geometric information of each wheel into a geometric representation in a unified space can eliminate the problem of inconsistent input caused by differences in the three-dimensional structural shape and size of different wheels, construct a standardized geometric input form, and enable the surrogate model to stably identify the structural features of different wheels, laying a unified foundation for subsequent feature learning.

[0024] By converting each working condition parameter into a conditional feature vector, working condition information such as impact mass and impact height can be quantified into a standard feature form, realizing standardized input of working condition information. This enables the surrogate model to accurately perceive the differences between different impact working conditions and supports joint mapping learning between geometry and working conditions.

[0025] Using geometric representations and conditional feature vectors as inputs and the corresponding stress field distribution as outputs, a surrogate model based on geometric information neural operators is trained to obtain a stress prediction model. The intrinsic mapping relationship between wheel structure, working conditions and stress field can be directly learned through data-driven methods, eliminating the need to repeatedly perform tedious mechanical simulations and fundamentally simplifying the stress analysis process, while ensuring that the predicted laws closely match the actual stress response characteristics.

[0026] Based on the three-dimensional geometric information of the target wheel and the target working condition parameters, the stress field distribution of the target is predicted using a stress prediction model. The target data can be directly input and the stress field results can be quickly output, eliminating the simulation steps such as modeling, mesh generation, and iterative solution, which greatly shortens the time spent on stress analysis and meets the needs of rapid screening of large-scale schemes in the early stage of design.

[0027] This application achieves efficient and accurate prediction of wheel impact stress field through a coherent process of sample accumulation, geometric unified characterization, working condition feature quantification, surrogate model training, and rapid prediction. It effectively simplifies the stress analysis process, improves prediction efficiency, and adapts to the actual engineering needs of the wheel design stage.

[0028] 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

[0029] 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.

[0030] Figure 1 A flowchart of a wheel impact stress prediction method provided in Embodiment 1 of this application is shown; Figure 2 This illustration shows a schematic diagram of the construction and reasoning process of a wheel impact stress prediction model provided in Embodiment 1 of this application; Figure 3 A flowchart of a wheel three-dimensional geometric information conversion method provided in Embodiment 1 of this application is shown; Figure 4 A flowchart of a method for calculating the symbolic distance function value provided in Embodiment 1 of this application is shown; Figure 5 A flowchart of a working condition parameter conversion method provided in Embodiment 1 of this application is shown; Figure 6 A flowchart of a stress prediction model training method provided in Embodiment 1 of this application is shown; Figure 7 A flowchart of a method for determining the target stress field distribution results provided in Embodiment 1 of this application is shown; Figure 8 This illustration shows a schematic diagram of a rapid evaluation and design iteration process for a wheel hub solution provided in Embodiment 1 of this application; Figure 9 This illustration shows a comparison and verification diagram of a predicted stress field and a real stress field provided in Embodiment 1 of this application; Figure 10 This paper shows a schematic diagram of the structure of a wheel impact stress prediction device provided in Embodiment 2 of this application; Figure 11 A schematic diagram of the structure of a computer device provided in Embodiment 3 of this application is shown. Detailed Implementation

[0031] 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.

[0032] Example 1 To facilitate understanding of this application, the following is combined with... Figure 1 The flowchart of the wheel impact stress prediction method provided in Embodiment 1 of this application illustrates Embodiment 1 in detail.

[0033] See Figure 1 As shown, Figure 1 A flowchart of a wheel impact stress prediction method provided in Embodiment 1 of this application is shown, wherein the method includes steps S101 to S105: S101: Obtain the three-dimensional geometric information, working parameters, and corresponding stress field distribution results of various different wheels.

[0034] Specifically, the core data comes from a large number of historical finite element simulation samples accumulated over a long period of time during the company's wheel R&D process. These samples cover wheel models from different design stages, different structural schemes, and different impact conditions. The samples need to be systematically organized to form a one-to-one correspondence between structural scheme, working condition, and stress field results, ensuring data correlation and guaranteeing that each set of geometric, working condition, and stress data is traceable and matchable.

[0035] It is necessary to conduct sample validity checks and quality screening to remove invalid samples with incomplete data, ambiguous working parameters, abnormal stress results, or insufficient structural representativeness, so as to ensure the reliability of training samples and avoid low-quality data interfering with the model's learning patterns.

[0036] The sample needs to cover wheels with different numbers of spokes, spoke orientations, flange transition shapes, and center mounting area structures. It should also include multiple sets of impact conditions with different impact masses, impact heights, and impact angles, so as to provide sufficient basic data for the model to learn comprehensive and universal structure-condition-stress mapping laws.

[0037] S102: Convert the three-dimensional geometric information of each wheel into a geometric representation in a unified space.

[0038] Specifically, this move aims to address the core pain points of complex three-dimensional wheel structures, significant differences in geometric shapes among different schemes, and inconsistent input formats. It abandons inefficient expressions using simple geometric parameters, local discrete points, or feature sections, as these expressions cannot fully reflect the spatial coupling relationships and detailed features of the wheel.

[0039] Unified geometric representation has three core functions: First, it completely preserves the overall contour information of the wheel, allowing the model to perceive the global morphological differences of different wheels; second, it accurately reflects local geometric changes such as spoke thickness, rim transition, and local reinforcement areas; and third, it incorporates scattered local geometric information into the same spatial system, establishing the correlation between the overall structure and local stress response.

[0040] This characterization forms the basis for subsequent continuous stress field modeling. It allows the surrogate model to extract global features in a unified space and then map them to the target location, enabling continuous prediction of the wheel stress field and avoiding prediction bias caused by local fitting.

[0041] S103: Convert each of the operating condition parameters into a conditional feature vector.

