Pedestrian leg collaborative optimization method and system based on AI

Abstract

The invention provides an AI-based pedestrian leg collaborative optimization method and system. The system comprises an intelligent evaluation model and an intelligent optimization model. Comprising the following steps: constructing a leg graph structure; calculating an adjacent matrix, a degree matrix and a Laplacian matrix based on the leg graph structure; constructing a feature matrix based on the vehicle head finite element simulation model; constructing observation data based on the pedestrian leg model and the vehicle head finite element simulation model; constructing a training set and a test set based on the feature matrix and the observation data; training an intelligent evaluation model based on the training set and the test set; setting a scoring system, optimizing an objective function and constraint conditions, and constructing an intelligent optimization model by adopting a gradient ascending algorithm; and performing intelligent evaluation optimization on the new vehicle model. The method has the beneficial effects that the single-working-condition optimization period of the vehicle pedestrian leg protection design is shortened to be within 1 hour from 24-48 hours.

Description

Technical Field

[0001] This invention belongs to the field of automotive passive safety technology, and in particular relates to an AI-based method and system for optimizing pedestrian leg coordination. Background Technology

[0002] In global automotive safety testing systems, such as the New Car Assessment Program (NCAP) in various countries, quantitative indicators are required for pedestrian leg protection.

[0003] In current automotive development, pedestrian leg protection design commonly employs finite element method (FEM) simulation to predict collision damage. This technology involves constructing detailed models of the vehicle's front end and the pedestrian's legs, placing sensor models at key leg assessment points, and simulating the force transmission and injury generation mechanisms during a collision. The process employs a "simulation calculation - curve analysis - animation backtracking - structural optimization - simulation verification" approach. Engineers must first locate the point of exceeding the limit using the damage curve (e.g., the peak bending moment of the lower leg occurs at 50ms), then manually replay the collision animation to find the contact area at that moment, and rely on experience to determine the structural components that need optimization. Coarse-grained strategies may be employed, such as weakening protruding structural parts or increasing / decreasing material thickness by 0.5mm. Each optimization requires resubmitting the full model for simulation verification, resulting in an average single-condition optimization cycle of 24-48 hours.

[0004] This "simulation calculation - curve analysis - animation backtracking - structural optimization - simulation verification" work mode inherently suffers from core flaws such as the fragmentation of spatiotemporal analysis (requiring manual switching between time-dimensional curves and spatial-dimensional animations) and the inability to quickly and collaboratively develop and optimize multiple optimization parts of the leg, making it difficult to balance development efficiency and optimization quality. For automakers, the finite element simulation method is ill-suited to the rapid development and iteration needs of automobiles. Relying heavily on engineers' experience and repeated trial adjustments is not only inefficient but also fails to accurately meet increasingly stringent safety regulations. Therefore, developing an AI-based pedestrian leg optimization method and system is of great significance for improving the pedestrian protection performance of automobiles and accelerating the automotive R&D process. Summary of the Invention

[0005] In view of this, the present invention aims to propose an AI-based pedestrian leg coordination optimization method and system to solve the problems of low efficiency of existing technologies and difficulty in accurately meeting increasingly stringent safety requirements, which makes it difficult to adapt to the rapid development and iteration needs of automobiles.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows: In the first aspect, the present invention proposes an AI-based pedestrian leg coordination optimization system, including an intelligent evaluation model and an intelligent optimization model; The intelligent evaluation model is used to predict the value of the leg observation position; the intelligent optimization model is used to calculate the scores of each indicator and the comprehensive score, to determine whether the new model meets the requirements, and if the new model does not meet the requirements, to optimize the feature parameters of the new model in a targeted manner, and output a design scheme that meets the requirements. The intelligent evaluation model includes graph convolution and result output; graph convolution is used for feature extraction; the result output includes Dropout and fully connected layers; the Dropout layer is used to prevent overfitting during training; and the fully connected layers are used to map the feature vectors output by the graph convolution to the final result.

