Automobile hub structure optimization method based on parametric modeling and neural network integration

Abstract

The invention provides an automobile hub structure optimization method based on parametric modeling and neural network integration, comprising the following steps: selecting an initial hub model as an optimization reference, constructing a hub parametric modeling platform, and defining design variables; constructing a'structure-performance 'data set; constructing a weighted integration agent model based on a particle swarm optimization algorithm; and carrying out multi-objective optimization on the structural parameters of the hub by using an NSGA-II algorithm. The method has the beneficial effects that the parameterized automatic modeling, the finite element analysis, the integrated agent model based on intelligent algorithm weighting and the multi-objective optimization algorithm are systematically fused, and a complete and automatic'design-simulation-prediction-optimization 'intelligent optimization technical route is constructed; seamless connection of geometric model adjustment, performance simulation verification, accurate performance prediction and multi-target collaborative optimization on the basis of an initial hub is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of new energy vehicle parts design and intelligent optimization, in particular to a vehicle hub structure optimization method based on parameterized modeling and neural network integration. BACKGROUND

[0002] In the field of new energy vehicles, the hub as a key load-bearing component, its design needs to meet multiple extreme performance requirements such as "lightweight, high strength, high modal frequency (NVH performance)". However, the current hub structure design mainly relies on the following traditional methods, and the existing technology has the following significant problems: (1) Experience dependence and trial and error cycle: The design process highly depends on the personal experience of engineers, and iterates through the physical trial and error cycle of "design-simulation-manufacturing-testing". The cycle is long and costly, and it is difficult to systematically explore the complex design space.

[0003] (2) Local optimization and low efficiency: Traditional optimization methods often conflict with each other in the optimization of multiple goals such as lightweight, safety and comfort of the hub, and are easily trapped in local optimal solution, and each iteration needs to call time-consuming finite element analysis, the calculation efficiency is extremely low, and it is impossible to realize fast global optimization.

[0004] (3) Unclear "structure-performance" mapping relationship: There is a complex nonlinear coupling relationship between hub structure parameters and multiple performance indicators. Traditional methods lack efficient and accurate quantitative prediction models, and when the training data has noise or sample distribution is uneven, the model performance will decrease significantly. SUMMARY

[0005] Therefore, the application aims to provide a vehicle hub structure optimization method based on parameterized modeling and neural network integration to solve at least one of the above problems existing in the prior art.

[0006] To achieve the above purpose, the technical scheme of the application is as follows: A vehicle hub structure optimization method based on parameterized modeling and neural network integration, comprising the following steps: S1, constructing a hub parameterized modeling platform and defining design variables; S2, constructing a "structure-performance" dataset based on the hub parameterized modeling platform and the design variables; S3, constructing a weighted integrated proxy model based on particle swarm optimization algorithm based on the "structure-performance" dataset and multiple different base learners; S4, performing multi-objective optimization of hub structure parameters based on the weighted integrated proxy model and using NSGA-II algorithm; In step S1, the design variables include: hub radius, hub width, rim thickness, rim width, spoke thickness, array hole number, hollow length and hollow width.

[0007] Further, in step S2, based on the hub parameterization modeling platform and the design variables, a "structure-performance" dataset is constructed, including: S21, selecting from the defined design variables, based on the selected design variables, designing a sample space to obtain multiple sets of structure parameter combinations; S22, based on the multiple sets of structure parameter combinations, using the hub parameterization modeling platform to obtain multiple hub structure models; S23, respectively performing performance simulation on each hub structure model to construct a "structure-performance" dataset.

[0008] Further, in step S3, based on the "structure-performance" dataset and multiple different base learners, a weighted ensemble proxy model based on the particle swarm optimization algorithm is constructed, including: S31, selecting multiple different base learners, independently training the "structure-performance" dataset to construct multiple "structure-performance" prediction models; S32, based on the multiple "structure-performance" prediction models and the particle swarm optimization algorithm, developing a weighted ensemble strategy; S33, based on the weighted ensemble strategy, constructing a weighted ensemble proxy model.

