沃尔沃XC90空气动力学CFD网格划分

网格大师 2023年7月20日浏览：1726 评论：1

沃尔沃XC90空气动力学CFD网格划分的图1

With the aim to shorten the lead times for the aerodynamics simulations with CFD programs of Ansys Fluent significantly, the Volvo Car Corporation has analyzed, optimized and partially automated all process steps and tools. One of the changes was the use of the meshing program Harpoon from CEI as an alternative to the programs ICEM-CFD and TGrid, uses so far. To show that these changes were not made at the expense of quality, a comparison was made between simulations for a Volvo XC 90 on the base of an ICEM-CFD/TGrid mesh and a Harpoon mesh.

1 Introduction

In recent years the pressure on the automotive industry has increased extremely. The necessity to develop more and better products in a shorter time span has led to an increased deployment of numerical methods like the CFD simulation. Computational Fluid Dynamics (CFD) not only improves the knowledge and understanding of flow phenomena, but it is a crucial tool to minimize the use of prototypes and test cars. Sincemore than a decade CFD is used for the vehicle development at Volvo cars in different phases. The broad spectrum reaches from aerodynamics and aero acoustics to under hood flow for cooling performance and climate control. In the early nineties, up to four months were needed for the analysis of quite simple models. Increased computing power and more efficient meshing techniques have reduced the time for a detailed aerodynamic simulation to fourup to five weeks (2004). Amongst other things, a further reduction was prevented by a much higher geometrical complexity.To decrease the project times and the change cycles, the complete simulation cycle was analyzed in detail and improved where ever possible during the last two years. Especially the time consuming steps for the geometrical conversion of the meshing were in the centre of interest, Figure 1.

沃尔沃XC90空气动力学CFD网格划分的图2

2 Building the CFD Model

Traditionally, at Volvo Cars two meshing programs were used. ICEM-CFD for an Octree mesh for the upper body, and TGrid for an unstructured Delaunay mesh for the under body. Today, meshing is done consequently with Harpoon from CEI, an octree mesher, which generates a hex-dominant mesh. To evaluate the influence of the two different meshes on the simulation results, two variants of the SUV Volvo XC 90 (with and without under body panels (UBP), Figure 2) were calculated. It is an exterior model with rear view mirrors, fully detailed under body, fairly detailed engine compartment and open rim wheels.

2.1 Find Vehicle and Extract CAD Data

For the management of the CAD data Volvo Cars uses the PDM solution TeamCenter from UGS. Users now can find and extract easily the CAD models of a specific car (for example left hand drive, diesel etc.). The CAD data from TCe is saved in the JT-format, which is universal and contains both CAD surfaces (Nurbs) and tessellated surfaces. In the CFD group at Volvo, the vehicle is divided into modules. Before the vehicle is assembled, these modules are either cleaned, surface wrapped, or simplified.

沃尔沃XC90空气动力学CFD网格划分的图3

沃尔沃XC90空气动力学CFD网格划分的图4

2.2 CAD Data Clean-up

For the clean-up of the CAD data – simplifying, removal of gaps and intersections – different methods are used, dependent of the content of the module. The biggest time savings can be achieved by using surface wrapping. With this method the base geometry is shrinked with a surface mesh. Holes will be closed, intersections removed and, if needed, geometrical details can be removed. Theoretically, the whole model could be generated with surface wrapping. Practically, the meshes become too big and the numerous surfaces cannot be handled very well in Ansa. Usually, a mixture of surface wrapping und manual editing is used, Table 1:

– Body/Exterior: unnecessary details in the exterior surfaces are removed and gaps are closed manually with Ansa.

– Front and rear chassis, engine and wheels: Simplifing and correction with surface wrapping. The wrapper of CD-Adapco is executed in the batch modus via a fitting script

– Under body: generation of a continuous tessalated surfaces. Based on this construction of new surfaces in Catia V5. Hundreds of surfaces are reduced to ten.

– Cooling system: manual cleanup by the thermodynamic CFD team.

