Data processing method, device and equipment of multi-body suspension system, medium and product

Abstract

The embodiment of the invention provides a data processing method and device of a multi-body suspension system, equipment, a medium and a product. The method comprises the following steps: constructing a historical database; the historical database comprises input features and output labels corresponding to the input features, the input features comprise topological parameters and design targets of the suspension model, and the output labels comprise initial state vectors obtained through convergence solving according to the Newton-Raphson method; according to the input features and the output labels, training a machine learning model to obtain a trained machine learning model; and predicting a prediction state vector of the new design task based on the trained machine learning model, and taking the prediction state vector as an initial value of a Newton-Raphson method. The method is used for improving the iteration effect of the Newton-Raphson method.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, apparatus, equipment, medium and product for a multibody suspension system. Background Technology

[0002] The vehicle suspension system is a crucial subsystem affecting the vehicle's handling stability, ride comfort, and safety. As automotive R&D moves towards intelligence, platformization, and rapid iteration, reverse engineering of the suspension has become a key aspect of chassis development.

[0003] Current mainstream suspension reverse engineering methods typically combine multibody dynamics modeling with numerical iterative algorithms (such as the Newton-Raphson method). The process includes: first, establishing a multibody dynamics model based on the suspension topology, including parameters such as hardpoint coordinates and spring / bushing mounting positions; second, the user inputs the design objectives; then, engineers manually set the initial state parameters based on empirical formulas or historical cases; finally, using these initial values, the Newton-Raphson method is used for iterative solving until the residuals converge.

[0004] However, when determining initial values, existing methods suffer from poor iterative results due to their heavy reliance on human experience and lack of automation and generalization capabilities. Summary of the Invention

[0005] The data processing methods, apparatus, equipment, media, and products for multibody suspension systems provided in this application are intended to improve the iterative effect of the Newton-Raphson method.

[0006] In a first aspect, embodiments of this application provide a data processing method for a multi-body suspension system, including:

[0007] Construct a historical database; the historical database includes input features and output labels corresponding to the input features, wherein the input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method;

[0008] The machine learning model is trained based on the input features and output labels to obtain the trained machine learning model.

[0009] The predicted state vector for the new design task is predicted based on the trained machine learning model, and the predicted state vector is used as the initial value for the Newton-Raphson method.

[0010] In one possible implementation, a historical database is constructed, including:

[0011] Automatically extract input features and output labels after completing the suspension design task using the Newton-Raphson method;

[0012] The input features and output labels after completing the suspension design task are stored in the historical database.

[0013] In one possible implementation, when the amount of data in the historical database is less than a preset amount of data, the method further includes:

[0014] Based on preset physical rules and preset design goals, generate initial guess values;

[0015] Based on a preset data range, the initial guess value is perturbed by parameters to obtain the initial target guess value;

[0016] The initial target guess is input into the parametric suspension model to obtain the output result;

[0017] Based on the output results, the initial target guess values ​​are filtered to determine the target guess values ​​that meet the preset operating conditions;

[0018] The target guess value and the preset design target are used as input features, and the target output result corresponding to the target guess value is used as output label and stored in the historical database.

[0019] In one possible implementation, the initial target guess values ​​are filtered based on the output results to determine target guess values ​​that meet preset operating conditions, including:

[0020] Based on the preset operating conditions, determine the multidimensional satisfaction vector of the output results;

[0021] Perform Pareto sorting on the multidimensional satisfaction vector to determine the non-dominated solutions in the multidimensional satisfaction vector;

[0022] Based on the non-dominated solution, determine the target guess value that satisfies the preset operating conditions.

[0023] In one possible implementation, the predicted state vector for the new design task is predicted based on the trained machine learning model, and the predicted state vector is used as the initial value for the Newton-Raphson method, including:

[0024] Extract input features from the historical database that correspond to the target input features in the new design task;

[0025] The target input features and input features are fed into the trained machine learning model to obtain the predicted state vector of the machine learning model for the new design task, and the predicted state vector is used as the initial value of the Newton-Raphson method.

[0026] In one possible implementation, when the number of training iterations of the machine learning model is less than a preset number,

[0027] The target input features and input features are fed into the trained machine learning model to obtain the predicted state vector of the machine learning model for the new design task. The predicted state vector is then used as the initial value for the Newton-Raphson method, including:

[0028] The target input features and input features are fed into the trained machine learning model to obtain the initial prediction state vector of the machine learning model for predicting the new design task.

[0029] Based on the preset feasibility criteria, the credibility of the initial predicted state vector is evaluated;

[0030] If the initial predicted state vector satisfies the feasibility criterion, then the initial predicted state vector is used as the predicted state vector, and the predicted state vector is used as the initial value of the Newton-Raphson method.

[0031] If the initial predicted state vector does not meet the feasibility criterion, the initial predicted state vector is adjusted based on the verified state vector closest to the initial predicted state vector in the historical database, and the adjusted initial predicted state vector is used as the initial value of the Newton-Raphson method.

[0032] In one possible implementation, there are multiple machine learning models, and the different machine learning models have different model types and different weights.

[0033] The trained machine learning model predicts the state vector for the new design task, and uses the predicted state vector as the initial value for the Newton-Raphson method, including:

[0034] Obtain the predicted state vector for each machine learning model in predicting the new design task;

[0035] Based on the weights corresponding to the machine learning model, the predicted state vector is weighted and summed, and the weighted summed predicted state vector is used as the initial value of the Newton-Raphson method.