[0042] Specifically, operating condition parameters are the core variables affecting the stress field distribution of the wheel, including key indicators such as impact mass, impact height, impact intensity, and stress level, which directly determine the kinetic energy, impact intensity, and effect of the impact load. Operating condition features are not simply added labels, but modeling inputs that are as important as geometric information. Through unified conversion and feature encoding, operating condition parameters are transformed into a format that the model can recognize, avoiding the weakening of operating condition information.

[0043] By jointly modeling geometric information and working condition characteristics, the model can simultaneously perceive structural differences (different wheels) and working condition differences (different impact conditions on the same wheel), learn the stress response law under the combined action of structure and working condition, enhance the model's flexibility and generalization, and enable it to adapt to various working condition scenarios.

[0044] S104: Using the geometric representation and the conditional feature vector as inputs and the corresponding stress field distribution as outputs, train a surrogate model based on geometric information neural operators to obtain a stress prediction model.

[0045] Specifically, the improved geometry-informed neural operator IGINO is used as a surrogate model. This model is suitable for engineering problems with complex three-dimensional structures, irregular geometric inputs, and continuous output fields. Unlike traditional single-point regression models, which can only output a single stress value and cannot restore the full field distribution, this model is different.

[0046] From the perspective of operator learning, the model essentially learns the mapping operator from the input function space to the output function space, and the core formula is:

[0047] Formula character definition: This is the mapping operator from geometry-working condition to stress field; The wheel geometry information is described by the SDF (Signed Distance Function). Characteristics of impact load conditions; The three-dimensional stress field distribution to be predicted.

[0048] This operator is approximated by a combination of multiple integral operators, and the approximation formula is:

[0049] Formula character definition: For parameterized mapping operators; The set of trainable parameters for the model; For output point mapping (Multilayer Perceptron MLP); , For the first (Network Layer Number) Layer, Layer 1 continuous kernel integral operator; (Operator Composition Operation) refers to operator composition operations.

[0050] The IGINO core uses a kernel integral operator to propagate spatial information, characterizing wheel spoke-rim coupling, the influence of local reinforcement zones on stress paths, and the modulation of local responses by global morphology. The feature update formula is:

[0051] Formula character definition: For the first Layered networks in spatial location Feature representation at (Spatial Position); For the first Layers can learn kernel functions to characterize position. (Spatial Position) The impact; For the first Layer in position Features of the location; It is a spatial integral infinitesimal element.

[0052] In practical calculations, the integral is discretized in a local neighborhood to reduce computational complexity. The discretization formula is as follows:

[0053] Formula character definition: For Center, radius The neighborhood region of (Neighborhood Radius); These are discrete sampling points within the neighborhood; Sampling points Discrete weights.

[0054] The model achieves global information exchange through FNO, efficiently capturing long-distance structural dependencies and solving the problem that local information cannot cover the global force influence. The core calculation formula is:

[0055] Formula character definition: Fourier transform and inverse Fourier transform; These are learnable kernel parameters in the frequency domain; It is a linear mapping matrix; It is a non-linear activation function; For spatial location The characteristics of the location.

[0056] Operating condition characteristics are embedded into the latent space through an embedding mapping, participating in operator calculations, and ultimately establishing a unified mapping from (geometry, operating condition) to stress field, rather than a simple information superposition, ensuring that the impact of operating conditions on global stress is fully learned. The model output is a three-dimensional continuous stress field, rather than a single numerical value, which can fully reflect the stress state at various locations of the wheel, meeting the designers' needs for assessing high-stress areas and stress distribution trends, and compensating for the limitations of two-dimensional prediction.

[0057] S105: Based on the three-dimensional geometric information of the target wheel and the target working condition parameters, the target stress field distribution of the target wheel is predicted using the stress prediction model.

[0058] Specifically, the prediction process is perfectly matched to the input format of the training phase, requiring no additional adaptation. It can directly and quickly evaluate the new wheel structure, avoiding errors caused by format conversion. The model prediction efficiency is extremely high, with a single wheel prediction taking only about 500 milliseconds, while traditional finite element simulation takes about 20 minutes, resulting in a speedup of up to 2400 times. It is suitable for large-scale scheme screening in the early stages of design, significantly shortening the iteration cycle.

[0059] The model's prediction accuracy meets engineering requirements, with L2 error controllable within 10%. It can accurately reproduce the three-dimensional stress field distribution of the wheel, ensuring that the prediction results can be directly used for engineering evaluation. The output is a complete three-dimensional continuous stress field, covering areas missed by traditional two-dimensional predictions, such as the side of the impact zone and non-disc surfaces, comprehensively reflecting the three-dimensional structural stress state of the wheel and avoiding missed stress detection in critical areas.

[0060] It supports continuous evaluation of the same wheel under multiple working conditions, reuses geometric representation results, eliminates the need for repeated preprocessing, and further improves the efficiency of multi-working-condition evaluation. When the wheel structure is large and there are many locations of stress field to be output, the prediction process can be organized in segments and output step by step. After completion, the complete stress field can be obtained by summarizing, which improves the stability of calculation and avoids the problems of excessive memory consumption and calculation lag caused by large-scale calculation.

