A method and system for modeling automotive multibody dynamics

Abstract

The application discloses a kind of automobile multi-body dynamics modeling method and system, method includes according to the vibration model of automobile and Lagrange equation, establishes initial multi-body dynamics model;Through the analysis method of working condition transmission path, precision index is established;According to the load spectrum data and iteration condition collected in each road condition, initial multi-body dynamics model is virtually iterated and checked, and whether each local model and initial multi-body dynamics model meets the requirement of precision index is judged respectively, if all meet the requirement, initial model is used as optimal model;For any local model that does not meet the requirement, iteration is continued until all local models meet the precision requirement, then overall optimization is carried out;Local meets the requirement, and initial multi-body dynamics model does not meet the requirement, then overall optimization is carried out, and orthogonal latin hypercube experiment and response surface optimization evaluation are carried out;Finally, optimal multi-body dynamics model is established.This embodiment effectively improves the precision, reliability and applicability of modeling.

Description

Technical Field

[0001] This invention relates to the field of multibody dynamics modeling, and more particularly to a method and system for multibody dynamics modeling of automobiles. Background Technology

[0002] Vehicle dynamics simulation directly impacts a vehicle's handling stability, ride comfort, and reliability. Furthermore, the accuracy of ride comfort performance analysis results directly depends on the accuracy of the established dynamics simulation model. To obtain accurate dynamic responses and improve overall vehicle dynamics performance, multibody dynamics modeling is crucial. Multibody dynamics modeling shortens product design cycles, saves on new product development costs, and reduces development risks. The accuracy, real-time performance, applicability, and reliability of the model directly determine whether the simulation model can be applied in product development.

[0003] Although the research and application of multibody dynamics modeling has been developing for many years, the structure and function of the modeling objects are constantly evolving and becoming increasingly complex. This has led to problems such as nonlinearity and uncertainty in dynamics modeling that are currently difficult to solve. Existing multibody dynamics modeling methods only take a global perspective and make simple iterative corrections to replace manual trial and error. However, they cannot establish accurate multibody dynamics models for automobiles, which have more complex structures and functions. As a result, the accuracy of the established multibody dynamics models is difficult to meet the actual needs of engineering. The accuracy, applicability and precision of the established multibody dynamics models need to be improved. Summary of the Invention

[0004] This invention provides a method and system for modeling multibody dynamics of automobiles, which improves the accuracy, reliability and applicability of multibody dynamics model establishment and increases modeling efficiency.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for modeling multibody dynamics of a vehicle, including...

[0006] Based on the vibration model of the automobile and the Lagrange equation, an initial multibody dynamics model is established;

[0007] Based on the load spectrum data of key points in the vibration transmission path collected by the road condition test and the first preset iteration termination condition, the initial multibody dynamics model is virtually iterated. Based on the first iteration results and accuracy index, it is determined whether each local model and the initial multibody dynamics model meet the preset accuracy index requirements. The initial multibody dynamics model consists of several local models. The accuracy index is established based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path. The response measurement points are one or more selected from the key points.

[0008] If all requirements are met, the initial multibody dynamics model will be used as the optimal multibody dynamics model.

[0009] If all local models meet the requirements, but the initial multibody dynamics model does not, then based on the results of the first iteration and the preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is evaluated by response surface optimization to establish the optimal multibody dynamics model.

[0010] If any local model fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the first iteration result until all local models meet the preset accuracy index requirements. Then, the second iteration result is obtained, and the initial multibody dynamics model is optimized and evaluated based on the second iteration result and the preset orthogonal Latin hypercube experiment to establish the optimal multibody dynamics model.

[0011] By implementing the embodiments of the present invention, an initial multibody dynamics model is established based on the vibration model of the automobile and the Lagrange equation. The Lagrange equation and the vibration model are combined for modeling. The accuracy index is established using the vibration transmissibility method under working conditions. The local model is iteratively checked using load spectrum data. Then, the accuracy of the complete model is evaluated, completing the index decomposition from global to local. For those models that meet the reliability and accuracy index, the initial multibody dynamics model is used as the final model, which greatly improves the efficiency of model building. For local models that do not meet the accuracy requirements, iteration is performed. Through the iteration results and the design of orthogonal Latin hypercube tests, response surface optimization evaluation is carried out to establish the optimal multibody dynamics model that meets the reliability and accuracy index, effectively improving the accuracy, reliability and applicability of modeling.

[0012] As a preferred option, an initial multibody dynamics model is established based on the vehicle's vibration model and the Lagrange equations, specifically as follows:

[0013] Based on the structural combination, vibration excitation and response of the automotive system, mass damping is abstracted to establish a vibration model;

[0014] Based on the Lagrange equation and the vibration model, the dynamic equation is established, and the component parameters of the vibration model are substituted into the dynamic equation for calculation to establish an initial multibody dynamic model.

[0015] As the preferred option, the accuracy index is established based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path, specifically:

[0016] Based on the load-transmission path analysis method, a transmission equation is established. Using this equation, the first transmissibility matrix is ​​calculated. Based on this first transmissibility matrix and the transmissibility at the response measurement points along the vibration transmission path, an accuracy index is established.

[0017] As a preferred approach, based on the load spectrum data of key points in the vibration transmission path collected from the road condition test and the first preset iteration termination condition, the initial multibody dynamics model is virtually iterated. Based on the first iteration results and accuracy indicators, it is determined whether each local model and the initial multibody dynamics model meet the preset accuracy requirements, specifically:

[0018] Sensors were placed at key points along the vibration transmission path in the vibration model to measure the road condition test response and obtain load spectrum data.

[0019] Select the output point of the component in the vibration transmission path;

[0020] Based on the load spectrum data and the Newton-Rafaelson formula, a virtual iteration is performed as follows:

[0021] u k+1 (s)=u k (s)+f -1 (s)(y d (s)-y k (s));

[0022] k = 1, 2, 3, ..., n-1;

[0023] Among them, u k (s) represents the k-th excitation signal, y k u1(s) is the response signal iterated by the k-th excitation signal, u1(s) is the initial excitation signal, and f -1 (s) is the inverse propagation function, y d (s) represents the load spectrum data;

[0024] The virtual iteration stops when the first preset iteration termination condition is met.

