Vehicle moving component and physical attribute analysis method and device thereof

Abstract

The invention relates to the technical field of vehicle dynamics simulation and virtual testing, in particular to a vehicle motion component and a physical attribute analysis method and device thereof. The method comprises the steps of S1, receiving and storing historical dynamic data of a vehicle under multiple scenes and multiple moments, completing component division based on a vehicle three-dimensional model, and extracting dynamic features and physical attribute features of components in combination with the historical dynamic data; s2, constructing tensor representation of a vehicle motion component based on historical dynamic data and a component division result; s3, constructing a tensor representation learning framework based on Tucker decomposition, and outputting an optimizable parameter set; s4, receiving an optimizable parameter set, integrating physical constraints about dynamic interaction of the components, and constructing a target loss function; and S5, based on the optimized hidden features and the dynamic interaction relation of the components, combining the topological information of the vehicle structure to synthesize a whole vehicle motion sequence, and the method can be widely applied to vehicle dynamics simulation, automatic driving systems and other fields related to vehicle dynamic simulation.

Description

Technical Field

[0001] This invention relates to the technical field of vehicle dynamics simulation and virtual testing, specifically to a method and apparatus for analyzing the physical properties of moving vehicle components. Background Technology

[0002] With the rapid development of intelligent connected vehicles and autonomous driving technologies, the simulation of vehicle dynamic behavior in complex and multi-scenario environments has become a crucial aspect of the research and development process. Accurately simulating the coordinated motion of various vehicle components is of great significance for overall vehicle performance evaluation, safety verification, and control system design. As a complex multibody dynamics system, a vehicle exhibits close kinematic coupling and mechanical interactions among its components.

[0003] Currently, many researchers have proposed vehicle motion simulation methods based on traditional dynamic equations or finite element analysis. These methods predict the motion state of components by establishing detailed physical models. However, methods based on traditional physical modeling suffer from high computational complexity, difficulty in parameter tuning, and difficulty in extracting the implicit motion coordination laws between components from massive amounts of historical operating data. Especially when dealing with vehicle dynamic responses under multiple scenarios and operating conditions, traditional methods often fail to effectively capture the complex interaction effects between components, leading to significant deviations between simulation results and actual observations.

[0004] On the other hand, while purely data-driven machine learning methods can learn motion patterns from historical data, they often lack explicit constraints on physical laws, resulting in component motion sequences that are physically unreasonable or lack coordination. How to learn coordinated motion patterns between components from data while ensuring physical plausibility, and generate vehicle-level motion sequences that conform to real physical laws, has become a critical problem urgently needing to be solved in the field of vehicle virtual testing.

[0005] Another prominent problem in existing technologies is that the motion characteristics and physical properties of vehicle components are often modeled and analyzed separately, ignoring the inherent relationship between them. This separate approach makes it impossible to fully utilize the constraints imposed by the physical properties of components on their motion behavior, and also makes it difficult to reveal the motion coordination mechanisms between components with different physical properties.

[0006] Therefore, there is an urgent need for a joint analysis method that can simultaneously consider the motion characteristics, physical properties, and interactions between components, so as to automatically learn coordinated motion patterns from historical operating data and generate physically reasonable and coordinated vehicle motion sequences. Summary of the Invention

[0007] The present invention aims to provide a method and apparatus for analyzing the moving components of a vehicle and their physical properties, which can automatically learn coordinated motion patterns from historical operating data and generate a physically reasonable and coordinated whole vehicle motion sequence.

[0008] To address the aforementioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for analyzing the physical properties of a vehicle's moving components, comprising the following steps: S1: Receive and store historical dynamic data of the vehicle in multiple scenarios and at multiple times, complete component division based on the vehicle's 3D model, and extract the dynamic features and physical attribute features of the components by combining historical dynamic data; S2: Construct tensor representations of vehicle motion components based on historical dynamic data and component partitioning results; S3: Construct a tensor representation learning framework based on Tucker decomposition, and output an optimizable parameter set including the core tensor and factor matrix; S4: Receive the optimizable parameter set and integrate the physical constraints on the dynamic interaction of components to construct the target loss function; S5: Based on the optimized hidden features and the dynamic interaction relationship of components, the vehicle motion sequence is synthesized by combining the vehicle structural topology information.