[0007] Secondly, based on the same concept, this invention also proposes an AI-based pedestrian leg coordination optimization method, comprising the following steps: S1. Construct the leg diagram structure; S2. Based on the leg graph structure, the adjacency matrix and degree matrix are calculated, and the Laplacian matrix is ​​calculated based on the adjacency matrix and degree matrix. S3. Obtain the finite element simulation model of the vehicle front, and construct the feature matrix based on the finite element simulation model of the vehicle front; S4. Based on the pedestrian leg model and the finite element simulation model of the vehicle front, construct observation data; S5. Based on the feature matrix and observation data, construct the training set and the test set; S6. Train the intelligent evaluation model based on the training and test sets; S7. Based on the intelligent evaluation model, set the scoring system, optimize the objective function and constraints, and use the gradient ascent algorithm to construct an intelligent optimization model; S8. Utilize intelligent evaluation and intelligent optimization models to perform intelligent evaluation and optimization on new vehicle models.

[0008] Furthermore, in step S1, a leg diagram structure is constructed. S11. Set leg nodes: Based on the leg observation position of the leg impact point, set leg nodes in the order of thigh-knee-lower leg with the knee as the center. S12. Construct the edge relationships of leg nodes: Based on the connection method of human leg bones and the axial transmission path, the association of collision damage of each leg node is abstracted into edge relationships, forming a set of directed edges; S13. The expression for the leg diagram structure is as follows: ; In the formula, The diagram shows the structure of the legs. For the set of leg nodes, Let be the set of directed edges for the leg nodes.

[0009] Furthermore, in step S11, the leg observation positions include 3 thigh bending moments, 4 calf bending moments, and knee ligament extension. In step S12, the edge relationships include edge relationships based on axial conduction, edge relationships based on bidirectional energy interaction of the knee joint, edge relationships based on distal energy convergence, and edge relationships based on damage radiation effects.

[0010] Furthermore, in step S3, the feature matrix includes features in five dimensions: front bumper area, cabin trim area, front bumper light assembly area, windshield wiper assembly area, and leg impact point.

[0011] Furthermore, in step S4, based on the pedestrian leg model and the vehicle front finite element simulation model, observation data is constructed, including: S41. Based on the pedestrian leg model and the vehicle front finite element simulation model, use simulation software to calculate the acceleration curves at different leg observation positions at the leg impact point. S42. Calculate the peak values ​​of thigh bending moment, lower leg bending moment and knee ligament extension during a specified time period from the moment the leg first contacts the vehicle, and use these as simulation observation values ​​for the leg node, and construct observation data based on the simulation observation values.

[0012] Furthermore, in step S6, based on the training set and the test set, an intelligent evaluation model is trained, including: S61. Using the adjacency matrix, degree matrix, and Laplacian matrix as the topological structure, combined with the feature matrix, the graph convolutional layer updates the node features layer by layer according to the topological relationship. The calculation formula for the graph convolutional layer is as follows: ; In the formula, For the first The feature matrix of the layer, This is the standardized adjacency matrix. For the first The weight matrix of the layer, It is the ReLU activation function; The expression for the standardization of the adjacency matrix is ​​as follows: ; In the formula, It is the identity matrix. It is an adjacency matrix. For degree matrix, This is the standardized adjacency matrix; S62. Using mean squared error as the loss function and observed data as the monitoring signal, the calculation expression of the loss function is as follows: ; In the formula, For loss function, Let i be the feature matrix of the i-th data. For the i-th data point, The graph convolutional layer predicts the output value of the i-th data point. S63. The weight matrix of each layer is optimized using the Adam algorithm to minimize the error between the predicted data and the observed data of the intelligent evaluation model. Furthermore, in step S7, a scoring system, an optimization objective function, and constraints are set, and a gradient ascent algorithm is used to construct an intelligent optimization model, including: S71. Based on the known rating levels and rating limit standards, set... , and The scoring rules for the indicators define the rules for calculating the total score for leg impact points. S72. Define the optimization objective function: For a vehicle model, the sum of the comprehensive scores of the leg impact points is the total leg score, and the objective is to maximize the total leg score; S73. Set constraints; S74. Using the gradient ascent algorithm, the feature matrix is ​​continuously adjusted until the optimization objective function is satisfied, as shown in the following expression: ; In the formula, For the current feature matrix, This is the updated feature matrix for the next iteration. For gradient operators, The total score for the legs. Let be the gradient of the total leg score with respect to the features. This is the learning rate.

