Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/340471846

An OpenFOAM solver for multiphase and turbulent flow

Article  in  Physics of Fluids · April 2020


DOI: 10.1063/1.5145051

CITATIONS READS
29 1,982

3 authors:

Viet Bac Nguyen Vu Do-Quoc


School of Transportation Engineering University of Illinois, Urbana-Champaign
3 PUBLICATIONS   40 CITATIONS    10 PUBLICATIONS   77 CITATIONS   

SEE PROFILE SEE PROFILE

Sang V. Pham
Hanoi University of Science and Technology
32 PUBLICATIONS   600 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Next Generation Brine Desalination and Management for Efficiency, Reliability, and Sustainability View project

Vietnam National Foundation for Science and Technology Development (NAFOSTED) under Grant No. 107.03-2016.11 View project

All content following this page was uploaded by Sang V. Pham on 16 May 2020.

The user has requested enhancement of the downloaded file.


AN OPENFOAM SOLVER FOR MULTIPHASE AND TURBULENT FLOW
Viet-Bac Nguyen, Quoc-Vu Do, Van–Sang Pham*
Department of Fluid Mechanics and Ship Engineering, Hanoi University of Science and Technology,
No. 1 Dai Co Viet Street, Hanoi 100000, Vietnam

ABSTRACT equipped meshing tool and mesh processing routines, and


In this work, an OpenFOAM-integrated numerical linkable to outside codes. In a developed version of
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

solver is developed using the Immersed Boundary OpenFOAM, FOAM-extend [12], an Immersed
Method (IBM) to model incompressible multiphase and Boundary Method was implemented, however, the library
turbulent flow over a solid obstacle. The solver combines has not been used popularly, making its reliability and
the Pressure Implicit Split Operator (PISO) algorithm performance remained questionable.
with the 𝑘 − 𝜔 Shear Stress Transport ( 𝑘 − 𝜔 𝑆𝑆𝑇 ) In this work, we put efforts for continuing to develop
model for the velocity-pressure coupling in Navier- a solver, namely IBMFoam, which combines advantages
Stokes equations and resolving turbulent flow near the of OpenFOAM platform and immersed boundary method.
obstacle’s surface. To strengthen the performance of the IBMFoam is equipped with an Adaptive Mesh
solver, an Adaptive Mesh Refinement – AMR scheme is Refinement (AMR) scheme for improving the accuracy
developed for efficient mesh using of IBM. The problem of the IBM via refining mesh cells in the supporting area
of penetration of multiphase flow through immersed of an immersed boundary [13,14]. IBMFoam is able to
boundary is considered and resolve with a method of solve turbulent flow via an IBM-based scheme for
modifying the forcing term at solid cells. The solver has turbulent variables. The solver is validated via well-
been validated via benchmark problems and applied in documented cases and used for modeling complicated
simulating complicated fluid flow problems. problems. This novel solver fills the gap in the
OpenFOAM platform, handles the fluid-structure
INTRODUCTION interaction problems such as ship hydrodynamics and
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

In Computational Fluid Dynamics (CFD), the maneuvering, ocean and coastal engineering,
utilization of body-fitted (BF) mesh is the most common development of multiphase flow under the presence of
approach in modeling various problems in creating obstacles, dynamic properties of transport vehicles, etc.
prospective designs and testing new technological In the following parts, we begin with a mathematical
solutions [1,2]. Generating BF mesh for complex model and numerical method for turbulent fluid flow.
geometry domains are challenging, especially in moving An adaptive mesh refinement is then presented and
objects problem due to time-consuming and huge following by the implementation of flow solver in
computational resources requirement [3]. Immersed OpenFOAM environment. Validations for the solver are
Boundary Method (IBM) is an alternative numerical provided with a discussion on the problem of liquid
procedure that enables to efficiently handle motions of phase penetrating across an impermeable boundary, and
complex geometry while maintaining high accuracy. IBM method to resolve the problem.
was first proposed by Peskin to simulate the blood flow
in the mitral valve and the heart at two-dimensional (2D) NUMERICAL METHOD
flow and a low Reynolds number [4]. In his method, the Governing equations
desired velocity is imposed over the surface of objects by Considering the Navier-Stokes equations for the
using a body force field. IBM has been subsequently incompressible, unsteady, viscous fluid flow:
developed by many researchers, Fadlun et al. [5], 𝜕𝐔
𝜌 + 𝜌𝐔 ⋅ 𝛁𝐔 = −𝛁𝑃 + 𝜇𝛁2 𝐔 + 𝜌𝐟 𝐢𝐛 (1)
Constant et al. [6], and Uhlmann [7] introduced a direct 𝜕𝑡
formulation of the force term. 𝛁⋅𝐔=0 (2)
In our previous work, Pham [8], Do et al. [9,10], we where 𝜌, 𝜇, 𝐔, 𝑡, 𝑃, 𝐟 𝐢𝐛 denote the density, dynamic
developed an IBM solver in OpenFOAM open-source viscosity, velocity of fluid, time, pressure, and volume
framework [11]. The solver makes use of the direct force field, respectively.
forcing approach and has shown its robustness in dealing Motion of objects in fluid-flow force is governed by
with the laminar flow around solid obstacles and the Newton’s equations:
interaction of multi-objects. However, the lack of an 𝑑𝐔𝐩
(𝜌𝑝 − 𝜌𝑓 )𝑉𝑝 = (𝜌𝑝 − 𝜌𝑓 )𝑉𝑝 𝐠 − 𝐅 (3)
appropriate turbulent model limits its use to small 𝑑𝑡
Reynolds number flows only. In this work, this limitation 𝑑𝛚𝐩
𝐼𝑝 =𝜏 (4)
will be addressed as a 𝑘 − 𝜔 Shear Stress Transport (𝑘 − 𝑑𝑡
𝜔𝑆𝑆𝑇) turbulent model, enabling the solver to cope with where𝑉𝑝 , 𝐼𝑝 , 𝛚𝐩 , 𝜌𝑝 , 𝜏are the volume, moment of inertial,
high Reynolds number flows. angular velocity, density and torque vector of object;  f
OpenFOAM is a brilliant and well known CFD open
is the fluid density; 𝐠 is the vector of gravitational
source with numerous advantages, such as no license cost,
acceleration; 𝐅 is hydrodynamic force.
flexible code customization, fully MPI parallelization,
Turbulent model
*
Corresponding author: sang.phamvan@hust.edu.vn
We make use of the most popular approach for To compute turbulent viscosity, we choose a two-
turbulent flow problems - RANS model. The key idea of equation turbulence model 𝑘 − 𝜔 𝑆𝑆𝑇 which is the most
this model is decomposing the flow variables into time- popular application in industry. This turbulence model is
mean value component ( 𝐔, 𝑝) and fluctuating employed by Menter [16]. It combines the advantages of
components (𝐔 ′ , 𝑝′ ): the 𝑘 − 𝜀 turbulence model for the free-stream flow and
the standard 𝑘 − 𝜔 turbulence model by Wilcox [17] for
𝐔 = 𝐔 + 𝐔 ′ ; 𝑃 = 𝑝 + 𝑝′ (5)
the boundary layer flow. The 𝑘 − 𝜔 𝑆𝑆𝑇 model
The fluctuating quantities are assumed to be zero employed in OpenFOAM v5.0 [11] is based on a blend of
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