[0009] By adopting the above technical solution, in the process of analyzing the moving components and their physical properties of a vehicle, the historical dynamic data of the corresponding vehicle is first acquired and preprocessed, and the vehicle is divided into components to obtain the dynamic and physical property characteristics of the components. Since the historical dynamic data is obtained from multiple scenarios and at multiple times, it contains corresponding scenario (acceleration, braking, turning) and temporal information. After dividing the vehicle into multiple components, scenario and temporal information can be assigned to each component, thereby unifying the dynamic and physical property characteristics of the components. Then, by constructing a tensor representation learning framework to output an optimizable parameter set, and combining it with the dynamic interaction relationships of the components to construct a target loss function, the entire vehicle motion sequence can be synthesized based on the optimized latent features and the dynamic interaction relationships of the components, combined with the vehicle's structural topology information. In this process, because the dynamic and physical property characteristics of the components are unified, coordinated motion patterns can be automatically learned from historical operating data, which can be widely applied to vehicle dynamics simulation, autonomous driving systems, and other fields related to vehicle dynamics simulation.

[0010] Optionally, S2 includes: The historical dynamic data and component partitioning results are organized into a fourth-order vehicle dynamic attribute tensor. : in, Indicates the number of components; The feature dimension includes motion parameters describing the motion state of each component and physical property parameters describing the mechanical properties of the material. Indicates the number of time slices; Indicates the number of scenes.

[0011] Optionally, S3 includes: S31: Initialize the core tensors and factor matrices involved in the Tucker decomposition process: Component factor matrix Set as a full-rank square array; S32: Combine the constructed vehicle component tensor and the initialized parameters to construct the loss function and the vehicle dynamic attribute tensor. The pattern, decomposed into a product of a core tensor and four factor matrices, satisfies the following approximate relationship: in, Represents the product of a tensor and the nth modulus of a matrix; The basic loss function is defined as: in, These are the reconstructed values ​​calculated using the Tucker decomposition model. For each regularization term, (the hyperparameters are...) Denotes the Frobenius norm; S33: Initialize the optimizable parameter set and the basic loss function for output construction.

[0012] Optionally, S4 includes: S41: Calculate the interaction parameters and initialization optimization parameters, including: Initialize learning rate The weights of the losses under the three constraints Total number of iterations And the threshold for early cessation and the weight parameters of the basic loss function ; To perform interaction parameter calculations, the tensor is first reconstructed: Extracting the component factor matrix Latent feature vectors: in, ; The ratio of the normalized norm of the row vectors of the component factor matrix to the maximum norm of all component row vectors is defined as the component's motion participation degree. Based on the latent feature vectors of all components Calculate the dynamic interaction adjacency matrix between components. The matrix elements are: in, It is a cosine similarity metric function; Constructing component motion feature vectors: from Extract key motion feature components including position coordinates, velocity, and acceleration. Motion feature vectors are constructed by aggregating the components using mean pooling. : Calculate the dynamic interaction matrix: in, Represents the motion characteristic sequence of component i; S42: Perform loss calculation and gradient update; S43: Obtain the optimized latent feature representation under dynamic interaction consistency constraints.

[0013] Optionally, S42 includes: S421: The loss for constructing physical and temporal constraints helps optimize the tensor reconstruction loss, which includes three parts: contrastive learning loss, temporal smoothing loss, and consistency loss for latent feature space interaction and motion behavior space interaction. Contrastive learning loss is: in, The cosine similarity function; Temperature parameter; N is the total number of components; Positive samples; Negative samples; The time smoothing loss is: The consistency loss between latent feature space interaction and motion behavior space interaction is: S422: The final total loss is: S423: Core Tensor Update: Factor matrix update: S424: The part that determines whether the iteration satisfies the termination condition. When it satisfies: If the training stops, then return to S424 to continue iterating; otherwise, return to S424 to continue iterating.

[0014] Optionally, S5 includes: S51: Reconstructing the final tensor based on the optimized core tensor and factor matrix: For each component i, extract the feature sequence from the corresponding time scene slice. ; S52: Based on the feature sequence, the coordinate dimension is obtained. According to the rigidity assumption of vehicle components, the geometric posture of each component in the world coordinate system is mapped. Based on the geometric posture of all components at each moment, the real-time structure of the whole vehicle is dynamically calculated and restored according to the multibody dynamics model of the vehicle.