[0025] The excitation signal of the kth iteration is input into the initial multibody dynamics model to obtain the model response signal at the output point; wherein, the excitation signal of the kth iteration is the excitation signal that satisfies the first preset iteration termination condition;

[0026] Based on the model response signal, establish the second transitivity matrix;

[0027] Based on the first transmissibility matrix, the second transmissibility matrix, and the accuracy index, determine whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index.

[0028] By implementing the embodiments of the present invention, only the response signal of the road condition test is needed in the entire evaluation process of the reliability and accuracy index, which greatly reduces the experimental cost and shortens the test cycle.

[0029] As a preferred approach, if any local model fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the first iteration result until all local models meet the preset accuracy requirements. Then, the second iteration result is obtained. Based on the second iteration result and the preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is optimized and evaluated using response surface techniques to establish the optimal multibody dynamics model. Specifically:

[0030] If any local model fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the results of the first iteration and the Newton-Rafaelson formula, as follows:

[0031] u l+1 (s)=u l (s)+f -1 (s)(y b (s)-y l (s));

[0032] l = 1, 2, 3, ..., n-1;

[0033] Among them, u l (s) represents the k-th excitation signal, y l u1(s) is the response signal iterated by the k-th excitation signal, u1(s) is the initial excitation signal, and f -1 (s) is the inverse propagation function, y b (s) represents the result of the first iteration;

[0034] Based on the relative error value of the response curve, modify the preset iteration termination condition. When the modified iteration termination condition is met, stop the virtual iteration.

[0035] The l-th excitation signal is input into the initial multibody dynamics model to obtain the modified model response signal at the output point; where the l-th excitation signal is the excitation signal that satisfies the modified iteration termination condition;

[0036] Based on the modified model response signal, a modified transitivity matrix is ​​established, and the modified iteration results are obtained.

[0037] Based on the first transferability matrix, the modified transferability matrix, and the accuracy index, determine whether each local model meets the preset accuracy index requirements;

[0038] If any local model fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the modified iteration results until all local models meet the preset accuracy requirements. Then, the second iteration results are obtained, and the initial multibody dynamics model is optimized and evaluated based on the second iteration results and the preset orthogonal Latin hypercube test to establish the optimal multibody dynamics model.

[0039] In implementing this embodiment of the invention, if any local model does not meet the requirements, the iteration termination condition is further modified based on the relative error value of the response curve. The initial multibody dynamics model is then continuously virtual-iterated to ensure that the local model meets the accuracy index. Based on the iteration results, the elements to be corrected can be identified, which helps to further optimize the global model (initial multibody dynamics model).

[0040] As a preferred approach, based on the iteration results and a pre-set orthogonal Latin hypercube experiment, the initial multibody dynamics model is evaluated through response surface optimization to establish the optimal multibody dynamics model. The iteration results are either the first iteration result or the second iteration result, specifically:

[0041] Based on the iteration results, the elements to be corrected are determined. Based on the elements to be corrected, sensitivity and contribution analysis are performed to obtain the design variables. Based on the design variables and the pre-set first orthogonal Latin hypercube experiment, the first design variable matrix is ​​established.

[0042] Based on the first design variable matrix, design variables, and accuracy indicators, the initial multibody dynamics model is evaluated by response surface optimization to establish the optimal multibody dynamics model.

[0043] As a preferred option, based on the first design variable matrix, design variables, and accuracy indicators, the initial multibody dynamics model is evaluated through response surface optimization to establish the optimal multibody dynamics model, specifically as follows:

[0044] The first design variable matrix is ​​combined with the initial multibody dynamics model to obtain the first response;

[0045] Based on the design variable values, the first response value, and the response surface approximation function, establish the first response surface model;

[0046] The accuracy of the first response surface model is verified to evaluate whether the first response surface model meets the preset response surface accuracy requirements.

[0047] If so, then based on the design variables, the first response, and the convolutional neural network, an optimized response surface model is established, and based on the optimized response surface model, an optimal multibody dynamics model is established.

[0048] If not, a second orthogonal Latin hypercube experiment is preset. Based on the second orthogonal Latin hypercube experiment and design variables, a second response surface model is established until the second response surface model meets the preset response surface accuracy requirements. Then, the design of the orthogonal Latin hypercube experiment is stopped, and the subsequent steps of meeting the preset response surface accuracy requirements are entered.

[0049] By implementing the embodiments of the present invention, the accuracy of the response surface model is verified, the model with the highest similarity between the dynamic model and the response surface model is found, and the response surface model that meets the accuracy requirements is optimized by a convolutional neural network to establish the optimal multibody dynamic model, thereby effectively improving the accuracy and precision of the multibody dynamic model.

[0050] As a preferred approach, a second orthogonal Latin hypercube experiment is pre-designed. Based on the second orthogonal Latin hypercube experiment and design variables, a second response surface model is established. This process continues until the second response surface model meets the pre-defined response surface accuracy requirements. At this point, the design of the orthogonal Latin hypercube experiment is stopped, and the process proceeds to the next step where the response surface model meets the pre-defined response surface accuracy requirements. Specifically:

[0051] A second orthogonal Latin hypercube experiment is pre-set, and a second design variable matrix is ​​established based on the second orthogonal Latin hypercube experiment and the design variables;

[0052] The second design variable matrix is ​​combined with the initial multibody dynamics model to obtain the second response;

[0053] The second response surface model is established by using the design variable values, the second response values, and the response surface approximation function.

[0054] The accuracy of the second response surface model is verified to evaluate whether the second response surface model meets the preset response surface accuracy requirements.

[0055] If the second response surface model does not meet the preset response surface accuracy requirements, modify the preset orthogonal Latin hypercube experiment to obtain the third response, and verify the accuracy of the third response surface model until the third response surface model meets the preset response surface accuracy requirements, and then stop the design of the orthogonal Latin hypercube experiment.

[0056] The next step is to meet the preset response surface accuracy requirements of the response surface model. Based on the design variables, the modified response, and the convolutional neural network, an optimized response surface model is established. Based on the optimized response surface model, an optimal multibody dynamics model is established. The modified response is either the second or third response.

[0057] By implementing the embodiments of the present invention, the response surface model is continuously modified through multiple designs of the second orthogonal Latin hypercube experiment, thereby establishing a highly accurate response surface model and laying a solid foundation for establishing an accurate multibody dynamics model.