[0061] See Figure 2 As shown, Figure 2 The diagram illustrates the construction and inference process of a wheel impact stress prediction model according to Embodiment 1 of this application. The diagram fully demonstrates the entire process from model training to deployment and inference, as explained below: Model training phase: The model takes "3D wheel hub geometry" and "external load conditions" as inputs. The external load conditions include parameters such as the mass, height, and impact point of the impactor at a 13-degree impact. For the "3D wheel hub geometry," a "unified representation construction of wheel hub geometry" is performed, representing the wheel hub geometry through a signed distance function, converting the three-dimensional wheel hub geometry into a unified spatial geometric representation. For the "external load conditions," a "feature construction of working conditions" is performed, generating pattern-recognizable condition feature codes. Based on the above representations, "rapid proxy training of structural response based on neural operators" is conducted. The training process is divided into three parts: encoding: geometric information reception and representation; FNO (Fourier Neural Operator): latent space feature modeling and calculation; decoding: target position response recovery. The neural operator module takes the wheel hub structure and working conditions as inputs, and through latent space feature evolution, outputs the stress response results at the target position.

[0062] Deployment and inference phase: Taking "three-dimensional geometric information of the wheel hub to be evaluated" and "target working condition parameters" as input, it supports "unified representation execution of wheel hub geometry" and "geometric preprocessing reuse mechanism", reusing the geometric representation logic of the training phase to reduce redundant calculations; input to "trained operator proxy model", which is deployed on the server side and supports CPU and GPU inference. It can output Von Mises stress distribution results through GPU millisecond-level inference.

[0063] In an optional implementation, see Figure 3 As shown, Figure 3The flowchart of a method for converting three-dimensional geometric information of a wheel according to Embodiment 1 of this application is shown, wherein the step of converting the three-dimensional geometric information of each wheel into a geometric representation in a unified space includes steps S301 to S305: S301: Perform global normalization on the three-dimensional geometry of each wheel so that the geometry of different wheels has a consistent spatial reference datum under a unified coordinate system.

[0064] Specifically, global normalization includes three core operations: geometric center alignment, principal scale normalization, and spatial extent standardization.

[0065] Geometric center alignment maps the geometric centers of all wheels to the origin of the coordinate system, eliminating spatial positional deviations and ensuring that all wheels have consistent references; principal scale normalization scales the wheels according to their maximum profile size, eliminating size ratio differences and keeping large and small wheels at the same scale; spatial range standardization confines the wheels to a fixed spatial range, unifying spatial boundaries and avoiding feature deviations caused by inconsistent boundaries.

[0066] By normalizing, wheels with large differences in size, proportion, and shape can have a consistent spatial reference benchmark, laying the foundation for subsequent unified embedding and ensuring that subsequent geometric representations can be directly compared and learned.

[0067] S302: Construct a cubic latent space of a predetermined size, and embed the normalized wheel geometry into the cubic latent space.

[0068] Specifically, the cubic latent space is a standard cube of fixed size, which serves as a unified embedding carrier for all wheel geometries, eliminating the input format chaos caused by differences in geometric shapes and ensuring the uniformity of input data structure.

[0069] The cubic latent space is regularly discretized along three dimensions, forming uniformly distributed sampling points, voxel units, or grid nodes. This constructs a unified data structure that adapts to model input requirements and facilitates subsequent feature extraction. The normalized wheel geometry is completely embedded within this cube, ensuring that all wheel geometry features are extracted within the same spatial range, without spatial overflow or missing issues.

[0070] S303: Set multiple sampling points within the cubic hidden space, and calculate the signed distance function value from each sampling point to the geometric surface of the wheel.

[0071] Specifically, SDF is used to characterize the spatial relationship between sampling points and the boundary of the wheel entity. It is a core parameter of unified geometric representation and can accurately depict three-dimensional geometric boundaries.

[0072] SDF value rules: When the sampling point is located inside the wheel body, the SDF value is negative; when it is located outside the wheel body, the SDF value is positive; when it is located on the wheel's geometric surface, the SDF value is 0 or close to 0. The SDF value can continuously reflect the relative position of the sampling point to the wheel boundary, accurately depict complex structural features such as spokes, rims, center discs, and local reinforcement areas, and capture subtle geometric changes.

[0073] S304: The spatial coordinates of each sampling point and its signed distance function value together constitute the geometric feature vector of that sampling point.

[0074] Specifically, the geometric feature vector of a single sampling point (Geometric Feature Vector) contains 4 dimensions, and the formula is:

[0075] Formula character definition: For the first hidden space of the cube (Sampling Point Index) geometric feature vectors of sampling points; These are the three-dimensional spatial coordinate components of the sampling point in the cubic coordinate system; This represents the SDF value corresponding to the sampling point. The dimension is fixed. (Geometric Feature Dimension) That is, 3 spatial coordinate dimensions + 1 SDF value dimension, which accurately carries the local geometric information of a single sample, including both location information and boundary relationship information.

[0076] S305: The geometric representation is formed by arranging the geometric feature vectors of all sampling points in the spatial order of the sampling points in the latent space of the cube.

[0077] Specifically, the final geometric representation The Geometric Unified Representation Matrix (GUM) is a two-dimensional matrix, and its formula is:

[0078] Formula character definition: This is the unified geometric representation matrix for wheels; (Total Sampling Points) represents the total number of sampling points in the latent space of the cube; The geometric feature dimension of a single sampling point ( ); for OK The columns are a real matrix space. The matrices are concatenated according to the spatial arrangement of the sampling points, fully carrying all the geometric information of the overall outline and local details of the wheel, without any omissions. This serves as the geometric input to the IGINO model, ensuring that the model can comprehensively learn geometric features.

[0079] In an optional implementation, see Figure 4 As shown, Figure 4 The flowchart of a method for calculating the symbolic distance function value provided in Embodiment 1 of this application is shown, wherein the calculation of the symbolic distance function value from each sampling point to the geometric surface of the wheel includes steps S401 to S403: S401: Obtain the surface mesh model of the three-dimensional geometry of the wheel.