[0009] Further, in step S4, based on the weighted ensemble proxy model, using the NSGA-II algorithm to perform hub structure parameter multi-objective optimization, including: S41, using the weighted ensemble proxy model to drive the NSGA-II algorithm to perform multi-objective optimization to obtain a Pareto optimal solution set; S42, based on the Pareto optimal solution set, designing an optimal scheme; S43, using finite element simulation to verify the effectiveness of the optimal scheme.

[0010] Further, in step S21, the selected design variables include: spoke thickness, hollow length and hollow width.

[0011] Further, in step S23, the performance simulation includes: Performing radial fatigue tests, bending fatigue tests, 13° impact tests and modal analysis; Recording target parameters, including maximum stress, hub lowest first-order natural frequency and hub mass.

[0012] Further, in step S31, the multiple different base learners include: There are three base learners with different principles: BP, RBF, and BNN.

[0013] Furthermore, in step S32, a weighted integration strategy is formulated, including: With the goal of minimizing the overall prediction error of the ensemble surrogate model, the weight allocation of the base learners is dynamically optimized to construct a weighted ensemble model with the optimal weights.

[0014] Furthermore, the base learner can be a neural network, a Gaussian process regression, or an extreme learning machine; the particle swarm optimization algorithm can be replaced by the differential evolution algorithm, the simulated annealing algorithm, or the gradient-based optimization algorithm; and the NSGA-II algorithm can be replaced by the NSGA-III algorithm or the MOEA / D algorithm.

[0015] Compared with existing technologies, the automobile wheel hub structure optimization method based on parametric modeling and neural network integration described in this invention has the following advantages: (1) This invention systematically integrates parametric automatic modeling, finite element analysis, integrated proxy model based on intelligent algorithm weighting and multi-objective optimization algorithm to construct a complete and automated "design-simulation-prediction-optimization" intelligent optimization technology route, realizing the seamless connection of wheel hub from geometric model generation, performance simulation verification, accurate performance prediction to multi-objective collaborative optimization.

[0016] (2) This invention proposes a new method for constructing a high-precision, high-fit-goodness, and high-robust surrogate model. In order to solve the prediction problem of the “structure-performance” mapping of complex engineering structures, three base learners, BP, RBF and BNN, are used for preliminary prediction. The PSO algorithm is innovatively used to globally optimize the output weights of the base learners to obtain an integrated prediction model with significantly better prediction accuracy, fit-goodness and robustness than any single model.

[0017] (3) This invention uses the PSO weighted ensemble surrogate model as a fast evaluator and combines it with the NSGA-II multi-objective genetic algorithm to provide an efficient global optimization method for automobile wheel hub structures, thus protecting the specific application of this particular “prediction-optimization” coupling mode in multi-objective optimization of wheel hub lightweighting. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the technical route described in the embodiments of the present invention; Figure 2 This is a schematic diagram of the main design parameters of the wheel hub according to an embodiment of the present invention; Figure 3 This is a tree-like diagram of the modeling code architecture described in an embodiment of the present invention; Figure 4 This is a schematic diagram of the wheel hub parametric automatic modeling platform described in an embodiment of the present invention; Figure 5 This is a schematic diagram of the finite element model and stress cloud of the radial fatigue test of the wheel hub according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the finite element model and stress cloud of the wheel hub bending fatigue test according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the finite element model and stress cloud of the 13° impact test of the wheel hub according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the modal array and natural frequency cloud of the hub elastomer as described in an embodiment of the present invention; Figure 9 This is a schematic diagram of the training loss curve of the ensemble model described in an embodiment of the present invention; Figure 10 This is a scatter plot of the actual and predicted values ​​of the integrated model described in this embodiment of the invention. Figure 11 This is a schematic diagram comparing the MSE of the integrated model and the single model under different noise levels according to an embodiment of the present invention; Figure 12 This is a schematic diagram of the training loss curves of the ensemble model under different noise levels as described in an embodiment of the present invention; Figure 13 This is a schematic diagram comparing the R² values ​​of the integrated model and the single model under different noise levels according to an embodiment of the present invention. Figure 14 This is a schematic diagram of the scatter plot of actual values ​​versus predicted values ​​of the integrated model under different noise levels as described in the embodiments of the present invention; Figure 15 This is a schematic diagram of the Pareto optimal solution set for multi-objective optimization according to an embodiment of the present invention; Figure 16 This is a schematic diagram of the optimal model performance cloud map according to an embodiment of the present invention. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0020] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0021] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] like Figures 1 to 16 As shown, a method for optimizing automobile wheel hub structure based on parametric modeling and neural network integration is presented.