[0015] In a second aspect, the present invention provides a vehicle motion component and its physical property analysis device, used to implement the vehicle motion component and its physical property analysis method described in the first aspect, comprising: a vehicle component dynamic data module, a vehicle component tensor construction module, a vehicle component tensor decomposition module, a joint optimization module, and a structural topology aggregation module connected in sequence, wherein the vehicle component dynamic data module is used to acquire historical dynamic data and is connected to the structural topology aggregation module.

[0016] Optionally, the vehicle component tensor decomposition module includes a first initialization unit, a basic loss construction unit, and an output unit connected in sequence. The first initialization unit is connected to the vehicle component dynamic data module, and the output unit is connected to the joint optimization module.

[0017] Optionally, the joint optimization module includes a second initialization unit, a loss calculation and gradient update unit, and a result storage unit connected in sequence. The second initialization unit is connected to the output unit, and the result storage unit is connected to the structure topology aggregation module.

[0018] Optionally, the structural topology aggregation module includes interconnected feature reconstruction units and motion state mapping units. The feature reconstruction units are connected to the result storage unit, and the motion state mapping unit is connected to the vehicle component dynamic data module.

[0019] In summary, the present invention has at least the following beneficial technical effects: This invention unifies and jointly optimizes the dynamic and physical properties of components, effectively outputting a coordinated and consistent vehicle motion sequence. This facilitates the application of this invention in vehicle dynamics simulation, autonomous driving systems, and other fields related to vehicle dynamics simulation. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for analyzing the physical properties of a vehicle moving component in an embodiment of the present invention. Figure 2 This is a flowchart of a vehicle motion component and its physical property analysis system according to an embodiment of the present invention.

[0021] Figure reference numerals: 110, Vehicle component dynamic data module; 111, Data receiving unit; 112, Component partitioning and corresponding feature extraction unit; 113, Data receiving and partitioning result storage unit; 114, Multi-component linkage simulation result storage unit; 120, Vehicle component tensor construction module; 130, Vehicle component tensor decomposition module; 131, First initialization unit; 132, Basic loss construction unit; 133, Output unit; 140, Joint optimization module; 141, Second initialization unit; 142, Loss calculation and gradient update unit; 143, Result storage unit; 150, Structural topology aggregation module; 151, Feature reconstruction unit; 152, Motion state mapping unit. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0023] The terminology used in the following embodiments of the present invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification and appended claims of the present invention, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in the present invention refers to and includes any or all possible combinations of one or more of the listed items. The terms “first” and “second” are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as “first” or “second” may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of the present invention, unless otherwise stated, “a plurality” means two or more.

[0024] This invention provides a method for analyzing the physical properties of moving components in a vehicle.

[0025] refer to Figure 1 A method for analyzing the moving components of a vehicle and their physical properties includes the following steps: S1: The vehicle component dynamic data module 110 receives and stores historical dynamic data of the vehicle in multiple scenarios and at multiple times, completes component division based on the vehicle's three-dimensional model, and extracts the dynamic features and physical attribute features of the components by combining the historical dynamic data.

[0026] S1 includes the following steps: S11: The data receiving unit 111 receives historical dynamic data from physical sensors and / or simulation software. This historical dynamic data covers a variety of driving scenarios (such as acceleration, braking, turning, etc.) and multiple continuous time steps.

[0027] S12: Component Division and Corresponding Feature Extraction Unit 112 decomposes the entire vehicle into N independent components based on the component assembly relationship of the vehicle's 3D model and assigns each component a unique component identification number. Based on the 3D model, physical properties such as mass, local stiffness, and damping of the components are collected, and kinematic characteristics (position coordinates, velocity, acceleration, etc.) of the components under different times and scenarios are collected based on historical dynamic data. Component-level feature vectors are constructed: S13: Data reception and partitioning. The storage unit 113 stores the received vehicle dynamic data and components in a structured format as quadruples. The quadruple component data is represented as follows: Where i represents the component identifier, j represents the feature vector obtained from the previous unit, k represents the time slice index, and l represents the scene number.