mean value. high Reynolds number versions of 𝑘 − 𝜀 and 𝑘 − 𝜔


̅̅̅
𝐔′ = 𝑝̅′ = 0 (6) turbulence models. It implies that the 𝑘 − 𝜔 𝑆𝑆𝑇 model
Substituting these components into the Navier-Stokes is built on the assumption that the flow field is fully
and the continuity equations, and taking the time average turbulent, and therefore the turbulence model has no
to obtain the Reynolds averaged forms [15]: damping function for low Reynolds number flows.
However, as for the Reynolds numbers considered in this
𝜕𝑢𝑖 𝜕𝑝 𝜕 𝜕𝑢𝑖 𝜕𝜌𝑢𝑖 ′𝑢𝑗 ′ paper, the boundary layer on the surface of the obstacle
𝜌 + 𝜌𝑢𝑖 ⋅ 𝛁𝑢𝑖 = − +𝜇 ( )− transits from laminar to turbulent. Thus, proper wall
𝜕𝑡 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗
𝐢𝐛
functions are needed to solve the turbulence kinetic
+𝜌𝐟 (7)
energy 𝑘, specific turbulence dissipation 𝜔, and turbulent
The above RANS equations introduce six additional viscosity 𝜈𝑡 . The governing equations of the 𝑘 − 𝜔 𝑆𝑆𝑇
unknowns (the components of the tensor 𝜌𝑢𝑖 ′𝑢𝑗 ′, known model are as follows:
as the Reynolds stress tensor 𝜏 𝑅 ) to the momentum
𝜕𝜌𝑘 𝜕(𝜌𝑢𝑗 𝑘) 𝜕 𝜕𝑘
equations. The expended forms of the 𝜏 𝑅 is given by: + = 𝑃𝑘 − 𝛽 ∗ 𝜌𝜔𝑘 + [(𝜇 + 𝜎𝑘 𝜇𝑡 ) ]
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗
𝑢′𝑢′ 𝑢′𝑣′ 𝑢′𝑤′
𝑘
𝜏 𝑅 = −𝜌𝑢𝑖 ′𝑢𝑗 ′ = −𝜌 ( 𝑣′𝑢′ 𝑣′𝑣′ 𝑣′𝑤′ ) (8) +𝑓 (15)
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

𝑤′𝑢′ 𝑤′𝑣′ 𝑤′𝑤′ 𝜕𝜌𝜔 𝜕(𝜌𝑢𝑗 𝜔) 𝛾 𝜕 𝜕𝜔


+ = 𝑃𝑘 − 𝛽𝜌𝜔2 + [(𝜇 + 𝜎𝜔 𝜇𝑡 ) ]
To close the system of equations, Reynolds stress 𝜕𝑡 𝜕𝑥𝑗 𝜈𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗
tensor is calculated using a turbulence model. Several
𝜌𝜎𝜔2 𝜔
turbulence models have been proposed based on the +2(1 − 𝐹1 ) 𝛁𝑘: 𝛁𝜔 + 𝑓 (16)
Boussinesq hypothesis. The Reynolds stress tensor is 𝜔
𝑘 𝜔
assumed to be a linear function of the mean velocity where 𝑓 and 𝑓 is body force field depicts interaction
between solid object and turbulent flow.
gradients and has been expressed as  R :
T 2 𝛷 = 𝐹1 𝛷1 + (1 − 𝐹1 )𝛷2 (17)
𝜏 𝑅 = −𝜌𝑢𝑖 ′𝑢𝑗 ′ = 𝜇𝑡 {𝛁𝐔 + (𝛁𝐔) } − 𝜌𝑘𝐈 (9)
3 The empirical constants ( 𝜎𝑘 , 𝜎𝜔 , 𝛾, 𝛽 ) of the 𝑘 −
where k is the turbulent kinetic energy defined by: 𝜔 𝑆𝑆𝑇 model are represented by 𝛷 which is computed
1 from Eq. (17) using 𝐹1 function and the following model
𝑘 = 𝑢𝑖 ′𝑢𝑖 ′ (10) constants [16]:
2
𝜎𝑘1 = 0.85, 𝜎𝜔1 = 0.5, 𝛾1 = 𝛽1 /𝛽 ∗ − 𝜎𝜔1 𝜅 2 /√𝛽 ∗
𝐈 is the identity matrix:
(18)
𝜎𝑘2 = 1.0, 𝜎𝜔2 = 0.856, 𝛾2 = 𝛽1 /𝛽 ∗ − 𝜎𝜔2 𝜅 2 /√𝛽 ∗
1 0 0
𝐈 = [0 1 0] (11) ∗
𝛽1 = 0.075, 𝛽2 = 0.0828, 𝛽 = 0.09
0 0 1 (19)
𝜅 = 0.41, 𝑎1 = 0.31
 t the turbulent eddy viscosity (in analogy with
molecular viscosity  ). Substitute Reynolds stress 𝐹1 and 𝐹2 are blending functions:
4
tensor into Eq. (7) we have: √𝑘 500𝜈 4𝜌𝜎𝜔2 𝑘
𝐹1 = 𝑡𝑎𝑛ℎ ((𝑚𝑖𝑛 [𝑚𝑎𝑥 ( ∗ , ), ]) )
𝜕𝑢𝑖 𝜕𝑝 𝜕 2 𝛽 𝜔𝑦 𝑦 2 𝜔 𝐶𝐷𝑘𝜔 𝑦 2
𝜌 + 𝜌𝑢𝑖 ⋅ 𝛻𝑢𝑖 = − + [2(𝜇 + 𝜇𝑡 )𝐒𝑖𝑗 − 𝜌𝑘𝐈]
𝜕𝑡 𝜕𝑥𝑖 𝜕𝑥𝑗 3 (20)
𝐢𝐛 (12)
+𝜌𝐟 1 𝜕𝑘 𝜕𝜔
𝐶𝐷𝑘𝜔 = 𝑚𝑎𝑥 (2𝜌𝜎𝜔2 , 10−20 ) (21)
where 𝑺𝑖𝑗 is the mean strain rate tensor: 𝜔 𝜕𝑥𝑗 𝜕𝑥𝑗
1 𝜕𝑢𝑖 𝜕𝑢𝑗
𝐒𝑖𝑗 = ( + ) (13) where 𝑦 is the distance from each computational cell to
2 𝜕𝑥𝑗 𝜕𝑥𝑖 the nearest wall.
For incompressible flows the term −2/3𝜌𝑘𝐈 in the The turbulent viscosity in the 𝑘 − 𝜔 𝑆𝑆𝑇 is calculated
Reynolds stress is usually combined with the pressure by:
gradient term by defining a turbulent pressure 𝑝̅ as: 𝜌𝑎1 𝑘
2 𝜇𝑡 = (22)
𝑝̅ ← 𝑝̅ + 𝜌𝑘 (14) 𝑚𝑎𝑥( 𝑎1 𝜔, √2𝑆𝐹2 )
3
where 𝑆 is the magnitude of the strain rate defined in Eq.
(13), and 𝐹2 is given by:
  2 k 500 
2