[0024] like Figure 1 The technical route is illustrated below: 1. Basic verification and parametric modeling include: initial model performance simulation verification, construction of parametric design platform, and definition of design variables.

[0025] 2. Sample generation and dataset construction include: using Latin hypercube sampling to generate 80 sets of experimental samples, parameterized modeling and performance simulation of the samples, and construction of a "structure-performance" dataset.

[0026] 3. Training and validation of the predictive performance of the agent model includes: training and comparing single agent models, and constructing an integrated agent model with PSO optimization and weighting.

[0027] 4. Multi-objective optimization solution includes: defining optimization objectives and constraints, obtaining the optimal solution set through NSGA-II multi-objective optimization, and verifying the performance of the optimal solution.

[0028] The specific implementation method is as follows: Example 1: Step 1: Rapid generation of wheel hub model based on parametric automatic modeling platform For a specific new energy vehicle wheel hub, a parametric automatic modeling platform based on NX secondary development was built. A user-friendly interface was implemented using MenuScript and Block UI Styler. Through a core modeling algorithm based on the NXOpen API, the platform enables rapid generation and driving of wheel hub models with parameters such as hub radius, hub width, rim thickness, flange width, spoke thickness, number of array holes, cutout length, and cutout width. By inputting the design variable parameters, the corresponding wheel hub model can be quickly generated, completely changing the inefficient manual model modification method. This lays a solid foundation for subsequent large-scale parametric sample generation, finite element analysis, performance prediction, and multi-objective optimization.

[0029] Step 2: Constructing a "Structure-Performance" Dataset Based on Finite Element Simulation Three spoke dimensions with the most significant impact on the wheel hub—spoke thickness (X1), cutout length (X2), and cutout width (X3)—were selected as design variables. Within the range of these design variables, a sample space design was performed using the Latin hypercube multi-parameter sample setting method, resulting in 80 combinations of structural parameters. The "Wheel Hub Parametric Automatic Modeling Platform" from step one was then used to quickly obtain 80 wheel hub structural models. Radial fatigue tests, bending fatigue tests, 13° impact tests, and modal analysis were conducted on each wheel hub model. The target parameters for the maximum stress (Y1), the lowest first-order natural frequency (Y2), and the wheel hub mass (Y3) during multi-condition simulations of each model were recorded, constructing a "structure-performance" dataset for the wheel hub.

[0030] Step 3: Construction of a high-precision surrogate model based on a weighted neural network ensemble using the Particle Swarm Optimization (PSO) algorithm Three base learners with different principles—BP, RBF, and BNN—were selected and trained independently on the dataset to establish several preliminary "structure-performance" prediction models.

[0031] To address the shortcomings of individual models, a weighted ensemble strategy based on Particle Swarm Optimization (PSO) is proposed. With the goal of minimizing the overall prediction error of the ensemble surrogate model, the weight allocation of the base learners is dynamically optimized, constructing a system with optimal weights. The weighted ensemble model [0.31925717, 0.3428353, 0.33790758] has a mean squared error (MSE) as low as 0.0017 and a coefficient of determination (R²) of 0.9728. This makes the weighted ensemble model superior to the single model in terms of prediction accuracy, goodness of fit, and robustness on the validation set, achieving a significant leap in prediction performance.