[0028] S14: Multi-component linkage simulation result storage unit 114 stores the simulation sequence results of the final multi-component joint motion.

[0029] S2: Vehicle Dynamic Tensor Construction Module 120 constructs tensor representations of vehicle motion components based on historical dynamic data and component partitioning results.

[0030] The vehicle dynamics tensor construction module 120 structures discrete data into high-order tensors to capture complex relationships between multiple dimensions, specifically: The stored quadruplet component data is organized into a fourth-order vehicle dynamic attribute tensor. : in, Indicates the number of components. This represents a feature dimension, which simultaneously includes motion parameters describing the motion state of each component and physical property parameters describing the mechanical properties of the material. Indicates the number of time slices. Indicates the number of scenes.

[0031] Elements in a tensor Indicates the first The component in the first The first time slice, the first In the first scenario, The values ​​of each feature dimension, namely: .

[0032] Feature dimension values ​​include kinematic quantities such as position, velocity, and acceleration, as well as physical properties such as mass, stiffness, and damping; for unobserved data locations, preprocessing is performed by setting them to null or zero values; and sets are used for... Represents the vehicle dynamic attribute tensor The set of all known data indices.

[0033] S3: Vehicle Dynamics Tensor Decomposition Module 130 constructs a tensor representation learning framework based on Tucker decomposition (higher-order singular value decomposition, HOSVD), and outputs a set of optimizable parameters such as core tensors and factor matrices.

[0034] S3 includes the following steps: S31: First initialization unit 131 initializes the core tensors and factor matrices involved in the Tucker decomposition process: Component factor matrix It is set as a full-rank square matrix to maintain the independent identity of all components.

[0035] S32: Basic loss construction unit 132 combines the constructed vehicle component tensor and the initialized parameters to construct the loss function and vehicle dynamic attribute tensor. The pattern, decomposed into a product of a core tensor and four factor matrices, satisfies the following approximate relationship: in, This represents the product of the tensor and the nth modulus of the matrix; the decomposition is set to full rank in the component dimension. The physical meaning of this setting is to preserve the independent identity of all components and encode the motion correlation and physical property coupling information between components together in the core tensor G and the component factor matrix. The resulting joint implicit feature space provides a complete feature basis for subsequent simultaneous analysis of component interaction relationships and mechanical transmission.

[0036] The basic loss function is defined as: in, These are the reconstructed values ​​calculated using the Tucker decomposition model. For each regularization term, (the hyperparameters are...) This represents the Frobenius norm.

[0037] S33: Output unit 133 outputs the set of optimizable parameters and the constructed basic loss function obtained from the initialization part. It is the starting point for subsequent joint optimization, thus providing input for subsequent joint optimization training.

[0038] S4: The joint optimization training module 140 receives the set of optimizable parameters and integrates the physical constraints on the dynamic interaction of components to construct a comprehensive target loss function, thereby achieving the collaborative analysis of motion characteristics and physical properties.

[0039] S4 includes the following steps: S41: The parameter initialization unit 141 receives the set of optimizable parameters output by the output unit 133, such as the factor matrices of each dimension of the core tensor, calculates the interaction parameters and initializes the optimization parameters. The specific workflow includes: The initialization parameters specifically include: learning rate. The weights of the losses under the three constraints Total number of iterations And the threshold for early cessation and the weight parameters of the basic loss function output by S33 .

[0040] Next, the interaction parameters are calculated, starting with the reconstruction of the tensor: Next, extract the component factor matrix. Latent feature vectors: in, .

[0041] The ratio of the normalized norm of the row vectors of the component factor matrix to the maximum norm of all component row vectors is defined as the component's motion participation degree. Its physical significance lies in the fact that the stronger the representation ability of a component in the latent feature space, the higher its degree of participation in the overall motion.

[0042] Based on the latent feature vectors of all components Calculate the dynamic interaction adjacency matrix between components. This is used to characterize the dynamic correlation strength between components, where the matrix elements are: in, It is a cosine similarity metric function; the matrix A directly reflects the similarity of the feature vectors of each component in the latent space, that is, the coupling strength of their motion patterns.