[0036] In one possible implementation, after predicting the predicted state vector for the new design task based on the trained machine learning model and using the predicted state vector as the initial value for the Newton-Raphson method, the method further includes:

[0037] Obtain and display the output solution of the Newton-Raphson method obtained from the predicted state vector.

[0038] Secondly, embodiments of this application provide a data processing device for a multi-body suspension system, comprising:

[0039] The module is used to build a historical database. The historical database includes input features and output labels corresponding to the input features. The input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method.

[0040] The training module is used to train the machine learning model based on the input features and output labels to obtain the trained machine learning model.

[0041] The prediction module is used to predict the state vector of the new design task based on the trained machine learning model, and uses the predicted state vector as the initial value of the Newton-Raphson method.

[0042] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0043] The memory stores the instructions that the computer executes;

[0044] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0045] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0046] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0047] The data processing method, apparatus, device, medium, and product for multi-body suspension systems provided in this application embodiment record the correspondence between input features and output labels of each successful design case in a historical database, thereby forming structured data. A machine learning model then learns the nonlinear mapping relationship between input features and output labels based on this historical database. When a user inputs new suspension model parameters and design goals, the model can extract the corresponding input features and input them into the trained machine learning model to obtain the initial state vector predicted by the model output. This initial state vector is then used as the initial value for the Newton-Raphson method to initiate the iterative solution process. Thus, by making the implicit engineering experience in historical data explicit into the capabilities of the machine learning model, the reverse design of suspensions is automated and efficient. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0049] Figure 1 A schematic diagram illustrating a scenario for the data processing method of the multibody suspension system provided in this application;

[0050] Figure 2 Flowchart of the data processing method for the multibody suspension system provided in this application Figure 1 ;

[0051] Figure 3 Flowchart of the data processing method for the multibody suspension system provided in this application Figure 2 ;

[0052] Figure 4 A schematic diagram of the data processing device for the multibody suspension system provided in this application;

[0053] Figure 5 A schematic diagram of the structure of the electronic device provided in this application.

[0054] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0055] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0056] First, let me explain the terms used in this application:

[0057] The Newton-Raphson method is a classic numerical iterative algorithm for solving nonlinear equations. Its core idea is to locally linearize the nonlinear problem by performing a first-order Taylor expansion of the objective function at the current estimation point, and then use this linear approximation to quickly approximate the roots of the equations. In reverse engineering of vehicle suspensions, this method is often used to solve nonlinear constraint equations derived from multibody dynamics models (such as inferring hardpoint coordinates from a given wheel center position). It continuously updates state variables (such as joint displacements and spring deformations) until the system residuals meet a preset convergence accuracy.

[0058] Machine learning models refer to mathematical algorithm systems that automatically learn patterns from data and use them for prediction or decision-making. The core idea is to optimize internal parameters to make the model approximate the target output as accurately as possible under given input conditions, without explicitly programming the rules. Depending on the learning method, machine learning models can be categorized into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Forms of machine learning models can include linear regression, decision trees, support vector machines, neural networks, random forests, Gaussian processes, etc. After training, the model can make generalized predictions on new, unseen inputs and is widely used in tasks such as classification, regression, clustering, and dimensionality reduction.

[0059] The suspension system can refer to the mechanical structure in a vehicle chassis used to support the wheels and buffer road impacts, such as MacPherson strut suspension and double wishbone suspension.

[0060] In existing technologies, although the suspension reverse design method based on the Newton-Raphson method has solved the accuracy problem of multi-objective coupled solutions, its convergence speed and success rate depend to some extent on the initial guesses given by engineers. A poor initial value may lead to an increase in the number of iterations, or even iteration divergence. Currently, the setting of initial values ​​still relies on engineers' rough estimation experience, which has become a bottleneck in the automation process. This problem is particularly prominent when dealing with suspension models with entirely new architectures, where there is a lack of empirical data for reference.

[0061] The data processing method, apparatus, device, medium, and product for multi-body suspension systems provided in this application utilize the implicit experience from historical successful cases. Through a machine learning model, it learns the complex mapping relationship between input features and output labels, and uses the predicted results as initial values ​​for the hot start of the Newton-Raphson method. This achieves automation, efficiency, and improved robustness in suspension reverse design. Thus, engineering experience from historical data is made explicit into model capabilities, and the system's continuous evolution is achieved through a data closed-loop mechanism. This solves the problems of initial value dependence on experience, low computational efficiency, and difficulty in cold start in traditional methods, thereby improving the iterative effect of the Newton-Raphson method.

[0062] Figure 1 A schematic diagram illustrating a scenario for the data processing method of the multi-body suspension system provided in this application, such as... Figure 1As shown, the specific application scenario of this application is a data processing system. This system can be a server, which can be a computer, tablet, or other device. This application does not restrict the executing entity of the data processing method for the multi-body suspension system, as long as it can build a historical database. The historical database includes input features and output labels corresponding to the input features. The input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method. The machine learning model is trained based on the input features and output labels to obtain the trained machine learning model. The predicted state vector can be used as the initial value of the Newton-Raphson method.

[0063] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0064] Figure 2 Flowchart of the data processing method for the multibody suspension system provided in this application Figure 1 ,like Figure 2 As shown, the method includes:

[0065] S201. Construct a historical database; the historical database includes input features and output labels corresponding to the input features. The input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method.

[0066] The historical database can refer to a structured data set accumulated during the suspension design process, used to store relevant information about completed design tasks. In this embodiment, the historical database includes input features and corresponding output labels.

