Refinement Zones for Passenger Vehicles

This section explains the resolution best practice for passenger-type vehicles. The recommendations are divided into:

General best practices
Hatchbacks/Compacts
Sedans/Coupes
SUVs/Vans
Pickup Trucks

Far field element size and used refinement levels

The simulation uses Refinement Levels (denoted RL), in which the cell size is halved each time the RL increases. This increases the cell count in each direction by a factor of two, and thus a factor 8 in 3d space. As each halved cell also doubles the time-step count per same physical time, the total computational load for a section by increasing the RL level is a factor 16. The far field element size (unit m) defines the largest element size in the model and is set in the .xml file or the inside the VWT. All additional mesh controls are defined relative to the far field size.

With a recommended far field voxel size set to 0.192 meters (192 mm), the following refinement levels will suffice for external aerodynamic simulations:

Refinement Level Voxel Size (mm)

RL 0: 192 mm
RL 1: 96 mm
RL 2: 48 mm
RL 3: 24 mm
RL 4: 12 mm
RL 5: 6 mm
RL 7: 1.5 mm
RL 8: 0.75 mm

RL6 (3 mm) is the Refinement Level that is used on the whole vehicle for standard passenger vehicles, with local detailing sometimes at a higher RL than that. Therefore, RL6 and RL7 are important reoccurring Refinement Levels.

Types of refinement

Refinements in the simulation are defined in 3 different ways:

Refinement boxes
Custom refinement zones
Offset refinement zones
- Overall offset
- Offset per part

Refinement boxes are axes aligned and defined by minimum and maximum length in each direction (x,y,z).

Custom zones are defined using a closed part in the .stl geometry file, with resolution set inside that part. These custom zones need to be present in the main .stl file that contains the vehicle. If the custom zones are not defined but the part is present in the .stl file, any custom zone part is regarded as having a solid wall. Check that all custom zones are properly defined in the .xml deck or in Virtual Wind Tunnel.

Offset zones are defined using an offset length and Resolution Level.ultraFluidX generates a minimum of 4 voxels regardless of the offset length setting. If no part is assigned for an offset, it is applied to all solid surfaces. If a part is assigned, the offset is only applied to that particular surface. An offset instance can have multiple parts, and a part can have multiple offsets overlaid on each other. The finest RL setting has precedence in that case, so if multiple layers are build up with decreasing RL, the coarser ones need to be defined with an additional offset length of all underlying higher RLs. For example, if you want 4 layers of 1.5 mm RL7 and 12 layers of 3 mm RL6 on top, the offset lengths would be 4*1.5 = 6 mm for RL7 and 6 mm + 12*3 mm = 42 mm for RL6.

In the sections below, these 3 types of Refinement will be used to capture relevant physics around the vehicle.

Note that all passenger car sections use versions of the DrivAer model for showing the resolutions (source TUMunich, https://www.mw.tum.de/en/aer/research-groups/automotive/drivaer/, allowance of usage greatly appreciated).

Vehicle refinement for outer regions (RL 1 through 3)

This sections goes through the resolution setup in steps, working inward from the outermost resolution.

The outer 3 refinement regions around the vehicle are boxes, with a shift towards the rear of the tunnel to capture the wake structures.

It is recommended that RL1 is at least:

X: 3 car lengths downstream, 1 car length upstream of the vehicle
Y: at least one vehicle width off to the sides
Z: at least one vehicle height above the vehicle

RL2 and RL3 are scaled down from there. It is recommended that RL3 end one car length behind the vehicle.

Refinement zones closer to the vehicle (RL4 and RL5)

Near the vehicle further, RL4 and RL5 are defined to capture:

the underbody flow
the wake
the wheels
the upstream boundary layer (in case of a VWT floor with a defined BL suction location)

The figure below indicates the location of refinement for the underbody flow and wake.

It is possible to apply resolution in these regions using either box or custom refinement zones. Recommended practice is to create custom refinement zones using loose wraps of the car that are scaled and repeated with a shifting towards the downstream side to encapsulate the wake.