[0043] For each component Time slice Scene We can construct the component motion feature vector: Next from Extract key motion feature components such as position coordinates, velocity, and acceleration. The next step is to perform mean pooling aggregation based on the components to construct motion feature vectors. : Therefore, the dynamic interaction matrix is ​​calculated: in, The sequence of motion characteristics of component i reflects the similarity of the motion behavior of the components.

[0044] S42: Loss calculation and gradient update unit 142 performs loss calculation and gradient update.

[0045] S42 includes the following steps: First, the loss calculation and gradient update unit 142 obtains the base loss function output by the output unit 133. We use it as the most basic tensor reconstruction loss.

[0046] S421: Loss Calculation and Gradient Update Unit 142 constructs physical and temporal constraints to aid in the optimization of tensor reconstruction loss, comprising three parts: contrastive learning loss, temporal smoothing loss, and consistency loss for latent feature space interactions and motion behavior space interactions. Details are as follows: The calculation of contrastive learning loss requires first constructing positive and negative sample pairs. Each positive sample pair is formed by using itself as the positive sample in each construct. Negative sample pairs: Negative samples are formed between different components, i.e. .in, Therefore, the contrastive learning loss is: in, The cosine similarity function; Temperature is a parameter that controls the sharpness of the similarity distribution; N is the total number of components. The gradient solution method of the existing Information Noise Contrast Estimation Loss (InfoNCE) is adopted, followed by backpropagation via the chain rule. The purpose of this loss is to ensure that the same component maintains a stable motion representation in different optimization iterations, promoting the separation of motion representations of different components in the motion feature space, and enhancing the intra-class compactness and inter-class discriminability of motion features.

[0047] The time smoothing loss calculates the continuity loss in the time dimension using the time factor matrix (constructing a time smoothing loss pair based on the differences between adjacent rows of the time factor matrix). The gradient is used to automatically suppress drastic changes in the time dimension during the update process. The structure is as follows: The interaction consistency loss is calculated by combining the consistency loss between the latent feature space interaction matrix and the motion behavior interaction matrix, along with the participation level. This means that if two components highly involved in the overall motion (e.g., the car body and wheels) exhibit inconsistencies in the latent space interaction matrix and the motion behavior interaction matrix, a greater loss will be incurred. In summary, this loss term is designed to constrain the latent space interaction matrix to maintain consistency with the physical space motion behavior interaction matrix, ensuring that the learned latent features corroborate the physical behavior. S422: The final total loss consists of the previous four parts: The tensor reconstruction loss is set to 1, and the weights of other constraint losses cannot be greater than the reconstruction loss value. Parameters no greater than 1 are used for optimization and updating.

[0048] S423: In the iterative update section, the gradient update is based on the chain rule to calculate the partial derivatives of the total loss function with respect to the core tensor and each factor matrix, and then uses the gradient descent algorithm to iteratively update the core tensor and factor matrices. Core tensor update: Factor matrix update: in, This is the learning rate.

[0049] S424: The part that determines whether the iteration satisfies the termination condition. When it satisfies: This indicates that there are two scenarios where the training iteration process reaches the termination condition: First, training stops when the value of the control variable t for the number of training iterations reaches the maximum number of training iterations; second, during training, the absolute value of the difference between the total loss value calculated after the current iteration and the value of the previous iteration is less than the convergence termination threshold. If the termination condition is not met, training stops. If not, return to S424 to continue iteration.

[0050] S43: The final result It is the optimized latent feature representation under dynamic interaction consistency constraints, which is stored in storage unit 143.

[0051] S5: The structural topology aggregation module 150 synthesizes a coordinated vehicle motion sequence based on the optimized hidden features and the dynamic interaction relationship of components, combined with the vehicle structural topology information.

[0052] S5 includes the following steps: S51: Feature reconstruction unit 151 reconstructs the final tensor based on the optimized core tensor and factor matrix: For each component i, extract the feature sequence from the corresponding time scene slice. This is the final motion feature sequence of the component after model coordination. This sequence fully describes the component's motion characteristics (such as displacement, velocity, and acceleration) and physical properties (such as mass and force) at different times and in different scenarios.