[0013] Furthermore, in step S8, the new vehicle model is intelligently evaluated and optimized using the intelligent evaluation model and the intelligent optimization model, including: S81. Obtain the finite element simulation model of the front of the new vehicle model and construct the feature matrix of the new vehicle model; S82. Input the feature matrix into the intelligent evaluation model to obtain the predicted value of the leg observation position; S83. Calculate the predicted value based on the leg observation position. , and The system calculates the scores for each indicator and the total score for each leg, and then determines whether the new model meets the requirements. S84. If the new model does not meet the requirements, the parameters of the front structure will be optimized in a targeted manner, and a design scheme that meets the requirements will be output.

[0014] Compared with existing technologies, the AI-based pedestrian leg coordination optimization method and system described in this invention have the following beneficial effects: (1) By using node design based on mechanical correlation and graph neural network, accurate collaborative analysis of damage to key parts of pedestrian legs is achieved, enabling intelligent assessment and optimization of pedestrian protection for new vehicle models.

[0015] (2) Compared with traditional simulation methods, the single-condition optimization cycle of vehicle and pedestrian leg protection design is shortened from 24-48 hours to less than 1 hour, which greatly improves the R&D efficiency.

[0016] (3) It can be directly applied to the development process of new vehicle models, guide the design and improvement of the front structure, effectively reduce the risk of pedestrian leg injury, enable the vehicle's pedestrian protection performance to meet and exceed regulatory target requirements, provide innovative solutions for the development of automobile pedestrian safety technology, and has significant economic and social benefits. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the leg structure according to an embodiment of the present invention; Figure 2 The adjacency matrix of the leg graph structure described in the embodiments of the present invention. Degree matrix and Laplace matrix Schematic diagram; Figure 3 This is a schematic diagram of the overall architecture of the intelligent evaluation model described in an embodiment of the present invention; Figure 4 This is a schematic diagram of the loss function curve of the intelligent evaluation model described in an embodiment of the present invention; Figure 5 This is a schematic diagram of the intelligent evaluation and optimization process for new vehicle models as described in an embodiment of the present invention; Figure 6 This is a schematic diagram of the pedestrian leg observation area according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the pedestrian leg model and the finite element simulation model of the vehicle front as described in an embodiment of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0019] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0020] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] like Figures 1 to 7 As shown, an AI-based pedestrian leg coordination optimization method includes the following steps: S1. Construct the leg diagram structure; S2. Based on the leg graph structure, the adjacency matrix and degree matrix are calculated, and the Laplacian matrix is ​​calculated based on the adjacency matrix and degree matrix. S3. Obtain the finite element simulation model of the vehicle front, and construct the feature matrix based on the finite element simulation model of the vehicle front; S4. Based on the pedestrian leg model and the finite element simulation model of the vehicle front, construct observation data; S5. Based on the feature matrix and observation data, construct the training set and the test set; S6. Train the intelligent evaluation model based on the training and test sets; S7. Based on the intelligent evaluation model, set the scoring system, optimize the objective function and constraints, and use the gradient ascent algorithm to construct an intelligent optimization model; S8. Utilize intelligent evaluation and intelligent optimization models to perform intelligent evaluation and optimization on new vehicle models.

[0023] Specifically as follows: I. Constructing the leg diagram structure.