Note: Creation of scaled and translated custom zones can be achieved in HyperMesh. In short: Wrap the entire body & wheels with a 12.0 mm loose wrap, scale 105%, copy it multiple times till 1.5m aft (for the wake), and a few times down (for the underbody). Then, wrap all these copies with a 12.0 mm convex wrap. Import this wrap and assign as resolution level of 5 (RL5).

For very 'boxy' cars, or when voxel count is less critical, it is possible to set box refinement zones for RL5 and RL4. In that case follow the above figure for RL5, and add an extra RL4 box enclosing the entire vehicle, stretching 1 car length behind. The width of the RL5 sections should be about 15% wider than the vehicle.

VWT floor refinements

In the case of wind tunnel modeling, optional belts and a boundary layer suction location are present. In order to adequately capture the flow development in those regions, resolution needs to be placed correctly.

For the boundary layer suction location, not only the region itself needs to be resolved, but also the development of the tunnel floor boundary layer past this element. It is recommended to put one RL6 box around the suction location, spanning two vehicle widths and extending 8 voxels (24 mm) upstream, downstream, and in height. Then, to grow the boundary layer, create an RL5 box that runs into the vehicle RL5 zone, extending two vehicle widths, 6 voxels (36 mm) in height.

In the case of wheel belts, extra refinement zones just around the belts are recommended, RL6 4 voxels (12 mm) high, spanning past the extents of these belts. This is not needed for main center belts or T-belts, as the focus is solely to capture the squeezed flow in front of the tire patch. Refinement boxes at the tire/road interface are also recommended in the absence of belts (open road simulation). In that case, place resolution starting around 100 mm in front of the tire/floor contact location and for 100 mm downstream as well, 50 mm beyond the tire extent on the sides.

Custom refinements

In order to capture complex regions of high shear and separated flow, specific custom refinements are recommended around parts of the vehicle.

The recommended zones include (but not limited to):

Nose of vehicle
Air dams/spoilers
Cowls
A-pillar vortices
Mirror wakes
Wheels

In the sections below, each custom refinement zone will be explained in detail.

Mirror custom refinement: The mirror needs special attention and two refinement zones need to be created. First, start by wrapping the mirror with a 12 mm convex wrap. Move the wrap rearward 100 mm and wrap the two wraps with a 12 mm convex wrap then scale this wrap by 105%. This will be a refinement zone of level RL6. Repeat the process by scaling the first mirror wrap by 120% and copying the wrap rearward 1.0 m and wrapping them with a 12 mm convex wrap. This wrap will become a resolution zone of RL5.
Nose custom refinement: Select the nose region and add surfaces in order to close the region. Loose wrap around the region, scaling and extruding down to the floor. This covers the engine bay and some wakes. It is recommended to set to RL5 (6 mm) to capture the engine bay and the surrounding flow.
Cowl and A-pillar custom region: Create a simplified shape that generously covers any cowl vortex and separation/reattachment on the windscreen. RL6 is recommended.

Offset refinements

Now that separated flow regions are covered using custom zones and/or boxes, offsets regions are used to capture the boundary layer flow in regions of interest.

As mentioned, offset regions can be general on all parts (when no specific part is indicated) or on separate sections. The offsets are used to set resolution in the boundary layers on the vehicle in order to capture the necessary physics. As boundary layers are growing over the vehicle, you see more placement of resolution on the upwind parts of the vehicle, where boundary layers are typically thinner. Also, separations and the effects of pressure gradients are important. Therefore, one will also see placement of resolution on parts that (possibly) separate, and rounded corners where large pressure gradients occur.

General offset refinement on whole body

As a starting point, for the general offset, 4 voxels (4x3 mm = 12 mm) of RL6 is recommended on the whole vehicle. In case of scaling the far-field resolution, the offset distance needs to be scaled to match 4 voxels again. This is true for both general and part offsets.

