Automobile wind resistance optimization design method based on machine learning model

Abstract

The invention relates to the technical field of automobile design, and particularly discloses an automobile wind resistance optimization design method based on a machine learning model, and the method comprises the steps: collecting data of different automobile types, and carrying out the cleaning and standardization processing; constructing a wind resistance prediction model, training the model by using the preprocessed data, and minimizing a prediction error by adjusting model parameters; key design variables are selected, the variation range of the key design variables is determined, and a parameterized model is constructed; selecting sample points in a design variable space, generating a corresponding geometric model, calculating a wind resistance coefficient and establishing a sample point database; dividing sample point data into a training set and a test set, optimizing a machine learning model, searching an optimal solution in combination with an optimization algorithm, and iteratively updating the model by increasing sample points until the precision requirement is met; an optimal design variable combination is determined, an optimized automobile geometric model is generated, CFD simulation verification is carried out, and a design scheme containing detailed parameters and optimization results is output.

Description

Technical Field

[0001] This invention relates to the field of automotive design technology, and more specifically to a method for optimizing automotive aerodynamics based on a machine learning model. Background Technology

[0002] In the automotive industry, wind resistance has a crucial impact on vehicle performance. Data shows that the energy consumed to overcome aerodynamic drag during vehicle operation accounts for approximately 20% of the total energy consumption. Reducing the drag coefficient not only improves fuel economy but also significantly increases the driving range for electric vehicles. Furthermore, it helps reduce wind noise, enhancing cabin quietness and ride comfort.

[0003] Traditionally, reducing the drag coefficient of automobiles has primarily relied on wind tunnel testing. However, wind tunnel testing resources are scarce, each test is costly, and the entire R&D cycle is lengthy, making it difficult to meet the needs of modern automobile's rapid iterative development. With the development of computer technology, CFD (Computational Fluid Dynamics) simulation technology has gradually become an important tool in the automotive development and design process, which can shorten the R&D cycle and reduce R&D costs to some extent. However, optimizing the drag coefficient of automobiles involves numerous design variables, and these design variables have complex coupling relationships, and their impact on the drag coefficient is not a simple linear superposition. Traditional CFD simulation methods heavily rely on the designer's engineering experience, often searching for an optimization solution through extensive trial and error. This approach is inefficient and rarely yields a globally optimal solution directly.

[0004] In recent years, surrogate models and optimization algorithms have been increasingly applied to the optimization of automotive drag coefficients due to their ability to fit data patterns and search for optimal solutions. Surrogate models can be categorized into static surrogate models and adaptive surrogate models based on their construction strategies. Static surrogate models require acquiring a large number of sample points at once during construction to ensure model accuracy, but this results in low optimization efficiency. While adaptive surrogate models can update the model by gradually increasing the number of sample points, requiring fewer points for the same accuracy requirements and thus improving optimization efficiency, they still suffer from low efficiency, complex operation, and long processing times when dealing with the complex flow field structure of automobiles. Meanwhile, utilizing AI models for vehicle external flow field prediction and aerodynamic shape optimization is gradually becoming an innovative approach in vehicle design, but related technologies still need further improvement and refinement to better meet the rapidly evolving needs of the automotive industry. Summary of the Invention

[0005] The purpose of this invention is to provide a vehicle drag optimization design method based on a machine learning model, so as to improve the efficiency and accuracy of vehicle drag optimization, reduce R&D costs and time, and quickly find a better vehicle drag optimization solution.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A vehicle drag optimization design method based on a machine learning model includes the following steps:

[0008] Data Acquisition and Preparation: Collect geometric parameters, vehicle operating status data, and corresponding drag coefficient data for different vehicle models, and clean and standardize the data.