F2 = tanh   max  * , 2    (23)
   y y   
 
1 𝜕𝑢𝑖 𝜕𝑢𝑗 1 𝜕𝑢𝑖 𝜕𝑢𝑗
𝑆 = √( ( + )) ( ( + )) (24)
2 𝜕𝑥𝑗 𝜕𝑥𝑖 2 𝜕𝑥𝑗 𝜕𝑥𝑖

Definition of the forcing term


This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

The IBM developed in this work, firstly reported in Figure 1. Quadtree discretization and its corresponding tree
[5,6], has the surface of a solid object represented by a representation where the original hexahedral mesh cell (parent
set of Lagrangian points. Mutual interactions between cell) is subdivided into four smaller hexahedral cells (children
the object and fluid flow are modeled via a forcing field cells).
𝐟 𝐢𝐛 added into the momentum equation (12). The force
field is calculated to reproduce the desired velocity Adaptive Mesh Refinement
value on the object surface. OpenFOAM provided a robust adaptive meshing tool
After solving for the intermediate velocity field, the that works perfectly with 3D simulations but does not
forcing term 𝐅 𝐢𝐛 on the object surface is calculated from support 2D problems. Thus, we modified the original tool
the predicted velocity (𝐔 𝐥𝐠 ) and desired velocity (𝐔 𝐝 ) at based on the AMR technique described in [20] to obtain
the object surface: a new dynamic refinement mesh library for simulating
two-dimensional planar and axisymmetric geometries.
𝐔𝐝 − 𝐔𝐥𝐠 The library can locally refine mesh at the regions of
𝐅 𝐢𝐛 = (25)
𝛥𝑡 interest, e.g., liquid-air interface and surface of the
immersed objects. The technique splits an original
The 𝐅 𝐢𝐛 is then transferred back to the Eulerian mesh
hexahedral mesh element (parent cell) into four and eight
via a transferring function to obtain forcing term 𝐟 𝐢𝐛 [7].
smaller hexahedral elements (children cells) in 2D and 3D,
The forcing term 𝐟 𝐢𝐛 then is added to the RHS of Eq.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