[0067] The input features in the historical database can refer to the set of parameters used for the suspension model and design objectives. This set can include topology parameters and design objectives. Topology parameters can refer to the geometric and connectivity properties that characterize the layout of the suspension mechanical structure, such as control arm length, hard point coordinates, spring mounting angles, etc., which are used to define the basic configuration of the multibody dynamics model. Design objectives are the performance indicators that the user expects to achieve, such as wheel center motion trajectory, wheel positioning angle, roll center height, or load distribution.

[0068] The output labels in the historical database can refer to the initial state vector obtained by convergence through the Newton-Raphson method, which can include parameters such as spring pre-deformation, bushing pre-displacement, pre-rotation, and four-wheel alignment angles in the suspension system.

[0069] Historical databases can be constructed by systematically collecting and structuredly storing input and output data from suspension design tasks. Specifically, during each design iteration or simulation verification, user-defined design objectives and suspension topology parameters are used as input features, while the initial state vector successfully converged using the Newton-Raphson method is stored as the output label in the database to obtain the historical database.

[0070] Optionally, the data in the historical database can be historical data.

[0071] In this embodiment, after a suspension design is successfully completed using the Newton-Raphson method, the system automatically records a data sample. Each data sample consists of two parts:

[0072] 1. Input Features (X): Used to describe the suspension model and design goals. Includes:

[0073] Suspension topology features: such as suspension type (MacPherson strut, double wishbone, etc.), key hardpoint coordinates, and the installation positions of springs and bushings.

[0074] Design target values: target wheel load F_z_target, target wheel center coordinates (X, Y, Z), target camber angle Camber_target, target toe angle Toe_target.

[0075] 2. Output Labels (Y): This is the initial state vector obtained by the final convergence solution of the Newton-Raphson method. This vector includes the initial deformation of the spring, the pre-displacement / pre-rotation of each bushing node, and the four-wheel positioning angles, etc.

[0076] S202. Based on the input features and output labels, train the machine learning model to obtain the trained machine learning model.

[0077] This process involves several steps. First, a large number of validated data samples are extracted from historical databases, each sample including input features and corresponding output labels. Second, the data samples are divided into training, validation, and test sets, and corresponding model structures (such as multilayer perceptron, Gaussian process regression, or graph neural networks) are selected. Third, a loss function (such as mean squared error) is defined to measure the deviation between the model's predicted initial state vector and the true convergent initial value, and an optimization algorithm (such as stochastic gradient descent or Adam) is used to iteratively update the model parameters to minimize this loss. During training, model performance is monitored through the validation set to prevent overfitting, and hyperparameters are adjusted based on metrics such as convergence and generalization ability. Finally, the model that performs best on the test set is selected as the trained machine learning model for automatic prediction of high-quality initial values ​​in newly designed tasks.

[0078] In the embodiments of this application, the machine learning model may be a gradient boosting decision tree (such as XG Boost, LightGBM) or a deep neural network (DNN) that can handle complex nonlinear relationships.

[0079] S203. Based on the trained machine learning model, predict the predicted state vector of the new design task, and use the predicted state vector as the initial value of the Newton-Raphson method.

[0080] Among them, new design tasks can refer to design requirements that are proposed for the first time during the suspension development process and have not yet been solved or verified. Specifically, they are a new set of input conditions, including specific suspension topologies (such as new double wishbone layouts) and user-defined performance targets (such as target wheel center motion trajectories, desired camber angle variation ranges, or load distribution requirements).

[0081] The predicted state vector can refer to a set of physical parameter estimates output by the trained machine learning model for a new design task, used to characterize the initial working state of the suspension system under the premise of meeting the design objectives. The vector can include continuous variables such as spring pre-compression, shock absorber initial stroke, bushing pre-deformation displacement, and joint initial rotation angle. Although its value is a prediction result, it should have physical feasibility and numerical stability, and be used as the initial value for the Newton-Raphson method iterative solution to accelerate convergence and avoid divergence.

[0082] The data processing method for multi-body suspension systems provided in this application utilizes techniques such as constructing a historical database and training a machine learning model. This structured storage of the historical database makes implicit experiences in suspension design cases (such as the mapping relationship between spring preload and wheel load) explicit, providing a reliable data foundation for the machine learning model. The machine learning model can then learn the complex mapping relationship between input features and output labels through algorithms, enabling it to quickly generate initial guesses that approximate the true solution. Thus, by transforming engineering experience into algorithmic capabilities, a paradigm shift from experience-driven to data-driven suspension reverse design is achieved, improving computational efficiency and automation while reducing reliance on individual engineer experience. This solves the bottleneck problem of experience-dependent initial values ​​in suspension reverse design.

[0083] Optionally, a historical database is constructed, including:

[0084] Automatically extract input features and output labels after completing the suspension design task using the Newton-Raphson method;

[0085] The input features and output labels after completing the suspension design task are stored in the historical database.

[0086] Among them, the suspension design task can refer to the complete process of calculating the initial state vector of the suspension system through physical simulation methods based on the suspension model parameters and design objectives input by the user.

[0087] After the suspension design task is successfully completed, the system automatically extracts the input features (such as suspension type and target wheel load) and output labels (such as spring pre-deformation and bushing pre-displacement) of the task and stores them in a structured manner in the historical database. This process can be linked with simulation tools through a programmatic interface to ensure the integrity and consistency of data records.