[0058] As a preferred approach, an optimized response surface model is established based on the design variables, the response, and the convolutional neural network. Based on this optimized response surface model, an optimal multibody dynamics model is then established. The response is either the initial response or the modified response, specifically:

[0059] Based on the design variables and response, train a convolutional neural network. Based on the training results of the convolutional neural network, establish an optimized response surface model. Based on the optimized response surface model, establish an optimal multibody dynamics model. Continue until the optimal multibody dynamics model meets the preset accuracy requirements, then stop training the convolutional neural network to obtain the optimal response surface model. Based on the optimal response surface model, establish an optimal multibody dynamics model.

[0060] By implementing the embodiments of the present invention, a convolutional neural network is trained to continuously optimize the response surface model. Training stops when the accuracy requirements of the response surface are met, thereby obtaining the optimal response surface model and performing accurate model building and correction.

[0061] To address the same technical problem, embodiments of the present invention also provide an automotive multibody dynamics modeling system, comprising: an initial module, a precision index module, an evaluation module, a precision satisfaction module, a local optimization module, and a global optimization module;

[0062] The initial module is used to establish an initial multibody dynamics model based on the vibration model of the car and the Lagrange equation. The initial multibody dynamics model consists of several local models.

[0063] The accuracy index module is used to establish accuracy indexes based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path; the response measurement points are one or more selected from the key points.

[0064] The evaluation module is used to perform virtual iteration on the initial multibody dynamics model based on the load spectrum data of key points in the vibration transmission path collected by the road condition test and the first preset iteration termination condition. Based on the first iteration result and accuracy index, it determines whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index.

[0065] The accuracy module is used to select the initial multibody dynamics model as the optimal multibody dynamics model if all local models and the initial multibody dynamics model meet the accuracy requirements.

[0066] The local optimization module is used to perform virtual iteration on the initial multibody dynamics model again based on the first iteration result if any local model does not meet the accuracy index requirements, until all local models meet the preset accuracy index requirements, and then obtain the second iteration result.

[0067] The global optimization module is used to optimize and evaluate the response surface of the initial multibody dynamics model and establish the optimal multibody dynamics model if the initial multibody dynamics model does not meet the accuracy requirements, based on the iteration results and the preset orthogonal Latin hypercube experiment. Attached Figure Description

[0068] Figure 1: A flowchart illustrating an embodiment of the multibody dynamics modeling method for automobiles provided by the present invention;

[0069] Figure 2 : A simplified flowchart for establishing a multibody dynamics model, which is an embodiment of the multibody dynamics modeling method for automobiles provided by the present invention;

[0070] Figure 3 : A multibody dynamic vibration model diagram of a cargo truck, representing an embodiment of the multibody dynamics modeling method for automobiles provided by the present invention;

[0071] Figure 4 : A flowchart illustrating the response surface optimization evaluation process of an embodiment of the multibody dynamics modeling method for automobiles provided by this invention;

[0072] Figure 5 : A schematic diagram of an embodiment of the automotive multibody dynamics modeling system provided by the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Example 1

[0075] Please refer to Figure 1 This is a flowchart illustrating a multibody dynamics modeling method for automobiles provided in an embodiment of the present invention. Figure 2 A simplified process for establishing a multibody dynamics model is provided in this embodiment. The modeling method utilizes the load-condition vibration transmissibility method to iteratively verify the model, optimize the response surface, and conduct multiple accuracy assessments to establish an optimal multibody dynamics model, effectively improving the accuracy, reliability, and applicability of the modeling. This modeling method includes steps 101 to 104, each step as follows:

[0076] Step 101: Establish an initial multibody dynamics model based on the vibration model of the car and the Lagrange equation.

[0077] Step 101 includes steps 1011 to 1012, and the specific details of each step are as follows:

[0078] Step 1011: Based on the structural combination, vibration excitation and response of the vehicle system, perform mass damping abstraction and establish a vibration model;

[0079] In this embodiment, the structural combination of the automotive research system is determined, then the vibration excitation and response of the research system are determined, and finally, based on the vibration transmission path of the system, the mass block, damper, and spring are abstracted to determine a simplified diagram of the vibration model. Taking a truck as an example, the multibody dynamics vibration model of a truck is as follows: Figure 3 As shown in Table 1, the relevant parameters in the vibration model of the truck are as follows.

[0080] Table 1. Parameter Table for Cargo Trucks

[0081]

[0082]

[0083]

[0084] Step 1012: Based on the Lagrange equation and the vibration model, establish the dynamic equation, and substitute the component parameters of the vibration model into the dynamic equation for calculation to establish the initial multibody dynamic model.

[0085] In this embodiment, the dynamic equations are determined based on the Lagrange equations and the vibration model of the vehicle. The Lagrange equations are as follows:

[0086]

[0087] In the formula, E k E p E d For the system's kinetic energy, potential energy, and dissipated energy; q i , For the system's generalized coordinates and generalized velocity; F i To be with q i The corresponding generalized force.

[0088] The dynamic model parameters are the mass parameters, stiffness, damping, and geometric dimensions of each component in the vibration model. Therefore, specific parameters (mass, moment of inertia, product of inertia, stiffness, damping, and geometric parameters) can be substituted into the dynamic equations for specific calculations to establish an initial multibody dynamic model.

[0089] Step 102: Establish accuracy indicators based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path.

[0090] Step 102 specifically involves: establishing a transmission equation based on the working condition transmission path analysis method; calculating the first transmissibility matrix based on the transmission equation; and establishing an accuracy index based on the first transmissibility matrix and the transmissibility at the response measurement points in the vibration transmission path.

[0091] In this embodiment, based on the load transfer path analysis method, the OPTA equation (transfer equation) is established as follows:

[0092]

[0093] In the formula, Let A be the response quantity of the target response point set A. Let B be the response quantity of the target response point set B, where the superscript (i) indicates the frequency domain signal of the measured i-th data block. The transmissibility matrix can be calculated by the following formula:

[0094]

[0095] In the formula, the superscript "+" indicates the pseudo-inverse of the matrix, V is an r×r unitary matrix, and U is a B×B unitary matrix. This is the B×r Victic singularity matrix obtained after removing smaller singular values, where all off-diagonal elements are zero, and fc is the external force acting on the system.