respectively. The data structure is organized


(12) to resolve the final velocity field. Mass flux velocity hierarchically as a quadtree (octree in 3D), with root cell
field and pressure are then corrected via correction is the base of the tree and leaf cells are cells without any
procedures. child. Each cell has its corresponding level, which starts
The wall function (Bredberg [18]) is chosen. If the from zero for all elements at the beginning of the
turbulent model is activated, the solver computes body simulation. The level of the children is then determined
force term 𝑓 𝑘 and 𝑓 𝜔 which is later added to the RHS of by adding one to their parent’s level. A schematic
the Eq. (15) and Eq. (16), respectively, to resolve for illustrating the spatial discretization and corresponding
turbulence kinetic energy and specific dissipation rate: tree representation is shown in Fig. 1.
This new library is compiled independently and can
kd − klg be integrated it into other solvers for specific purposes. As
𝑓𝑘 = (26)
𝛥𝑡 shown in [10], using AMR reduce computation time in
ωd − ωlg comparison with fixed mesh case.
𝑓𝜔 = (27)
𝛥𝑡 IMPLEMENTATION IN OPENFOAM
We use fixed value condition for the turbulence We developed the solver based on the pisoFoam, an
kinetic energy 𝑘 = 0 at the surface of immersed object. available transient solver in OpenFOAM that uses the
The value of specific turbulence dissipation 𝜔 is PISO algorithm [15] for pressure-velocity coupling. Fig.
calculated through: 2 describes our algorithm that combining IBM and AMR
2
𝜔 = √𝜔𝑣𝑖𝑠 2
+ 𝜔𝑙𝑜𝑔 (28) into pisoFoam. In the figure, the black blocks denote the
original procedures of pisoFoam, whereas the red blocks
in which, 𝜔𝑣𝑖𝑠 and 𝜔𝑙𝑜𝑔 are specific turbulence dissipation represent our additional procedures responsible for
in viscous and logarithmic regions, respectively. 𝜔𝑣𝑖𝑠 and running IBM and AMR related tasks.
𝜔𝑙𝑜𝑔 are expressed as: The solver tracks cells surrounding Lagrangian
points to compute all indicators which represent the
6𝜈 𝑘 0.5 region of interest. Information about those cells is
𝜔𝑣𝑖𝑠 = 𝛽 2 ; 𝜔𝑙𝑜𝑔 = 𝐶 0.25𝜅𝑦 (29)
1 𝑦𝑝 𝜇 𝑝
imported into the new dynamicFvMesh library. This
library checks all marked cells and the indicators, splits
where 𝛽1 is constant with value of 0.075, 𝐶𝜇 is constant
valid cells, and updates the value of all variables for
with value of 0.09, 𝜅 is Von Karman constant with value
newborn cells.
of 0.41, 𝑦𝑝 is the distance of the first cell center to the
We compute the body force term as describe above,
wall [19]. use PISO algorithm to solve the pressure equation and
correct the velocity field. After that, the solver checks the
turbulent properties, if the turbulent flow is activated; the
solver runs the code to calculate turbulent viscosity. This
term is determined by using Eq. (22). First, we compute
𝑘 − 𝜔 𝑆𝑆𝑇 equations for the first time. Then, we compute boundaries, adaptive meshing is off. For pisoFoam, as
body force term 𝑓 𝑘 and 𝑓 𝜔 by using Eq. (26) and Eq. well validated in literature [21], we used orthogonal
(27). Finally, we resolve 𝑘 − 𝜔 𝑆𝑆𝑇 to obtain the mesh with clustering of mesh points near the solid upper
desired value of turbulence kinetic energy and specific and lower wall. A uniform velocity of 𝑈∞ = 1𝑚/𝑠 is
dissipation rate. The process continues for the next time driving into the channel at the inlet boundary,
step. corresponding Reynolds number is 20000. We denote
the cases of running with IBMFoam are M1 (coarse
mesh) and M2 (fined mesh), the case of pisoFoam is M3.
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

The mesh parameters are shown in Table 1.

Figure 3. The modeling domain of 2D turbulent flow in a


straight channel with 2cm in width and 20cm in length. The
domain is discretized using orthogonal mesh. No-slip wall
condition is enforced at the top and bottom boundary.

Mesh M1 M2 M3
y min
p 15.8 7.88 0.788
Total cells 9600 32500 14400
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

Table 1: Meshs for wall functions study.

Figure 2. Flow chart of the IBMFoam algorithm. The black


blocks denote the original procedures of pisoFoam, whereas the
red blocks represent developed procedures responsible for
executing IBM and AMR tasks.

RESULTS AND DISCUSSION


In this part, the IBMFoam solver will be validated
using several well-documented cases. We first consider Figure 4: The turbulent velocity profiles in the straight
the case of turbulent flow in a 2D strait channel. The channel for the M1, M2, and M3 cases. These profiles
represent the power-law velocity profile for Newtonian
results obtained with IBMFoam will be compared with
fluid which are significantly flatter than the paraboloidal
the validated ones from native pisoFoam solver. The shape of laminar flow in the literature (Dodge and Metzner
turbulent characteristic is continued validated with the [22]).
case of high Reynolds number flow over a cylinder.
IBMFoam is then validated with two cases of two-phase
flow, the dam-break problem with obstacle blocking, and Fig. 4 shows the time-average velocity profile of the
the water exit problem. turbulent flow in the straight channel. All profiles
represent correctly the power-law velocity profile for a
Turbulent flow in strait channel Newtonian fluid. The turbulent profiles are significantly
To validate the accuracy of IBMFoam when resolving flatter than the paraboloidal shape of laminar flow in the
turbulent flow, we are going to compare the results of literature [22]. That means the immersed walls generate
turbulent flow obtained with IBMFoam, and turbulent good results as the solid walls.
flow results obtained with the body-fitted mesh solver, Fig. 5 shows the profiles of turbulence kinetic energy,
pisoFoam. Considering turbulent flow in a 2D straight specific turbulence dissipation and turbulent viscosity
channel, which is 2𝑐𝑚 in width and 20𝑐𝑚 in length. The measured across the channel at 𝑥 = 5𝐻. The 𝑘, 𝜔, and
channel is discretized using orthogonal mesh. For 𝜈𝑡 converge with refining the mesh.
IBMFoam, both coarse and a fined meshes are tested,
the top and bottom walls are modeled with immersed
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

Figure 6. The root mean squared, mean velocity,


dimensionless velocity, and 𝑦 + profiles across the channel
at 𝑥 = 5𝐻 for the M1 (circle marker line), M2 (square
marker line), M3 (solid line) cases, Moser et al. [23] (dash
line), Ma et al. [24] (triangular marker line), and Goldstein
et al. [25] (cross marker line).
Figure 5. The profiles of turbulence kinetic energy, specific
turbulence dissipation and turbulent viscosity measured
across the channel at 𝑥 = 5𝐻: IBMFoam (circle marker-
M1, rectangle marker-M2) and pisoFoam (black line-M3).