[0024] Eight leg observation positions were set up: three thigh bending moments, four calf bending moments, and the knee ligament extension MCL. With the knee as the center, the leg observation positions were defined as leg nodes in the order of thigh-knee-calf, i.e., V={thigh bending moment 3, thigh bending moment 2, thigh bending moment 1, lumbar ligament MCL, calf bending moment 1, calf bending moment 2, calf bending moment 3, calf bending moment 4}, and the leg nodes were labeled as v1-v8, as shown in Table 1.

[0025] Table 1

[0026] Based on the connection method and axial transmission path of human leg bones, the correlation of collision damage at various leg nodes is abstracted as edge relationships. A single arrow represents unidirectional energy transfer, and a double arrow represents bidirectional energy interaction, forming a set of directed edges. The methods for constructing edge relationships include: edge relationships based on axial conduction, edge relationships based on bidirectional energy interaction of the knee joint, edge relationships based on distal energy convergence, and edge relationships based on injury radiation effects.

[0027] During a car collision, energy is primarily conducted axially. In the human leg, the femur and tibia are longitudinally aligned, allowing energy to be transmitted along the long axis of the bones. The knee joint is the core mechanical connection between the thigh and lower leg, and its medial collateral ligament (MCL) is the only soft tissue node that "crosses the thigh and lower leg." Energy transmitted axially from the femur to the knee joint is then transmitted "across the bones" to the proximal tibia via the MCL. Conversely, the axial energy borne by the tibia is also fed back to the femur via the MCL, making the MCL a "cross-hub" for energy transmission in the lower limbs. Therefore, the constructed directed edge set... As shown below: 1. Based on axial transmission Establish a unidirectional transmission edge from thigh bending moment 3 to thigh bending moment 2 to thigh bending moment 1; Establish a unidirectional transmission edge for lower leg bending moment 1 → lower leg bending moment 2 → lower leg bending moment 3 → lower leg bending moment 4; 2. Based on bidirectional energy interaction of the knee joint Establish a bidirectional coupling edge for thigh moment 1 ↔ knee ligament extension (MCL) ↔ lower leg moment 1; 3. Based on remote energy convergence Establish unidirectional transmission sides for thigh moment 3 → knee ligament extension (MCL) and lower leg moment 4 → knee ligament extension (MCL); 4. Based on the damage radiation effect Establish the unidirectional influence sides of knee ligament extension (MCL) → lower leg bending moment 2 and knee ligament extension (MCL) → lower leg bending moment 3.

[0028] like Figure 1 As shown, the leg graph structure formed by the leg nodes and their edge relationships is expressed as follows: ; In the formula, This is a diagram of the leg structure. For the set of leg nodes, Let be the set of directed edges for the leg nodes.

[0029] 2. Based on the leg diagram structure, calculate the adjacency matrix. degree matrix According to the adjacency matrix degree matrix The Laplace matrix representing the energy conduction characteristics of collision damage was calculated. .

[0030] Among them, the adjacency matrix is ​​based on the number of nodes. Used to quantify node connection relationships, matrix The elements in are If node arrive If a propagation edge exists, then =1, otherwise, =0.

[0031] Degree matrix Used to represent the number of connected edges of a node, diagonal element Represents a node The number of connected edges (degrees).

[0032] Laplace matrix Used to characterize the conduction properties of a graph structure, describing the "energy / force" transfer relationship at nodes, the calculation expression is as follows: ; In the formula, For Laplace matrix, For degree matrix, Let be an adjacency matrix. A positive value indicates that the direction of force transmission is from the node. .on the contrary, Negative values ​​indicate that the direction of force transmission is from the node. of.

[0033] like Figure 2 As shown, the adjacency matrix of the leg graph structure Degree matrix and Laplace matrix .

[0034] 3. Obtain the finite element simulation model of the vehicle front, and construct the feature matrix based on the finite element simulation model of the vehicle front; construct the observation data based on the pedestrian leg model and the finite element simulation model of the vehicle front; construct the training set and the test set based on the feature matrix and the observation data.