[0096] To establish a vibration transmissibility accuracy index, for the transmissibility at two different response measurement points p and q, the transmissibility accuracy index in the frequency band [f1, f2] is:

[0097]

[0098] In the formula, D pq (f1,f2) is the accuracy index. This is the transferability matrix from the actual vehicle test. This is the transmissibility matrix (first transmissibility matrix) of the dynamic model.

[0099] Step 103: Based on the load spectrum data of key points in the vibration transmission path collected by the road condition test and the first preset iteration termination condition, perform virtual iteration on the initial multibody dynamics model, and determine whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index based on the first iteration result and accuracy index; wherein, the initial multibody dynamics model is composed of several local models.

[0100] In this embodiment, the local models specifically refer to the engine sub-model, cab sub-model, and frame sub-model of the initial multibody dynamics vibration model. The global model represents the overall multibody dynamics model. The ultimate goal of establishing the dynamics model is to ensure that the global model meets engineering requirements. If the final global model meets all requirements, the errors in the local models will not have an impact. Similarly, if all local models meet the requirements, and the global model also meets the preset accuracy index (reliability index), then no further optimization is needed to directly establish the model, and the initial multibody dynamics model is taken as the optimal multibody dynamics model. If the local models do not meet the basic engineering error requirements, the local real-world constraints cannot be ignored. Therefore, if each local model does not meet the requirements, iterative verification is required to find the variables that need to be modified and to design and optimize the global model. Alternatively, if the local models meet the accuracy requirements but the global model does not, the global model also needs to be designed and optimized.

[0101] Step 103 specifically involves: deploying sensing devices at key points along the vibration transmission path in the vibration model, conducting road condition test response measurements, and obtaining load spectrum data;

[0102] Select the output point of the component in the vibration transmission path;

[0103] Based on the load spectrum data and the Newton-Rafaelson formula, a virtual iteration is performed as follows:

[0104] u k+1 (s)=u k (s)+f -1 (s)(y d (s)-y k (s));

[0105] k = 1, 2, 3, ..., n-1;

[0106] Among them, u k (s) represents the k-th excitation signal, y k u1(s) is the response signal iterated by the k-th excitation signal, u1(s) is the initial excitation signal, and f -1 (s) is the inverse propagation function, y d (s) represents the load spectrum data;

[0107] The virtual iteration stops when the first preset iteration termination condition is met.

[0108] The excitation signal of the kth iteration is input into the initial multibody dynamics model to obtain the model response signal at the output point; wherein, the excitation signal of the kth iteration is the excitation signal that satisfies the first preset iteration termination condition;

[0109] Based on the model response signal, establish the second transitivity matrix;

[0110] Based on the first transmissibility matrix, the second transmissibility matrix, and the accuracy index, determine whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index.

[0111] In this embodiment, sensors are arranged at key points of the vibration transmission path in the vibration model to measure the road condition test response and obtain the load spectrum data of the test. The load spectrum data is the response signal. Taking a truck as an example, the key points of the response are arranged at the driver's seat, the connection between the driver's seat and the frame, the connection between the frame and the front suspension, the connection between the frame and the rear suspension, the connection between the engine and the frame, and the connection between the tire and the axle.

[0112] Based on the load spectrum data and the initially established dynamic model, iterative verification is performed. Output points of components along the transmission path are selected (including reference output points and non-reference output points), where both reference and non-reference output points are corresponding points of the actual objects in the vibration model. These points are chosen on any component along the vibration transmission path, typically at the connection point of any two components. Then, the transmission path matrix of the experimental data is established. In the dynamic model, the system input is first iterated using the Newton-Rafaelson formula.

[0113] u k+1 (s)=u k (s)+f -1 (s)(y d (s)-y k (s))

[0114] k = 1, 2, 3, ..., n-1

[0115] In the formula, y k (s) represents the k-th excitation signal u k The response signal iterated by (s), u1(s) is the initial excitation signal, and f -1 (s) is the inverse pass function, y d (s) represents the load spectrum data (measured response signal).

[0116] When the response signal under the current iteration excitation is basically consistent with the response signal under the experiment in the time domain and frequency domain, and the relative error RMS value is less than the preset value (such as 0.2), that is, when the first preset iteration termination condition is met, the virtual iteration stops.

[0117] Next, the response at the reference output point and the response at the non-reference output point are determined by inputting the excitation into the dynamic model. The transfer rate matrix of the dynamic model is established, and the accuracy of the initially established model is evaluated by referring to the transfer rate accuracy index. The excitation signal of the kth iteration is input into the initial multibody dynamic model to determine the response at the reference output point and the response at the non-reference output point, i.e., the model response signal at the output point; where the excitation signal of the kth iteration is the excitation signal that satisfies the first preset iteration termination condition.

[0118] Based on the model response signal at the output point, the transfer rate matrix (second transfer rate matrix) for the actual vehicle test is established using the transfer rate matrix calculation method in step 102.

[0119] Using the transfer rate accuracy index formula in step 102, the accuracy index value is calculated based on the transfer rate matrix of the dynamic model (first transfer rate matrix) and the transfer rate matrix of the actual vehicle test (second transfer rate matrix). By comparing the preset accuracy index value with the calculated accuracy index value, it is determined whether each local model and the initial multibody dynamic model meet the requirements of the preset accuracy index (preset confidence value).

[0120] Step 104: If any local model does not meet the requirements, then based on the first iteration result, the initial multibody dynamics model is virtually iterated again until all local models meet the preset accuracy index requirements. Then, the second iteration result is obtained, and based on the second iteration result and the preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is optimized and evaluated on the response surface to establish the optimal multibody dynamics model.

[0121] Step 104 specifically involves: If any local model does not meet the requirements, based on the first iteration result and the Newton-Rafaelson formula, performing a virtual iteration again on the initial multibody dynamics model, as shown in the following equation:

[0122] u l+1 (s)=u l (s)+f -1 (s)(y b (s)-y l (s));

[0123] l = 1, 2, 3, ..., n-1;

[0124] Among them, u l (s) represents the k-th excitation signal, y l u1(s) is the response signal iterated by the k-th excitation signal, u1(s) is the initial excitation signal, and f -1 (s) is the inverse propagation function, y b (s) represents the result of the first iteration;

[0125] Based on the relative error value of the response curve, modify the preset iteration termination condition. When the modified iteration termination condition is met, stop the virtual iteration.