In Fig. 6, the root mean squared, mean velocity,


dimensionless velocity, and 𝑦 + profiles across the
channel at 𝑥 = 5𝐻 are shown. As can be seen, the result
represents correctly Law of the wall (Fig. 6a-6b). From
Fig. 6c we see that the dimensionless velocity profile,
𝑢+ , is the same as that obtained by Moser et al. [23], Ma
et al. [24], and Goldstein et al. [25], both for the bulk of Figure 7. The orthogonal computational mesh and immersed
the flow and for the flow close to the wall. The 𝑦 + and boundary representation of 2D cylinder.
𝑢+ is expressed as:
𝑦𝑝 𝐶𝜇0.25 √𝑘𝑝 1
𝑦+ = ; 𝑢+ = 𝜅 𝑙𝑛( 𝐸𝑦 + ) (30) Flows around a fixed smooth circular cylinder at Re
𝜈
of 63100
where 𝐸 is constant with value of 9.8. Flow around a fixed smooth circular cylinder is
chosen for validation. The cylinder has a diameter of 𝐷 =
0.206𝑚 located in, as sketched in Fig. 7, a water flow
with 10𝐷 𝑎𝑛𝑑 20𝐷 of height and length, respectively. At
the inlet, a steady uniform velocity, 𝑈𝑖𝑛 is imposed along
with a zero-pressure gradient condition. A mass
conservation condition is imposed at the outlet. Free-flow
boundary condition is enforced at the top and bottom
boundary. The cylinder is represented by 64 and 128
Lagrangian points according to M1 and M2 mesh, Table 3 shows the results for average and root-mean-
respectively (Table 2). square coefficients of lift and drag forces acting onto the
Simulation is carried out for the Reynolds number, cylinder and Strouhal number in comparison with other
𝑅𝑒 = 𝑈𝑖𝑛 𝐷/𝜈 = 63100 . The value of 𝑦 + is shown in experimental and numerical results. As can be seen, the
Table 2. In addition, the residual criteria of all physical obtained results are in good agreement with the published
quantities are set to be 1𝑒 − 6. results [26,27,28]. In comparison to other modeling
The drag coefficient 𝐶𝐷 , lift coefficient 𝐶𝐿 and results [26,27], while the average lift force and Strouhal
Strouhal number 𝑆𝑡 are investigated in this paper and number in this study are closer to the experimental data
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

their definitions are shown in Eq. (31) [28], the deviation on drag force coefficient is higher, this
might due to the coarser used in this study (32500 cells in
𝐅𝐷 𝐅𝐿 𝑓𝑠 𝐷
𝐶𝐷 = 2 𝐷 ; 𝐶𝐿 = 2 𝐷 ; 𝑆𝑡 = (31) comparison to 96785 cells [26] and 11.3 million cells
0.5𝜌𝐔𝑖𝑛 0.5𝜌𝐔𝑖𝑛 𝐔𝑖𝑛
[27]).
in which, 𝐅D , 𝐅L and 𝑓𝑠 are the drag force, lift force, and Fig. 9-11 show the comparisons between the contours
vortex-shedding frequency, respectively. of velocity, vorticity and 𝐶𝑝 at 𝑅𝑒 = 63100. The results
are good agreement with the published results [26]. It,
Mesh M1 M2 therefore, demonstrates that the combination of the IBM
𝒚+ 88.64 44.32 approach and the present numerical algorithm can well
Number of Lagrangian points capture important global flow features in the critical flow
64 128 regime.
representing cylinder
Total cells 20000 32500
Table 2: Meshes size and corresponding 𝑦 + and number of
Lagrangian points used for the validation of flows around a
fixed smooth circular cylinder.

Ye Lee Wilde and


PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

and and Huijsmans


M1 M2
Wan Yang [28]
[26] [27]
𝑪𝑫,𝒂𝒗𝒆 1.59 1.56 1.1157 1.09 1.04
𝑪𝑳,𝒂𝒗𝒆 0.0406 0.022 0.0053 0.0335 0.0145
𝑪𝑳,𝒓𝒎𝒔 0.7636 0.7283 0.8285 0.179 0.24
Figure 9. The contours of the normalized velocity
𝑺𝒕 0.224 0.221 0.2385 0.278 0.195
magnitude for θ = 0 and 𝑅𝑒 = 63100 . The separation of
Table 3: The average and root-mean-square coefficients of lift the shear layer can be seen at the point at which the
and drag forces acting onto the cylinder and Strouhal number separation occurs moves upstream. The narrow wakes
in comparison with other experimental and numerical results behind the cylinder are formed as same as the results in
published by Ye and Wan [26], Lee and Yang [27], and Widle [26].
and Hujismans [28]. The drag coefficient 𝐶𝐷 and the lift
coefficient 𝐶𝐿 over time using meshes M1 and M2 at 𝑅𝑒 =
63100 (Fig. 8).

Figure 10. The contours of pressure coefficient, 𝐶𝑝 , for


𝜃 = 0 at 𝑅𝑒 = 63100. The picture is in good agreement
with the result obtained in [26]

Figure 8: Lift (a) and drag (b) force coefficient - 𝐶𝐿 , 𝐶𝐷 for


meshes M1 and M2.
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

Figure 12. Simulation model of the dam break problem.

The volume of fluid (VOF) method is utilized in


resolving two-phase flow. A volume fraction indicator
function  is used to mark the individual fluids and the
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

interface is defined as the transitional region between the


fluids. A unity value represents a cell full of fluid; a zero
Figure 11. The contours of vorticity for Re = 63100 . The
value represents a cell full of air, whereas cells with a
vortex shedding for a shedding period agrees well with the
value between zero and one contain a free surface.
published results [26]. Considering two immiscible fluids as one single effective
fluid throughout the domain, its physical properties can
be defined as a weighted average:
Dam break with an obstacle in the flow 𝐔 = 𝛾𝐔𝐥 + (1 − 𝛾)𝐔𝐠
Our work contributes to the field by presenting a
solver capable of capturing a moving interface between 𝜌 = 𝛾𝜌𝑙 + (1 − 𝛾)𝜌𝑔 (32)
two immiscible fluids on an arbitrary Eulerian mesh. This 𝜇 = 𝛾𝜇𝑙 + (1 − 𝛾)𝜇𝑔
enables the modeling of large deformations of interfaces
The transport equation of the indicator function is
and to predict the flow of immiscible fluids. The merging
solved together with the mass and momentum
or breakup of the interface is also handled in a natural way.
conservative equations:
Especially, this validation proves our solver has the
ability to interact with the incoming fluid by using 𝜕𝛾
+ 𝛻 ⋅ (𝐔𝛾) + 𝛻 ⋅ [𝐔𝐫 𝛾(1 − 𝛾)] = 0 (33)
immersed obstacles. In other words, our solver is able to 𝜕𝑡