[0088] Therefore, by automatically recording the input features and output labels of successful cases, the historical database is dynamically expanded. This ensures that the system can continuously accumulate high-quality training data, providing a stable data source for machine learning models and thus improving the generalization ability of model predictions.

[0089] In this embodiment of the application, when the amount of data in the historical database is less than the preset amount of data, the method further includes:

[0090] Based on preset physical rules and preset design goals, generate initial guess values;

[0091] Based on a preset data range, the initial guess value is perturbed by parameters to obtain the initial target guess value;

[0092] The initial target guess is input into the parametric suspension model to obtain the output result;

[0093] Based on the output results, the initial target guess values ​​are filtered to determine the target guess values ​​that meet the preset operating conditions;

[0094] The target guess value and the preset design target are used as input features, and the target output result corresponding to the target guess value is used as output label and stored in the historical database.

[0095] Among them, the preset physical rules can refer to the constraints or empirical formulas established based on the basic principles of vehicle dynamics, mechanism kinematics and material mechanics, such as the suspension lever ratio relationship, spring-damping matching criterion, hard point geometric feasibility constraints, etc., to ensure that the generated parameters are physically achievable and conform to common sense in engineering.

[0096] Preset design objectives can refer to performance indicators or boundary conditions explicitly specified by the user in the suspension reverse design task, including but not limited to wheel center motion trajectory, wheel positioning angles (such as toe angle, camber angle), roll center height, tire dynamic load limit or ride comfort threshold, in order to define the desired system behavior.

[0097] Based on preset physical rules and preset design goals, the generated initial guess values ​​can refer to a set of preliminary suspension parameters that characterize the topological features of the suspension, generated through analytical estimation or rule-based reasoning, based on a combination of physical rationality and design intent.

[0098] Parameter perturbation can refer to applying controlled random or structured changes to the initial guess value within a preset data range to generate a diverse set of candidate parameter combinations (i.e., the initial target guess value). In the embodiments of this application, the initial guess value can be perturbed by adding noise using a Gaussian distribution, uniform sampling, Latin hypercube sampling (LHS), or a mutation strategy based on an evolutionary algorithm.

[0099] Parametric suspension models can refer to high-fidelity multibody dynamics simulation systems that take key suspension design parameters as input variables. The model can receive the initial target guess value generated by parameter disturbance as a configuration command, automatically construct the corresponding virtual suspension structure, and perform simulation calculations under standard working conditions (such as specific road surface excitation or steering action), and output the corresponding dynamic or static performance indicators (such as vehicle acceleration, tire dynamic load, wheel alignment angle changes, etc.).

[0100] The target guess value that meets the preset operating conditions can refer to the parameter combination that meets the preset operating conditions (such as vehicle acceleration RMS≤0.6m / s² and tire bounce rate<2%) in key performance indicators (such as comfort, safety and durability) after simulation verification by the parametric suspension model.

[0101] In this embodiment, a set of physically reasonable initial guesses are generated by pre-setting physical rules and design goals; then, these guesses are perturbed within the allowable parameter range to form diverse initial target guesses; next, these guesses are input into a parameterized suspension model for simulation to obtain the corresponding performance output results; then, the results are evaluated and filtered based on preset operating conditions, and target guesses that meet the requirements are retained; finally, these effective parameters and their corresponding design goals are used as input features and target output results are used as output labels, and are structured and stored in a historical database to form a knowledge accumulation closed loop that can be used for machine learning training.

[0102] Therefore, if the amount of data in the historical database is less than the preset amount before training the machine learning model, initial guesses can be generated based on physical rules, and synthetic samples can be generated using parameterized perturbations. Synthetic samples expand the coverage of the training dataset by adjusting the hardpoint coordinates and the boundary values ​​of the target parameters, thus solving the problem of data scarcity in the cold start phase. This ensures that the model can still generate effective initial values ​​when data is insufficient in the early stages, shortening the model training cycle and improving prediction accuracy.

[0103] In this embodiment of the application, the initial target guess value is filtered based on the output result to determine the target guess value that meets the preset operating conditions, including:

[0104] Based on the preset operating conditions, determine the multidimensional satisfaction vector of the output results;

[0105] Perform Pareto sorting on the multidimensional satisfaction vector to determine the non-dominated solutions in the multidimensional satisfaction vector;

[0106] Based on the non-dominated solution, determine the target guess value that satisfies the preset operating conditions.

[0107] The multidimensional satisfaction vector refers to a vector formed by mapping multiple performance indicators (such as vehicle acceleration RMS, tire dynamic load coefficient, suspension travel, etc.) output by the parametric suspension model to continuous scores within the interval [0, 1] according to preset operating conditions (such as comfort threshold and safety limit). Each dimension corresponds to the degree of satisfaction of a design goal, and the closer the value is to 1, the better it meets the operating conditions. Thus, the multidimensional satisfaction vector can transform the traditional hard "meets / does not meet" judgment into a differentiable and comparable soft comprehensive evaluation.

[0108] Pareto sorting can refer to a non-dominated solution selection method in multi-objective optimization, which is used to hierarchically sort candidate solutions in a multi-dimensional satisfaction space: if a solution is better than another solution in at least one satisfaction dimension and not worse than the solution in all other dimensions, then the latter is dominated by the former; by iteratively eliminating dominated solutions, the final set of undominated solutions is the Pareto front.