[0053] S52: The motion state mapping unit 152 extracts coordinate dimensions such as displacement angle based on the feature sequence, maps the geometric posture of each component in the world coordinate system according to the rigidity assumption of the vehicle components, and then dynamically calculates and restores the real-time structure of the whole vehicle based on the geometric posture of all components at each moment and the multi-body dynamics model of the vehicle (such as the connection relationship between parent and child components and the joint type).

[0054] This invention also provides a device for analyzing the physical properties of a vehicle moving component.

[0055] refer to Figure 2 A device for analyzing the physical properties of a vehicle moving component includes: a vehicle component dynamic data module 110, a vehicle component tensor construction module 120, a vehicle component tensor decomposition module 130, a joint optimization module 140, and a structural topology aggregation module 150 connected in sequence. The vehicle component dynamic data module 110 is also connected to the structural topology aggregation module 150.

[0056] The vehicle component dynamic data module 110 includes: a data receiving unit 111, a component division and corresponding feature extraction unit 112, a data receiving and division result storage unit 113, and a multi-component linkage simulation result storage unit 114.

[0057] The vehicle component tensor decomposition module 130 includes: a first initialization unit 131, a basic loss construction unit 132, and an output unit 133.

[0058] The joint optimization module 140 includes: a second initialization unit 141, a loss calculation and gradient update unit 142, and a result storage unit 143.

[0059] The structural topology aggregation module 150 includes: a feature reconstruction unit 151 and a motion state mapping unit 152.

[0060] The data receiving unit 111, the component partitioning and corresponding feature extraction unit 112, the data receiving and partitioning result storage unit 113, the vehicle component tensor construction module 120, the first initialization unit 131, the basic loss construction unit 132, the output unit 133, the second initialization unit 141, the loss calculation and gradient update unit 142, the result storage unit 143, the feature reconstruction unit 151, the motion state mapping unit 152, and the multi-component linkage simulation result storage unit 114 are connected in sequence.

[0061] Various variations and specific examples of the methods provided in the above embodiments are also applicable to the vehicle motion component and its physical property analysis device in this embodiment. Through the foregoing detailed description of the vehicle motion component and its physical property analysis method, those skilled in the art can clearly understand the implementation method of the vehicle motion component and its physical property analysis device in this embodiment. For the sake of brevity, it will not be described in detail here.

[0062] The above description of the embodiments is only used to provide a detailed introduction to the technical solution of the present invention. However, the description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention, and should not be construed as a limitation of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for analyzing the physical properties of moving components of a vehicle, characterized in that, The method comprises the following steps: S1: receiving and storing historical dynamic data of a vehicle in multiple scenes and at multiple times, completing component division based on a three-dimensional model of the vehicle, and extracting dynamic characteristics and physical attribute characteristics of the components in combination with the historical dynamic data; S2: constructing a tensor representation of a vehicle motion component based on the historical dynamic data and the component division result; S3: constructing a tensor representation learning framework based on Tucker decomposition, and outputting an optimizable parameter set including a core tensor and factor matrices; S4: receiving the optimizable parameter set, integrating physical constraints on dynamic interactions of the components, and constructing an objective loss function; S5: synthesizing a vehicle motion sequence in combination with vehicle structure topology information based on the optimized hidden features and the dynamic interaction relationship of the components.

2. The vehicle motion member and method of analyzing physical properties thereof according to claim 1, characterized by, The S2 comprises: Organizing historical dynamic data and component division results into a four-order vehicle dynamic property tensor : wherein, represents the number of components; represents the feature dimension, including the motion parameters describing the motion state of each component and the physical attribute parameters describing the material mechanical properties; represents the number of time slices; represents the number of scenes.

3. The vehicle motion member and method of analyzing physical properties thereof according to claim 1, characterized by, The S3 comprises: S31: initializing the core tensor and the factor matrices involved in the Tucker decomposition process: component factor matrix is set to a full rank matrix; S32: Construct a loss function combined with the vehicle component tensor of the configuration and the initialized parameters, vehicle dynamic attribute tensor The product mode decomposed into a core tensor and four factor matrices satisfies the following approximate relationship: wherein denotes the n-mode product of a tensor with a matrix; The basic loss function is defined as: wherein, is a reconstruction value calculated by a Tucker decomposition model, is a hyperparameter of each regularization term, denotes the Frobenius norm; S33: initializing the optimizable parameter set and outputting the constructed basic loss function.

