You are on page 1of 3

Performance of subgrid-scale models in large eddy simulation of a

laminar separation bubble


Francois Cadieux & Julian A. Domaradzki
Department of Aerospace & Mechanical Engineering
University of Southern California
Los Angeles, California 90089
E-mail: cadieux@usc.edu

March 4, 2015
The flow over blades and airfoils at moderate angles of attack and Reynolds numbers ranging from 10, 000
to 200, 000 often undergoes separation-induced transition and forms a laminar separation bubble due to
the adverse pressure gradient generated by surface curvature. This phenomenon significantly impacts the
aerodynamic performance of small UAVs and low pressure turbines. Fast and accurate CFD prediction
tools are sought at the design stage to produce more efficient airfoil shapes, test flow control schemes,
and better predict high cycle fatigue (HCF) for turbomachinery blades. To avoid numerical issues associated with body-fitted meshes, an equivalent problem is formulated on a flat plate (figure 1) by imposing
boundary conditions that lead to a pressure distribution and Reynolds number that are similar to those on
airfoils (Spalart & Strelets, J. Fluid Mech., 403, 2000). Large eddy simulation (LES) of laminar separation
bubble flow over a flat plate are performed with different subgrid-scale (SGS) models at drastically reduced
resolution 1% of direct numerical simulation (DNS) resolution for a fast time-to-solution and are found
to improve on Reynolds-averaged Navier-Stokes (RANS) predictions when compared to DNS data for key
metrics such as time-averaged pressure and skin friction shown in figure 2a and 2b. The dynamic Smagorinsky model (DSM), the -model, and the truncated Navier-Stokes model with automatic filtering (TNS-A)
show remarkable agreement with the DNS benchmark whereas the no-model simulations predictions do
not because of excessive energy accumulating at the small scales contaminating the solution due to the
lack of numerical dissipation. The performance of the models is delineated in their ability to capture mean
velocity and RMS fluctuation profiles seen in figure 3, as well as their ability to predict the reattachment
point and shape factor. TNS-A demonstrates the best agreement and is the least costly, followed closely
by the -model whereas DSM is nearly twice as costly due to its extensive use of test-filters.

Figure 1: Physical domain, boundary and inlet conditions used to investigate laminar separation bubble
flow.

0.5

x 10

0.4

Cp

Cf

0.3

0.2

0.1

4
1

(a)

4
5
x

Cp = pp
1
2 .
2 U

(b) Cf =

4
x

5
U
y
1
2
2 U

Figure 2: Time-averaged Cf and Cp at the wall for LES at 1% of DNS resolution. Circles: Spalart &
Strelets (2000) DNS; dotted line: under-resolved DNS; dashed line: -model ; dash-dotted line: dynamic
Smagorinsky; line: truncated Navier-Stokes with automatic filtering.

0.2

0.18

0.18

0.16

0.16

0.14

0.14

0.12

0.12

0.1

0.1

0.2

0.08

0.08

0.06

0.06

0.04

0.04

0.02

0.02

0
0

0.5

1.5

0.5

1.5

2.5

3.5

4.5

2.5

3.5

4.5

(a) TNS-A.

0
0

0.5

1.5

2.5

3.5

4.5

(b) DSM.

0.2
0.18
0.16
0.14

0.12
0.1
0.08
0.06
0.04
0.02
0
0

(c) -model.

(d) DNS [1].

Figure 3: Line: U ; dashed line: u0 /u0max ; dash-dotted line: |Uy |/Uy,max . From left to right, plots spaced
by 1.2: x = 2, x = 2.5, x = 3, x = 3.5. [1] Spalart & Strelet (2000)

You might also like