[0009] Machine learning model construction: Select a machine learning algorithm to build a wind resistance prediction model, train the model using preprocessed data, and minimize the prediction error by adjusting the model parameters;

[0010] Design variable selection and parametric model construction: Based on automotive aerodynamics principles and engineering experience, key design variables are selected, their range of variation is determined, and parametric models are constructed using mesh deformation software or parametric modeling tools;

[0011] Sample point selection and wind resistance calculation: Sample points are selected in the design variable space using experimental design method, corresponding geometric models are generated based on parametric models, wind resistance coefficients are calculated through CFD simulation, and a sample point database is established;

[0012] Model optimization and iteration: The sample point data is divided into training set and test set, the machine learning model is optimized, the optimal solution is searched by combining optimization algorithm, and the model is iteratively updated by adding sample points until the accuracy requirements are met;

[0013] Determining and Outputting the Optimization Scheme: Determine the optimal combination of design variables, generate the optimized vehicle geometric model, verify it through CFD simulation, and output the design scheme containing detailed parameters and optimization results.

[0014] As a further aspect of the present invention: in the data acquisition and preparation, the geometric shape parameters include vehicle length, width, height, body line curvature, front shape parameters, rear shape parameters, rearview mirror shape and position parameters, and wheel and wheel arch parameters; vehicle operating status data includes speed, acceleration, and driving posture; the drag coefficient data comes from wind tunnel tests or high-precision CFD simulations.

[0015] As a further aspect of the present invention: in the construction of the machine learning model, the machine learning algorithm is selected from deep neural networks (DNN), convolutional neural networks (CNN) or recurrent neural networks (RNN); the training process uses the mean squared error loss function, and the optimization algorithm is selected from the stochastic gradient descent algorithm and its variants Adagrad, Adadelta or Adam.

[0016] As a further aspect of the present invention: the specific steps in selecting the design variables are as follows:

[0017] Based on aerodynamic theory, a pool of potential variables was initially screened, covering basic vehicle body parameters, surface features, aerodynamic accessories and detailed structures;

[0018] The global sensitivity analysis algorithm was used to calculate the main effect value and total effect value of the variables, and highly sensitive variables with a main effect value > 5% and a total effect value > 10% were selected.

[0019] Redundant variables with coupling coefficients greater than 0.8 are eliminated by calculating mutual information entropy.

[0020] The number of core design variables is determined by combining production process constraints, and the dynamic range of change is set by interval division method.

[0021] As a further aspect of the present invention: in the construction of the parameterized model, adaptive density control points are set up in key aerodynamic regions, and a mapping relationship between control points and geometric features is constructed through a non-uniform B-spline surface interpolation algorithm to form a topological correlation matrix; flow field constraints are embedded, and when the adjustment of design variables causes the local flow field to show signs of separation, the control point compensation mechanism is automatically triggered; and the mesh smoothing algorithm ensures that the mesh distortion rate after deformation is <5%.

[0022] As a further aspect of the present invention: in the selection of sample points:

[0023] Based on the design variable sensitivity weight stratified sampling, enhanced Latin hypercube sampling is used for highly sensitive variables, standard Latin hypercube sampling is used for moderately sensitive variables, and uniform sampling is used for low-sensitivity variables.

[0024] A space filling rate monitoring index is introduced. When the ratio of the minimum Euclidean distance to the average distance between sample points is less than 0.3, additional sample points are added. Combined with the cross-validation error of the machine learning model, if the mean square error of the 5-fold cross-validation is greater than 5%, additional samples are added in the variable interval with the largest prediction error.

[0025] The initial sample points were quickly calculated using a simplified flow field model. Potential samples with drag coefficients lower than 15% of the initial model were then selected. A high-precision full flow field model was then used for detailed simulation. An error correction model was constructed based on the deviation between wind tunnel test data and CFD results.

[0026] As a further aspect of the present invention: in the model optimization and iteration, the optimization algorithm is selected from genetic algorithm or particle swarm optimization algorithm; the number of sample points is increased in the following ways:

[0027] Select the optimal solution with the largest minimum distance from the sample point and its corresponding drag coefficient as the first addition point;

[0028] Select the remaining optimal solution with the largest absolute value of the difference between the drag coefficient and the drag coefficient of the sample point, and the corresponding drag coefficient, as the second addition point.

[0029] Add point 1 and point 2 to the sample point database, retrain the model and iterate until the error between the model's predicted drag coefficient and the CFD simulation calculated drag coefficient is within a preset range, or the drag coefficient improvement is less than a threshold after multiple iterations of the optimization algorithm.