[0080] Specifically, the surface mesh model is a discretized and precise representation of the wheel's three-dimensional geometry. It is composed of a large number of triangular / quadrilateral mesh patches and can reproduce the shape of the wheel's outer surface without error.

[0081] The mesh model can accurately reproduce key structural details such as the outer surface contour of the wheel, the direction of the spokes, the transition curvature of the rim, the shape of the center mounting hole, and the curved surface of the local reinforcement area, providing a precise geometric benchmark for SDF calculation and avoiding errors in SDF values ​​caused by geometric deviations.

[0082] S402: For each sampling point, calculate the shortest Euclidean distance from that sampling point to the surface mesh model.

[0083] Specifically, the shortest Euclidean distance refers to the minimum straight-line distance between the sampling point and all vertices and faces of the surface mesh model, and is the core indicator for measuring the distance between the sampling point and the surface.

[0084] By traversing all faces of the mesh model using a spatial geometry algorithm, the distance from the sampling point to each face is calculated, and the minimum value is taken as the shortest Euclidean distance. This accurately quantifies the proximity of the sampling point to the wheel surface, ensuring the accuracy of the SDF value calculation.

[0085] S403: Based on whether the sampling point is located inside or outside the surface mesh model, assign a positive or negative sign to the shortest Euclidean distance to obtain the sign distance function value.

[0086] Specifically, a spatial topology judgment algorithm is used to determine the spatial position of the sampling point relative to the surface mesh model. This algorithm can accurately distinguish between inside and outside without judgment error. Position judgment rules: Sampling point inside the mesh model → shortest distance is assigned a negative sign; sampling point outside the mesh model → shortest distance is assigned a positive sign; sampling point on the surface of the mesh model → assigned a value of 0.

[0087] By using symbols to distinguish the spatial relationship between sampling points and wheel entities, SDF values ​​with geometric semantics are formed to support subsequent geometric feature construction and ensure that the features contain spatial location semantics.

[0088] In an optional implementation, see Figure 5 As shown, Figure 5 The flowchart of a working condition parameter conversion method provided in Embodiment 1 of this application is shown, wherein converting each working condition parameter into a condition feature vector includes steps S501-S503: S501: Extract the impact mass and impact height from the operating parameters.

[0089] Specifically, impact mass and impact height are the core operating parameters of wheel impact tests, directly determining the kinetic energy, impact intensity, and effect of the impact load. They are the two most critical variables affecting the stress field. In addition to the core parameters, auxiliary operating parameters such as impact angle, impact velocity, and stress area can be extracted to enrich the dimensions of operating characteristics, adapt to complex impact scenarios, and improve the model's generalization ability.

[0090] S502: Normalize the impact mass and the impact height respectively.

[0091] Specifically, the purpose of normalization is to eliminate differences in the numerical range and units of impact mass and impact height, and to avoid model learning bias caused by differences in units. The max-min normalization method is adopted to map all working condition parameters to the [0,1] interval, so that different working condition parameters are on the same numerical scale, which facilitates the model to learn the working condition characteristics in a unified manner and ensures the balance of working condition feature weights.

[0092] S503: Combine the normalized values ​​into a multi-dimensional vector, which serves as the conditional feature vector.

[0093] Specifically, conditional feature vectors The formula for the Working Condition Feature Vector is:

[0094] Formula character definition: The feature vector represents the operating conditions. For the working condition feature dimension (when extracting 2 core parameters) ); for A 3D real vector space. Vector dimension and extracted parameter index. The number of parameter indices is the same, and they serve as independent input items, along with the geometric representation matrix. After fusion, they are jointly input into the IGINO model for training and prediction, ensuring that the working condition features are completely transmitted.

[0095] In an optional implementation, see Figure 6 As shown, Figure 6 The flowchart of a stress prediction model training method provided in Embodiment 1 of this application is shown. The method involves training a surrogate model based on a geometric information neural operator, using the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain a stress prediction model. The steps include S601-S606. S601: Construct an improved geometric information neural operator model as a surrogate model.

[0096] Specifically, the IGINO model is designed for continuous field prediction of complex three-dimensional structures. It integrates the advantages of kernel integral operators and Fourier neural operators to solve the pain points of traditional models that cannot take into account global structural cognition, local response prediction, and multi-condition coupling.

[0097] The model consists of three parts: a geometric information receiving and representation module, a latent spatial feature modeling module, and a target position response recovery module, realizing end-to-end mapping from geometry-working conditions to stress field without intermediate processing steps.

[0098] S602: Input the geometric representation and the conditional feature vector into the surrogate model to obtain the predicted stress field.

[0099] Specifically, the model input-output mapping formula is as follows:

[0100] Formula character definition: To predict the stress field distribution; For parameterizing the IGINO mapping operator; The geometric representation matrix of the wheel; The model first integrates geometric and working condition features, then completes feature evolution through multi-layer kernel integral operators and Fourier neural operators, and finally outputs a three-dimensional continuous stress field covering the entire wheel space to ensure full-field stress prediction.

[0101] S603: Calculate the error between the predicted stress field and the corresponding stress field distribution.

[0102] Specifically, using the real stress field obtained from historical finite element simulations as a benchmark, the L2 norm error is used to quantify the deviation between the predicted and actual values. This error can comprehensively measure the prediction accuracy across the entire field. The core training objective is to control the L2 error within 10%, ensuring that the model's prediction accuracy meets the actual needs of engineering design and strength assessment, and guaranteeing that the prediction results can be directly used for engineering decision-making.

[0103] S604: Update the parameters of the proxy model based on the error.

[0104] Specifically, the backpropagation algorithm is used to propagate errors from the output layer to the input layer and iteratively update the model parameters. This algorithm is the core optimization method for neural network training.