[0032] Step 4: Multi-objective optimization of wheel hub structure parameters based on NSGA-II algorithm The optimization objectives were clearly defined as lightweight, low stress, and high frequency. Considering the strength constraints of the materials, the high-precision PSO weighted ensemble surrogate model trained in step three was used to drive the NSGA-II algorithm for efficient multi-objective optimization. The trade-offs between different objectives were explored, and a high-quality Pareto optimal solution set was obtained. The optimal parameters were: spoke thickness 21.69 mm, cutout length 120.23 mm, and cutout width 27.84 mm. Finally, the effectiveness of the optimal solution was verified by finite element simulation. The results show that the solution reduces the maximum stress of the wheel hub by 12.43%, increases the first-order natural frequency by 2.97%, and reduces the mass by 7.33%, effectively balancing the requirements of lightweighting, safety, and comfort, improving the overall performance of the wheel hub, and the prediction error of the surrogate model is controlled within 7%.

[0033] It should be noted that in step three, other machine learning models can be selected as the base learner, such as other neural networks, Gaussian process regression, extreme learning machines, etc., as long as their principles are different, they can all be used as the basis for ensemble. The PSO algorithm can also be replaced by other metaheuristic optimization algorithms, such as differential evolution algorithm, simulated annealing algorithm, or gradient-based optimization algorithm to complete the weight optimization, which can also achieve the technical effect of optimizing the weight of the ensemble model. In step four, the NSGA-II algorithm can also be replaced by other multi-objective evolutionary algorithms, such as NSGA-III, MOEA / D, etc., which can also achieve multi-objective optimization based on surrogate models.

[0034] The innovative aspects and beneficial effects of this invention are as follows: (1) Methodological Innovation: The core innovation of this invention lies in the systematic integration of parametric automatic modeling, finite element analysis, integrated proxy model based on intelligent algorithm weighting, and multi-objective optimization algorithm to construct a complete and automated intelligent optimization technology route of "design-simulation-prediction-optimization". This method realizes the seamless connection of wheel hub from geometric model generation, performance simulation verification, accurate performance prediction to multi-objective collaborative optimization.

[0035] (2) Core Model Innovation: This invention proposes a new method for constructing a surrogate model with high accuracy, high goodness of fit, and high robustness. To solve the prediction problem of the "structure-performance" mapping of complex engineering structures, three base learners, BP, RBF, and BNN, are used for preliminary prediction. The PSO algorithm is innovatively used to globally optimize the output weights of the base learners to obtain an integrated prediction model with significantly better prediction accuracy, goodness of fit, and robustness than any single model.

[0036] (3) System Integration Innovation: This invention uses the PSO weighted ensemble surrogate model as a fast evaluator, combined with the NSGA-II multi-objective genetic algorithm, to provide an efficient global optimization method for automotive wheel hub structures. It protects the specific application of this particular "prediction-optimization" coupling mode in multi-objective optimization for lightweight wheel hubs.

[0037] The technical problem solved by this invention: (1) It solves the problems of low design efficiency and long R&D cycle. The parametric modeling platform based on NX secondary development realizes one-click automatic generation from design variables to three-dimensional models, completely changing the inefficient mode of manually modifying models. The constructed high-precision integrated proxy model replaces the computationally expensive finite element simulation in the optimization stage, enabling thousands of performance evaluations to be completed instantly, solving the bottleneck of simulation calculation time.

[0038] (2) The problem of difficulty in coordinating multiple objectives in the wheel hub optimization process is solved. The NSGA-II multi-objective genetic algorithm can effectively search the entire design space and obtain a uniformly distributed Pareto optimal solution set. The final solution does not sacrifice one performance to improve another, but finds an optimal point for coordinated improvement.