[0109] Non-dominated solutions in a multidimensional satisfaction vector can refer to the initial target guess value (i.e., suspension parameter combination) corresponding to the multidimensional satisfaction vector, whose satisfaction vector is not dominated by any other satisfaction vector in the Pareto ranking. Although such solutions may not fully meet the hard threshold on a single indicator, they can have irreplaceable trade-off advantages in the comprehensive performance of multiple objectives. Therefore, they can be regarded as high-quality target guess values ​​that meet the preset operating conditions.

[0110] For example, the output result corresponding to the i-th initial target guess value satisfies:

[0111] ;

[0112] Where m represents the performance index.

[0113] The preset operating conditions characterize the threshold values ​​of various performance indicators specified. The preset operating conditions satisfy the following:

[0114] ;

[0115] Therefore, the multidimensional satisfaction vector satisfies:

[0116] ;

[0117] in,

[0118] ;

[0119] in, This is a preset attenuation coefficient used to control the rate of decrease in satisfaction after exceeding the limit.

[0120] The set of multidimensional satisfaction vectors corresponding to all initial target guesses:

[0121] ;

[0122] Perform Pareto sort.

[0123] For any two satisfaction vectors and If the following conditions are met:

[0124] ,and ;

[0125] but Dominate .

[0126] Therefore, it can be determined The Pareto order is determined, and the non-dominated solutions representing the Pareto front in the multidimensional satisfaction vector are identified.

[0127] And based on the non-dominated solution, determine the corresponding target guess value.

[0128] Optionally, the predicted state vector for the new design task is predicted based on the trained machine learning model, and the predicted state vector is used as the initial value for the Newton-Raphson method, including:

[0129] Extract input features from the historical database that correspond to the target input features in the new design task;

[0130] The target input features and input features are fed into the trained machine learning model to obtain the predicted state vector of the machine learning model for the new design task, and the predicted state vector is used as the initial value of the Newton-Raphson method.

[0131] The target input features can refer to the input features given by the user in the new design task. For example: target track width, wheelbase, sprung mass, desired camber angle variation range, target roll center height, vehicle purpose (such as sport or comfort).

[0132] The input features corresponding to the target input features in the new design task can refer to the input variables contained in the existing vehicle sample that is similar to the current new task in terms of design requirements and is selected from the historical database. For example, similar engineering parameters such as vehicle weight, wheelbase, track width, target camber angle range or suspension type.

[0133] The predicted state vector refers to a set of estimated initial design parameters output by a machine learning model based on input features, corresponding to the internal state or key geometric variables of the suspension system. In the suspension design scenario, it can include: the initial coordinates (X, Y, Z) of each hardpoint, spring preload, damper mounting angle, initial positioning angle, etc. This vector is not the final solution, but rather provides a physically reasonable and near-convergent starting point for the subsequent Newton-Raphson method.

[0134] In this embodiment, the system can first retrieve existing cases that are most similar to the target input features of the current new task from the historical database (e.g., by feature distance measurement), and extract their corresponding input features as a reference; then, input the target input features and input features of the new task into the trained machine learning model, and the model predicts the initial state vector; and use the prediction result as the initial guess value of the Newton-Raphson method.

[0135] In this embodiment of the application, when the number of training iterations of the machine learning model is less than a preset number, the machine learning model may still have inaccurate results when used. Therefore:

[0136] The target input features and input features are fed into the trained machine learning model to obtain the predicted state vector of the machine learning model for the new design task. The predicted state vector can also be used as the initial value for the Newton-Raphson method.

[0137] The target input features and input features are fed into the trained machine learning model to obtain the initial prediction state vector of the machine learning model for predicting the new design task.

[0138] Based on the preset feasibility criteria, the credibility of the initial predicted state vector is evaluated;

[0139] If the initial predicted state vector satisfies the feasibility criterion, then the initial predicted state vector is used as the predicted state vector, and the predicted state vector is used as the initial value of the Newton-Raphson method.

[0140] If the initial predicted state vector does not meet the feasibility criterion, the initial predicted state vector is adjusted based on the verified state vector closest to the initial predicted state vector in the historical database, and the adjusted initial predicted state vector is used as the initial value of the Newton-Raphson method.

[0141] Feasibility criteria refer to a set of predefined judgment conditions based on physical laws, engineering constraints, and design specifications, used to evaluate whether a state vector is practically feasible. Examples include: hard point coordinates must not cause mechanism interference; spring preload must be within the material strength range; the outward tilt angle variation curve must be monotonic and not exceed ±2°; and the tilt center height should be within a reasonable range (e.g., 50-200mm). These criteria ensure that the predicted results are not only mathematically valid but also engineering-applicable.

[0142] Credibility assessment refers to the process of verifying the reasonableness of the initial predicted state vector output by a machine learning model. It can be done by checking whether it falls within the range of historical data distribution, whether it meets basic geometric or mechanical constraints, or by using auxiliary discriminant models (such as anomaly detectors) to determine its fidelity.

[0143] If the initial predicted state vector satisfies the feasibility criterion, it means that the parameters such as hard point configuration and preload predicted by the machine learning model, after preliminary verification, meet all the preset engineering and physical constraints. Therefore, it can be trusted as a high-quality starting point and can be used as the initial value of high-precision solvers such as the Newton-Raphson method without correction, so as to accelerate convergence to the final feasible solution.