[0105] The parameters to be updated include: the learnable kernel of the kernel integral operator. The linear mapping matrix of Fourier neural operators Parameters of nonlinear activation function, discrete weights The model incorporates elements such as weights and biases in point mapping MLPs. Through multiple iterations, the error between the predicted and actual stress fields is gradually reduced, optimizing the model's mapping capability and improving prediction accuracy.

[0106] S605: Repeat the above input, calculation and update steps until the preset training stop condition is met.

[0107] Specifically, the preset training stopping conditions include three categories, and the training can be stopped if any one of them is met: First, the L2 error converges to a preset threshold (e.g., ≤10%); second, the number of model iterations reaches a preset upper limit; and third, the generalization performance of the model on the validation set tends to stabilize without significant improvement.

[0108] Setting a validation set can prevent model overfitting, ensure that the trained model has good generalization prediction ability for unknown new wheel structures, and guarantee the practicality of the model.

[0109] S606: Use the trained surrogate model as the stress prediction model.

[0110] Specifically, the trained model has established a stable (geometry, working condition) → stress field mapping relationship, and has fully learned the stress response laws under different wheel structures and different impact conditions.

[0111] The model does not require retraining and can be deployed directly. It is used for rapid prediction of the three-dimensional stress field of unknown new wheel structures under arbitrary target impact conditions. It is plug-and-play and adaptable to the needs of rapid engineering deployment.

[0112] In an optional implementation, see Figure 7 As shown, Figure 7 The flowchart illustrates a method for determining the target stress field distribution result provided in Embodiment 1 of this application. The step of predicting the target stress field distribution result of the target wheel using the stress prediction model based on the target wheel's three-dimensional geometric information and target working condition parameters includes steps S701-S704: S701: Convert the three-dimensional geometric information of the target wheel into a target geometric representation.

[0113] Specifically, the conversion process is completely consistent with the training phases S301~S305, ensuring that the input format matches the trained model and avoiding prediction failures due to format incompatibility.

[0114] Support for preprocessing reuse mechanism: When conducting multi-condition evaluation of the same wheel structure, there is no need to repeatedly perform operations such as normalization, cube embedding, SDF calculation, and feature vector concatenation. The generated geometric representation can be directly reused, which greatly reduces the preprocessing time and improves the efficiency of multi-condition evaluation.

[0115] S602: Convert the target operating condition parameters into a target condition feature vector.

[0116] Specifically, the conversion process is completely consistent with the training phases S501 to S503, extracting and normalizing the impact mass and impact height to generate a multi-dimensional vector that matches the input format of the model, ensuring the uniformity of the working condition feature format.

[0117] Ensure that the feature vector of the target working condition is consistent with the dimension of the working condition feature vector during the training phase to avoid prediction failure due to differences in input format and ensure prediction stability.

[0118] S703: Input the target geometric representation and the target condition feature vector into the stress prediction model.

[0119] Specifically, the model eliminates the need for complex steps in traditional finite element simulation, such as mesh generation, boundary condition setting, material parameter definition, and iterative solving, saving significant manual operations and computation time. The model directly performs feature fusion and operator calculations on the input geometric representation and working condition characteristics, quickly mapping the input to a three-dimensional stress field and simplifying the prediction process.

[0120] S704: Obtain the three-dimensional continuous stress field distribution result output by the stress prediction model, and use it as the target stress field distribution result.

[0121] Specifically, the output is a complete three-dimensional continuous stress field, covering all locations including wheel spokes, rims, center disc, transition zone, and mounting holes, with no stress information omissions. It can accurately reflect the stress state on the sides and non-disc surfaces of the impact area, overcoming the shortcomings of existing two-dimensional stress prediction methods in terms of insufficient characterization and incomplete assessment, and avoiding misjudgment of stress in key areas.

[0122] The output includes equivalent stress values ​​and stress distribution cloud maps at each location, which can be directly used for subsequent design evaluation without secondary processing.

[0123] See Figure 8 As shown, Figure 8This paper presents a schematic diagram of a rapid evaluation and design iteration process for a wheel hub design according to Embodiment 1 of this application, demonstrating the complete application chain of this method in engineering design, as explained below: Input process: The input consists of "multiple wheel hub candidate schemes" (including scheme A, scheme B, scheme C and other schemes) and "parameter and working condition inputs" (including punch, impact load mass, impact load height and impact load position); Rapid screening process: Through "operator-based rapid screening", the total design and evaluation cycle can be reduced from several weeks to hours; the core is the "stress field prediction unit", which quickly outputs the stress field results of each scheme through the process of encoding → operator → decoding. Evaluation and decision-making process: First, "high stress area identification" is performed to identify the peak stress areas of the wheel hub for Scheme A, Scheme B, and Scheme C; then, "scheme comparison results" are output, and the merits of the schemes are judged by comprehensively considering factors such as peak stress, stress distribution, and wheel hub weight. Iteration and output phase: Based on the comparison results, “design iteration and decision support” are carried out, and the best wheel hubs are finally selected through “final design evaluation output”; at the same time, the evaluation results are supported to be combined with high-precision FEM simulation to form a “rapid prediction + key review” R&D model.

[0124] In an optional implementation, the method further includes: The high-stress areas or stress concentration points in the target wheel are identified using the target stress field distribution results.

[0125] Specifically, high-stress areas are concentrated at the spoke root, rim transition zone, around mounting holes, center disc connection area, and locations of local geometric abrupt changes; these are the wheel's weakest points in terms of stress distribution. By using stress field numerical threshold screening, stress concentration points can be quickly located, potential structural weaknesses can be identified, safety risks can be mitigated in advance, and precise targets can be provided for structural optimization.