[0030] As a further aspect of the present invention: the output of the optimization scheme includes vehicle geometric parameters, drag coefficient optimization results, optimization process data and analysis reports, which are applicable to the overall drag optimization of vehicle shape and the drag optimization design of local components such as rearview mirrors and spoilers.

[0031] The beneficial effects of this invention are as follows: By constructing a machine learning model, the drag coefficient of different car design schemes can be predicted quickly and accurately. Compared with traditional CFD simulation methods, this significantly shortens the calculation time and improves the efficiency of drag optimization. In some cases, using a trained machine learning model for drag prediction can be several times or even tens of times faster than traditional CFD simulation, enabling rapid evaluation of a large number of design schemes in the early stages of car design and providing timely feedback to designers.

[0032] By combining optimization algorithms with machine learning models, the approach can more efficiently search for the global optimum in a complex design variable space, avoiding the blindness of traditional trial-and-error methods. Through continuous iterative optimization, a superior aerodynamic optimization scheme for automobiles can be found compared to traditional methods, effectively reducing the drag coefficient.

[0033] This invention has good versatility and scalability. It is not only applicable to the overall aerodynamic optimization of automobiles, but also to the aerodynamic optimization design of local components of automobiles (such as rearview mirrors, spoilers, etc.). Attached Figure Description

[0034] The invention will now be further described with reference to the accompanying drawings.

[0035] Figure 1 This is a flowchart illustrating a vehicle drag optimization design method based on a machine learning model, according to the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] See also Figure 1 As shown, this invention is a vehicle drag optimization design method based on a machine learning model, comprising the following steps:

[0038] Data Acquisition and Preparation: A large amount of vehicle data from different models was collected, including geometric parameters (such as body length, width, height, body line curvature, front and rear shape parameters, rearview mirror shape and position parameters, and wheel and wheel arch parameters), vehicle operating status data (speed, acceleration, driving posture, etc.), and drag coefficient data obtained from wind tunnel tests or high-precision CFD simulations. The collected data was cleaned to remove outliers and errors, and standardized to ensure that data for different parameters were on the same scale, facilitating subsequent model training.

[0039] Machine learning model construction: Select appropriate machine learning algorithms, such as deep neural networks (DNN), convolutional neural networks (CNN), or recurrent neural networks (RNN), to construct a wind resistance prediction model. Taking deep neural networks as an example, the model structure may include multiple hidden layers. The input layer receives processed vehicle geometry parameters and operating status data. The hidden layers extract and learn data features, and the output layer outputs the predicted wind resistance coefficient.

[0040] The model is trained using a prepared dataset. By continuously adjusting the model's weights and biases, the error between the model's predictions and the actual drag coefficient data is minimized. During training, appropriate loss functions (such as mean squared error loss function) and optimization algorithms (such as stochastic gradient descent and its variants Adagrad, Adadelta, Adam, etc.) are employed to improve the model's training efficiency and convergence speed.

[0041] Design variable selection and parametric model construction: Based on automotive aerodynamics principles and engineering experience, analyze the main factors affecting vehicle drag, select key design variables such as body roll angle, windshield tilt angle, body underside flatness, spoiler size and shape, and determine the range of variation for each design variable.

[0042] Based on an initial vehicle model, a parametric model is constructed by associating selected design variables with corresponding geometric features in the model using mesh deformation software or parametric modeling tools. For example, in mesh deformation software, control volumes are created in the regions corresponding to design variables. By changing the values ​​of the design variables, the control points are moved, thereby achieving mesh deformation and obtaining vehicle mesh models with different combinations of design variables.