[0144] If the initial predicted state vector does not meet the feasibility criterion, it means that although the machine learning model provides a prediction result, the result has obvious problems. For example, the hard point position causes the suspension to experience link crossing during the jumping process, the spring force exceeds the reasonable range, or the predicted camber angle changes drastically. This indicates that the initial value may be located in a physically infeasible region. If it is directly used in the Newton-Raphson method, it is very easy to cause iterative divergence or obtain invalid solutions. Therefore, it needs to be corrected.

[0145] The verified state vector closest to the initial predicted state vector in the historical database can refer to the state vector of one or more real and valid samples that are closest to the current infeasible predictable state in an existing suspension design sample library that has been verified by simulation or real vehicles, found by distance metrics (such as Euclidean distance, weighted feature distance, or embedding space similarity).

[0146] Adjusting the initial predicted state vector can refer to correcting an infeasible initial predicted state vector towards the nearest validated state vector. For example, it can be done by using linear interpolation, projecting to the feasible region, or replacing the outlier components in the original prediction with some parameters from neighboring samples, thereby generating a new initial value that retains the model's prediction trend while satisfying basic feasibility.

[0147] For example, the nearest verified state vector can be determined using the weighted Euclidean distance formula, which satisfies the following:

[0148] ;

[0149] in, Characterizing the first The weighted distance between each historical sample and the predicted value; Dimension index (from 1 to 0) representing state variables; Characterizing the prediction of the machine learning model One state component; Characterizing the first In the historical sample, the first The true value of each state component; Characterizing the first The weight coefficients of each dimension.

[0150] Therefore, for key hard points that violate the feasibility criterion constraints, the values ​​of historical neighboring samples can be used as substitutes; the remaining values ​​are retained as predicted values.

[0151] Optionally, there are multiple machine learning models, and different machine learning models have different model types and different weights.

[0152] The trained machine learning model predicts the state vector for the new design task, and uses the predicted state vector as the initial value for the Newton-Raphson method, including:

[0153] Obtain the predicted state vector for each machine learning model in predicting the new design task;

[0154] Based on the weights corresponding to the machine learning model, the predicted state vector is weighted and summed, and the weighted summed predicted state vector is used as the initial value of the Newton-Raphson method.

[0155] The system integrates multiple machine learning models of different types (such as neural networks, gradient boosting trees, Gaussian process regression, etc.). Each model is independently trained based on historical suspension design data and is used to predict the initial state vector (such as hard point coordinates) under the new design task. In the prediction stage, the prediction results of each model for the same input task are first obtained. Then, according to the pre-set model weights (reflecting the accuracy, stability or applicable scenarios of each model on the validation set), these predicted state vectors are weighted and summed to finally generate a comprehensive initial value that integrates the opinions of multiple models. This value is then used as the initial guess value for numerical solvers such as the Newton-Raphson method.

[0156] For example, if the system includes three models:

[0157] MLP neural network (weight 0.5): It excels at fitting complex nonlinear relationships, but is sensitive to small sample sizes;

[0158] XGBoost tree model (weight 0.3): sensitive to engineering rules and stable generalization;

[0159] Gaussian process regression (GPR) with a weight of 0.2: provides an estimate of uncertainty and is more conservative in marginal tasks.

[0160] When predicting the front suspension hard point, the three output the Y coordinates of the inner point of the upper control arm as follows: -310mm, -295mm, and -300mm, respectively.

[0161] We can get the result by weighted summation:

[0162] y=0.5×(−310)+0.3×(−295)+0.2×(−300)=−303.5mm.

[0163] Since a single model may make inaccurate predictions due to overfitting, data bias, or structural limitations, weighted fusion can smooth individual errors and leverage the strengths of each model (such as tree models being good at handling nonlinear rules and neural networks capturing high-dimensional patterns), thereby generating initial values ​​that are closer to the true feasible solution. This not only improves the convergence success rate of the Newton-Raphson method but also reduces the number of iterations and accelerates the entire reverse design process.

[0164] Optionally, after predicting the predicted state vector for the new design task based on the trained machine learning model and using the predicted state vector as the initial value for the Newton-Raphson method, the method further includes:

[0165] Obtain and display the output solution of the Newton-Raphson method obtained from the predicted state vector.

[0166] The process involves obtaining an initial state vector predicted and weighted by a machine learning model, which is then used as an initial value input to the Newton-Raphson method for iterative solution. The state vector is gradually corrected by continuously linearizing the nonlinear kinematic equations of the suspension system (such as hard point geometric constraints and target positioning angle conditions) until the preset convergence accuracy is met. The final output solution is a set of hard point coordinates or system state parameters that accurately satisfy the physical constraints and design objectives. This solution can be directly used for subsequent detailed modeling, simulation verification, or engineering release, and its corresponding suspension structure, bounce trajectory, and four-wheel alignment parameter change curves can be displayed through a visual interface.

[0167] The data processing method for the multibody suspension system provided in this application significantly reduces the number of iterations of the Newton-Raphson method by using high-quality initial values ​​predicted by a machine learning model, thereby significantly improving computational efficiency and reducing solution time from minutes to seconds. Simultaneously, the automatic setting of initial values ​​by the machine learning model effectively replaces the traditional manual parameter tuning process that relies on engineer experience, lowering the technical threshold and automating the design process. Even in the early stages of a project when real data is scarce, the system can robustly handle the "cold start" challenge by combining a physics-based initialization mechanism with synthetic data augmentation strategies. As design samples are continuously fed back for verification during use, the system continuously optimizes the model through data closure, achieving self-evolution in prediction accuracy and generalization ability. Ultimately, because the initial values ​​are closer to the feasible region, the risk of divergence in numerical solutions is significantly reduced, and overall robustness, convergence stability, and engineering success rate are comprehensively improved.