[0126] And / or, use the target stress field distribution results to compare and rank the stress fields of multiple candidate wheel designs.

[0127] Specifically, the ranking criteria include maximum equivalent stress, area of ​​the critical region, stress gradient change, distribution of high-stress areas, and wheel structure quality, etc., to comprehensively evaluate the merits of the schemes from multiple dimensions. In the early stages of design, hundreds of candidate schemes can be quickly ranked and screened out, eliminating schemes with high risk and poor mechanical performance, reducing unnecessary simulation computing power investment, and shortening the scheme selection cycle.

[0128] And / or, the target stress field distribution results are used to evaluate the response stability of the same wheel under multiple different working conditions.

[0129] Specifically, it can quickly obtain the stress field distribution of the same wheel under different impact masses, impact heights, and impact angles, covering a variety of real-world impact scenarios. By analyzing changes in stress peak values, high-stress zone locations, and stress distribution uniformity, it can determine the structure's adaptability to fluctuations in operating conditions and its mechanical stability, ensuring the wheel's safety and reliability under complex operating conditions.

[0130] And / or, the stress field changes before and after local structural modifications are compared using the target stress field distribution results to determine the optimization effect.

[0131] Specifically, it adapts to local modification scenarios such as spoke thickening, transition fillet optimization, center disc reinforcement, local opening adjustment, and rim curvature optimization, covering common structural optimization methods. By comparing the reduction in peak stress, the reduction in the range of high-stress areas, and the uniformity of stress transmission paths before and after modification, the effectiveness of structural optimization can be quickly verified, accelerating design iteration and reducing optimization trial-and-error costs.

[0132] And / or, the target stress field distribution results are used as a preliminary screening basis for subsequent high-precision finite element analysis.

[0133] Specifically, a collaborative R&D model combining rapid prediction and focused review is adopted: First, a large number of solutions are quickly screened through this application, filtering out obviously unqualified solutions. Only a small number of solutions with relatively good prediction results, requiring focused verification, and posing boundary risks are submitted for traditional finite element simulation, where computing power is concentrated for high-precision verification. This significantly reduces the workload of high-precision simulation, shortens the wheel development cycle, lowers R&D costs, and meets the enterprise's need for efficient R&D.

[0134] See Figure 9 As shown, Figure 9 This diagram illustrates a comparison between the predicted stress field and the actual stress field provided in Embodiment 1 of this application. It visually demonstrates the model's prediction accuracy. The diagram includes three sets of comparisons of hub stress fields under different working conditions / structures. Each set includes three columns of results: "actual" stress field, "predicted" stress field, and "error" field. The legend on the right side of the diagram is a Von Mises stress (vmis) color scale, with a value range of 50 to 250. The intensity of the color represents the stress magnitude. As can be seen from the error field, the deviation between the model's prediction results and the actual stress field is extremely small, verifying that the prediction accuracy of the method in this application meets engineering requirements.

[0135] Example 2 See Figure 10 As shown, Figure 10 This illustration shows a schematic diagram of a wheel impact stress prediction device according to Embodiment 2 of this application, wherein the device includes: The wheel data acquisition module 1001 is used to acquire the three-dimensional geometric information, working parameters and stress field distribution results of various different wheels. The wheel three-dimensional geometric information conversion module 1002 is used to convert the three-dimensional geometric information of each wheel into a geometric representation in a unified space; The working condition parameter conversion module 1003 is used to convert each of the working condition parameters into a condition feature vector; The stress prediction model determination module 1004 is used to train a surrogate model based on geometric information neural operators, with the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain the stress prediction model. The stress field distribution prediction module 1005 is used to predict the target stress field distribution of the target wheel based on the target wheel's three-dimensional geometric information and target working condition parameters using the stress prediction model.

[0136] In an optional implementation, the conversion of the three-dimensional geometric information of each wheel into a geometric representation in a unified space includes: The three-dimensional geometry of each wheel is globally normalized so that the geometry of different wheels has a consistent spatial reference datum under a unified coordinate system. Construct a cubic latent space of a predetermined size, and embed the normalized wheel geometry into the cubic latent space; Multiple sampling points are set within the cubic hidden space, and the signed distance function value from each sampling point to the geometric surface of the wheel is calculated; The spatial coordinates of each sampling point and its signed distance function value together constitute the geometric feature vector of that sampling point; The geometric representation is formed by arranging the geometric feature vectors of all sampling points in the spatial order of the sampling points in the latent space of the cube.

[0137] In an optional implementation, calculating the signed distance function value from each sampling point to the wheel geometry includes: A surface mesh model for obtaining the three-dimensional geometric information of the wheel; For each sampling point, calculate the shortest Euclidean distance from that sampling point to the surface mesh model; Depending on whether the sampling point is located inside or outside the surface mesh model, the shortest Euclidean distance is assigned a positive or negative sign to obtain the sign distance function value.

[0138] In an optional implementation, converting each of the operating condition parameters into a conditional feature vector includes: Extract the impact mass and impact height from the aforementioned operating parameters; The impact mass and the impact height are respectively normalized numerically. The normalized values ​​are combined into a multidimensional vector, which serves as the conditional feature vector.

[0139] In an optional implementation, the step of training a surrogate model based on a geometric information neural operator, using the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain a stress prediction model, includes: An improved geometric information neural operator model is constructed as a surrogate model; The geometric representation and the conditional feature vector are input into the surrogate model to obtain the predicted stress field; Calculate the error between the predicted stress field and the corresponding stress field distribution; The parameters of the proxy model are updated based on the error; Repeat the above input, calculation, and update steps until the preset training stopping condition is met; The trained surrogate model is used as the stress prediction model.

