Sie sind auf Seite 1von 10

Proceedings of the ASME 2012 31st International Conference on Ocean, Offshore and Arctic Engineering

OMAE2012

July 10-15, 2012,Rio de Janeiro, Brazil

OMAE2012-83880

Computational Fluid Dynamics Study for Modeling Vortex Induced Vibrations Using High Fidelity Turbulence Models

Madhusuden Agrawal 1 and Mohammad A. Elyyan 1

  • 1 ANSYS, Inc.

Abstract

Flow over smooth cylinder at very high Reynolds number, Re D = 2x10 6 , is simulated using the unsteady Scale Adaptive Simulation (SAS) turbulence model. Flow structures and vortex shedding were accurately captured. Grid sensitivity study was performed to compare averaged drag coefficient for a conformal fine mesh as well as non- conformal coarse mesh. Predicted value of drag coefficient was within 8% of the experimental value and Strouhal number compared well with the experimental observations.

Introduction and Literature Survey

Vortex-induced vibration (VIV) of structures is of great practical importance in industry such as deepwater oil and gas drilling, bridges, and heat exchanger tubes. Due to its engineering importance and ease of setup, hundreds of papers have discussed the problem of flow over circular cylinder [1]. In ultra-deepwater, the suspended lengths of risers can be in excess of 15,000 feet, creating challenges in both physical testing and computer simulation used to improve their design. Without an accurate method of measuring the forces exerted on them, risers have to be designed and built to very high margins of safety, which increases

their cost and creates practical limitations on the depths at which they can be used. Moreover, it is common to have flow with Reynolds number in the

critical and post-critical regions, Re D = 2x10 5 3x10 6 , which adds to the modeling challenges,

where the boundary layer on the surface of the cylinder itself transitions to turbulence.

CFD analysis of the flow over cylinder problem phenomenon has been a rich topic for research for some time now, where a number of flow instabilities exist, i.e. wake transition, shear layer transition, and boundary layer transition. Modeling such a complex phenomenon can be challenging where turbulence model selection plays a critical role in accurately capturing the vortex shedding from the cylinder. Among the most commonly turbulence modeling techniques are the Reynolds- Averaged Navier-Stokes (RANS) models, which have known limitations in modeling such complex flows. On the other hand, Large Eddy Simulation (LES), and Detached Eddy Simulation (DES) models can accurately model the vortex shedding phenomena, but at a high computational cost even in today’s standards. Extensive experimental and numerical studies of flow past stationary cylinder have been conducted, Williamson (1996) [1], Zdravkovich (1997) [2], and Rajani et al. (2012) [3] give a good

discussion of the

physics

involved

and

a

comprehensive literature survey.

Due to modeling challenges of the laminar to turbulent transition associated with high Reynolds number flows, most of the numerical studies were mainly confined to low Reynolds number applications [3]. Some of the high Reynolds number flows were studied by Dong and Karniadakis, 2005, [3] who used DNS to study flow past a cylinder at Re D = 10 000, which was a significant increase in Reynolds number, compared to previous DNS studies. Catalano et al, 2003, used wall modeled LES to model flow past circular cylinder at Re D = 5x10 5 , 1x10 6 , and 2x10 6 ; although their results compared well with experiments at Re D = 5x10 5 , 1x10 6 , the computational results were inaccurate at Re D = 2x10 6 and the Reynolds number dependency is not captured. Travin et al, 1999, [5] used DES to simulate flow past circular cylinder at Re D = 5x10 4 , 1.4x10 5 , and 3x10 6 . They had partial success in capturing the laminar to turbulent flow transition, but overall agreement with drag, shedding frequency, pressure, and skin friction were reported. Rajani et al, 2009, [3] used URANS for modeling flow past 2D cylinder over the range of flow Reynolds number varying from 10 4 to 10 7 ; they reported that URANS models compared reasonably well in the subcritical flow regime, Re D <10 5 , but were inadequate in the critical and post critical regime.

In

this

paper,

we

test the effectiveness of a

relatively new turbulent model, Scale Adaptive Scaling (SAS) model in modeling the canonical

problem of flow over stationary cylinder problem in

the

challenging

post-critical

region,

Re D =2x10 6 .

Although we will analyze the instantaneous structures in this model, we will focus more on the mean quantities of the flow, i.e. mean drag coefficient, Strouhal number, mean pressure

coefficient, and mean lift coefficient, as they are of great importance to the industrial community.