Specific part offsets

Next, specific part offsets need to be defined. The following regions are recommended for extra offset refinement:

Area around front bumper and nose (thin developing boundary layers)
A-pillar and roof line (accelerated flow, separation capturing)
Under the nose and around front wheels: air-dams, spoilers, splitter areas, deflectors
Back/tail end of vehicle: C/D pillars, side of bumper, roof spoilers, and so on
Tires
Mirrors
Any other elements that are being specifically investigated

All these areas are recommended to have: 4 voxels in RL7 (4*1.5 mm = 4 mm) plus 12 voxels offset of RL6 over that (+12x3 mm = +36 mm for a total of 40 mm RL6 offset)

On the last point, it is recommended to place additional resolution when investigating the effect of a specific part, for example a spoiler or lip.

The figure below shows some typical resolution placement for the front of the vehicle. The mirrors have thin boundary layers, changes in pressure gradients and separations, and are therefore part of the increased resolution sections. Next, the A-pillars are set to increased resolution, as formation of an A-pillar vortex is prone here. Note that the resolution definition stretches beyond the actual A-pillar part in both the upstream and downstream direction. It is recommended to isolate and include these sections to have the resolution interface further from where the relevant physics occur. The nose of the vehicle is also covered, as there is a thin developing boundary layer all around, as well as pressure gradients. The tires are also set to a higher resolution, in part due to their curvature into the flow. The tire parts in purple should have a minimum of RL7 offset. Using RL8 on the tires will improve the solution, but this significantly increase simulation time. Note that the lower bumper edge is indicated in a separate color. If the lower bumper has a sharp edge, separation may occur. It is recommended to further extend the RL7 offset on this part to capture the separated flow. The grill (in red) also needs attention, and is covered separately below.

Depending on the geometry, grills could need an offset of RL7 (1.5 mm) or even RL8 (0.75 mm) on grill elements to ensure enough voxels fit through the gaps. A bare minimum voxel count of 4 voxels is needed across a gap, with a recommended minimum of 6. This can be achieved using a simple offset on an isolated grill part.

The separation location on the back end of the vehicles is important for capturing the right flow physics. It is recommended that the areas around the tail of the vehicle have an RL7 offset of 4 layers. The specific shape and location of these offset regions might differ by vehicle type. Specific vehicle types are elaborated below.

Wheels and tires

Parts that rotate are identified as wheels. There are two methods currently available to describe rotation in ultraFluidX. They are:

Rotating walls
Moving Reference Frame (MRF)

For simulation of the flow between the rim spokes of passenger vehicles, a Moving Reference Frame (MRF) approach is recommended. The MRF can be created by applying rotational symmetry around a spoke. This creates a snug closure between the spokes. The MRF is allowed to intersect the spokes; individual pocket surfaces are thus not needed.

Special preparations for fully treaded, deformed tires

Standard best practice is to use non-treaded tires that are undeformed, and intersect with the ground plane, with longitudinal stripes (axisymmetric) remaining in the part.

Scanned detailed threaded tires can be implemented using the rotating wall condition. In case of features that are not fully parallel to the rotation direction, the rotating wall condition allows for a crude model of cavity flow: partly perpendicular will have an inflow/outflow component using this boundary condition. This prevents artificial separations around the tire sidewall.

For more detailed cavity flows (like in for example wheel/tire studies), an MRF filling the tread cavities can be used. The MRF zone can be created using rotational symmetry, as long as the MRF surface does not stick through the tire surfaces next to the cavities. This usage is not recommended in the general Best Practice for vehicles, only for special detailed studies on tires.

If you use scanned deformed tires that are flattened at the ground intersect, there is experience in setting up the detailed ground/tire boundary. For example, specific patches are created using a cut-plane 1 mm above the ground, and grouping any elements that are inside this intersect into a separate part. Remember this usage is not recommended in the general Best Practice for vehicles, only for special detailed studies on tires.

Porous media and cooling flow

Porous media are defined using an inflow plane, outflow plane, and sides. See the Wind Tunnel Setup topic for more information. To capture the correct cooling flow physics, a box resolution region is recommended that is loose around the porous media zone. This zone captures the flow from grills as well as the outflow along the front of the engine block past the porous media. RL5 is a recommended setting for this.

