CFD in built environment: Assessment of turbulence models

CFD simulations of flow over urban built environment is commonly used for wind climate assessments. These simulations can also be used for wind load calculations on buildings or on structures/equipment mounted on roofs/facades. However, the flow around a blunt object (like a building) is strongly dependent on the turbulence modelling. The present study compares several commonly used turbulence modelling approaches, presents the pros and cons of different modelling approaches and the best turbulence modelling practices for various applications.

To test various turbulence models for flow over a building, a standard scaled down case of a building, i.e. surface mounted cube, has been used. Experimental data for such a case is widely available in scientific literature, making it convenient to compare the results obtained from CFD simulations.

Several commonly used turbulence models for external flow in the built-environment have been tested:

  • RNG: This is a k-e based turbulence model, derived using renormalized group theory. This refines and improves the standard k-e in case of rapidly strained flows.
  • SST: This is a hybrid model with both k-omega and k-e. The k-omega model is well suited for simulating flow in the viscous sub-layer. The k-epsilon model is ideal for predicting flow behavior in regions away from the wall. Therefore it can account for the transport of the turbulent shear stress and can accurately predict flow separations under adverse pressure gradients.
  • SST-RM: The SST model (like most RANS models) can underpredict turbulent stresses in separating shear layers. This can lead to lower mixing and hence longer separation zones. The SST model is capable of predicting the separation onset well but to improve the prediction of the separated shear layer modification are made to the original SST model. This model is called SST-RM (RM: Reattachment Modification), it introduces additional production of turbulence in the separated shear layer.
  • SAS: The Scale-Adaptive Simulation (SAS) is derived by introducing the von Karman length-scale into the turbulence scale equation. The Scale-Adaptive Simulation can then dynamically change to resolved structures in a URANS simulation, as a consequence an LES-like behavior can be observed in unsteady regions of the flow field while still keeping RANS capabilities in stable flow regions.
  • DES: This model is developed to overcome the shortcomings of RANS by using LES in certain areas away from the wall boundary layer and using RANS in the near wall regions. This allows the use of LES without having to refine the near wall mesh layer. A blending function is used in the solver which acts as a switch between RANS and LES resolved regions.

A comparison of the results obtained using the above mentioned models with experimental data is presented in the slider below.

  • A cube is mounted on a surface with top of the domain as a no slip wall, the flow enters from the left (inlet colored green), the sides (colored yellow) are given a symmetry boundary condition.
  • The flow velocity at the center line: 4 m/s and Reynolds number: 40,000
Velocity contours side
  • The velocity contours and streamlines from the side view show that DES shows best agreement in wake length to the experimental and LES results from literature. All RANS models like RNG, SST and SST-RM predict a much longer wake, typical for RANS models.
  • The streamlines and the velocity curl contours from the top view show the horse shoe vortex around the cube. It could be observed that the DES and SAS demonstrate higher vorticity compared to RANS models. However, amongst the RANS models the SST-RM model under-predicts vorticity in the horse show vortex.
Velocity profiles

Velocity profile: Line 1

  • As marked in the figure, line 1 is at 0.5*H from the windward side of the cube.
  • In the separated boundary layer close to the wall (z/H < 1.1, where z/H =1 is the wall), all turbulence models fail to predict the velocity profile.
  • For z/H > 1.1, most turbulence models predict the velocity well. However, it can be observed that SST-RM captures the profile best in the region 1.1 < z/H < 1.3. The SST-RM model introduces additional turbulence production, which helps the reattachment of the separated boundary layer on the top surface of the cube and hence produced better results.
  • The results from DES simulations do not match well to the experimental values close to the wall, this is expected as DES only uses LES away from the walls. Close to the wall a RANS model is used, also the switch between LES and RANS is based on a blending function. Hence the overall effect of a RANS model and a switch function does not help in resolving the flow near the wall accurately.

Velocity profile: Line 2

  • As marked in the figure, line 2 is at the trailing edge of the cube.
  • In the reattached boundary layer zone close to the wall (z/H < 1.1, where z/H =1 is the wall), SST-RM predicts the profile with high accuracy as it is designed to capture reattachment better.
  • For 1.1 < z/H < 1.4, all turbulence models under-predict predict the velocity well. However, it can be observed that SST-RM captures the profile with the least error.

Velocity profile: Line 3

  • As marked in the figure, line 3 is in the wake at distance H from the trailing edge of the cube.
  • In the reattached boundary layer zone close to the wall (z/H < 0.5, where z/H =0 is the surface), SAS model predicts the profile with high accuracy as it is designed to capture reattachment better. Out of the RANS models, SST-RM seems to be the most reasonable in this regards.
  • For 0.6 < z/H < 1.4, all turbulence models under-predict the velocity. However, it can be observed that SST-RM captures the profile with the least error

The following conclusion are drawn from the comparison of several turbulence models in terms of its use in built-environment:

  • Wind climate assessment simulations are generally conducted for large areas, 300-500m radius around the building of interest. At such scales the use of DES or SAS models are computationally very expensive and hence not practical. Amongst the RANS models, all models predict longer wakes but it would be desirable to capture the windward side corners more accurately in order to assess the pedestrian comfort in the high velocity regions (upstream corner) where the climate is more critical. Although the SST-RM models predict the velocity on the top of the cube and wake better, it fails to predict a realistic horse shoe vortex around the corners. For wind studies conducted for large areas, use of RNG or SST turbulence models seems to be the optimal option considering accuracy in predicting relevant aspects and computational cost.
  • Built-environment CFD modelling is often used to calculate wind loads on buildings or structures on and around the building. Based in the comparative study conducted for the surface mounted cube, several tips can be used for different scenarios and are listed below:
  • In case of objects/structures placed on the windward side of the building like a canopy, a rooftop railing or signage near the leading edge of the building, the use of RNG or SST turbulence models can provide sufficient accuracy in predicting wind loads on these objects.
  • For objects mounted on top of the building such as rooftop equipment, a more accurate velocity field can be calculated using the SST-RM turbulence model.
  • For buildings or structures in the wake of another building, the wind loads on building/structure in the wake would be highly error prone if RANS models are used. In such cases either an LES/DES/SAS simulations should be performed (if computationally feasible) or a reasonable factor of safety should be assumed to correct the under-estimated forces due to the longer wake that RANS models predict.