[0126] The l-th excitation signal is input into the initial multibody dynamics model to obtain the modified model response signal at the output point; where the l-th excitation signal is the excitation signal that satisfies the modified iteration termination condition;

[0127] Based on the modified model response signal, a modified transitivity matrix is ​​established, and the modified iteration results are obtained.

[0128] Based on the first transferability matrix, the modified transferability matrix, and the accuracy index, determine whether each local model meets the preset accuracy index requirements;

[0129] If any local model fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the modified iteration results until all local models meet the preset accuracy requirements. Then, the second iteration results are obtained, and the initial multibody dynamics model is optimized and evaluated based on the second iteration results and the preset orthogonal Latin hypercube test to establish the optimal multibody dynamics model.

[0130] In this embodiment, if any local model does not meet the requirements, the result of the first iteration is used as the initial response signal for the new round of iteration according to the same virtual iteration method. At the same time, the termination condition of the new round of iteration is modified according to the specific value of the relative error of the response curve in the previous virtual iteration. For example, when the relative error RMS value of the response curves under the current iteration excitation and the response signal under the experiment is less than the preset value (e.g., 0.1), the virtual iteration is stopped when the preset iteration termination condition is met. The same method in step 103 is used to determine whether the local model meets the preset accuracy index requirements. After all local models meet the preset accuracy index requirements, the second iteration result is obtained. Then, the initial multibody dynamics model (global model) is optimized. Based on the second iteration result and the preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is evaluated for response surface optimization, and the optimal multibody dynamics model is established.

[0131] Step 105: If all local models meet the requirements, but the initial multibody dynamics model does not meet the requirements, then based on the results of the first iteration and the preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is evaluated by response surface optimization to establish the optimal multibody dynamics model;

[0132] Step 105 includes steps 1051 to 1053, and the specific details of each step are as follows:

[0133] Step 1051: Based on the iteration results, determine the elements to be corrected, and based on the elements to be corrected, perform sensitivity and contribution analysis to obtain design variables;

[0134] In this embodiment, based on the actual evaluation process, the iteration result is either the first iteration result or the second iteration result. Based on the iteration result, the vibration transmission path and component to be corrected are determined. The sensitivity formula is as follows:

[0135]

[0136]

[0137] In the formula, x is the output response at the target point, k is the stiffness, c is the damping, and l is the number of components on the transmission path.

[0138] According to the response contribution formula, the formula is as follows:

[0139]

[0140] In the formula, and This represents the value after the local stiffness and damping have changed by the same amount; and This represents the stiffness and damping of the original model; and Let x and x represent the values ​​of the original model after the overall change. These are the displacement and velocity responses, respectively.

[0141] The design variables for model correction are determined using the sensitivity formula, contribution formula, and initial multibody dynamics model (dynamic equations).

[0142] Step 1052: Establish the first design variable matrix based on the design variables and the preset first orthogonal Latin hypercube experiment.

[0143] In this embodiment, q independent random permutations of 1, 2, ..., p are taken, and then the permutations are used as column vectors to form a p×q design matrix. Specifically, any two columns in the design are orthogonal, and the square column of any column and the dot product column of any two columns are orthogonal to the other columns of the design. Given the number of experiments as 2^(c+1) (c≥1) and the number of factors as 2^c, the super Latin square experiment construction method is LHD(2^(c+1),2^c).

[0144] The super Latin square experiment construction method is adopted. Based on the design variables and the preset orthogonal Latin hypercube experiment, the sample points determined by the orthogonal Latin hypercube experiment design are used to form the design variable matrix. If the first orthogonal Latin hypercube experiment is preset, the first design variable matrix is ​​established.

[0145] Step 1053: Based on the first design variable matrix, design variables, and accuracy index, perform response surface optimization evaluation on the initial multibody dynamics model to establish the optimal multibody dynamics model.

[0146] Step 1053 is as follows: Figure 4 The response surface optimization evaluation process is shown, including steps 401 to 404, each step of which is detailed below:

[0147] Step 401: Combine the first design variable matrix with the initial multibody dynamics model to obtain the first response;

[0148] In this embodiment, the design variable matrix is ​​determined by orthogonal Latin hypercube experimental design, and the response is obtained by combining the design variable matrix with the kinetic equation. Based on the first design variable matrix, the first response is obtained.

[0149] Step 402: Establish the first response surface model based on the design variable values, the first response value, and the response surface approximation function;

[0150] In this embodiment, an incomplete fourth-order polynomial is used as the response surface approximation function, as shown in the following formula:

[0151]

[0152] In the formula, Let x be the response surface approximation function. i Design variables for the model, β (0,i.ij,ii,iii,iiii) These are coefficients to be determined.

[0153] Substitute the design variable values ​​and response values ​​into the response surface approximation function to determine the undetermined coefficient β. (0,i.ij,ii,iii,iiii) This allows us to establish a response surface model, which is essentially an approximate function of the response surface with determined coefficients. Based on the design variable values, the first response value, and the approximate response surface function, a first response surface model is then established.

[0154] Step 403: Verify the accuracy of the first response surface model and evaluate whether the first response surface model meets the preset response surface accuracy requirements;

[0155] In this embodiment, the accuracy of the established response surface model is verified using the root mean square residual (RMSE) and the coefficient of determination (R²). 2 The verification is performed using the following formula:

[0156]

[0157]

[0158] In the formula, k is the number of samples; For the response surface calculation, yi These are the initial values ​​calculated for the multibody dynamics model. This represents the average value of the calculation results from the initial multibody dynamics model. A low RMSE indicates a small response surface error; a low RMSE indicates a small response surface error. 2 A value approaching 1 indicates a high similarity between the response surface and the kinetic model, and the preset RMSE value and R0 of the response surface model meet the accuracy requirements. 2 The value is verified by the calculated RMSE value and R. 2 The value is compared with the preset value. If the calculated RMSE value is less than the preset RMSE value, R is calculated. 2 Value greater than preset R 2 The value meets the preset requirements, which means that the response surface model meets the preset response surface accuracy requirements.

[0159] Step 404: If the first response surface model does not meet the preset response surface accuracy requirements, then a second orthogonal Latin hypercube experiment is preset. Based on the second orthogonal Latin hypercube experiment and design variables, a second response surface model is established until the second response surface model meets the preset response surface accuracy requirements. Then, the design of the orthogonal Latin hypercube experiment is stopped, and the subsequent steps of ensuring that the response surface model meets the preset response surface accuracy requirements are entered.