[0035] Based on C-NCAP regulations, 150 versions of the front-end finite element simulation models from the development process of 30 historical vehicle models were obtained, resulting in a total of 2200 leg impact points. The front-end finite element simulation models were divided into four regions. Based on these four regions and the leg impact points, five-dimensional feature data was constructed. This feature data underwent preprocessing (Z-score normalization was applied to numerical data to eliminate the influence of dimensions; material names were used as type data and mapped to one-heat encoding), resulting in a 1×32-dimensional feature matrix of the leg impact points. ;like Figure 7 As shown, a pedestrian leg model and a vehicle front finite element simulation model are presented. The pedestrian leg model is a finite element simulation model of the pedestrian leg impactor. C-NCAP regulations require the use of the aPLI leg model. Based on the pedestrian leg model and the vehicle front finite element simulation model, simulation software was used to calculate the peak values ​​of thigh bending moment, lower leg bending moment, and knee ligament extension within a 60ms time period from the moment the leg first contacts the vehicle. These values ​​were used as simulation observations of the leg nodes, and observation data were constructed based on these simulation observations. .

[0036] Feature matrix The system comprises five dimensions, with each region representing one dimension. The four regions are the front bumper area, engine compartment trim area, front bumper light assembly area, and wiper assembly area. The characteristic parameters for each region include the spatial arrangement parameters of the components, material name, material thickness, and material properties. The leg impact point is the fifth dimension, and its characteristic parameters include the position of the leg impact point relative to the front of the vehicle. The total number of characteristic parameters is 32.

[0037] The feature parameters for the five dimensions are as follows: 1. Front bumper area: name of anti-collision beam material, thickness of anti-collision beam material, density of anti-collision beam material, X-direction distance from anti-collision beam to front bumper skin, name of front bumper cover material, thickness of front bumper cover material, density of front bumper cover material, name of central lower grille material, thickness of central lower grille material, density of central lower grille material, name of pedestrian lower leg protection bracket material, thickness of pedestrian lower leg protection bracket material, density of pedestrian lower leg protection bracket material; 2. Cabin trim area: Air guide material name, air guide material thickness, air guide material density; front frame material name, front frame material thickness, front frame material density; cabin upper lining material name, cabin upper lining material thickness, cabin upper lining material density. 3. Front bumper light assembly area: Name of the material of the through light body, thickness of the through light body, density of the through light body, name of the material of the headlight body, thickness of the headlight body, density of the headlight body; 4. Wiper Assembly Area: Name of wiper rotating material, thickness of wiper rotating material, density of wiper rotating material; 5. Leg impact point: distance in the y direction relative to the center line of the front of the vehicle.

[0038] The material name, thickness, and density were obtained by reading the finite element simulation model of the vehicle front using CAE simulation software.

[0039] Table 2 shows partial observation data for leg bending moment (unit: N×m) and knee ligament extension (MCL). .

[0040] Table 2

[0041] Feature matrices based on a large number of vehicle models With observation data To form a training dataset , The total number of samples is 2200. The training set and the test set are divided according to an 8:2 ratio.

[0042] IV. Constructing an intelligent evaluation model based on graph neural networks.

[0043] like Figure 3 As shown, the overall architecture of the intelligent evaluation model includes graph convolution and result output; graph convolution is used for feature extraction; the result output includes Dropout and fully connected layers; the Dropout layer is used to prevent overfitting during training; and the fully connected layers are used to map the feature vectors output by the graph convolution to the final result.

[0044] Based on the constructed training dataset We will construct an intelligent evaluation model based on graph neural networks.

[0045] 1. Using the adjacency matrix, degree matrix, and Laplacian matrix as the topological structure, combined with the feature matrix, the graph convolutional layer updates the node features layer by layer according to the topological relationship; The calculation formula for the graph convolutional layer is as follows: ; In the formula, For the first The feature matrix of the layer, For the first The feature matrix of the layer, This is the standardized adjacency matrix. For the first The weight matrix of the layer, It is the ReLU activation function; The expression for the standardization of the adjacency matrix is ​​as follows: ; In the formula, It is the identity matrix. It is an adjacency matrix. For degree matrix, This is the standardized adjacency matrix.

