Advanced LES, Hybrid Models (DES), and Post-Processing in ANSYS Fluent

Kiarash Kord
Reading time: 9 min

In Blog 2, we set up a complete Large Eddy Simulation (LES) case. We created a high-quality mesh, initialized with RANS, and configured the WALE SGS model. For academic flows or low-Reynolds number internal flows (like a flow inside a small pipe), this setup works perfectly. However, if you try to apply that same setup to an airplane wing, a car body, or a wind turbine blade, you will face a major problem: Computational Cost. There are 2 major issues that we need to consider:

Contents

The Boundary Layer Challenge

The problem lies in the physics of the boundary layer. Near a solid wall, turbulent eddies become very small. In a pure LES, we must resolve these eddies. This means our mesh cells () must be extremely small.

Specifically, the requirements for a Wall-Resolved LES are:

$y^+ \approx 1$ : The first cell height must be in the viscous sublayer.
$\Delta x^+ \approx 50 \text{ and } \Delta z^+ \approx 20:$ : The streamwise and spanwise spacing must also be very fine to capture the turbulent streaks.

As the Reynolds Number (Re) increases, the boundary layer becomes thinner, and the eddies become smaller. To capture them, the number of mesh cells ( $\Delta x, \Delta y, \Delta z$ ) must increase drastically. The cost scales roughly as $N \propto Re^{1.8} \text{ to } Re^{2.4}.$ .

Figure 1: Chart showing the computational cost & complexity of LES increasing compared to RANS.

For a typical industrial flow (Re>10e+6), a pure LES might require billions of cells and months of calculation time. This is simply not practical for most engineering projects.

The Solution: Hybrid RANS-LES

So, how do we simulate high-Reynolds number flows without waiting until 2045? The answer is to stop trying to resolve the tiny eddies near the wall.

We know that RANS (Reynolds-Averaged Navier-Stokes) models, like the K-w SST, are excellent at predicting attached boundary layers at a very low cost. LES, on the other hand, is excellent at predicting large, separated turbulent structures in the wake.

Hybrid RANS-LES models combine the best of both worlds:

Near the Wall: The model acts like RANS. It models the boundary layer turbulence instead of resolving it. This allows us to use much larger mesh cells (higher aspect ratios) near the wall.
Away from the Wall: The model switches to LES. It resolves the large unsteady eddies in the core flow and separated regions.

Figure 2: Diagram of Detached Eddy Simulation (DES) showing RANS usage near the wall and LES usage in the wake.

This approach reduces the mesh count from billions to millions, making LES simulation feasible for industrial aeronautics and automotive applications today. The most popular family of these hybrid models is called Detached Eddy Simulation (DES). If you want to see these concepts applied in real scenarios, check out these advanced tutorials from CFDLAND: CFD Simulation of an Impinging Jet Spray Using the VOF to DPM Method and VOF to DPM CFD Simulation, Jet in Crossflow for Aeronautical Applications.

Figure 3: Advanced Hybrid RANS-LES examples from CFDLAND: Impinging Jet Spray (VOF to DPM) and Jet in Crossflow.

LES vs. DES (The Comparison)

As we established in the previous section, resolving the boundary layer with pure LES is too expensive for high-speed flows. This brings us to the industry-standard solution: Detached Eddy Simulation (DES).

What is Detached Eddy Simulation (DES)?

DES is a Hybrid RANS-LES model. It was explicitly designed to combine the low cost of RANS simulations for wall-bounded flows with the high accuracy of LES for separated flows.

In a DES simulation, the solver divides the domain into two distinct regions:

RANS Region (Near-Wall): In the boundary layer, the model acts like a standard RANS model (typically Spalart-Allmaras or k-w SST). It models the turbulence, allowing for a coarser mesh.
LES Region (Detached Flow): In the wake or regions of massive separation, the model switches to LES mode. Here, it lowers the eddy viscosity to allow the formation of resolved eddies.

Figure 4: Diagram showing the RANS and LES regions in a Hybrid model Simulation around an airfoil.

Comparison: LES vs. DES

Feature	LES (Large Eddy Simulation)	DES (Detached Eddy Simulation)
Physics	Resolves large eddies everywhere (even at the wall).	Models eddies at the wall (RANS); Resolves eddies in the wake (LES).
Mesh Cost	Extremely High ( $(N \approx Re^{1.8}).$ ).	Moderate (Independent of Re in the wake; RANS-like at wall).
Wall Spacing	Strict: $y^+ \approx 1, \Delta x^+ \approx 40, \Delta z^+ \approx 20.$	Relaxed: $y^+ \approx 1, \text{ but } \Delta x \text{ and } \Delta z \text{ can be much larger.}$
Best For	Low-Re flows, Acoustics, Complex Mixing.	High-Re Aerodynamics (Wings, Cars, Wind Turbines).

The Switching Mechanism

How does ANSYS Fluent know when to use RANS and when to use LES? It uses a mathematical switch based on the Turbulent Length Scale ( $L_t$ ) and the Grid Size ( $\Delta_{max}$ ).

In RANS mode: The turbulence length scale is determined by the turbulence model (e.g., $L_t = \frac{k^{1/2}}{\omega}$ ).
In LES mode: The length scale is determined by the grid size ( $\Delta_{max}$ ), acting as a sub-grid filter.

The DES model calculates the distance to the nearest wall (d).

If $d < C_{DES}\Delta_{max}$ (Near Wall), the model uses RANS.
If $d > C_{DES}\Delta_{max}$ (Far Field), the model switches to LES.

Important Variant: Delayed DES (DDES)

The original DES model had a flaw called Grid Induced Separation (GIS). If the engineer created a mesh that was too fine inside the boundary layer, the model would accidentally switch to LES mode too early. Since the mesh was fine but not fine enough for pure LES, the turbulent viscosity would drop artificially, causing the flow to separate incorrectly.