2.3 Model Assembly

The different modules are assembled using Ansa. Since octree-meshing is used, the structure doesn’t need to be inspected for intersections and gaps. Only bigger holes with a size significantly bigger than the mesh size have to be considered.

2.4 Meshing

Harpoon was used fort the meshing. Harpoon is an octree mesher that produces a hex-dominant mesh. At the boundaries, the cut cells are converted into pyramids,tetrahedrons, or prisms in order to get the best quality. The computational domain is 50 m x 9.5 m x 9.6 m, Figure 3. For the vehicle structure the edge length is set to 5 mm. Only for critical domains, where separations were expected to occur, for example in the rear and on very fine details such as the bars of the grille, the size is set to 2.5 mm.

The maximum element size for the surrounding volume mesh was set to 320 mm. Around the entire vehicle, a refinement zone with size 40 mm was placed to assure that the cells do not grow too fast away from the surface. Mesh refinements were placed in the rear (20 mm), under the car (10 mm) and around the cooling package, to capture the wake, to resolve the narrow area to the ground, and to assure a constant mesh size within the cooling components (radiator, condenser, and charge air cooler).

The transition between differently dimensioned hex-zones was maintained by pyramids.

It is a conformal mesh, which means that transitions between two sizes of hexahedra are maintained by pyramids (conformal mesh). The final mesh consists of approximately 30 million cells and takes less than two hours to generate on a computer HP C8000.

沃尔沃XC90空气动力学CFD网格划分的图5

3 Numerical Method

A pressure based coupled algorithm was chosen, which gives a robust and efficient solution for steady-state flows. For the convection terms in the momentum equations, a second-order upwind scheme was used while a first order upwind scheme was used for the turbulent properties. For turbulence modelling, the „Realizable kappa-epsilon Model“ is used with standard wall functions.

The simulations are carried out on a Linux cluster of 32 CPUs (AMD Opteron 2.2 GHz). The typical run time is 36 hours for 2500 iterations. Convergence of the simulation is assumed when the residuals have decreased by at least three orders of magnitude and Cd shows a stable value or small oscillation (±0.001).

3.1 Boundary Conditions

At the entry into the computational domain a constant velocity of 140 km/h is set with a turbulence intensity of 0.1 % and a viscosity ratio of 200. At the outlet, zero gradients in the flow direction were specified for all variables. The ground is moving with the same speed as the freestream flow, and the rotation of the wheels is consistent with the free-stream velocity. Fluids between the spokes of the wheels are rotating with the same velocity as the wheels. The cooling components are modelled as porous regions and are included in the solution via source terms. For the fan a specified pressure rise was set as a function of the velocity. The sides and the top of the domain are treated with symmetry boundary conditions.

4 Results

The simulations, done with the different meshes (ICEM CFD/TGrid and Harpoon) show in general a good correlation, but present clear differences in certain areas. Evaluating the quality of the Harpoon mesh, the results were compared with calculations on the base of a conventionally generated mesh: an Octree-mesh (ICEM-CFD) for the upper body, and a Delaunay mesh (TGrid) for the floor.

The ICEM-CFD/TGrid-Netz has prismatic layers on the upper body surface and on the wheels. The first prismatic layer has a height of 1.5 mm and an edge length of approximately 10 mm, which gives a y+ value of less than 100 for a velocity of 38.88 m/s.

Even Harpoon has the capabilities to make prismatic layers. But due to the relatively poor quality of the cells this option was not choosen. To get an acceptable low y+ value, a finer mesh was generated (cell size at 5 mm, in critical zones at 2.5 mm).

4.1 Effects of Under Body Panels

For both calculated variants (without/with UBP) the Cd values of the Harpoon mesh are by 0.016 below that ones of the ICEM-CFD/TGrid mesh. The difference of the Cd values between UBP and no UBP is 0.005 for both meshes, see Table 2.