[0046] 2. Using mean squared error as the loss function and observed data as the monitoring signal, the calculation expression of the loss function is as follows: ; In the formula, For loss function, Let i be the feature matrix of the i-th data. For the i-th data point, Let be the predicted output value of the convolutional layer for the i-th data point.

[0047] like Figure 4 As shown in the loss function curve, the intelligent evaluation model converges.

[0048] 3. The weight matrix W of each layer is optimized using the Adam algorithm to minimize the error between the predicted data and the observed data of the intelligent evaluation model; Table 3 shows the hyperparameter settings and values ​​during the training process.

[0049] Table 3 Hyperparameter settings Setting value The output dimension of the convolutional layer is shown. 64 Output layer dimension 8 Dropout rate 0.3 Optimizer type Adam Batch Size 32 Loss function type MSE Learning rate 0.001 Number of iterations 200 4. On the test set, use the coefficient of determination R. 2 The accuracy of the prediction data for the leg nodes is evaluated, as shown in Table 4. The coefficient of determination R for the prediction data of the eight leg nodes is... 2 All values ​​are greater than 0.7, indicating that the intelligent evaluation model has high prediction accuracy.

[0050] Table 4

[0051] V. Based on the intelligent evaluation model, a scoring system, an optimized objective function and constraints are set, and an intelligent optimization model is constructed using the gradient ascent algorithm.

[0052] As shown in Tables 5 and 6, the evaluation grading standards and scoring limits for leg observation positions are as follows, according to C-NCAP regulations.

[0053] Table 5 rating level FEMUR-MOMENT TIBIA-MOMENT MCL-Elongation good FEMUR<351 TIBIA<248 MCL<24 Points lost 351<FEMUR<396 248<TIBIA<288 24<MCL<28 Zero points FEMUR>396 TIBIA>288 MCL>28 Table 6 ZONE FemurMoment(Nm) TibiaMoment(Nm) MCL(mm) Limit 351~396 275~320 24~28 Score 0.4~0 0.4~0 0.2~0 1. Based on the known rating levels and rating limit standards, set... , and The rules for calculating indicator scores are as follows: when hour, The indicator score is 0.4; when When linear interpolation is used for calculation Indicator score; when hour, The index score is 0; linear interpolation calculation , as well as The expression for the indicator score is as follows: ; ; ; In the formula, for Indicator scores for Indicator scores for Indicator scores This represents the bending moment value of the thigh. This represents the bending moment value of the lower leg. This refers to the knee ligament extension value; Note: 4 lower leg flexion moments and 3 thigh flexion moments, based on the worst-case scenario. and Calculate the indicator scores; The formula for calculating the total score for the legs is as follows: ; In the formula, The total score for the legs. for Indicator scores for Indicator scores for Indicator score; 2. Define the optimization objective function: For a given vehicle model, the sum of the comprehensive scores of the leg impact points is the total leg score, and the objective is to maximize the total leg score; 3. Set constraints: Meet C-NCAP constraints, and each observation value does not exceed the maximum limit specified in the regulations; 4. Employ the gradient ascent algorithm to continuously adjust the feature matrix until the optimization objective function is satisfied, as shown in the following expression: ; In the formula, For the current feature matrix, This is the updated feature matrix for the next iteration. For gradient operators, The total score for the legs. Let be the gradient of the total leg score with respect to the features. This is the learning rate.