To fix this, Delayed DES (DDES) was created. DDES adds a “Shielding Function” ( $f_d$ ) that detects the boundary layer. It forces the model to stay in RANS mode inside the boundary layer, regardless of the grid size. In modern ANSYS Fluent versions, you should almost always use DDES (or IDDES) instead of standard DES to ensure safety against grid-induced errors.

Figure 5: The effect of shielding function on flow separation behavior modeling

Other Hybrid Strategies (SAS & WMLES)

While DES is the standard for external aerodynamics, ANSYS Fluent offers two other powerful hybrid strategies for specific engineering problems.

Figure 6: Selecting the Detached Eddy Simulation (DES) model in the ANSYS Fluent Viscous Model panel.

Scale-Adaptive Simulation (SAS)

SAS is known as the “safe” hybrid model. Unlike DES, which relies strictly on the grid size ( $\Delta$ ) to switch modes, SAS uses the von Karman length scale ( $L_{t,K}$ ). The model looks for flow instabilities (wobbling or fluctuating flow).

Stable Flow: If the flow is steady (like inside a straight pipe), SAS remains in RANS mode.
Unstable Flow: If the flow features massive separation (like behind a cylinder or a bluff body), SAS adapts the length scale and switches to LES-like resolution.

It is safer than DES because it does not suffer from Grid Induced Separation. If your mesh is too coarse, SAS simply reverts to a standard RANS solution, whereas DES might give a wrong answer. It is excellent for flows with strong instabilities, such as flow past a cylinder or heat exchanger tube banks.

Figure 7: Visualization of von Karman vortex street behind a cylinder using Scale-Adaptive Simulation (SAS).

Wall-Modeled LES (WMLES)

WMLES is a strategy designed to reduce the cost of wall-bounded flows, such as flow inside a high-speed channel or pipe. In a pure Wall-Resolved LES, we need to resolve the inner boundary layer ( $y^+ \approx 1$ ) and use very small cells in the streamwise direction. WMLES relaxes this. It models the inner part of the boundary layer (the viscous sublayer and log-layer) using an algebraic formula, while resolving the outer part with LES.

It drastically reduces the cell count required. According to the references, WMLES requires a constant grid density ( $(N)_y \approx 40$ ) regardless of the Reynolds number, whereas pure LES mesh requirements explode as speed increases.

Model	Best Application	Key Advantage
LES (WALE)	Low Re, Aeroacoustics	Most accurate, resolves all scales.
DDES / IDDES	Ext. Aerodynamics (Wings, Cars)	Shielded boundary layer protection.
SAS	Unstable Industrial Flows	“Safe” fallback to RANS on coarse meshes.
WMLES	High Re Internal Flows	Reduces cell count for boundary layers.

Post-Processing Masterclass (Data Sampling, Q-Criterion)

Running the simulation is only half the work. LES and DES results change constantly with time. You cannot look at a single picture of velocity to make a decision. You must collect statistical data over time. This section explains the three main steps for LES analysis: Data Sampling, Vortex Visualization, and Quality Validation.

Data Sampling: Getting Statistically Steady Results

In a steady RANS simulation, you run the calculation until the residuals stay flat. In LES, the residuals never stay flat. The flow always changes. You need a different method to stop the simulation.

Step 1 – Purge the Transient: When you switch from RANS to LES, the initial flow is “messy.” The RANS turbulence must convert into resolved eddies. You must run the simulation for 2 to 3 Flow Through Times. A “Flow Through Time” is the time it takes for a particle to travel from the inlet to the outlet.
Step 2 – Enable Sampling: Monitor the Drag or Lift coefficient. Wait until the values oscillate around a constant average. This is the “Statistically Steady State. Then, go to the Run Calculation task page in ANSYS Fluent and check the box “Data Sampling for Time Statistics.” Continue the simulation for another 3 to 5 Flow Through Times. This ensures your average values are smooth.

As outputs, use Mean Velocity/Pressure for standard contour plots. They should look smooth, similar to RANS results. Also, RMS (Root Mean Square) value shows the intensity of the fluctuations. High RMS Velocity means there is strong mixing or vibration in that area.

Figure 8: Enabling Data Sampling for Time Statistics in the ANSYS Fluent calculation settings.

Visualizing Turbulence: The Q-Criterion

Velocity contours do not show turbulent structures well. They hide the rotation inside the flow. To see the 3D eddies, you must use the Q-Criterion. The Q-Criterion identifies areas where the fluid rotation (vorticity) is stronger than the deformation (strain). To do so, go to the Results tab, create an Iso-Surface. For “Surface of Constant,” select Mesh… -> Q-Criterion. Enter a positive value and Color the Surface with Velocity Magnitude or vorticity to color the iso-surface. For instance, see Figure 9 which shows Supersonic Jet Flow LES CFD output.

Figure 9: Q-Criterion Iso-Surface colored by vorticity showing 3D turbulent structures in a supersonic jet

Conclusion

This completes our trilogy on Large Eddy Simulation. Blog 1 explained the theory of eddies and filtering. Blog 2 showed the step-by-step setup for a standard LES using the WALE model. Blog 3 introduced Hybrid RANS-LES (DES and SAS). These models allow you to simulate high-speed flows like airplanes and cars.

You now have the knowledge to move beyond steady-state RANS. You can predict noise, vibrations, and complex aerodynamic flows. In case you are interested in outsourcing your project, you can use Order Project service to leave it to CFDLAND experts.

Advanced LES, Hybrid Models (DES), and Post-Processing in ANSYS Fluent