Vehicle Simulations

This section presents the following recommended practices to model external automobile aerodynamics:

Choosing Between an Unsteady or Steady State Model

Steady State

Steady state simulations are excellent tools for optimization studies to reduce drag. The advantages of running a simulation with steady state include:
  • Quick turnaround times.
  • Accuracy in absolute values and trends.

Unsteady

Unsteady simulations must be used when looking at cases such as vehicle handling, overtaking, or fleet modeling. In these cases, understanding the unsteady flow characteristics can improve the aerodynamics. Optionally, unsteady simulations can also be used in cases where transient eddy effects are important, such as aero-acoustics or wheel well modeling.

Unsteady simulations take longer to run. However, the overall turnaround time of an unsteady simulation can be reduced as follows:
  • The time step can be increased using transient SIMPLE. However, if the time step is raised too high, the solution becomes non-physical.
  • Run transient simulations with higher under-relaxation factors than steady simulations. The higher the relaxation faction, the fewer the inner sweeps that are required.
  • Initialize transient simulations using a steady flow field. Use the K-Omega SST model to initialize the flow field and reduce the jump between steady and DES simulations.
  • You can use either coupled or segregated solvers to initialize the flow field. The coupled solver has the advantage of reduced run time for larger case studies.

Recommended Turbulence Model

  • Either SST (Menter) K-Omega or Spalart-Allmaras Detached Eddy Simulation model is recommended for vehicle aerodynamics. Both seem to give similar solutions.
  • For vehicle aerodynamics, DES offers improved information on large eddy structures in the wake. For wake flow, DES is more accurate than URANS. So, for all studies of vehicle passing or wake flow, DES is recommended.
  • Aero-acoustics studies have shown some cases where Spalart-Allmaras offers slightly better trends. Typically, for passenger cars, the SST K-Omega turbulence model is used.
  • Large time periods are more critical than local flow structure for thermal transient simulations. Therefore, URANS is appropriate for this class of problem.

Recommended Mesh Settings

  • The grid density must be refined enough to capture local physics. However, over-refinement slows down performance. Use volumetric sources to refine the mesh only in the required areas.
  • The grid size must be fine enough to capture the turbulent length scale.
  • Resolve the wake of vehicles and tires when modeling passenger cars. Smaller length scales are less critical. The wake is generally refined for half of the car length to the rear.

    A more realistic estimation is to calculate the frontal area. Based upon the frontal area of the vehicle, extend the same distance in the wake.

  • Examining the results of a steady state analysis can help determine grid resolution for DES simulation using the following field functions:

    Name

    Function Name

    Definition

    Purpose

    Turbulent Length Scale

    TurbulentLengthScale

    pow(0.09,-1)*sqrt($TurbulentKineticEnergy)/$SpecificDissipationRate

    Determines Turbulence Length Scale for SST (Menter) K-Omega model

    Grid Size

    GridSize

    pow($Volume,1.0/3.0)

    Estimate Edge Length of cell (based upon cubes)

    Length Ratio

    LengthRatio

    $TurbulentLengthScale/(2*$GridSize)

    To resolve turbulence properly, length ratio > 1 would be needed

  1. Run a preliminary steady simulation.
  2. Use the TurbulentLengthScale and GridSize field functions to calculate the Length Ratio, as defined in the table.

    The turbulent length scale is based upon the turbulent dissipation. The grid size is estimated based upon the volume of each cell

  3. Use a mesh that yields a length ratio greater than one in the preliminary steady simulation for the transient analysis.

Recommended Number of Inner Iterations

  • Lowering inner iterations to speed up the solution is a better strategy than increasing the time step. Large time steps adversely affect results.
  • In most cases, 8 inner iterations are enough to converge a time step. Lowering inner iterations between 3–5 helps in reducing turnaround time.
  • The overall drag value is not sensitive to the number of inner iterations. The grid density and refinement locations are more critical in getting accurate drag values.

Recommended Time-Step

  • Time-step size depends on the physics that is being resolved.
    • For aero-acoustics, the time step must be small enough to resolve the desired frequency.
    • For aerodynamics, larger time steps are used to capture the large eddies in the flow field.
    • For general drag values, resolve the large wake structures behind the tires/vehicle. A good estimate for a time step is approximately 100 time steps for a particle to reach the span of the vehicle:
      Δt=⁢Length of Vehicle100×Vehicle Speed
  • The courant number must be below 1 for most of the flow field. Small local peaks from flow accelerating around a sharp corner are fine if they spike over 1.

Recommended Monitors and Reports

Use the following monitors to check for convergence in inner iterations:
  • Force on vehicle: this monitor reaches a steady value at convergence.
  • Pressure probes in tire and vehicle wake

    80% convergence to final value is fine for drag prediction

  • Velocity magnitude

    The peak velocity must be below 150 m/s. If the peak velocity is higher, raise the time scale coefficient for upwind blending (CdesTimeLim) expert property of the SST (Menter) K-Omega model to improve stability.