deal with the velocity field with a large gradient at the The last convective term in the Eq. (33) is referred to
surface of an immersed object. The results are compared as the compressive term, with the compressive velocity
with the experimental data. defined as 𝐔𝐫 = 𝐔𝐥 – 𝐔𝐠 . This term is active only in the
The simulation model is depicted in Fig. 12 with free surface region (where γ is neither one nor zero) and
parameters that are similar to those introduced by Ubbink intends to “compress” the interface to achieve a higher
[29]. The domain has the size of [0.584 × 0.584]𝑚. interface resolution, thus avoiding the need to employ a
The dynamic viscosity and density of water are special scheme for sharp interfaces.
10−3 𝑘𝑔/𝑚𝑠 and 1000 𝑘𝑔/𝑚3 , respectively, while
those of air are 1.8 × 10−5 𝑘𝑔/𝑚𝑠 and 1 𝑘𝑔/𝑚3 . The The penetration of water phase on the immersed
gravity acceleration 𝑔 = −9.81 𝑚/𝑠 2 . No-slip boundary boundary
condition is applied to all boundaries except the top
surface which is assumed to be an open boundary. The Immersed boundary method with direct forcing can
initial mesh resolution is 25 × 25 and the free surface work well with single-phase flow and wall boundary
and the obstacle surface are refined up to the fourth level [6,7,8], however, when dealing with two-phase flow, the
depending on liquid motion during the simulation. non-zero density gradient at immersed boundary causing
the inaccurate velocity boundary condition. Zhang [30]
used a volume of solid (VOS) function which takes values
of either zero or one for cells within the fluid or the solid solid volume, with non-zero velocity in the solid volume.
region, respectively. The forcing term was specified with: With the additional forcing term, the liquid phase is
stopped effectively from passing through the
𝐔𝐬 − 𝐔𝐧
𝐟 𝐢𝐛 = 𝜂( − 𝒓𝒉𝒔) (34) impermeable surface.
Δ𝑡
The η field takes the value of 0 for fluid and 1 for solid.
𝐔𝐬 is the desired velocity of at the body surface, 𝐔𝐧 is the
calculated velocity, and 𝒓𝒉𝒔 includes the viscosity
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

terms, the pressure terms, the advection terms and the


external acceleration terms. To Zhang’s method has a
simple implementation; however, the desired velocity is
satisfied at the solid cells near the immersed boundary
rather than on the boundary. That causes the penetration
of liquid through the solid surface which is physically
impermeable.
Uhlmann [7], Constant [6], and Roma et al. [31] used
interpolation kernel 𝛿ℎ to interpolate velocity at the
immersed boundary and transfer the forcing term 𝑓𝑖𝑏
from Lagrangian mesh to Eulerian mesh (as mentioned
Figure 14. The temporal evolution of the collapsing water
above). When applying the delta function, we found that column obstructed by an obstacle. (a) The formation and
it does not work well with two-phase problems. The destruction of the water tongue. (b) The adaptive mesh at
liquid phase is penetrated through the obstacle as shown. 𝑡 = 0.3𝑠 (~30.000 cells).
To resolve this problem, we gave an improvement to
resolve this problem by improving Uhlmann’s procedure Simulation results at different time instants are shown
with additional forcing to the interior of an obstacle. We in Fig. 14a. From the beginning until 𝑡 = 0.3𝑠 the
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

applied a smooth function form to the forcing term 𝑓𝑖𝑏 collapsing water column obstructed by the obstacle
at cell centers inside the obstacle: bounces up from the upper left corner of the obstacle,
forming a tongue-shaped flow and heading to the opposite
𝐔𝐝 − 𝐔𝐢𝐧
𝐟 𝐢𝐛 = 𝛾( ) (35) wall. After the tongue reaches the right wall at 𝑡 = 0.4𝑠,
∆𝑡 water starts to fall under the effect of gravity and
The smooth function, 𝛾 , is zero at the immersed compresses the tapped air beneath it. The air bubble
boundary and increases gradually to 1 into the obstacle. eventually bursts through the water sheet while the water
𝐔𝐢𝐧 is a current velocity at the cell center. The forcing continues its downward motion due to gravity, creating a
term 𝐟 𝐢𝐛 in [6,7,8,9,10] is only active near the surface of mixing region of water and air. The phenomena observed
the object so that our solver computes supplementary in the current study are in excellent agreement with that
force for inner cells of the obstacle. produced by Ubbink [29]. Using AMR in the current
solver help to reduce a significant number of cells
(~30000 cells) in comparison with Zhang et al.’s model
(100000 cells) [30]. With fewer mesh cells, the solver is
still able to capture the clearer peak of the tongue-shaped
wave [29,30].

Water exit of a circular cylinder

Figure 13: Velocity and phase fraction field at the moment


liquid acting onto the obstacle surface. (a) Result obtained
without additional forcing term: the liquid phase penetrates
the solid volume, with non-zero velocity in the solid volume.
(b) Result obtained with the additional forcing term: the
liquid phase is stopped effectively from passing through the
impermeable surface.
Figure 15. Simulation model of the water exit problem. The
Fig. 13 shows the velocity and phase fraction field at domain size is 20𝐷 × 12𝐷, filled with water of 10𝐷 in
the moment when liquid acts onto the obstacle surface: a) depth, 𝐷 = 2𝑚 is the diameter of the cylinder.
without additional forcing term, b) with the additional
forcing term. As can be seen, without the forcing term at
cells inside the obstacle, the liquid phase penetrates the
The problem of water exit is closely related to significantly reduce 68% of the required mesh size, thus
practical applications in marine operation during which save about 59.4% simulation time compares with the
objects are either lowered into water or lifted out of the case without AMR. As can be seen in Fig. 16, the air-
water, which involves complicated physical processes water interface profiles at different dimensionless time
such as the breakup of free surface and interaction instants in the water exit problem are in excellent
between solid body, free-surface, and vortices [32,33,34]. agreement with the simulation results using the
Here, we study the influence of moving solid object on boundary element method of [36]. In addition, the
the interface between water and air. The simulation setup evolution of the free surface interacting with the rising
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