[0160] Step 404 specifically involves: pre-setting a second orthogonal Latin hypercube experiment, and establishing a second design variable matrix based on the second orthogonal Latin hypercube experiment and the design variables;

[0161] The second design variable matrix is ​​combined with the initial multibody dynamics model to obtain the second response;

[0162] The second response surface model is established by using the design variable values, the second response values, and the response surface approximation function.

[0163] The accuracy of the second response surface model is verified to evaluate whether the second response surface model meets the preset response surface accuracy requirements.

[0164] If the second response surface model does not meet the preset response surface accuracy requirements, modify the preset orthogonal Latin hypercube experiment to obtain the third response, and verify the accuracy of the third response surface model until the third response surface model meets the preset response surface accuracy requirements, and then stop the design of the orthogonal Latin hypercube experiment.

[0165] The next step is to meet the preset response surface accuracy requirements of the response surface model. Based on the design variables, the modified response, and the convolutional neural network, an optimized response surface model is established. Based on the optimized response surface model, an optimal multibody dynamics model is established. The modified response is either the second or third response.

[0166] In this embodiment, when the accuracy of the first response surface model does not meet the requirements, it is necessary to continue modifying the orthogonal Latin hypercube experimental design, changing the preset sample points, thereby changing the design variable matrix, and obtaining the second design variable matrix. After obtaining the second design variable matrix, the second response surface model is established according to the same method as in step 402. According to the same method as in step 403, it is determined whether the second response surface model meets the preset response surface accuracy requirements. If the second response surface model still does not meet the preset response surface accuracy requirements, the preset orthogonal Latin hypercube experiment is modified again, and the accuracy of the modified response surface model (third response surface model) is verified. By continuously modifying the orthogonal Latin hypercube experimental design and verifying the accuracy of the response surface model, until the third response surface model meets the preset response surface accuracy requirements, the modified response (second response or third response) and the modified design variable matrix can be obtained. At the same time, the design of the orthogonal Latin hypercube experiment is stopped, and the subsequent steps of ensuring that the response surface model meets the preset response surface accuracy requirements are continued. That is, the response surface model that meets the accuracy requirements is then optimized by a neural network.

[0167] Step 405: If the response surface model meets the preset response surface accuracy requirements, then an optimized response surface model is established based on the design variables, response, and convolutional neural network. Based on the optimized response surface model, an optimal multibody dynamics model is established. The response surface model includes a first response surface model, a second response surface model, and a modified response surface model. The response is either the first response or the modified response.

[0168] Step 405 specifically involves: training a convolutional neural network based on the design variables and response; establishing an optimized response surface model based on the training results of the convolutional neural network; establishing an optimized multibody dynamics model based on the optimized response surface model; continuing until the optimized multibody dynamics model meets the preset accuracy index requirements; stopping the training of the convolutional neural network to obtain the optimal response surface model; and establishing the optimal multibody dynamics model based on the optimal response surface model.

[0169] In this embodiment, the calculation formula for the l-th layer of the convolutional neural network is as follows:

[0170]

[0171] g(.) = max(0,x)

[0172] In the formula, This represents the nth feature of the output value of the l-th layer. This represents the weight matrix of the nth convolutional kernel in the l-th layer. This represents the output of the (l-1)th layer. represents the paranoid term, and g(.) represents the activation function.

[0173] As an example of neural network design, the initial number of network layers is 2, the number of convolutional kernels is 12 with a size of 3, the pooling layers all use max pooling with a size of 2, the batch size is 300, the epoch is 2000, the learning rate is 0.01, the dropout is 0.5, and the simulation scenario is set to a signal-to-noise ratio (SNR) of 20dB and a sequence length of 10s.

[0174] Based on the input (design variables) of the orthogonal Latin hypercube experiment and the output response of the response surface model (the response is obtained by combining the design variable matrix with the initial multibody dynamics equations), a convolutional neural network is trained. The design variable matrix is ​​the matrix of design variables that ensures the response surface model meets the accuracy requirements; it can be the first, second, or third design variable matrix, etc. The response is the response used to verify that the response surface model meets the accuracy requirements; it can be the first response or a modified response, i.e., the first, second, or third response, etc. The convolutional neural network is trained, and the optimized design variable values ​​are fed into the initial multibody dynamics model to obtain the optimized multibody dynamics model. The optimized multibody dynamics model is then verified again to ensure it meets the preset accuracy requirements. If it does not meet the requirements, the neural network is continuously trained until the accuracy assessment of the optimized multibody dynamics model meets the requirements, at which point training stops. This yields the optimal response surface model optimized using the convolutional neural network algorithm, thus completing the establishment of the optimal multibody dynamics model.

[0175] Step 106: If all requirements are met, the initial multibody dynamics model is taken as the optimal multibody dynamics model;

[0176] In this embodiment, when it is determined that each local model and the initial multibody dynamics model meet the accuracy requirements, the multibody dynamics model can be directly established without subsequent optimization, reducing the steps of multiple iterative optimizations, improving modeling efficiency, and using the initial multibody dynamics model as the optimal multibody dynamics model.

[0177] By implementing the embodiments of the present invention, an initial multibody dynamics model is established based on the vibration model of the automobile and the Lagrange equation. The Lagrange equation and the vibration model are combined for modeling. The accuracy index is established using the vibration transmissibility method under working conditions. The local model is iteratively checked using load spectrum data. Then, the accuracy of the complete model is evaluated, completing the index decomposition from global to local. For those models that meet the reliability and accuracy index, the initial multibody dynamics model is used as the final model, which greatly improves the efficiency of model building. For local models that do not meet the accuracy requirements, iteration is performed. Through the iteration results and the design of orthogonal Latin hypercube tests, response surface optimization evaluation is carried out to establish the optimal multibody dynamics model that meets the reliability and accuracy index, effectively improving the accuracy, reliability and applicability of modeling.

[0178] Example 2

[0179] Accordingly, see Figure 5 , Figure 5 This is a schematic diagram of a second embodiment of the automotive multibody dynamics modeling system provided by the present invention. Figure 5 As shown, the vehicle multibody dynamics modeling system includes: an initial module 501, an accuracy index module 502, an evaluation module 503, an accuracy satisfaction module 504, a local optimization module 505, and a global optimization module 506.