[0043] In this embodiment of the invention, a three-dimensional evaluation system of "sensitivity-coupling-engineering feasibility" is constructed, breaking through the traditional single-dimensional screening method that relies on experience. First, based on automotive aerodynamics theory, a pool of potential design variables related to wind resistance is initially screened, covering basic vehicle body parameters (such as the length / width / height ratio of the vehicle and wheelbase), surface features (such as the curvature of the hood, the slope of the roof arc, and the angle of the side waistline), aerodynamic accessories (such as the ground clearance of the front lip, the angle of attack of the rear diffuser, and the angle of attack of the rear wing), and detailed structural parameters (such as the depth of the door handle recess and the opening ratio of the wheel rim). Secondly, the latent variables are quantitatively evaluated using a global sensitivity analysis algorithm (such as the Sobol sequence method). The main effect value and total effect value of each variable on the drag coefficient are calculated, and highly sensitive variables with a main effect value > 5% and a total effect value > 10% are screened out. Then, the coupling strength between variables is analyzed by calculating mutual information entropy, and redundant variables with a coupling coefficient > 0.8 are eliminated (such as when there is a strong coupling between the roof arc slope and the rear window tilt angle, the latter with a clearer physical meaning is retained). Finally, combined with production process constraints (such as the feasibility of stamping dies and assembly clearance requirements), the final 8-12 core design variables are determined, and their dynamic change range is set by the interval division method (for example, the windshield tilt angle is set to 28°-42°, with a step accuracy of 0.5°).

[0044] Furthermore, unlike the fixed-dimensional mesh deformation methods in existing technologies, this invention, based on the initial CAD model of the vehicle, deploys adaptive density control points in key aerodynamic areas (such as the front windward surface, A-pillar corners, and rear contraction areas); and densifies the control points (up to 5 points / cm) in areas with drastic curvature changes (such as the A-pillar). 2 In flat areas (such as the middle of the roof), sparse control points (density of 1 point / 10cm) are used. 2 The system constructs a mapping relationship between control points and geometric features using a non-uniform B-spline (NURBS) surface interpolation algorithm, forming a topological correlation matrix. Matrix elements record the influence weights of control point displacements on vehicle surface parameters (such as normal vectors and radius of curvature). Simultaneously, flow field constraints are embedded: when adjustments to design variables lead to early signs of separation in the local flow field (e.g., a negative pressure zone area > 5% identified through CFD pre-simulation), a control point compensation mechanism is automatically triggered (e.g., increasing control point displacement by 5% in the rear diffuser region). Ultimately, real-time linkage between design variables and the geometric model is achieved—when any variable (e.g., rear wing angle of attack) changes, the system automatically updates the coordinates of associated control points through matrix operations and calls a mesh smoothing algorithm (e.g., Laplacian smoothing) to ensure mesh quality after deformation (distortion rate < 5%), generating a mesh model that meets aerodynamic simulation requirements without manual intervention.

[0045] Sample point selection and drag calculation: Using appropriate experimental design methods, such as Latin hypercube sampling (LHS) or orthogonal experimental design, a certain number of sample points are selected within the range of design variable variations. These sample points should cover the design variable space as evenly as possible to ensure the comprehensiveness of subsequent model training and optimization. For each sample point, a corresponding vehicle geometry model is generated based on the constructed parametric model, and the drag coefficient of this model is calculated using CFD simulation software. The design variable values ​​of the sample points and their corresponding drag coefficient calculation results are stored in a sample point database as the basic data for subsequent model training and optimization.

[0046] In this embodiment of the invention, firstly, the design variables are stratified based on their sensitivity weights (derived from global sensitivity analysis): highly sensitive variables (main effect value > 10%) are sampled using enhanced Latin hypercube sampling, generating dense samples within their value range (sampling density is 2-3 times that of low-sensitivity variables); moderately sensitive variables (main effect value 5%-10%) are sampled using standard Latin hypercube sampling; and low-sensitivity variables (main effect value < 5%) are sampled using uniform sampling.

[0047] Secondly, a spatial filling rate monitoring index is introduced: by calculating the ratio of the minimum Euclidean distance to the average distance between sample points (with a threshold set to 0.3), sample points are automatically added to sparse areas when the ratio is <0.3 (each addition is 5%-10% of the base sample size). Simultaneously, feedback adjustment is performed based on the cross-validation error of the machine learning model—if the mean squared error of the 5-fold cross-validation is >5%, a sample point addition mechanism is triggered, prioritizing the addition of samples to the variable interval with the largest prediction error. The final total sample size can be dynamically adjusted according to the number of variables (e.g., the initial sample size is set to 80 for 8 variables and increased to 120 for 12 variables). Compared to traditional fixed sampling methods, this reduces invalid samples by more than 30% while maintaining spatial coverage (>90%).