沃尔沃XC90空气动力学CFD网格划分的图6

Bigger differences are visible at the clvalues, Table 3. In comparison to the testing results, both meshes predict the front lift too low and rear lift, Figure 4, too high. The cl values for the Harpoon mesh are higher (0.109 without UBP, 0.086 with UBP). Even the difference with/without UBP is bigger for the Harpoon mesh. Responsible is the increase of rear lift.

沃尔沃XC90空气动力学CFD网格划分的图7

For the Harpoon mesh the pressure for the variant without UBP is slightly higher all over the base area. There is a larger region of low pressure on the rear lamp, Figure 5, for the ICEM-CFD/TGrid mesh compared to the Harpoon mesh. This indicates that separation occurs further downstream for the mesh made with ICEM-CFD/TGrid. Comparing to test results made in wind tunnel, the separation line for the Harpoon mesh looks more realistic.

沃尔沃XC90空气动力学CFD网格划分的图8

Because of the difference in separation in the rear, the wake structure is different. The bottom half looks the same for the two meshes. But on the top, the Harpoon mesh looks more irregular. A possible explanation is a vortex that is formed over the cat walk, which is stronger for the Harpoon mesh.Areas with a relative high pressure on the rear side of the front wheels, Figure 6, lower the Cd value in the Harpoon mesh. The part of the total Cd value that comes from the front wheels is 0.05 for the ICEMCFD/TGrid mesh and 0.03 for the Harpoon mesh. One reason for this difference could be the coarse mesh around the wheels in the Harpoon mesh. The y+-values (optimum 30–100) have an average of 200 significantly higher than fo the ICEM-CFD/TGrid mesh.

沃尔沃XC90空气动力学CFD网格划分的图9

5 Post Processing

EnSight from CEI is used for the post processing. This process step was automated by using a script, which starts EnSight in batch mode and generates a set of standard plots. The plots show pressure distributions on the vehicle seen from different angles, velocities in different cuts through the volume, Figure 7and Figure 8, and iso-contours of a specified constant velocity. However, close-up images of details in the flow are made manually.

沃尔沃XC90空气动力学CFD网格划分的图10

沃尔沃XC90空气动力学CFD网格划分的图11

6 Conclusions

A CFD simulation for a completely new vehicle can now be carried out in four weeks. Updates on an existing model can be made within a few days.

The main ingredients for this process acceleration are a faster and error free geometry generation. In this context the collection of CAD data with TeamCenter and the usage of surface wrapping should be mentioned explicitly. But even the faster mesh generation with Harpoon was one of the main reasons for this speed up. Making a volume mesh of 30 million cells using Harpoon takes less than two hours on the computer HP C8000.

The carried-out calculations show that the results for the Harpoon mesh are comparable to the previous method, where the mesh was made by using ICEM-CFD (Octree) and TGrid (Delaunay). Since no prismatic layers were used with the Harpoon mesh, yplus values were fairly high. Therefore, the mesh size should be decreased on the wheels.

For the development process it is important, that both meshes show the same tendencies. This is essential, since often the comparison between different variants are much more important than absolute values. Before this background the obtained results and realizations can be rated only positively.

为了利用Ansys Fluent的CFD程序显著缩短空气动力学仿真的交付周期，沃尔沃汽车公司对所有工艺步骤和工具进行了分析、优化和部分自动化。其中一个变化是使用CEI的网格划分程序Harpoon作为ICEM-CFD和TGrid程序的替代品，到目前为止。为了证明这些变化不是以牺牲质量为代价的，我们在ICEM-CFD/TGrid网格和鱼叉网格的基础上对沃尔沃XC 90的模拟进行了比较。