[0180] The initial module 501 is used to establish an initial multibody dynamics model based on the vibration model of the car and the Lagrange equation. The initial multibody dynamics model consists of several local models.

[0181] The accuracy index module 502 is used to establish accuracy indexes based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path; the response measurement points are one or more selected from the key points.

[0182] The evaluation module 503 is used to perform virtual iteration on the initial multibody dynamics model based on the load spectrum data of key points in the vibration transmission path collected by the road condition test and the first preset iteration end condition, and to determine whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index based on the first iteration result and accuracy index.

[0183] The accuracy module 504 is used to select the initial multibody dynamics model as the optimal multibody dynamics model if all local models and the initial multibody dynamics model meet the accuracy index requirements.

[0184] The local optimization module 505 is used to perform virtual iteration on the initial multibody dynamics model again based on the first iteration result if any local model does not meet the accuracy index requirements, until all local models meet the preset accuracy index requirements, and then obtain the second iteration result.

[0185] The global optimization module 506 is used to optimize and evaluate the response surface of the initial multibody dynamics model and establish the optimal multibody dynamics model if the initial multibody dynamics model does not meet the accuracy requirements, based on the iteration results and the preset orthogonal Latin hypercube experiment.

[0186] By implementing the embodiments of the present invention, an initial multibody dynamics model is established based on the vibration model of the automobile and the Lagrange equation. The Lagrange equation and the vibration model are combined for modeling. The accuracy index is established using the vibration transmissibility method under working conditions. The local model is iteratively checked using load spectrum data. Then, the accuracy of the complete model is evaluated, completing the index decomposition from global to local. For those models that meet the reliability and accuracy index, the initial multibody dynamics model is used as the final model, which greatly improves the efficiency of model building. For local models that do not meet the accuracy requirements, iteration is performed. Through the iteration results and the design of orthogonal Latin hypercube tests, response surface optimization evaluation is carried out to establish the optimal multibody dynamics model that meets the reliability and accuracy index, effectively improving the accuracy, reliability and applicability of modeling.

[0187] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for modeling multibody dynamics of a vehicle, characterized in that, include: Based on the vibration model of the automobile and the Lagrange equation, an initial multibody dynamics model is established; Based on the load spectrum data of key points in the vibration transmission path collected by the road condition test and the first preset iteration termination condition, the initial multibody dynamics model is virtually iterated. Based on the first iteration results and accuracy indicators, it is determined whether each local model and the initial multibody dynamics model meet the preset accuracy requirements. The initial multibody dynamics model consists of several local models. The accuracy indicators are established based on the working condition transmission path analysis method and the transmission rate at response measurement points in the vibration transmission path. The response measurement points are one or more selected from the key points. If all requirements are met, the initial multibody dynamics model is taken as the optimal multibody dynamics model. If all the local models meet the requirements, but the initial multibody dynamics model does not meet the requirements, then based on the first iteration results and the preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is evaluated by response surface optimization to establish the optimal multibody dynamics model; If any of the local models fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the first iteration result until all the local models meet the preset accuracy index requirements. Then, the second iteration result is obtained, and the initial multibody dynamics model is optimized and evaluated based on the second iteration result and the preset orthogonal Latin hypercube experiment to establish the optimal multibody dynamics model.

2. The vehicle multibody dynamics modeling method as described in claim 1, characterized in that, The initial multibody dynamics model is established based on the vibration model of the automobile and the Lagrange equations, specifically as follows: Based on the structural combination, vibration excitation, and response of the automotive system, mass damping abstraction is performed to establish the vibration model. Based on the Lagrange equation and the vibration model, a dynamic equation is established, and the component parameters of the vibration model are substituted into the dynamic equation for calculation to establish the initial multibody dynamic model.

3. The vehicle multibody dynamics modeling method as described in claim 1, characterized in that, The accuracy index is established based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path, specifically: Based on the working condition transmission path analysis method, a transmission equation is established. Based on the transmission equation, a first transmission rate matrix is ​​calculated. Based on the first transmission rate matrix and the transmission rate at the response measurement point in the vibration transmission path, the accuracy index is established.

4. The vehicle multibody dynamics modeling method as described in claim 3, characterized in that, The initial multibody dynamics model is virtually iterated based on the load spectrum data of key points in the vibration transmission path collected from the road condition test and the first preset iteration termination condition. Based on the first iteration results and accuracy indicators, it is determined whether each local model and the initial multibody dynamics model meet the preset accuracy requirements. Specifically: Sensing devices are deployed at key points along the vibration transmission path in the vibration model to perform road condition test response measurements and obtain the load spectrum data. Select the output point of the component in the vibration transmission path; Based on the load spectrum data and the Newton-Rafaelson formula, the virtual iteration is performed as follows: u k+1 (s)=u k (s)+f -1 (s)(y d (s)-y k (s)); k = 1, 2, 3, ..., n-1; Among them, u k (s) represents the k-th excitation signal, y k u1(s) is the response signal iterated by the k-th excitation signal, u1(s) is the initial excitation signal, and f -1 (s) is the inverse propagation function, y d (s) represents the load spectrum data; The virtual iteration stops when the first preset iteration termination condition is met. The excitation signal of the kth iteration is input into the initial multibody dynamics model to obtain the model response signal at the output point; wherein, the excitation signal of the kth iteration is the excitation signal that satisfies the first preset iteration termination condition; Based on the model response signal, establish the second transfer rate matrix; Based on the first transfer rate matrix, the second transfer rate matrix, and the accuracy index, determine whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index.