[0168] Figure 3 Flowchart of the data processing method for the multibody suspension system provided in this application Figure 2 ,like Figure 3 As shown, the method includes:

[0169] 1. Historical data accumulation stage:

[0170] Whenever a user successfully completes a suspension design using the Newton-Raphson exact solution module 400, the data sample of that successful case will be automatically extracted and sent to the historical database construction module 100.

[0171] The data sample contains two main elements: input features (i.e., the suspension model parameters and user-defined design objectives) and output labels (i.e., the optimal solution X of the initial state vector obtained by the solver at final convergence).

[0172] Historical Database 100 is responsible for storing, cleaning, and managing these massive amounts of high-quality success samples, providing data fuel for training AI models.

[0173] 2. AI Model Training and Update Phase:

[0174] The AI ​​prediction model training module 200 periodically or triggeredly retrieves training data from the historical database 100.

[0175] This module uses advanced machine learning algorithms (such as gradient boosting trees or neural networks) to train a model with input features X and output labels Y. The goal is to enable the model to learn the complex mapping relationship f: X→Y from "design requirements" to "approximate optimal solution".

[0176] After training, the AI ​​prediction model is generated or updated. This model has the ability to predict high-quality initial values ​​based on new inputs.

[0177] 3. Intelligent prediction and accurate solution stage:

[0178] The process begins when a new user enters new suspension model parameters and design goals.

[0179] Intelligent Initial Value Prediction and Application Module 300: First, it receives these input data and extracts the input features that are consistent with historical data.

[0180] Subsequently, the intelligent initial value prediction and application module 300 calls the pre-trained AI prediction model and inputs these features into the model. The model quickly outputs a predicted initial state vector X_predict.

[0181] The intelligent initial value prediction and application module 300 then uses the prediction vector X_predict as the initial value and passes it to the Newton-Raphson exact solution module 400.

[0182] 4. Physically accurate solution stage:

[0183] The Newton-Raphson exact solver module 400 begins iterative calculations starting with the received intelligent initial value X_predict. Because the starting point is very close to the true solution, the process usually converges quickly, obtaining accurate final design parameters that satisfy all design objectives in a very small number of iterations, and outputting them to the user.

[0184] 5. Data closed loop and self-evolution stage:

[0185] The successful cases solved in this study do not end here. All the data from this successful study (input features and the final exact solution X) will be fed back into the historical database building module 100 as a new sample.

[0186] This new data is then used for incremental training of the AI ​​model or in the next model update.

[0187] This ensures that the system accumulates more and more "experience" and the AI ​​prediction model becomes more and more accurate, thereby enabling the entire system to continuously optimize itself and improve its performance, becoming a truly learning-capable engineering intelligent system.

[0188] Figure 4 A schematic diagram of the data processing device for the multibody suspension system provided in this application is shown below. Figure 4 As shown, the data processing device 40 for the multi-body suspension system provided in this embodiment includes:

[0189] Module 401 is used to build a historical database. The historical database includes input features and output labels corresponding to the input features. The input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method.

[0190] The training module 402 is used to train the machine learning model based on the input features and output labels to obtain the trained machine learning model.

[0191] The prediction module 403 is used to predict the predicted state vector of the new design task based on the trained machine learning model, and uses the predicted state vector as the initial value of the Newton-Raphson method.

[0192] In one possible implementation, the building module 401 can also be specifically used for:

[0193] Automatically extract input features and output labels after completing the suspension design task using the Newton-Raphson method;

[0194] The input features and output labels after completing the suspension design task are stored in the historical database.

[0195] In one possible implementation, when the amount of data in the historical database is less than the preset amount of data, the construction module 401 can also be specifically used for:

[0196] Based on preset physical rules and preset design goals, generate initial guess values;

[0197] Based on a preset data range, the initial guess value is perturbed by parameters to obtain the initial target guess value;

[0198] The initial target guess is input into the parametric suspension model to obtain the output result;

[0199] Based on the output results, the initial target guess values ​​are filtered to determine the target guess values ​​that meet the preset operating conditions;

[0200] The target guess value and the preset design target are used as input features, and the target output result corresponding to the target guess value is used as output label and stored in the historical database.

[0201] In one possible implementation, the building module 401 can also be specifically used for:

[0202] Based on the preset operating conditions, determine the multidimensional satisfaction vector of the output results;

[0203] Perform Pareto sorting on the multidimensional satisfaction vector to determine the non-dominated solutions in the multidimensional satisfaction vector;

[0204] Based on the non-dominated solution, determine the target guess value that satisfies the preset operating conditions.

[0205] In one possible implementation, the training module 402 can also be specifically used for:

[0206] Extract input features from the historical database that correspond to the target input features in the new design task;

[0207] The target input features and input features are fed into the trained machine learning model to obtain the predicted state vector of the machine learning model for the new design task, and the predicted state vector is used as the initial value of the Newton-Raphson method.

[0208] In one possible implementation, when the number of training iterations of the machine learning model is less than a preset number, the training module 402 can also be specifically used for:

[0209] The target input features and input features are fed into the trained machine learning model to obtain the initial prediction state vector of the machine learning model for predicting the new design task.

[0210] Based on the preset feasibility criteria, the credibility of the initial predicted state vector is evaluated;

[0211] If the initial predicted state vector satisfies the feasibility criterion, then the initial predicted state vector is used as the predicted state vector, and the predicted state vector is used as the initial value of the Newton-Raphson method.

