Keywords: automobile; Pit-shaped non-smooth surface; Drag reduction; CFD simulation; Kriging model; Optimize
Analysis and optimization of drag reduction characteristics of automobile with concave non-smooth surface
Abstract: The drag reduction characteristics of concave non-smooth body surface are studied. Firstly, the drag reduction effects of rectangle, thombus and arithmetic pit arrangement are investigated, and the rectangle arrangement with better drag reduction effect is selected. Then, with the pit diameter and vertical and horizontal spacing as design variables and the minimum resistance as the goal, Latin hypercube experimental design is adopted to optimize. Next, the responses of different sample points are obtained by CFD simulation, and the Kriging model is established on this basis. Finally, after verifying the credibility of the metamodel, the metamodel is optimized globally. The results show that the maximum drag reduction effect of rectangular pits can reach 7.62%.
Keywords: automobile; Pit-shaped non-smooth surface; Drag reduction; CFD simulation; Kriging model; optimization
order
At present, the optimization of aerodynamic drag characteristics of automobile aerodynamics is mainly realized by streamlining and local improvement of the car body. As these methods become more and more mature, the space for drag reduction becomes smaller and smaller, and the drag reduction of automobiles enters a bottleneck period. In recent years, the research on drag reduction of concave non-smooth surface structure based on engineering bionics theory has developed rapidly. One of the most typical applications is the concave spherical surface of golf ball. During the flight of golf ball, due to the existence of pits, the boundary layer formed by air clings to the surface of the ball, which makes the smooth airflow move backward along the ball, delays the separation between the boundary layer and the ball, and reduces the wake area and front and rear pressure drag, thus making the golf ball with concave spherical surface fly farther than the golf ball with smooth spherical surface.
Inspired by it, this paper applies concave non-smooth surface to automobile surface, and studies its drag reduction effect through CFD numerical simulation. Firstly, the influence of different arrangement of pit units on vehicle drag reduction effect is studied. Then, based on the layout with the best drag reduction effect, the relevant design variables are selected, and the Latin square experimental design method is used to select the sample points. Then kriging approximate model-3-1is established; Finally, the multi-island genetic algorithm is used to optimize the approximate model globally.
CFD calculation and experimental verification of 1 original vehicle model
1. 1 Establishment of calculation model
The real vehicle model of 1: 1 is established by using UG software. The model is simplified appropriately, ignoring the door handle, wiper, rain ditch, etc., and the chassis is flattened to improve the analysis efficiency. The length× width× height of the car model are 5 088× 2 036× 1 497 (mm) respectively, and the whole car model is shown in figure 1.
Establishment and solution of 1.2 computational grid
The calculation domain of the whole vehicle is a cuboid around the car body. The population is three times the length of the vehicle from the front end of the model, and the exit is seven times the length of the vehicle from the back end of the model. The total height is five times the height and the total width is seven times the width. Unstructured tetrahedral mesh is generated by ANSYS ICEM CFD software, and the mesh is encrypted on the unsmooth body surface, so as to obtain the required flow field information more accurately. At the same time, the triangular prism grid parallel to the car body surface is stretched as the boundary layer to eliminate the influence of wall function. In order to avoid the influence of grid differences on the simulation results, the grid size of the same part of chess remains unchanged during the simulation process. The total number of meshes generated by each simulation is about 3.6 million.
The boundary conditions are set as follows: the inlet of the computational domain is set as the velocity population boundary, the velocity is 40m/s, the outlet of the computational domain is set as the pressure outlet boundary, the body surface is set as the non-sliding wall boundary condition, the bottom plate of the computational domain is set as the moving wall boundary condition, and the upper surface, left and right sides of the computational domain are all sliding wall boundary conditions. The realizable k-ε turbulence model is selected, and the second-order upwind scheme is used for discrete solution. The temperature in the calculation domain is the normal temperature calculated by CFD steady-state simulation.
1.3 wind tunnel test verification
Wind tunnel tests verify the accuracy of boundary conditions and turbulence model setting. According to the CAD model, the experimental model is processed into a model of 1: 3 by the NC machining center, which ensures the consistency between the physical model for the experiment and the CAD model for numerical simulation. The force test was carried out in the HD-2 wind tunnel of the Wind Engineering Experimental Research Center of Hunan University, and the aerodynamic force of the model was measured with a six-component floating force balance. The test wind speed is 40 m/s, and the ground boundary layer suction device is started, which eliminates the problem caused by
Influence of wind tunnel test on ground boundary layer. The wind tunnel test of the car model is shown in Figure 2.