Numerical Scheme:

The unsteady incompressible continuity and Navier- Stokes equations are solved using the commercial CFD code ANSYS-FLUENT:

…(1)

…(2)

Scale-Adaptive Simulation (SAS) model is used to model turbulent stresses, , in this study. SAS is an improved URANS formulation which is based on the concept of the introduction of the von Karman length scale into the turbulence scale equation [7]. The transport equation for the SAS-SST model implemented in ANSYS FLUENT are based on transforming the kL transport equation to k-(SST) approach and defined as:

…(4)

The transport equation of the SAS-SST model differs from those of the SST RANS model by the addition of the Q SAS in the transport equation for the

turbulence eddy frequency, . The additional source term, Q SAS , is given as

The SAS source term originates from a second order derivative term in Rotta’s transport equations. The model parameters in the SAS source terms are

2 = 3.51, = 2/3, C = 2.

Here L is the length scale of the modeled turbulence

…(6)

 

…(9)

Full derivation of the

SAS model

is available

in

Menter and Egorov, 2010 [8].

Second order discretization scheme is used for the pressure and momentum equations; and second order upwind scheme is used for the turbulent kinetic energy and dissipation equations. Moreover, second order implicit is used for transient formulation to allow for higher time step size.

The simulations were performed on 8 cores parallel processing machine, and it performed about 25 sec of transient simulation in one day run time.

 

Computational

Domain and Boundary

Conditions:

And the von Karman length scale, L kv , is a three- dimensional generalization of the classic boundary

In this study, we consider flow over a smooth

layer definition,

stationary cylinder of diameter, D, Figure 1, with the domain extending 15D upstream, 40D downstream

 

…(7)

of the cylinder, and 10D span-wise lengths (not shown). No-slip boundary condition with zero velocity is applied at the cylinder surface, fixed

The first derivative

is

represented in

velocity inlet, U in , is applied at the upstream and top

equation 5 by S, which is a scalar invariant of the strain rate tensor, S ij :

and bottom sides of the computational domain; pressure outlet B.C. is applied at the downstream surface of the domain. The span-wise sides of the

,

…(8)

Note, that the same S also directly participates in Q SAS , equation 4, and in the turbulence production term, P K = µ t S 2 . The second velocity derivative, U’’, is generalized to 3-D using the magnitude of velocity Laplacian:

domain are assigned as periodic.

In order to conduct grid independency of the model, the calculations were conducted using two mesh resolutions: conformal grid with 4 million cells and non-conformal grid with 1.2 million cells. Figure 2 shows the mesh resolution for the conformal and non-conformal grid layouts. In both mesh resolutions, the first grid height was kept at 2x10 -6 m, which gives a y 1 + < 1.0.

Table 1 shows the grid independency study results, where both mesh resolutions where considered. Flow field was first solved using steady settings with k-w SST turbulence model, then unsteady calculations were conducted using time step of 0.01 sec. Calculations were run until lift and drag coefficients exhibited statically stationary behavior for long enough time.

We compare the drag coefficient value for both coarse and fine mesh cases with that of the experimental results obtained from Roshko, 1962, [9] C D-exp = 0.50. Both grid resolutions show reasonable comparison with experiments with the maximum difference less than 8%. We considered the mean drag coefficient as it is the most available variable from experiments and most used in design. We also calculated the Strouhal Number, S, for both coarse and fine mesh cases. There is wide range of experimental data for Strouhal Number at Re D =2x10 6 , experimental results obtained from Roshko, 1962, [9], indicate Strouhal Number (S) of 0.35-0.45 at Re D =2x10 6 . Table 1 shows calculated Strouhal numbers for coarse and fine mesh cases. Our results compared very well with these numbers, as expected, fine mesh case predicts closer results. Although the coarse mesh resolution show comparable result to that of fine mesh, we will only discuss the results of finer mesh resolution in this study.

Table 1: Grid independency study

Model

Grid

C

Davg

C

D -

%Diff

S

 

Exp

Fine

4x10 6

0.519

0.50

 
  • 3.80 0.302

Mesh

Coarse

1.2

0.4622

0.35-

 
  • 7.56 0.265

Mesh

x10 6

0.45

Results and Discussion:

Figure 3

shows the line contours of the velocity

magnitude for fine mesh case at Re D =2x10 6 close to

the cylinder wall. Flow separation at about 105 o is clear from the velocity line contours. The vorticity contours in Figure 4 show the intense vortex shedding from the cylinder and how vortices dissipate as they convect downstream of the cylinder.

Figure 5 shows the iso-surfaces of the turbulent coherent structures using the Q-criterion, Q iso-surface = 0.001. The coherence structures show high level of turbulence available in the wake of the cylinder which is expected at this high Reynolds number. Periodic shedding of horse-shoe vortices can also be seen from Figure 5. The coherent structures not only appear in the wake of the cylinder, but also very close to the surface, indicating transition to turbulence and confirming the suitability of the SAS turbulence model for modeling such complex phenomenon.