[0140] In an optional implementation, the step of predicting the target stress field distribution of the target wheel using the stress prediction model based on the target wheel's three-dimensional geometric information and target working parameters includes: The three-dimensional geometric information of the target wheel is converted into a target geometric representation; Convert the target operating condition parameters into a target condition feature vector; The target geometric representation and the target condition feature vector are input into the stress prediction model; The three-dimensional continuous stress field distribution result output by the stress prediction model is obtained as the target stress field distribution result.

[0141] In an optional implementation, the apparatus further includes a predictive structure post-processing module for: The high-stress regions or stress concentration areas in the target wheel are identified using the target stress field distribution results; and / or, The stress field distribution results of the target stress field are used to compare and rank the stress fields of multiple candidate wheel designs; and / or, The target stress field distribution results are used to evaluate the response stability of the same wheel under multiple different operating conditions; and / or, The optimization effect can be judged by comparing the stress field changes before and after the local structural modification using the target stress field distribution results; and / or, The target stress field distribution results are used as a preliminary screening basis for subsequent high-precision finite element analysis.

[0142] Example 3 Based on the same application concept, see [link / reference] Figure 11 As shown, Figure 11This illustration shows a structural schematic diagram of a computer device provided in Embodiment 3 of this application, wherein, as shown... Figure 11 As shown, the computer device 1100 provided in Embodiment 3 of this application includes: The system includes a processor 1101, a memory 1102, and a bus 1103. The memory 1102 stores machine-readable instructions that can be executed by the processor 1101. When the computer device 1100 is running, the processor 1101 and the memory 1102 communicate via the bus 1103. When the machine-readable instructions are executed by the processor 1101, they perform the steps of the wheel impact stress prediction method shown in Embodiment 1 above.

[0143] Example 4 Based on the same concept, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the wheel impact stress prediction method described in any of the above embodiments.

[0144] 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.

[0145] The computer program product for predicting wheel impact stress 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.

[0146] The wheel impact stress 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.

[0147] 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.

[0148] 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.

[0149] 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.

[0150] 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.

[0151] 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.

[0152] 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 wheel impact stress, characterized in that, The method includes: Obtain the three-dimensional geometric information, working parameters, and corresponding stress field distribution results of various different wheels; The three-dimensional geometric information of each wheel is converted into a geometric representation in a unified space; Each of the aforementioned operating condition parameters is converted into a conditional feature vector; Using the geometric representation and the conditional feature vector as inputs and the corresponding stress field distribution as outputs, a surrogate model based on geometric information neural operators is trained to obtain a stress prediction model. Based on the three-dimensional geometric information of the target wheel and the target working condition parameters, the target stress field distribution of the target wheel is predicted using the stress prediction model. The process of training a surrogate model based on a geometric information neural operator, using the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain a stress prediction model includes: An improved geometric information neural operator model is constructed as a surrogate model; wherein: the improved geometric information neural operator model integrates kernel integral operators and Fourier neural operators, including a geometric information receiving and representation module, a latent space feature modeling module, and a target position response recovery module, used to learn a mapping operator from the geometric-condition input function space to the stress field output function space, and the mapping operator to be learned is: , For the mapping operator from geometry-working condition to stress field, The symbolic distance function describes the geometric information of the wheel. Characteristics of impact load conditions. The three-dimensional stress field distribution to be predicted; the mapping operator approximates it through a combination of multiple integration operators: , For parameterized mapping operators, For the set of trainable parameters of the model, For output point mapping, , The first Layer, first-layer continuous kernel integral operator, For operator composition operations; the improved geometric information neural operator model realizes spatial information propagation through the kernel integral operator, wherein, the first Layered networks in spatial location The feature at the location is represented as: , For the first Layered networks in spatial location Feature representation at location, For the first Layer learnable kernel function, For the first Layer in position The characteristics of the place, The integral element is a spatial integral; the integral of the kernel integral operator is discretized in the local neighborhood: , For Center, radius The neighboring region For discrete sampling points within the neighborhood region, Sampling points Discrete weights; the improved geometric information neural operator model achieves global information interaction through the Fourier neural operator, wherein the first... The formula for calculating the layer feature update is: , These are the Fourier transform and the inverse Fourier transform, respectively. For frequency domain learnable kernel parameters, It is a linear mapping matrix. It is a non-linear activation function; The geometric representation and the conditional feature vector are input into the surrogate model to obtain the predicted stress field; wherein: the conditional feature vector is incorporated into the latent space through an embedding mapping to participate in operator calculation, establishing an end-to-end mapping from geometry-working condition to stress field, and the model input-output mapping relationship is as follows: , To predict the stress field distribution, To parameterize the IGINO mapping operator, The geometric representation matrix of the wheel, The working condition feature vector is used as the surrogate model. The surrogate model first integrates geometric features and working condition features, and then completes feature evolution through multi-layer kernel integral operators and Fourier neural operators. Finally, it outputs a three-dimensional continuous stress field covering the entire space of the wheel, which is used as the predicted stress field. Calculate the error between the predicted stress field and the corresponding stress field distribution; The parameters of the proxy model are updated based on the error; Repeat the above input, calculation, and update steps until the preset training stopping condition is met; The trained surrogate model is used as the stress prediction model.

2. The method according to claim 1, characterized in that, The process of converting the three-dimensional geometric information of each wheel into a geometric representation in a unified space includes: The three-dimensional geometry of each wheel is globally normalized so that the geometry of different wheels has a consistent spatial reference datum under a unified coordinate system. Construct a cubic latent space of a predetermined size, and embed the normalized wheel geometry into the cubic latent space; Multiple sampling points are set within the cubic hidden space, and the signed distance function value from each sampling point to the geometric surface of the wheel is calculated; The spatial coordinates of each sampling point and its signed distance function value together constitute the geometric feature vector of that sampling point; The geometric representation is formed by arranging the geometric feature vectors of all sampling points in the spatial order of the sampling points in the latent space of the cube.