1 引言

近年来，汽车动力行业的压力急剧增加。在更短的时间内开发更多更好的产品的必要性导致了像CFD模拟这样的数值方法的部署增加。计算流体动力学（CFD）不仅提高了对流动现象的知识和理解，而且是最大限度地减少原型和测试车使用的关键工具。十多年来，CFD被用于沃尔沃汽车的车辆开发，处于不同的阶段。范围广泛，从空气动力学和空气声学到引擎盖下气流，用于冷却性能和气候控制。在九十年代初，分析非常简单的模型需要长达四个月的时间。计算能力的提高和更高效的网格划分技术将去尾空气动力学仿真的时间缩短到四到五周（2004 年）。除其他事项外，更高的几何复杂性阻止了进一步的减少。为了减少项目时间和变更周期，我们详细分析了整个仿真周期，并在过去两年中尽可能地进行了验证。特别是网格划分的几何对接耗时的步骤是人们关注的中心，如图1所示。

沃尔沃XC90空气动力学CFD网格划分的图12

2 构建差价合约模型

传统上，沃尔沃汽车使用两个网格划分程序。ICEM-CFD用于上半身的八叉树网格， TGrid用于身体下方的非结构化Delaunay网格。今天，网格化是用CEI的鱼叉完成的，CEI是一种八叉树网格器，它产生一个以六角为主的网格。为了评估两种不同网格对仿真结果的影响，计算了SUV Volvo XC 90的两种变体（带和不带车身下面板（UBP），图2）。这是一款带有后视镜、完全详细的车身、相当详细的发动机部件和开放式轮辋的外部车型。

2.1 查找车辆并提取CAD数据

对于CAD数据的管理，沃尔沃汽车使用UGS的PDM解决方案团队中心。用户现在可以轻松查找和提取特定汽车的CAD模型（例如左舵驾驶，柴油等）。来自 TCe 的 CAD 数据以通用格式保存，该格式包含 CAD 曲面（Nurbs）和细分曲面。在沃尔沃的CFD部门，车辆被划分为多个模块。在组装车辆之前，这些模块要么被清洁，要么被表面包裹，要么被简化。

沃尔沃XC90空气动力学CFD网格划分的图13

沃尔沃XC90空气动力学CFD网格划分的图14

2.2 CAD 数据清理

对于CAD数据的清理 - 简单检查，去除间隙和交叉点 - 根据模块的内容使用不同的方法。通过使用面部包装可以节省最大的时间。使用这种方法，使用曲面网格收缩基础几何体。孔将被关闭，交叉点将被移除，如果需要，可以移除几何细节。从理论上讲，整个模型可以通过表面缠绕生成。实际上，网格变得太大，并且在Ansa中无法很好地处理众多表面。通常，使用表面包装和手动编辑的混合方式，表1：

– 车身/外观：使用Ansa手动关闭外表面不必要的细节并关闭间隙。

– 前后底盘、发动机和车轮：通过表面包装进行简化和校正。CD-Adapco 的包装器通过拟合脚本以批处理方式执行

– 在车身下：产生连续的镶嵌表面。基于Catia V5中新表面的构造。数百个表面减少到十个。

– 冷却系统：由热力学CFD团队手动清理。

2.3 模型装配

不同的模块由我们组装而成。由于使用了八叉树网格划分，因此不需要检查结构的交叉点和间隙。只有尺寸明显大于网格尺寸的较大孔才需要考虑。

2.4 网格划分

鱼叉用于网格。Harpoon是一种八叉树网格，可产生以六角为主的网格。在边界处，切割的单元格被转换为金字塔，四面体或棱镜，以获得最佳质量。计算主图为50 m x 9.5 m x 9.6 m，图3。对于车辆结构，边缘长度设置为 5 mm。仅对于预计会发生分离的关键区域，例如在后部和非常精细的细节（例如格栅的杆）上，尺寸设置为 2.5 毫米。

周围体积网格的最大单元尺寸设置为320 mm。在整个车辆周围，放置了一个尺寸为40 mm的细化区，以确保细胞不会在远离表面的地方生长得太快。网格细化放置在后部（20 mm），汽车下方（10 mm）和冷却组件周围，以捕获尾流，将狭窄区域解析到地面，并确保冷却组件（散热器，冷凝器和增压空气冷却器）的网格尺寸恒定。