[0048] For the initial sample points, a simplified flow field model (ignoring minor factors such as wheel rotation and ground effect) is first used for rapid calculation (grid size 1-2 million, calculation time < 1 hour) to screen potential samples with drag coefficients lower than 15% of the initial model. For these potential samples, a high-precision full flow field model (including details such as moving mesh wheels, ground boundary layer, and rearview mirror vortices) is used for detailed simulation (grid size 5-8 million, calculation time 8-12 hours). Wall function optimization (using an enhanced k-ωSST model) is introduced to improve the near-wall flow field accuracy (y+ value controlled between 1-5). Simultaneously, an error correction model is constructed: based on the deviation between wind tunnel test data and CFD results, a correction formula (such as Cd) is established. 修正 =Cd CFD×(1+0.02×Ma), where Ma is the Mach number, is used to correct the simulation results for high-speed conditions (>120km / h). Ultimately, a balance between computational efficiency and accuracy is achieved—compared to full high-precision calculations, the total time is reduced by 40%-50%, while the drag coefficient calculation error is controlled within 3% (the error of traditional single-precision methods is typically 5%-8%).

[0049] Model optimization and iteration: The data in the sample point database is divided into training and testing sets. The training set is used to further train and optimize the constructed machine learning wind resistance prediction model, improving its accuracy and generalization ability. The model performance is validated on the testing set to observe its prediction error and stability, and to evaluate whether the model meets the design requirements.

[0050] Optimization algorithms (such as genetic algorithms and particle swarm optimization) are used to solve the wind resistance optimization problem based on machine learning models. Taking genetic algorithms as an example, design variables are encoded as chromosomes. Through genetic operations such as selection, crossover, and mutation, the optimal solution that minimizes the wind resistance coefficient is searched in the design variable space. In each iteration, the trained wind resistance prediction model is used to quickly calculate the predicted wind resistance coefficient values ​​corresponding to different combinations of design variables (i.e., chromosomes), which serve as the fitness values ​​of the genetic algorithm to guide the genetic operations and gradually approach the global optimum.

[0051] Calculate the distance between the current optimal solution and the sample points, and the absolute value of the difference between their corresponding calculated drag coefficients. Select the optimal solution with the largest minimum distance and its corresponding calculated drag coefficient as addition point one, and add addition point one to the sample point database. Calculate the absolute value of the difference between the calculated drag coefficients of the remaining optimal solutions and the calculated drag coefficients of the sample points, and select the optimal solution with the largest minimum absolute value and its corresponding calculated drag coefficient as addition point two, and add addition point two to the sample point database. Based on the sample point database after adding points, retrain and optimize the machine learning model and optimization algorithm, repeating the above process until the pre-set accuracy requirements are met (e.g., the error between the model's predicted drag coefficient and the CFD simulation's calculated drag coefficient is within a certain range, or the improvement in drag coefficient after multiple iterations of the optimization algorithm is less than a certain threshold).

[0052] Optimization Scheme Determination and Output: Once the accuracy requirements are met, the final optimized solution is determined, i.e., the optimal combination of vehicle design variables is obtained. Based on these optimal design variables, an optimized vehicle geometric model is generated, and CFD simulation is used for final drag verification to ensure the optimization effect.

[0053] The system outputs optimized vehicle design solutions, including detailed vehicle geometry parameters, corresponding drag coefficient optimization results, and relevant data and analysis reports from the optimization process, providing comprehensive reference for vehicle design and manufacturing.