3. The method according to claim 2, characterized in that, The calculation of the signed distance function value from each sampling point to the wheel geometry includes: A surface mesh model for obtaining the three-dimensional geometric information of the wheel; For each sampling point, calculate the shortest Euclidean distance from that sampling point to the surface mesh model; Depending on whether the sampling point is located inside or outside the surface mesh model, the shortest Euclidean distance is assigned a positive or negative sign to obtain the sign distance function value.

4. The method according to claim 1, characterized in that, The step of converting each of the operating condition parameters into a conditional feature vector includes: Extract the impact mass and impact height from the aforementioned operating parameters; The impact mass and the impact height are respectively normalized numerically. The normalized values ​​are combined into a multidimensional vector, which serves as the conditional feature vector.

5. The method according to claim 1, characterized in that, The step of predicting the target stress field distribution of the target wheel using the stress prediction model based on the target wheel's three-dimensional geometric information and target working parameters includes: The three-dimensional geometric information of the target wheel is converted into a target geometric representation; Convert the target operating condition parameters into a target condition feature vector; The target geometric representation and the target condition feature vector are input into the stress prediction model; The three-dimensional continuous stress field distribution result output by the stress prediction model is obtained as the target stress field distribution result.

6. The method according to claim 1, characterized in that, The method further includes: The high-stress regions or stress concentration areas in the target wheel are identified using the target stress field distribution results; and / or, The stress field distribution results of the target stress field are used to compare and rank the stress fields of multiple candidate wheel designs; and / or, The target stress field distribution results are used to evaluate the response stability of the same wheel under multiple different operating conditions; and / or, The optimization effect can be judged by comparing the stress field changes before and after the local structural modification using the target stress field distribution results; and / or, The target stress field distribution results are used as a preliminary screening basis for subsequent high-precision finite element analysis.

7. A wheel impact stress prediction device, characterized in that, The device includes: The wheel data acquisition module is used to acquire the three-dimensional geometric information, working parameters and corresponding stress field distribution results of various different wheels. A wheel 3D geometric information conversion module is used to convert the 3D geometric information of each wheel into a geometric representation in a unified space; The operating condition parameter conversion module is used to convert each of the operating condition parameters into a condition feature vector; The stress prediction model determination module is used to train a surrogate model based on geometric information neural operators, taking the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain the stress prediction model. The stress field distribution prediction module is used to predict the target stress field distribution of the target wheel based on the target wheel's three-dimensional geometric information and target working parameters using the stress prediction model. The process of training a surrogate model based on a geometric information neural operator, using the geometric representation and the conditional feature vector as input and the corresponding stress field distribution as output, to obtain a stress prediction model includes: An improved geometric information neural operator model is constructed as a surrogate model; wherein: the improved geometric information neural operator model integrates kernel integral operators and Fourier neural operators, including a geometric information receiving and representation module, a latent space feature modeling module, and a target position response recovery module, used to learn a mapping operator from the geometric-condition input function space to the stress field output function space, and the mapping operator to be learned is: , For the mapping operator from geometry-working condition to stress field, The symbolic distance function describes the geometric information of the wheel. Characteristics of impact load conditions. The three-dimensional stress field distribution to be predicted; the mapping operator approximates it through a combination of multiple integration operators: , For parameterized mapping operators, For the set of trainable parameters of the model, For output point mapping, , The first Layer, first-layer continuous kernel integral operator, For operator composition operations; the improved geometric information neural operator model realizes spatial information propagation through the kernel integral operator, wherein, the first Layered networks in spatial location The feature at the location is represented as: , For the first Layered networks in spatial location Feature representation at location, For the first Layer learnable kernel function, For the first Layer in position The characteristics of the place, The integral element is a spatial integral; the integral of the kernel integral operator is discretized in the local neighborhood: , For Center, radius The neighboring region For discrete sampling points within the neighborhood region, Sampling points Discrete weights; the improved geometric information neural operator model achieves global information interaction through the Fourier neural operator, wherein the first... The formula for calculating the layer feature update is: , These are the Fourier transform and the inverse Fourier transform, respectively. For frequency domain learnable kernel parameters, It is a linear mapping matrix. It is a non-linear activation function; The geometric representation and the conditional feature vector are input into the surrogate model to obtain the predicted stress field; wherein: the conditional feature vector is incorporated into the latent space through an embedding mapping to participate in operator calculation, establishing an end-to-end mapping from geometry-working condition to stress field, and the model input-output mapping relationship is as follows: , To predict the stress field distribution, To parameterize the IGINO mapping operator, The geometric representation matrix of the wheel, The working condition feature vector is used as the surrogate model. The surrogate model first integrates geometric features and working condition features, and then completes feature evolution through multi-layer kernel integral operators and Fourier neural operators. Finally, it outputs a three-dimensional continuous stress field covering the entire space of the wheel, which is used as the predicted stress field. Calculate the error between the predicted stress field and the corresponding stress field distribution; The parameters of the proxy model are updated based on the error; Repeat the above input, calculation, and update steps until the preset training stopping condition is met; The trained surrogate model is used as the stress prediction model.

8. A computer device, characterized in that, include: The system 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 wheel impact stress prediction method as described in any one of claims 1 to 6.

9. 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 wheel impact stress prediction method as described in any one of claims 1 to 6.

Images