A minimum number of 6 voxels is recommended across the thickness of any porous media, where possible. Therefore, the cooling pack itself would often receive a box refinement of RL6 tight around the pack.

Strategy for Large-Eddy Simulations

Introduction

ultraFluidX employs physics based modeling using a two-way coupled system, where the shear stress of the wall model projected onto the surface is continuously being matched with a shear force in the bulk flow. This provides a more accurate model versus one-way coupling, where the shear forces are based on modeled parameters. For thin boundary layers, where the resolution is too large to capture the small vortices, there are two techniques used to more accurately model the appropriate physics. These are as follows:

Wall model variants per part
Vortex generator to initiate turbulent eddies

Background for usage

ultraFluidX is a Large-Eddy Simulation solver that provides details of the flow, including the flow in the boundary layers. Changes to the two-way coupled model are needed where thin upstream boundary layers occur on curved regions, and the resolution in terms of voxels over boundary layer thickness is low. Typical regions on automotive vehicles are listed in order of importance:

Mirrors
Nose and front bumper
Leading edges of tires

For reach region a modeling strategy is employed to capture the appropriate physics.

Wall Model Variants

Application: Mirrors

The mirrors typically have a high curvature and thin boundary layers before separating over the cover edge. To model the correct physics on the mirror, a wall model variant is specified on the mirror part. The wall model variant on the mirrors uses the one-way coupled model instead of the default two-way coupled model. The one-way coupled model is more robust in predicting the flow and drag when there is not enough resolution and length to develop the boundary layer fluctuations.

Example of setting the wall model variant

Copy Code:

<wall_model_variant>
<wall_model_variant_instance>
<name>mirrorWallModel</name>
<coupling>one-way</coupling>
</wall_model_variant_instance>
</wall_model_variant>

Application of the wall model variant to a part

Copy code:

<static_instance>
<wall_model_variant>mirrorWallModel</wall_model_variant>
<parts>
<name>mirrors</name>
</parts>
</static_instance>

Turbulence Generator

Application: Nose and front bumpers

The nose and bumper of an automotive vehicle includes a stagnation area and a developing boundary layers around the curvature of the bumper and nose. This area creates the upstream conditions for the rest of the vehicle. In order to aid the development of the boundary layer, the usage of a turbulence generator is recommended. The turbulence generator needs to be close to the stagnation region to create turbulence in the thin boundary layers in the form of randomized vortices. A box resolution zone of RL6 is set around the turbulence generator that covers approximately 10 mm upstream of the generator and 300 mm downstream to cover the bumper curvature.

The turbulence generator can be placed based on the external dimensions of the vehicle:

x direction: 24 mm upstream from the xmin location
y direction: + and -80% of the width/2 in y direction (as defined from the symmetry line)
z direction: 10% to 60% of the height in z (as defined from the ground)

Sample of the turbulence generator setup

Copy code:

<turbulence>
<turbulence_instance>
<num_eddies>800</num_eddies>
<length_scale>0.012</length_scale>
<turbulence_intensity>0.01</turbulence_intensity>
<characteristic_velocity_intensity>0.3</characteristic_velocity_intensity>
<point>
<x_pos>-0.83</x_pos>
</point>
<bounding_box>
<y_min>-0.7</y_min>
<z_min>-0.16</z_min>
<y_max>0.7</y_max>
<z_max>0.44</z_max>
</bounding_box>
</turbulence_instance>
</turbulence>

Example of the RL6 box around the turbulence generator

Copy code:

<box_instance>
<name>turbgenbox</name>
<refinement_level>6</refinement_level>
<bounding_box>
<x_min>-0.84</x_min>
<y_min>-0.71</y_min>
<z_min>-0.17</z_min>
<x_max>-0.54</x_max>
<y_max>0.71</y_max>
<z_max>0.45</z_max>
</bounding_box>
</box_instance>