[0054] The workflow based on the intelligent optimization model is as follows: 1. Input the feature matrix of 15 leg impact points of a finite element model of a vehicle's front end. ; 2. Calculate and predict the leg biomechanical response using an intelligent evaluation model. ; 3. Calculation and gradient ; 4. Adjust features along the gradient ascent direction, with a convergence threshold of 0.005 and a maximum of 50 iterations; 5. Calculate the current score change rate. If the change rate is <0.005 for 5 consecutive rounds and If the number of iterations is less than 15, the iteration will terminate and the current best feature will be output. If the number of iterations is greater than or equal to 50, the iteration will be forcibly terminated and the current result will be output. Otherwise, the iteration will continue.

[0055] VI. Utilize intelligent evaluation and intelligent optimization models to conduct intelligent evaluation and optimization of new vehicle models.

[0056] like Figure 5 As shown, this illustrates the process of intelligent evaluation and optimization for new vehicle models.

[0057] 1. Obtain the finite element simulation model of the front of the new vehicle model and construct the feature matrix of the new vehicle model; 2. Input the feature matrix into the intelligent evaluation model to obtain predicted data on the leg observation position; 3. Based on the predicted data of the leg observation position, calculate... , and The system calculates the scores for each indicator and the total score for each leg, and then determines whether the new model meets the requirements. 4. If the new model does not meet the requirements, the parameters of the front structure will be optimized accordingly, and a design scheme that meets the regulatory objectives will be output.

[0058] The advantages and beneficial effects of this invention are as follows: (1) By using node design based on mechanical correlation and graph neural network, accurate collaborative analysis of damage to key parts of pedestrian legs is achieved, enabling intelligent assessment and optimization of pedestrian protection for new vehicle models.

[0059] (2) Compared with traditional simulation methods, the single-condition optimization cycle of vehicle and pedestrian leg protection design is shortened from 24-48 hours to less than 1 hour, which greatly improves the R&D efficiency.

[0060] (3) It can be directly applied to the development process of new vehicle models, guide the design and improvement of the front structure, effectively reduce the risk of pedestrian leg injury, enable the vehicle's pedestrian protection performance to meet and exceed regulatory target requirements, provide innovative solutions for the development of automobile pedestrian safety technology, and has significant economic and social benefits.

[0061] English definition: FEMUR: Thigh bending moment; TIBIA: Lower leg bending moment; MCL: Knee ligament stretch.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based pedestrian leg coordination optimization system, characterized in that: This includes intelligent evaluation models and intelligent optimization models; The intelligent evaluation model is used to predict the value of the leg observation position; the intelligent optimization model is used to calculate the index score and the comprehensive score, determine whether the new model meets the requirements, optimize the feature parameters of the new model that does not meet the requirements, and output the design scheme that meets the requirements. The intelligent evaluation model includes graph convolution and result output; graph convolution is used for feature extraction; the result output includes Dropout and fully connected layers; the Dropout layer is used to prevent overfitting during training; and the fully connected layers are used to map the feature vectors output by the graph convolution to the final result.

2. An AI-based pedestrian leg coordination optimization method, applied to the AI-based pedestrian leg coordination optimization system described in claim 1, characterized in that: Includes the following steps: S1. Construct the leg diagram structure; S2. Based on the leg graph structure, the adjacency matrix and degree matrix are calculated, and the Laplacian matrix is ​​calculated based on the adjacency matrix and degree matrix. S3. Obtain the finite element simulation model of the vehicle front, and construct the feature matrix based on the finite element simulation model of the vehicle front; S4. Based on the pedestrian leg model and the finite element simulation model of the vehicle front, construct observation data; S5. Based on the feature matrix and observation data, construct the training set and the test set; S6. Train the intelligent evaluation model based on the training and test sets; S7. Based on the intelligent evaluation model, set the scoring system, optimize the objective function and constraints, and use the gradient ascent algorithm to construct an intelligent optimization model; S8. Utilize intelligent evaluation and intelligent optimization models to perform intelligent evaluation and optimization on new vehicle models.

3. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S1, the leg diagram structure is constructed. S11. Set leg nodes: Based on the leg observation position of the leg impact point, set leg nodes in the order of thigh-knee-lower leg with the knee as the center. S12. Construct the edge relationships of leg nodes: Based on the connection method of human leg bones and the axial transmission path, the association of collision damage of each leg node is abstracted into edge relationships, forming a set of directed edges; S13. The expression for the leg diagram structure is as follows: ； In the formula, The diagram shows the structure of the legs. For the set of leg nodes, Let be the set of directed edges for the leg nodes.

4. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S11, the leg observation positions include 3 thigh bending moments, 4 lower leg bending moments, and knee ligament extension. In step S12, the edge relationships include edge relationships based on axial conduction, edge relationships based on bidirectional energy interaction of the knee joint, edge relationships based on distal energy convergence, and edge relationships based on damage radiation effects.

5. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S3, the feature matrix includes features in five dimensions: front bumper area, cabin trim area, front bumper light assembly area, windshield wiper assembly area, and leg impact point.

6. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S4, based on the pedestrian leg model and the vehicle front finite element simulation model, observation data is constructed, including: S41. Based on the pedestrian leg model and the vehicle front finite element simulation model, use simulation software to calculate the acceleration curves at different leg observation positions at the leg impact point. S42. Calculate the peak values ​​of thigh bending moment, lower leg bending moment and knee ligament extension during a specified time period from the moment the leg first contacts the vehicle, and use these as simulation observation values ​​for the leg node, and construct observation data based on the simulation observation values.

7. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S6, based on the training set and the test set, an intelligent evaluation model is trained, including: S61. Using the adjacency matrix, degree matrix, and Laplacian matrix as the topological structure, combined with the feature matrix, the graph convolutional layer updates the node features layer by layer according to the topological relationship. The calculation formula for the graph convolutional layer is as follows: ； In the formula, For the first The feature matrix of the layer, This is the standardized adjacency matrix. For the first The weight matrix of the layer, It is the ReLU activation function; The expression for the standardization of the adjacency matrix is ​​as follows: ； In the formula, It is the identity matrix. It is an adjacency matrix. For degree matrix, This is the standardized adjacency matrix; S62. Using mean squared error as the loss function and observed data as the monitoring signal, the calculation expression of the loss function is as follows: ； In the formula, For loss function, Let i be the feature matrix of the i-th data. For the i-th data point, The graph convolutional layer predicts the output value of the i-th data point. S63. The weight matrix of each layer is optimized using the Adam algorithm to minimize the error between the predicted data and the observed data of the intelligent evaluation model.

8. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S7, the scoring system, objective function, and constraints are set, and the gradient ascent algorithm is used to construct an intelligent optimization model, including: S71. Based on the known rating levels and rating limit standards, set... , and The scoring rules for the indicators define the rules for calculating the total score for leg impact points. S72. Define the optimization objective function: For a vehicle model, the sum of the comprehensive scores of the leg impact points is the total leg score, and the objective is to maximize the total leg score; S73. Set constraints; S74. Using the gradient ascent algorithm, the feature matrix is ​​continuously adjusted until the optimization objective function is satisfied, as shown in the following expression: ； In the formula, For the current feature matrix, This is the updated feature matrix for the next iteration. For gradient operators, The total score for the legs. The gradient of the total leg score with respect to the features. This is the learning rate.

9. The AI-based pedestrian leg coordination optimization method according to claim 2, characterized in that: In step S8, the new vehicle model is intelligently evaluated and optimized using the intelligent evaluation model and the intelligent optimization model, including: S81. Obtain the finite element simulation model of the front of the new vehicle model and construct the feature matrix of the new vehicle model; S82. Input the feature matrix into the intelligent evaluation model to obtain the predicted value of the leg observation position; S83. Calculate the predicted value based on the leg observation position. , and The system calculates the scores for each indicator and the total score for each leg, and then determines whether the new model meets the requirements. S84. If the new model does not meet the requirements, the parameters of the front structure will be optimized in a targeted manner, and a design scheme that meets the requirements will be output.

Images