[0039] (3) It solves the problem of insufficient intelligence. The entire process, from sample generation and model training to parameter optimization, is driven by the system and algorithms, which greatly reduces the reliance on human subjective experience. It forms a complete "design-simulation-prediction-optimization-verification" closed loop, ensuring the reliability of the results.

[0040] (4) The poor accuracy, goodness of fit and robustness of the surrogate model were solved. An integrated surrogate model based on PSO weighting was proposed. By combining the advantages of three neural networks, BP, RBF and BNN, the limitations of the single model were overcome and the prediction accuracy, goodness of fit and robustness were significantly improved.

[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing the structure of an automobile wheel hub based on the integration of parameterized modeling and neural networks, characterized in that it comprises the following steps: The method comprises the following steps: ​ S1, constructing a hub parameterized modeling platform and defining design variables; S2, constructing a "structure-performance" dataset based on the hub parameterized modeling platform and the design variables; S3, constructing a weighted ensemble proxy model based on a particle swarm optimization algorithm based on the "structure-performance" dataset and multiple different base learners; S4, performing multi-objective optimization of hub structure parameters based on the weighted ensemble proxy model using an NSGA-II algorithm; In step S1, the design variables include: hub radius, hub width, rim thickness, rim width, spoke thickness, array hole number, hollow length, and hollow width.

2. The method of claim 1, wherein: In step S2, the "structure-performance" dataset is constructed based on the hub parameterized modeling platform and the design variables, including: S21, selecting from the defined design variables, designing a sample space based on the selected design variables, and obtaining multiple sets of structure parameter combinations; S22, obtaining multiple hub structure models based on the multiple sets of structure parameter combinations using the hub parameterized modeling platform; S23, performing performance simulation on each hub structure model to construct a "structure-performance" dataset.

3. The method of claim 1, wherein: In step S3, the weighted ensemble proxy model based on the particle swarm optimization algorithm is constructed based on the "structure-performance" dataset and multiple different base learners, including: S31, selecting multiple different base learners, independently training the "structure-performance" dataset, and constructing multiple "structure-performance" prediction models; S32, developing a weighted ensemble strategy based on the multiple "structure-performance" prediction models and the particle swarm optimization algorithm; S33, constructing a weighted ensemble proxy model based on the weighted ensemble strategy.

4. The method of claim 1, wherein: In step S4, the multi-objective optimization of hub structure parameters is performed based on the weighted ensemble proxy model using the NSGA-II algorithm, including: S41, using the weighted ensemble proxy model to drive the NSGA-II algorithm for multi-objective optimization to obtain a Pareto optimal solution set; S42, designing an optimal scheme based on the Pareto optimal solution set; S43, verifying the effectiveness of the optimal scheme using finite element simulation.

5. The method of claim 2, wherein: In step S21, the selected design variables include: spoke thickness, hollow length, and hollow width.

6. The method of claim 2, wherein: In step S23, the performance simulation includes: Performing radial fatigue tests, bending fatigue tests, 13° impact tests, and modal analysis; Recording target parameters, including maximum stress, hub lowest first-order natural frequency, and hub mass.

7. The method of claim 3, wherein: In step S31, the base learners include: BP, RBF, and BNN, which are three different base learners.

8. The method of claim 3, wherein: In step S32, the weighted ensemble strategy is developed, including: Optimizing the weight distribution of the base learners dynamically to construct an optimal weight weighted ensemble model with the minimum overall prediction error of the ensemble proxy model.

9. The method of claim 3 or 4 or 7, wherein: The types of base learners can be neural networks, Gaussian process regression, or extreme learning machines; the particle swarm optimization algorithm can be replaced by differential evolution algorithm, simulated annealing algorithm, or gradient-based optimization algorithm; the NSGA-II algorithm can be replaced by NSGA-III or MOEA / D.

Images