is the same as those in [35]. The diameter of object is 𝐷 = cylinder at different dimensionless time step illustrated
2𝑚. The domain size is [40 × 24]𝑚, filled with water in Fig. 17, are well-matched with the results reported in
of 20𝑚 in depth as shown in Fig. 15. The dynamic [35].
viscosity and density of water are 10−3 𝑘𝑔/𝑚𝑠
and1000 𝑘𝑔/𝑚3 , respectively, while those of air are
1.8 × 10− 5 𝑘𝑔/𝑚𝑠 and 1 𝑘𝑔/𝑚3 . The gravity
acceleration 𝑔 = 1.0 𝑚/𝑠 2 .
A uniform mesh with 𝛥𝑥 = 𝛥𝑦 = 0.8𝑚 (cell level
0) is initially used for the entire domain, while cells in
the vicinity of the free surface and the cylinder surface
are refined to the fourth level (𝛥𝑥 = 𝛥𝑦 = 0.05𝑚).
The cylinder has radius 𝑅 = 1𝑚. Its center is initially
located at 𝑑 = 1.25𝑚 above the free surface and is
given a constant velocity 𝑉 = 0.39𝑚/𝑠 for the exiting
problem. For easier to compare our results with previous
studies, the dimensionless time 𝑇 = 𝑉𝑡/𝑑 is used in our
analysis.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

Figure 16. The free surface profile of the water exit (E1),
(E2) of a circular cylinder in comparison with previous
study [36].

With AMR
Without
AMR
M1 M2 M3

Min cell size (m) 0.05 0.1 0.05 0.025


No. of elements 49280 8142 15747 36552
Computation time
8507 1137 3456 18948 Figure 17. (a) – (h) Snapshots of the water exit problem at
(s) different dimensionless time instants. The results are in
good agreement with [35].
Table 4: Computation time for different mesh refinements of
the water exit problem. CONCLUSION
The simulation of the water exit problem is In this work, an OpenFOAM integrated numerical
conducted on four different meshes to compare the solver, IBMFoam was developed. IBMFoam is based on
difference in computation time. As can be seen from the immersed boundary method and an adaptive mesh
Table 4, the AMR technique in case M2 helps refinement algorithm to solve for turbulent, two-phase
flow. The solver was validated with the well-documented Pressure (N/ Volume fraction
cases: the turbulent flow in a 2D strait channel, high 𝑃, 𝑝̅ , 𝑝′ 𝛾
m2 ) indicator
Reynolds number flow over a cylinder, the dam-break
problem with obstacle blocking, and water exit problem. Moment of Velocity of air
𝐼𝑝 𝐔𝐠
The most important contributions of this work may inertial (kg m2 ) (m/s)
include i. Providing a numerical solver for turbulent, two- Turbulent kinetic Compressive
phase flow without using body-fitted mesh; ii. Combining 𝑘 𝐔𝐫
energy (m2 /s2 ) velocity (m/s)
the immersed boundary method with adaptive mesh
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

refinement algorithm to obtain efficient mesh ̅ Time-mean


𝑅𝑒 Reynolds number 𝐔
distribution; and, especially, iii. Presenting a method velocity (m/s)
dealing with the problem of water penetration at the Fluctuating
immersed surface that features the solver the capability of 𝑆𝑡 Strouhal number 𝐔′
velocity (m/s)
solving accurately problem of interaction between two-
phase flow and impermeable objects. Results obtained in Mean strain rate Non-dimensional
𝐒ij 𝑢+
this study support the capability of IBM to cope with the tensor velocity
complex and physiologically relevant flow without mesh Velocity of fluid Volume of object
generation tasks. This demonstrates the potential of IBM 𝐔 𝑉𝑝
(m/s) (m3 )
to perform various applications of flow simulation in
terms of a flapping wing, moving of multi-particles Object velocity
Non-dimensional
through periodic boundaries, the sediment of particles, 𝐔𝐩 𝑦+ distance
(m/s)
flow over a vehicle, etc. For more results, please visit our (= uτ y/ν)
Youtube channel https://bit.ly/2Q9LtTV Velocity of water
𝐔𝐥
(m/s)
NOMENCLATURE
𝐶𝐿 Lift coefficient 𝛽 Closure coefficient Acknowledgments
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

This research is funded by Vietnam National Foundation


Dynamic viscosity for Science and Technology Development
𝐶𝐷 Drag coefficient 𝜇
of fluid (kg/ms) (NAFOSTED) under grant number 107.03-2016.11
Characteristic Density of fluid
𝐷0 𝜌𝑓 REFERENCES
length scale (m) (kg/m3 )
[1] J. Slotnick, A. Khodadoust, J. Alonso, and D.
𝐸
Closure
𝜌𝑝
Density of object Darmofal, “CFD Vision 2030 Study: A Path to
coefficients (kg/m3 ) Revolutionary Computational Aerosciences,”
𝜎𝑘 Closure Nnasa/Cr-2014-218178, no. March 2014,
𝐅D Drag force (N) 2014.
𝜎𝜔 coefficients
[2] S. Hassan, “Teesside University BEng ( hons )
specific turbulence
𝐅L Lift force (N) 𝜔 Mechanical Engineering Aerodynamics
dissipation (1/s)
Investigation of Rear Vehicle ( Backlight
Vortex-shedding Angular velocity angle ) By : SAUD HASSAN,” no. April,
𝑓𝑠 𝛚𝐩
frequency (1/s) 2014.
[3] D. L. Quan, “A mesh-based technique for
Torque vector
F1 , F2 Blending function τ solving CFD problems with embedded
(Nm)
geometries : anisotropic adaptive ‘nearly’
𝐟 𝐢𝐛
Force field
τR
Reynolds stress body-fitted mesh approach,” 2014.
(m/s2 ) tensor [4] C. S. Peskin, “Flow patterns around heart
Force field for valves: A numerical method,” J. Comput.
𝑘 turbulence kinetic Interpolation Phys., vol. 10, no. 2, pp. 252–271, 1972.
𝑓 𝛿ℎ
energy equation kernel [5] E. A. Fadlun, R. Verzicco, P. Orlandi, and J.
(m2 /s3 ) Mohd-Yusof, “Combined Immersed-Boundary
Force field for Finite-Difference Methods for Three-
specific
Von Karman
Dimensional Complex Flow Simulations,” J.
𝑓𝜔 turbulence 𝜅 Comput. Phys., vol. 161, no. 1, pp. 35–60,
constant (0.41)
dissipation 2000.
equation (1/s2 )
[6] E. Constant, C. Li, J. Favier, M. Meldi, P.
Gravitational
Closure Meliga, and E. Serre, “Implementation of a
𝐠 acceleration 𝛾1 , 𝛾2 discrete Immersed Boundary Method in
coefficients
(m/s2 ) OpenFOAM,” Comput. Fluids, no. September,
2016.
[7] M. Uhlmann, “An immersed boundary method
with direct forcing for the simulation of [24] M. Ma, W. X. Huang, and C. X. Xu, “A
particulate flows,” J. Comp. Phys., vol. 209, dynamic wall model for large eddy simulation
pp. 448–476, 2005. of turbulent flow over complex/moving
[8] P. Van Sang, “An Immersed Boundary Method boundaries based on the immersed boundary
for Simulation of Moving Object in Fluid method,” Phys. Fluids, vol. 31, no. 11, 2019.
Flow,” J. Sci. Technol., vol. 127, no. 2354– [25] D. Goldstein, R. Handler, and L. Sirovich,
1083, pp. 040–044, 2018. “Modeling a No-Slip Flow Boundary with an
[9] D. Q. Vu, “A New Fluid-Structure Interaction External Force Field,” Journal of
This is the author’s peer reviewed, accepted manuscript. However, the online version of record will be different from this version once it has been copyedited and typeset.