4. The vehicle motion member and method of analyzing physical properties thereof according to claim 3, characterized by, The S4 comprises: S41: calculating the interaction relationship parameters and initializing the optimization parameters, including: initializing learning rate , weight of loss of three constraints , total number of iterations and early stopping threshold and weight parameter of base loss function ; The interaction parameter calculation is performed as follows: first, reconstructing the tensor: extracting the eigenvectors of the factor matrix of the hidden feature vector wherein ; The norm of the row vector of the component factor matrix is normalized, and the ratio to the maximum value in all component row vector norms is defined as the motion participation degree of the component: latent feature vectors of all components , a dynamic interaction adjacency matrix between components is calculated where the matrix elements are: wherein is a cosine similarity measure function; The component motion feature vector is constructed as follows: extracting from key motion feature components including position coordinates, velocity, acceleration , aggregating by means of mean-pooling, constructing motion feature vector : The dynamic interaction matrix is calculated as follows: wherein, represents the sequence of motion characteristics of the member i; S42: performing loss calculation and gradient update; S43: obtaining the optimized hidden feature representation under the dynamic interaction consistency constraint.

5. The vehicle motion member and method of analyzing physical properties thereof according to claim 4, wherein, The S42 comprises: S421: constructing a loss help tensor to optimize the reconstruction loss under physical and time constraints, including three parts of loss: contrastive learning loss, time smoothing loss, and consistency loss of hidden feature space interaction and motion row space interaction; The contrastive learning loss is as follows: wherein, is a cosine similarity function; is a temperature parameter; N is the total number of components; is a positive sample; is a negative sample; The time smoothing loss is as follows: The consistency loss of hidden feature space interaction and motion row space interaction is as follows: S422: finally constructing the total loss as follows: S423: core tensor update: Factor matrix update: S424: the part of determining whether the iteration meets the termination condition, when meeting: stop training; otherwise, return to S424 to continue iteration.

6. A vehicle motion member and a method of analyzing physical properties thereof according to claim 4, characterized in that, The S5 comprises: S51: reconstructing the final tensor based on the optimized core tensor and factor matrix: For each component i, extract the sequence of features under the corresponding time scene slice ​ S52: obtaining the coordinate dimension based on the feature sequence, mapping the geometric pose of each component in the world coordinate system according to the rigidity assumption of the vehicle components, and dynamically calculating the real-time structure of the whole vehicle based on the geometric poses of all components at each time according to the multi-body dynamics model of the vehicle.

7. A vehicle motion member and a device for analyzing physical properties thereof, characterized by comprising: A vehicle motion component and a physical attribute analysis method thereof are used to implement any one of claims 1-6, comprising: a vehicle component dynamic data module (110), a vehicle component tensor construction module (120), a vehicle component tensor decomposition module (130), a joint optimization module (140), and a structure topology aggregation module (150) connected in sequence, wherein the vehicle component dynamic data module (110) is used to acquire historical dynamic data and is connected with the structure topology aggregation module (150).

8. A vehicle motion member and a method of analyzing physical properties thereof according to claim 7, characterized in that, The vehicle component tensor decomposition module (130) comprises a first initialization unit (131), a basic loss construction unit (132), and an output unit (133) connected in sequence, the first initialization unit (131) is connected with the vehicle component dynamic data module (110), and the output unit (133) is connected with the joint optimization module (140).

9. The vehicle motion member and method of analyzing physical properties thereof according to claim 8, wherein, The joint optimization module (140) comprises a second initialization unit (141), a loss calculation and gradient update unit (142), and a result storage unit (143) connected in sequence, the second initialization unit (141) is connected with the output unit (133), and the result storage unit (143) is connected with the structure topology aggregation module (150).

10. The vehicle motion member and method of analyzing physical properties thereof according to claim 9, wherein, The structure topology aggregation module (150) comprises a feature reconstruction unit (151) and a motion state mapping unit (152) connected with each other, the feature reconstruction unit (151) is connected with the result storage unit (143), and the motion state mapping unit (152) is connected with the vehicle component dynamic data module (110).

Images