The wind drag coefficient CD of the model is measured by wind tunnel test, and the CFD simulation results are compared with the test, as shown in table 1. The relative error of drag coefficient is 3. 86%, within the allowable error of 5%, which verifies the reliability of numerical simulation.
2 Selection of non-smooth processing area and estimation of cell size
The non-smooth treatment area should be selected on the surface that can better control the wake area, so as to reduce the turbulent energy loss and pressure resistance, and the car body roof is the surface that has the greatest influence on the wake area. Therefore, this paper mainly studies the drag reduction effect of the car body roof after the pit non-smooth treatment, and the pit non-smooth area is shown in Figure 3.
Relevant research shows that the pressure resistance caused by air separation and the friction resistance caused by gas viscosity are related to the boundary layer and its thickness. The bionic non-smooth drag reduction method is realized by controlling the boundary layer to reduce the intensity of turbulence burst and the loss of turbulence kinetic energy. It can be seen that the choice of non-smooth structure should be related to the boundary layer, and the size, height or depth of non-smooth element should be less than the distance from the body surface to the logarithmic law region. At present, there is little research on pit drag reduction in the world, and no theoretical system has been formed. Therefore, in the early stage of the study, the size of the pit element is mainly determined by the thickness of the boundary layer.
The calculation formula of laminar boundary layer thickness of flat plate is
3 Design and layout of structural dimensions of foundation pit
3. Structural dimension design of1pit
When arranging the pit cells, the dimensions of the cells are mainly considered: diameter d, lateral spacing w, longitudinal spacing l and pit depth s, as shown in Figure 4. For the convenience of design and layout, the depth S is taken as half of the diameter D. According to the maximum boundary layer thickness, car roof size, car running speed and anti-interference requirements between pit units, the given values of D, W, L and S are [10/0,40], [60, 160] and [60,/60] respectively.
3.2 Influence of pit unit arrangement
According to a large number of bionic experiments, the reason why soil animals such as dung beetles can move freely in the soil is because of the shape of pits on its surface and the arrangement of pits on the other hand. Therefore, when studying the drag reduction performance of concave non-smooth surface, the influence of its arrangement should be considered. In this paper, three common arrangements are selected: rectangular arrangement, diamond arrangement and arithmetic arrangement, as shown in Figure 5.
In this paper, D= 15mm and Shape =120 mm ~ =120 mm are selected for CFD simulation, and the results are shown in Table 2.
As can be seen from Table 2, among the three pit-shaped unit arrangements, the rectangular row has the best drag reduction effect, and the drag reduction rate reaches 2. 13%.
Optimal design of non-smooth surface of four pits
4. 1 Optimization process and selection of design variables
According to the CFD simulation results of the three arrangements, the rectangular arrangement has the best drag reduction effect, so the concave non-smooth surface of the rectangular arrangement is taken as the optimization object. The whole process of analysis and optimization is as follows: (1) Latin square design method is used to determine design variables and select sample points; (2) Obtaining the response value of each sample point through CFD simulation, and constructing an approximate model based on the sample point and the response value; (3) Select three new groups of sample points to verify the accuracy of the approximate model, and if not, re-select the sample points; (4) On the basis of verifying the credibility of the approximate model E, the optimization algorithm is used to achieve global optimization in the region that meets the constraints, and the optimal solution is obtained. Finally, the simulation model is returned for checking calculation, as shown in Figure 6.
Taking d, w and l as design variables, the optimal combination is sought to achieve the maximum drag reduction effect, that is, the minimum CD value is obtained.
4.2 experimental design,
According to the range of design variables, Latin square sampling method is adopted. 20 groups of sample points were selected for CFD simulation calculation, and 20 groups of response values were obtained. The relationship between each design variable and the CD value is shown in Figure 7. D and so on represent the influence of a single design variable on CD, and D-W and so on represent the interactive influence of two variables on CD. d？ 0? 5 and so on. Represents the influence of the square of the design variable on the CD.
As can be seen from Figure 7, the design variables that have the greatest influence on CD are L, D and W ... D, with the most obvious interaction with shapes, followed by L and D, and the least interaction between shapes and \. Although the influence of W on aerodynamic drag is very small, the interaction between W and other parameters can not be ignored.