[0212] If the initial predicted state vector does not meet the feasibility criterion, the initial predicted state vector is adjusted based on the verified state vector closest to the initial predicted state vector in the historical database, and the adjusted initial predicted state vector is used as the initial value of the Newton-Raphson method.

[0213] In one possible implementation, there are multiple machine learning models, each with different model types and corresponding weights. The prediction module 403 can also be specifically used for:

[0214] Obtain the predicted state vector for each machine learning model in predicting the new design task;

[0215] Based on the weights corresponding to the machine learning model, the predicted state vector is weighted and summed, and the weighted summed predicted state vector is used as the initial value of the Newton-Raphson method.

[0216] In one possible implementation, the prediction module 403 can also be specifically used for:

[0217] Obtain and display the output solution of the Newton-Raphson method obtained from the predicted state vector.

[0218] The data processing device for the multibody suspension system provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0219] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0220] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0221] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0222] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0223] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0224] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0225] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0226] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0227] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0228] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0229] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0230] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0231] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0232] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0233] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0234] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A data processing method for a multi-body suspension system, characterized in that, include: Build a historical database; The historical database includes input features and output labels corresponding to the input features, wherein the input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method; The machine learning model is trained based on the input features and the output labels to obtain the trained machine learning model. The predicted state vector for the new design task is predicted based on the trained machine learning model, and the predicted state vector is used as the initial value for the Newton-Raphson method.

2. The method according to claim 1, characterized in that, The construction of the historical database includes: Automatically extract the input features and output labels after completing the suspension design task using the Newton-Raphson method; The input features and output labels after the suspension design task is completed are stored in the historical database.

3. The method according to claim 2, characterized in that, When the amount of data in the historical database is less than a preset amount of data, the method further includes: Based on preset physical rules and preset design goals, generate initial guess values; Based on a preset data range, the initial guess value is perturbed to obtain the initial target guess value; The initial target guess is input into the parameterized suspension model to obtain the output result; Based on the output results, the initial target guess values ​​are filtered to determine the target guess values ​​that meet the preset operating conditions; The target guess value and the preset design target are used as input features, and the target output result corresponding to the target guess value is used as an output label and stored in the historical database.

4. The method according to claim 3, characterized in that, The step of filtering the initial target guess values ​​based on the output results to determine target guess values ​​that meet preset operating conditions includes: Based on preset operating conditions, determine the multidimensional satisfaction vector of the output result; Perform Pareto sorting on the multidimensional satisfaction vector to determine the non-dominated solutions in the multidimensional satisfaction vector; Based on the non-dominated solution, determine the target guess value that satisfies the preset operating conditions.

5. The method according to claim 1, characterized in that, The method of predicting the predicted state vector of the new design task based on the trained machine learning model, and using the predicted state vector as the initial value of the Newton-Raphson method, includes: Extract input features from the historical database that correspond to the target input features in the new design task; The target input features and the input features are input into the trained machine learning model to obtain the predicted state vector of the machine learning model for predicting the new design task, and the predicted state vector is used as the initial value of the Newton-Raphson method.

6. The method according to claim 5, characterized in that, When the number of training iterations of the machine learning model is less than a preset number, The step of inputting the target input features and the input features into the trained machine learning model to obtain the predicted state vector of the machine learning model for predicting the new design task, and using the predicted state vector as the initial value of the Newton-Raphson method, includes: The target input features and the input features are input into the trained machine learning model to obtain the initial prediction state vector of the machine learning model for predicting the new design task; Based on preset feasibility criteria, the credibility of the initial predicted state vector is evaluated. If the initial predicted state vector satisfies the feasibility criterion, then the initial predicted state vector is used as the predicted state vector, and the predicted state vector is used as the initial value of the Newton-Raphson method. If the initial predicted state vector does not meet the feasibility criterion, the initial predicted state vector is adjusted according to the verified state vector closest to the initial predicted state vector in the historical database, and the adjusted initial predicted state vector is used as the initial value of the Newton-Raphson method.

7. The method according to claim 1, characterized in that, There are multiple machine learning models, and the different machine learning models have different model types and different weights. The method of predicting the predicted state vector of the new design task based on the trained machine learning model, and using the predicted state vector as the initial value of the Newton-Raphson method, includes: Obtain the predicted state vector for each of the machine learning models in predicting the new design task; Based on the weights corresponding to the machine learning model, the predicted state vector is weighted and summed, and the weighted summed predicted state vector is used as the initial value of the Newton-Raphson method.

8. The method according to any one of claims 1 to 7, characterized in that, After predicting the predicted state vector of the new design task based on the trained machine learning model and using the predicted state vector as the initial value for the Newton-Raphson method, the method further includes: Obtain and display the output solution of the Newton-Raphson method obtained based on the predicted state vector.

9. A data processing device for a multi-body suspension system, characterized in that, include: Modules for building historical databases; The historical database includes input features and output labels corresponding to the input features, wherein the input features include the topological parameters and design objectives of the suspension model, and the output labels include the initial state vector obtained by convergence solution according to the Newton-Raphson method; The training module is used to train the machine learning model based on the input features and the output labels to obtain the trained machine learning model. The prediction module is used to predict the predicted state vector of the new design task based on the trained machine learning model, and to use the predicted state vector as the initial value of the Newton-Raphson method.

10. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-8.

Images