金字塔维持了不同尺寸的六角区之间的过渡。

它是一个共形网格，这意味着两种尺寸的六面体之间的过渡由金字塔（共形网格）保持。最终的网格单元包含大约 30 万个单元，在计算机 HP C8000 上生成所需的时间不到两个小时。

沃尔沃XC90空气动力学CFD网格划分的图15

3 数值法

选择了基于压力的耦合算法，为稳态流动提供了稳健有效的解决方案。对于动量方程中的对流项，湍流特性采用二阶上风方案，一阶上风方案。对于湍流建模，“可实现的kappa-epsilon模型”与标准壁功能一起使用。

模拟是在由32个CPU组成的Linux集群（AMD皓龙2.2 GHz）上进行的。36 次迭代的典型运行时间为 2500 小时。当落差至少减少三个或两个数量级且Cd显示稳定值或小振荡（±0.001）时，假设模拟收敛。

3.1 边界条件

在进入计算主机时，设定了140 km/h的恒定速度，湍流强度为0.1%，粘度比为200。在出口处，为所有变量指定了流向上的零梯度。地面以与自由流相同的速度移动，车轮的旋转与自由流速度一致。车轮辐条之间的流体以与车轮相同的速度旋转。冷却组件被建模为多孔区域，并通过源项包含在解决方案中。对于风扇，指定的确定上升被设置为ve性能的函数。做主干的侧面和顶部采用对称约束条件处理。

4 结果

使用不同网格（ICEM CFD / TGrid和Harpoon）进行的模拟总体上显示出良好的相关性，但在某些区域存在明显差异。在评估鱼叉网格的质量时，将结果与基于常规生成的网格的计算进行了比较：上半身的八叉树网格（ICEM-CFD）和地板的Delaunay网格（TGrid）。

ICEM-CFD/TGrid-Netz在上半车身表面和车轮上都有多层。第一棱柱层的高度为1.5 mm，边缘长度约为10 mm，对于100.38 m/s的密度，y+值小于88。

甚至鱼叉也有能力制作棱柱形层。但由于细胞质量相对较差，因此没有选择这种选择。为了获得可接受的低y+值，生成了更细的网格（细胞大小为5 mm，关键区域为2.5 mm）。

4.1 车身下面板的影响

对于两个计算变体（没有/有UBP），鱼叉网格的C d值比ICEM-CFD/TGrid网格的C d值低0.016。对于两个网格，UBP 和无 UBP 之间的 Cd 值之差为 0.005，请参见表 2。

沃尔沃XC90空气动力学CFD网格划分的图16

在表 3 的 c l 值处可以看到更大的差异。与测试结果相比，两个网格都预测前升力太低，后升力（图 4）太高。鱼叉网格的 c l 值更高（不使用 UBP 时为 0.109，使用 UBP 时为 0.086）。即使有/没有UBP的差异对于Harpoon网格来说也更大。负责增加后升力。

沃尔沃XC90空气动力学CFD网格划分的图17

对于鱼叉网，没有UBP的变体在整个基部区域的压力略高。与鱼叉网相比，ICEM-CFD/TGrid网的尾灯上的低压区域更大，如图5所示。这表明用ICEM-CFD/TGrid制成的网格在下游发生分离。与在风洞中进行的测试结果相比，鱼叉网的分离线看起来更逼真。

沃尔沃XC90空气动力学CFD网格划分的图18

由于后部的分离不同，尾流结构是不同的。两个网格的下半部分看起来相同。但在顶部，鱼叉网看起来更不规则。一种可能的解释是在猫走道上形成的漩涡，这对于鱼叉网来说更强。前轮后侧压力相对较高的区域（图 6）会降低鱼叉网格中的 C d 值。来自前轮的总Cd值的一部分对于ICEMCFD / TGrid网格为0.05，对于鱼叉网格为0.03。造成这种差异的原因之一可能是鱼叉网格中车轮周围的粗网格。y+值（最佳30–100）的平均年龄为200，明显高于ICEM-CFD/TGrid网格。