[0054] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A vehicle drag optimization design method based on a machine learning model, characterized in that, Includes the following steps: Data Acquisition and Preparation: Collect geometric parameters, vehicle operating status data, and corresponding drag coefficient data for different vehicle models, and clean and standardize the data. Machine learning model construction: Select a machine learning algorithm to build a wind resistance prediction model, train the model using preprocessed data, and minimize the prediction error by adjusting the model parameters; Design variable selection and parametric model construction: Based on automotive aerodynamics principles and engineering experience, key design variables are selected, their range of variation is determined, and parametric models are constructed using mesh deformation software or parametric modeling tools; Sample point selection and wind resistance calculation: Sample points are selected in the design variable space using experimental design method, corresponding geometric models are generated based on parametric models, wind resistance coefficients are calculated through CFD simulation, and a sample point database is established; Model optimization and iteration: The sample point data is divided into training set and test set, the machine learning model is optimized, the optimal solution is searched by combining optimization algorithm, and the model is iteratively updated by adding sample points until the accuracy requirements are met; Determining and Outputting the Optimization Scheme: Determine the optimal combination of design variables, generate the optimized vehicle geometric model, verify it through CFD simulation, and output the design scheme containing detailed parameters and optimization results.

2. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, In the data acquisition and preparation, the geometric shape parameters include vehicle length, width, height, body line curvature, front shape parameters, rear shape parameters, rearview mirror shape and position parameters, and wheel and wheel arch parameters; vehicle operating status data includes speed, acceleration, and driving posture. The drag coefficient data comes from wind tunnel tests or high-precision CFD simulations.

3. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, In the construction of the machine learning model, the machine learning algorithm is selected from deep neural networks (DNN), convolutional neural networks (CNN), or recurrent neural networks (RNN); the training process uses the mean squared error loss function, and the optimization algorithm is selected from stochastic gradient descent algorithm and its variants Adagrad, Adadelta, or Adam.

4. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, The specific steps for selecting the design variables are as follows: Based on aerodynamic theory, a pool of potential variables was initially screened, covering basic vehicle body parameters, surface features, aerodynamic accessories and detailed structures; The global sensitivity analysis algorithm was used to calculate the main effect value and total effect value of the variables, and highly sensitive variables with a main effect value > 5% and a total effect value > 10% were selected. Redundant variables with coupling coefficients greater than 0.8 are eliminated by calculating mutual information entropy. The number of core design variables is determined by combining production process constraints, and the dynamic range of change is set by interval division method.

5. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, In the construction of the parameterized model, adaptive density control points are set up in key aerodynamic regions, and the mapping relationship between control points and geometric features is constructed by non-uniform B-spline surface interpolation algorithm to form a topological correlation matrix. Embedded flow field constraints automatically trigger a control point compensation mechanism when adjustments to design variables cause signs of separation in the local flow field; a mesh smoothing algorithm ensures that the mesh distortion rate after deformation is less than 5%.

6. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, In the selection of sample points: Based on the design variable sensitivity weight stratified sampling, enhanced Latin hypercube sampling is used for highly sensitive variables, standard Latin hypercube sampling is used for moderately sensitive variables, and uniform sampling is used for low-sensitivity variables. A space filling rate monitoring index is introduced. When the ratio of the minimum Euclidean distance to the average distance between sample points is less than 0.3, additional sample points are added. Combined with the cross-validation error of the machine learning model, if the mean square error of the 5-fold cross-validation is greater than 5%, additional samples are added in the variable interval with the largest prediction error. The initial sample points were quickly calculated using a simplified flow field model. Potential samples with drag coefficients lower than 15% of the initial model were then selected. A high-precision full flow field model was then used for detailed simulation. An error correction model was constructed based on the deviation between wind tunnel test data and CFD results.

7. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, In the model optimization and iteration, the optimization algorithm is selected from genetic algorithm or particle swarm optimization algorithm; Increase sample points in the following ways: Select the optimal solution with the largest minimum distance from the sample point and its corresponding drag coefficient as the first addition point; Select the remaining optimal solution with the largest absolute value of the difference between the drag coefficient and the drag coefficient of the sample point, and the corresponding drag coefficient, as the second addition point. Add point 1 and point 2 to the sample point database, retrain the model and iterate until the error between the model's predicted drag coefficient and the CFD simulation calculated drag coefficient is within a preset range, or the drag coefficient improvement is less than a threshold after multiple iterations of the optimization algorithm.

8. The vehicle drag optimization design method based on a machine learning model according to claim 1, characterized in that, The optimization scheme outputs vehicle geometric parameters, drag coefficient optimization results, optimization process data and analysis reports, and is applicable to overall vehicle drag optimization and drag optimization design of local components such as rearview mirrors and spoilers.

Images