Figure 6 shows distribution of static pressure on the surface of the cylinder. The waviness in pressure distribution on the leeward side of the cylinder, in the approximate range φ = 100 0 - 115 0 , also noticed in this figure which is explained by the oscillation of secondary eddies on the periphery of the cylinder wall. Figure 7, shows x-y plot of C p distribution on the cylinder From the Cp distributions, which is in good agreement with other published data [6].

Figure 8 shows the non-dimensional frequency

spectra of the lift coefficient against the non- dimensional frequency. As clear from this plot, the peak value power spectrum is obtained at 0.302

value of fD/U , which falls easily within the

reported experimental results [2, 9].

Summary and Conclusions:

The

Scale

Adaptive

Simulation,

SAS,

turbulence

model was used to model the flow over cylinder at

very high Reynolds number, Re D =2x10 6 .

Averaged

drag

coefficient

show

good

match

with

experimental values for the coarse as well as fine

mesh

resolutions

with

maximum

percentage

difference

less

than

8%.

Calculated

Strouhal

number

matches

well

with

experimental

observations.

 

Velocity and vorticity show that the SAS model accurately captured flow separation and vortex shedding from the cylinder surface. Coherent structures show high level of turbulence in the wake of cylinder, expected at this high Reynolds number.

References:

  • 1. Williamson, C., Vortex dynamics in the cylinder wake. Annu. Rev. Fluid Mech., 1996, vol. 28, pp. 477-539.

  • 2. Zdravkovich, M., Flow around circular cylinders. Oxford Science Publications, VI, 1997.

  • 3. Rajani, B., Kandasamy, A., Majumdar S., On the Reliability of Eddy Viscosity Based Turbulence Models in Predicting Turbulent Flow past a Circular Cylinder Using URANS Approach.

Journal of Applied Fluid Mechanics, Vol. 5, No. 1, pp. 67-79, 2012.

  • 4. Dong, S., Karniadakis, G. E., DNS of flow past a stationary and oscillating cylinder at Re=10 000. J. Fluid and Structures, vol. 20, pp. 519-531,
    2005.

  • 5. Travin, A., Shur, M., Strelets, M., Spalart, P., Detached-eddy simulations past a circular cylinder. Flow, Turbulence and Combustion, vol. 63, pp. 293-313, 1999.

  • 6. Catalano, P., Wang, M., Iaccarino, G., Moin, P., Numerical simulation of the flow around a circular cylinder at high Reynolds numbers. Int. J. Heat Fluid Flow, vol. 24, pp. 463-469, 2003.

  • 7. ANSYS 14.0 User Manual, 2011, Canonsburg, PA, USA.

  • 8. Menter, F., Egorov, Y., The scale-adaptive simulation method for unsteady turbulent flow predictions, part 1: theory and model description. Flow Turbulence Combust, vol 85, pp. 113-138, 2010.

  • 9. Roshko, A., Experiments on the flow past a circular cylinder at very high Reynolds number, Journal of Fluid Mechanics, vol. 10, pp. 345-356,
    1961.

Figure 1: Computational domain schematic

Figure 1: Computational domain schematic

(a) (b) Figure 2: Mesh resolution for: a) conformal grid model, Fine Mesh; b) non-conformal grid
(a) (b)
(a)
(b)

Figure 2: Mesh resolution for: a) conformal grid model, Fine Mesh; b) non-conformal grid model, Coarse Mesh

Figure 3: Velocity magnitude contours for fine mesh case, Re = 2x10 Figure 4: Contour lines

Figure 3: Velocity magnitude contours for fine mesh case, Re D = 2x10 6

Figure 3: Velocity magnitude contours for fine mesh case, Re = 2x10 Figure 4: Contour lines

Figure 4: Contour lines of vorticity magnitude for fine mesh case and Re D =2x10 6

Figure 5: Iso-surfaces of Q criterion at Q=0.001 for fine mesh case and Re =2x10 (contours

Figure 5: Iso-surfaces of Q criterion at Q=0.001 for fine mesh case and Re D =2x10 6 (contours of velocity magnitude)

Figure 5: Iso-surfaces of Q criterion at Q=0.001 for fine mesh case and Re =2x10 (contours

Figure 6: Contours of Static Pressure on Cylinder Surface for fine mesh case and Re D =2x10 6

Figure 7: Mean pressure distribution on the cylinder for fine mesh case and Re =2x10 Figure

Figure 7: Mean pressure distribution on the cylinder for fine mesh case and Re D =2x10 6

Figure 7: Mean pressure distribution on the cylinder for fine mesh case and Re =2x10 Figure

Figure 8: FFT plot of lift coefficient power spectrum for fine mesh case and Re D =2x10 6