4.3 Establishment of approximate model
Approximate model refers to a mathematical model with small calculation amount and short calculation period without reducing the calculation accuracy, but the calculation results are similar to those of numerical analysis or physical experiment. It is used to replace the simulation analysis software with high calculation cost, which greatly improves the analysis efficiency and eliminates the "calculation noise" of the simulation software. The methods used to construct approximate models mainly include response surface model, kriging model, radial basis function neural network model and Taylor series model.
Compared with other models, the approximate surface constructed by kriging model can cover all the sample points, and the approximate surface quality is very high, so kriging model is used to construct the approximate model.
In order to verify the fitting accuracy of the established approximate model, any three experimental points except the experimental design scheme are selected in the design space for CFD simulation, and the calculated results are compared with those of the approximate model, as shown in Table 3.
As can be seen from Table 3, the difference between the CFD value of the verification point and the approximate model value is less than 2%, which shows that the established approximate model can well describe the relationship between design variables and response values, with high reliability, and can replace direct CFD calculation.
4.4 Optimization Results and Analysis
Multi-Island Genetic Algorithm (MIGA) is based on traditional genetic algorithm. It is different from the traditional genetic algorithm in that the individuals of each population are divided into several subgroups called "islands": all operations such as selection, crossover and mutation of the traditional genetic algorithm are carried out on each island, and the selected individuals on each island regularly migrate to other islands, and then continue to carry out the traditional genetic algorithm operations. The migration process is controlled by two parameters: migration interval and mobility. The migration interval represents the generation of each migration, and the mobility determines the percentage of the number of individuals migrated on each island during the migration. The migration operation in multi-island genetic algorithm keeps the diversity of optimal solutions and improves the probability of containing global optimal solutions.
In this paper, multi-island genetic algorithm is used to optimize the approximate model. The initial population number is 50, the island number is 10, and the iterative algebra is 100. The optimal solutions of the final approximate model are D= 40mm, W= 100mm and L = 69 mm The optimal solutions are simulated by CFD, and the relative error is 0. 80%.
The maximum drag reduction rate can reach 7. 62% after the surface of the car body is not smooth, the specific values are shown in Table 4.
Figures 8 and 9 show the pressure nephogram and velocity streamline diagram of the original vehicle and the optimized vehicle respectively.
By comparing Figure 8 and Figure 9, it can be seen that the negative pressure area at the rear of the vehicle is obviously reduced and the positive pressure area is obviously increased after optimization, thus reducing the front and rear pressure drag, improving the vortex at the rear of the vehicle, reducing the aerodynamic resistance of the vehicle and reducing the fuel consumption of the vehicle.
5 conclusion
(1) The concave non-smooth treatment of the car body surface has a good drag reduction effect, which can effectively reduce the aerodynamic drag of the car, thus reducing fuel consumption and improving fuel economy.
(2) The drag reduction characteristics of non-smooth surfaces with pits are related to the arrangement of pit elements, and the rectangular arrangement is the best. The diameter, horizontal spacing and vertical spacing of rectangular pit units are selected as design variables for experimental design, and an approximate model is established, which is optimized by multi-island genetic algorithm. After optimization, the maximum drag reduction rate can reach 7. 62%.
(3) The combination of experimental design, approximate model and optimization algorithm can provide some engineering guidance for the research and optimization of drag reduction on concave non-smooth surface of car body.
refer to
[1] Gu Zhengqi. Automobile aerodynamics [M]. Beijing: People Communications Publishing House, 2005.
Han,,, et al. Regression analysis of micro-friction and wear test of pitted non-smooth surface [J]. Journal of Tribology, 2005,25 (6): 578-582.
Rong, Gu Zhengqi, Yang Yi, et al. Starting optimization of sports car tail shape based on Kriging model [J]. China Mechanical Engineering, 20 10/0,22 (2): 243-247.
Gu Zhengqi, He Yibin, et al. Numerical Simulation of External Flow Field of New Concept Vehicle [J]. China Mechanical Engineering, 2007,18 (14):1760-1763.
Xue Zusheng, Boundary Layer Theory [M]. Beijing: Water Conservancy and Electric Power Press, 1995.
Fang Kaitai, Ma Changxing, orthogonal uniform experimental design [M]. Beijing: Science Press, 200 1.
Xiao, Zhang, Quan and Zhang Lie. Structural shape optimization method based on Kriging agent model [J]. Mechanical Design, 2009,26 (7): 57-60.
Shi Xiuhua, Meng Xiangzhong, Du Xiangdang, et al. Optimal placement of vibration control sensors based on multi-island genetic algorithm [J]. Vibration Test and Diagnosis, 2008,28 (1): 62-65.
(Source: China Technician Network)