Consideration for Specific Vehicle Types

Considerations for Hatchbacks/Compacts/Estates

Hatchbacks, compacts, estates, and the like often have a geometrically defined separation line (at least for the roof section). This simplifies the separation modeling; however, some care needs to be given on these sections as well. Often there is a round-off and/or a spoiler element. Below is an example estate-back on the TUMunich DrivAer model.

source: https://www.mw.tum.de/en/aer/research-groups/automotive/drivaer/

It is recommended best practice to place RL7 on locations where separation is bound to occur. For spoilers, create an offset for the whole element. For round-offs, include 10-30cm upstream and 5-15cm downstream of the expected separation into the RL7 region. The sides and rear bumpers can still have geometrically undefined separation lines; it is, therefore recommended to also place sufficient RL7 on the sides in these regions for all the surface area where separation of significant thickening of the boundary layer under adverse pressure can occur. The figure below illustrates the strategy:

Note that the base does not need to be in RL7, which saves elements. Also, the RL7 region extends past the separation edge as mentioned before. The shape of the RL7 region might not span a single part, but can wrap sections of adjacent parts, as well. In the figure above for example, part of the rear glass is included in the region. Often taillights are in this zone. It is recommended to place the whole taillight unit in the RL7 region.

Considerations for Sedans/Coupes/Fastbacks

Sedans, coupes, fastbacks and the like often have a sweeping roof line where the boundary layer thickening under adverse pressure gradient. Also separations and reattachments can occur. Below are examples of a sedan and coupe model of the TUMunich DrivAer.

source: https://www.mw.tum.de/en/aer/research-groups/automotive/drivaer/

Separation lines (if any) on curved surfaces facing the rear are harder to predict than on geometrically sharp edges. As such, best practice is to apply RL7 on the rear window/trunk sections, as well as on any side surfaces where separations can occur. The RL7 regions can be divided into sections with known separations (depicted in green below) and sections where the boundary layer experiences an adverse pressure gradient and is prone to possible separation (depicted in blue below).

A final image highlighting the geometric details with RL7 recommendations is shown below. Note the grill (in red) may need additional refinement.

Considerations for SUVs/Pickup Trucks/Vans

Refinement zones

Due to the increased sizes typically of SUVs/pickups/vans, the far-field resolution might need to be increased to have a reasonable cell count for the simulation. RL6 is 3 mm on passenger cars, 3.2 mm or 3.5 mm RL6 is recommended depending on the size. For a 3.2 mm setup for example, the resolution regions would look like this:

Refinement Level Voxel Size (mm):

RL 0: 204.8 mm
RL 1: 102.4 mm
RL 2: 51.2 mm
RL 3: 25.6 mm
RL 5: 6.4 mm
RL 6: 3.2 mm
RL 7: 1.6 mm
RL 8: 0.8 mm

All defined offset regions need to be scaled accordingly to keep a similar voxel count. VWT has an entry for setting the number of voxels in a certain offset region; this can be used for scaling.

Custom refinement cowl/A-pillar/mirrors

When windscreens are more upright, a separation can occur upstream on the hood. Make sure that the cowl custom refinement region is enlarged to capture this phenomenon. A-pillar vortices and mirror wakes can be larger, enlarge the refinement regions and blend with overlap with the cowl and mirror wake regions.

Truck beds

Truck beds can have complex flows in them, where the flow from the cab interacts with the tailgate. It is recommended to add a region of RL5 surrounding the whole truck bed. Both the cab end, top of the tailgate area and sides of the truck around the taillights should have RL7 refinements as defined before (4x RL7 + 12x RL6), as the separation. The shear layer at the end of the cab should be captured in an RL6 custom refinement region for about 30cm, as well as the volume around the top of the tailgate.

In conclusion: General Mesh Setup for Passenger Cars

Based on the described best practice, a vehicle of 100-250 million voxels is created, depending on the vehicle details. A typical runtime of around 8 hours on Nvidia 8x V100 GPUs is estimated for 4 seconds of physical time using this scheme.