5. The vehicle multibody dynamics modeling method as described in claim 4, characterized in that, If any of the local models fails to meet the requirements, then based on the first iteration result, the initial multibody dynamics model is virtually iterated again until all the local models meet the preset accuracy index requirements. Then, a second iteration result is obtained. Based on the second iteration result and a preset orthogonal Latin hypercube experiment, the initial multibody dynamics model is optimized using response surface evaluation to establish the optimal multibody dynamics model. Specifically: If any of the local models does not meet the requirements, the initial multibody dynamics model is virtually iterated again based on the first iteration result and the Newton-Rafaelson formula, as follows: u l+1 (s)=u l (s)+f -1 (s)(y b (s)-y l (s)); l＝1,2,3,....,n-1； Among them, u l (s) represents the k-th excitation signal, y l u1(s) is the response signal iterated by the k-th excitation signal, u1(s) is the initial excitation signal, and f -1 (s) is the inverse propagation function, y b (s) represents the result of the first iteration; Based on the relative error value of the response curve, the preset iteration termination condition is modified, and the virtual iteration is stopped when the modified iteration termination condition is met. The l-th excitation signal is input into the initial multibody dynamics model to obtain the modified model response signal at the output point; wherein, the l-th excitation signal is the excitation signal that satisfies the modified iteration termination condition; Based on the modified model response signal, a modified transitivity matrix is ​​established, and the modified iteration result is obtained. Based on the first transfer rate matrix, the modified transfer rate matrix, and the accuracy index, determine whether each of the local models meets the preset accuracy index requirements; If any of the local models fails to meet the requirements, the initial multibody dynamics model is virtually iterated again based on the modified iteration results until all the local models meet the preset accuracy index requirements. Then, the second iteration result is obtained, and the initial multibody dynamics model is optimized and evaluated based on the second iteration result and the preset orthogonal Latin hypercube experiment to establish the optimal multibody dynamics model.

6. The vehicle multibody dynamics modeling method as described in claim 1, characterized in that, The initial multibody dynamics model is evaluated and optimized based on the iteration results and a preset orthogonal Latin hypercube experiment to establish an optimal multibody dynamics model. Specifically, the iteration result is either the first iteration result or the second iteration result. Based on the iteration results, the elements to be corrected are determined. Based on the elements to be corrected, sensitivity and contribution analysis are performed to obtain design variables. Based on the design variables and the preset first orthogonal Latin hypercube experiment, a first design variable matrix is ​​established. Based on the first design variable matrix, the design variables, and the accuracy index, the initial multibody dynamics model is evaluated by response surface optimization to establish the optimal multibody dynamics model.

7. The vehicle multibody dynamics modeling method as described in claim 6, characterized in that, The step of performing response surface optimization evaluation on the initial multibody dynamics model based on the first design variable matrix, the design variables, and the accuracy index, and establishing the optimal multibody dynamics model, specifically involves: The first design variable matrix is ​​combined with the initial multibody dynamics model to obtain the first response; Based on the design variable values, the first response value, and the response surface approximation function, establish the first response surface model; The accuracy of the first response surface model is verified to evaluate whether the first response surface model meets the preset response surface accuracy requirements. If so, then based on the design variables, the first response, and the convolutional neural network, an optimized response surface model is established, and based on the optimized response surface model, the optimal multibody dynamics model is established. If not, a second orthogonal Latin hypercube experiment is preset. Based on the second orthogonal Latin hypercube experiment and the design variables, a second response surface model is established until the second response surface model meets the preset response surface accuracy requirements. Then, the design of the orthogonal Latin hypercube experiment is stopped, and the subsequent steps of meeting the preset response surface accuracy requirements are entered.

8. The vehicle multibody dynamics modeling method as described in claim 7, characterized in that, The pre-set second orthogonal Latin hypercube experiment involves establishing a second response surface model based on the second orthogonal Latin hypercube experiment and the design variables. The design of the orthogonal Latin hypercube experiment is stopped when the second response surface model meets the pre-set response surface accuracy requirements, and the subsequent steps proceed to ensure the response surface model meets the pre-set accuracy requirements. Specifically: A second orthogonal Latin hypercube experiment is preset, and a second design variable matrix is ​​established based on the second orthogonal Latin hypercube experiment and the design variables; The second design variable matrix is ​​combined with the initial multibody dynamics model to obtain the second response; The second response surface model is established by using the design variable values, the second response value, and the approximate function of the response surface. The accuracy of the second response surface model is verified to evaluate whether the second response surface model meets the preset response surface accuracy requirements. If the second response surface model does not meet the preset response surface accuracy requirements, modify the preset orthogonal Latin hypercube experiment to obtain the third response, and verify the accuracy of the third response surface model until the third response surface model meets the preset response surface accuracy requirements, and then stop the design of the orthogonal Latin hypercube experiment. The next step after the response surface model meets the preset accuracy requirements is to establish an optimized response surface model based on the design variables, the modified response, and the convolutional neural network, and then establish the optimal multibody dynamics model based on the optimized response surface model; wherein the modified response is the second response or the third response.

9. The vehicle multibody dynamics modeling method as described in claim 7 or 8, characterized in that, The step involves establishing an optimized response surface model based on the design variables, response, and convolutional neural network, and then establishing the optimal multibody dynamics model based on the optimized response surface model. Specifically, the response is either the first response or the modified response. Based on the design variables and the response, a convolutional neural network is trained. Based on the training results of the convolutional neural network, an optimized response surface model is established. Based on the optimized response surface model, an optimal multibody dynamics model is established. The training of the convolutional neural network is stopped until the optimal multibody dynamics model meets the preset accuracy index requirements, thus obtaining the optimal response surface model. Based on the optimal response surface model, an optimal multibody dynamics model is established.

10. A multibody dynamics modeling system for automobiles, characterized in that, include: Initial module, accuracy index module, evaluation module, accuracy satisfaction module, local optimization module, and global optimization module; The initial module is used to establish an initial multibody dynamics model based on the vibration model of the vehicle and the Lagrange equation. The initial multibody dynamics model consists of several local models. The accuracy index module is used to establish accuracy indexes based on the working condition transmission path analysis method and the transmission rate at the response measurement points in the vibration transmission path; the response measurement points are one or more selected from key points. The evaluation module is used to perform virtual iteration on the initial multibody dynamics model based on the load spectrum data of key points in the vibration transmission path collected by the road condition test and the first preset iteration termination condition, and to determine whether each local model and the initial multibody dynamics model meet the requirements of the preset accuracy index based on the first iteration result and accuracy index. The accuracy module is used to select the initial multibody dynamics model as the optimal multibody dynamics model if all local models and the initial multibody dynamics model meet the accuracy index requirements. The local optimization module is used to perform virtual iteration on the initial multibody dynamics model again based on the first iteration result if any of the local models does not meet the accuracy index requirements, until all the local models meet the preset accuracy index requirements, and then obtain the second iteration result. The global optimization module is used to optimize and evaluate the initial multibody dynamics model based on the iteration results and the preset orthogonal Latin hypercube experiment if the initial multibody dynamics model does not meet the accuracy requirements, and to establish the optimal multibody dynamics model.

Images