Solver in OpenFOAM,” J. Sci. Technol., vol. Computational Physics, vol. 105, no. 2. pp.
135, no. 2354–1083, pp. 023–027, 2019. 354–366, 1993.
[10] V. Q. Do, B. V. Nguyen, P. Nguyen, and V. [26] H. Ye and D. Wan, “Benchmark computations
Pham, “An Immersed Boundary Method for flows around a stationary cylinder with
OpenFOAM Solver for Structure – Two-phase high Reynolds numbers by RANS-overset grid
Flow Interaction,” J. Sci. Technol., vol. 138, approach,” Appl. Ocean Res., vol. 65, pp. 315–
no. 2345–1083, pp. 028–032, 2019. 326, 2017.
[11] C. J. Greenshields, “The OpenFOAM User [27] K. Lee and K. S. Yang, “Large Eddy
Guide,” no. July. p. 235, 2017. simulation of turbulent flow past a circular
[12] H. Jasak, “Immersed Boundary Surface cylinder in the subcritical and critical regimes,”
Method in foam-extend,” 2018. J. Mech. Sci. Technol., vol. 31, no. 4, pp.
[13] S. J. Plimpton et al., “Direct simulation Monte 1729–1737, 2017.
Carlo on petaflop supercomputers and [28] J. J. De Wilde and R. H. M. Huijsmans,
beyond,” Phys. Fluids, vol. 31, no. 8, 2019. “Experiments for high Reynolds numbers VIV
[14] H. Z. Yuan, C. Shu, Y. Wang, and S. Shu, “A on risers,” Proc. Int. Offshore Polar Eng.
simple mass-conserved level set method for Conf., vol. 3, pp. 400–405, 2001.
PLEASE CITE THIS ARTICLE AS DOI:10.1063/1.5145051

simulation of multiphase flows,” Phys. Fluids, [29] O. Ubbink, “Numerical prediction of two fluid
vol. 30, no. 4, 2018. systems with sharp interfaces,” 1997.
[15] F. Moukalled, L. Mangani, and M. Darwish, [30] C. Zhang, W. Zhang, N. Lin, Y. Tang, and C.
The Finite Volume Method in Computational Zhao, “A two-phase flow model coupling with
Fluid Dynamics - An Advanced Introduction volume of fluid and immersed boundary
with OpenFOAM and Matlab, vol. 113. 2016. methods for free surface and moving structure
[16] F. R. Menter, “Two-equation eddy-viscosity problems,” Ocean Eng., vol. 74, pp. 107–124,
turbulence models for engineering 2013.
applications,” AIAA J., vol. 32, no. 8, pp. [31] A. M. Roma, C. S. Peskin, and M. J. Berger,
1598–1605, 1994. “An Adaptive Version of the Immersed
[17] W. D.C., Turbulent modeling for CFD. 1994. Boundary Method,” vol. 534, pp. 509–534,
[18] J. Bredberg, “On the Wall Boundary Condition 1999.
for Turbulence Models,” 2000. [32] H. Abdulmouti, “2D numerical simulation of
[19] M. Li and H. Nilsson, “A Thorough surface flowvelocity and internal flow structure
Description Of How Wall Functions Are generated by bubbles,” Multiph. Sci. Technol.,
Implemented In OpenFOAM,” 2016. vol. 28, no. 2, pp. 153–171, 2016.
[20] A. Baniabedalruhman, “Dynamic Meshing [33] H. Abdulmouti, “Numerical simulation and
Around Fluid-Fluid Interfaces With fundamental characteristics of surface flow
Applications To Droplet Tracking in generated by bubbly flows,” Int. J. Fluid Mech.
Contraction Geometries,” 2015. Res., vol. 45, no. 3, pp. 263–282, 2018.
[21] T. Mukha and M. Liefvendahl, “Large-Eddy [34] H. Abdulmouti, “Numerical simulation of
Simulation of a Turbulent Channel Flow,” Box bubble convection in two-phase stratified
337, SE-751 05 Uppsala, Sweden Technical, liquids,” Multiph. Sci. Technol., vol. 31, no. 1,
2015. pp. 1–17, 2019.
[22] D. W. Dodge and A. B. Metzner, “Turbulent [35] P. Lin, “A fixed-grid model for simulation of a
flow of non‐newtonian systems,” AIChE J., moving body in free surface flows,” Comput.
vol. 5, no. 2, pp. 189–204, 1959. Fluids, vol. 36, no. 3, pp. 549–561, 2007.
[23] R. D. Moser, J. Kim, and N. N. Mansour, [36] M. Greenhow, “Water-entry and -exit of a
“Direct numerical simulation of turbulent horizontal circular cylinder,” Appl. Ocean Res.,
channel flow up to Reτ=590,” Phys. Fluids, vol. 10, no. 4, pp. 191–198, 1988.
vol. 11, no. 4, pp. 943–945, 1999.
View publication stats

You might also like