Fluid Structure Interaction analysis using Abaqus

and FlowVision

A. Aksenov, A. Dyadkin, T. Luniewski, V. Pokhilko

Capvidia, Belgium

Abstract: The proposed fluid-structure interaction (FSI) approach is based on a two-way coupling
between finite-element code Abaqus and finite-volume code FlowVision. The FSI simulation is
possible due to a unique mesh generation method used in FlowVision. The method is called Sub-
Grid Resolution Method (SGRM). The SGRM connects seamlessly FE and CFD meshes without
introducing any intermediate structures into the FSI layer. It allows to link different mesh types
with different discretization levels and ensures a fully conservative (no data loss) bi-directional
data transfer between structure and fluid simulation domains. The approximation accuracy of the
governing equations is maintained in both simulation domains and the FSI layer. Sloshing water
in a flexible cylindrical tank is simulated to illustrate the described fluid structure interaction
approach.

1. Introduction

The explosive development of numerical methods in the past decades opened new possibilities to
conduct most sophisticated structural and fluid dynamics simulations. Complex scientific and
industrial problems can be efficiently solved today, but structural and fluid simulations are still
performed separately. It is no problem to calculate the aircraft wing bending under external load, it
is no problem to model the airflow over the wing, but it is difficult to simulate both effects
simultaneously.
A complex fluid-structure interaction approach is required to simulate dynamics of turbine blade
excited by the passing flow, flutter analysis of aircraft wings, hydroplaning of automotive tires,
aerodynamics of sails or airbags, etc. This category of problems requires a simultaneous solution
of governing equations as the physical phenomena are coupled and influenced by each other.
Coupling fluid dynamics and structure dynamics codes is difficult due to different domain
discretization (fluid grids vs. FE-meshes) and different numerical approach for solving the
governing equations. Fluid dynamics codes generally use a finite-volume approach; structure
dynamics codes use a finite-element approach. One solution to “conciliate” the codes is to
organize the data exchange via intermediate structures which interpolate and transfer the
information from one grid to another (MpCCI, http://www.scai.fraunhofer.de).

ABAQUS Users’ Conference 2004 Page1/9

In this paper another approach to solve fluid structure interaction problems is presented. This
approach is based on the Sub-Grid Resolution method (Aksenov et al, 1996, Aksenov et al, 1998).
In this method the Cartesian grid with local adaptation is introduced in the fluid flow
computational domain. The structure is described by a finite-element mesh. The external surface
of the finite-element mesh is used as a boundary for the fluid flow. The cells in the fluid domain
are intersected by the surface mesh describing the structure. The resulting cells are not simplified
and stored in the form of complex polyhedrons. The cell sides that are part of the surface mesh
have links to the original faces of the finite-element mesh. Solving fluid dynamics equations on
such a grid provides a natural data transfer mechanism allowing transfer of data from fluid to
structure computational domains without any intermediate interpolation.
The above described approach was tested using ABAQUS for the structural dynamic simulation
and FlowVision for fluid dynamic simulation. FlowVision uses the finite-volume approach and the
Sub-Grid Geometry Resolution Method to solve the governing fluid dynamics equations.
The numerical approach for solving fluid structure interaction problems using the Sub-Grid
Geometry Resolution Method is described in this paper. To demonstrate a fluid-structure
interaction problem, the simulation of sloshing water in a flexible tank is chosen.

2. Sub-Grid Geometry Resolution Method for fluid domain

A rectangular calculation grid is introduced in the fluid computational domain. To capture small
geometry details and solution variations a local automatic grid adoption is used. The local grid
adoption subdivides the initial cell into 8 different cells (1st level grid adoption).

Structure FE
mesh
j-th element

Elm Fij

Fluid grid m-th cell

Figure 1. Cutting finite-volume cell with finite-element mesh.

ABAQUS Users’ Conference 2004 Page2/9

The computational domain boundary is presented by a set of plane facets (surface mesh). This
surface mesh is formed by a set of external faces Fij of the finite-element mesh (Figure 1), where i
is the face number of element j. If a boundary crosses the grid cell the initial cell is cut by the
surface mesh transforming the original cell into a multifaceted polyhedron. A shape Sm of the m-th
cell can be presented as a sum of cell faces Elm (here l is number of face) and part of the j-th
element i-th face dFijm that was cut by the cell itself

(1) Sm = Elm + dFijm ,

l =1, L i =1, I

where L is the number of existing cell faces and I is the number of faces. All this information is
“stored” in the cell. The boundary cell “knows” the corresponding face dFij in the finite element
mesh. Using this link data can be easily transmitted to the FE code and back. For example, the
pressure of fluid Pij on face Fij is found by integrating pressure over all cells containing this face:

1
(2) Pij = Pm s (dFijm ) ,
s ( Fij ) m=1, M
where s(F) is the area of face F. The information from the finite-element code, for example,
velocity of face Fij , is easily transmitted to the CFD code.

3. FSI Numerical Method

3.1 Governing equations

The dynamic equilibrium (for the discrete model in finite-element system) is defined in terms of
the external applied forces (P), the internal element forces (I), and the nodal accelerations:

d 2u
(3) M =P I
dt 2
where M is the mass matrix of the finite element system, u – displacement of the nodes. P is the
sum of forces that act on the structure, including forces from the fluid.
Navier-Stokes equations are used for the description of the fluid flow (in this paper the
incompressible fluid is considered).

V P 1
(4) + (V V) = + (µ V) + g
t
(5) V=0
where V is the velocity vector, P is pressure, µ - effective viscosity (turbulent and molecular), –
density, g – gravity vector. The effective viscosity is calculated using a k- turbulent model (the

ABAQUS Users’ Conference 2004 Page3/9

turbulence model does not affect the fluid-structure interaction and is not further discussed in this
paper)t.

3.2 Boundary conditions on fluid-structure interface

Except ordinary boundary conditions for simulating structure deformations, boundary conditions
at fluid-structure interface are specified in terms of pressure loading from the fluid calculated in
(2) for each element face that is in contact with the fluid.
The general boundary conditions are applied to inlet, outlet and walls and special boundary
conditions are specified on the fluid-structure interface:

du nij
(6) Vij = wnij
n =1, N dt
where N is the number of nodes for element face Fij , wnij is weight coefficient for calculating face
velocity depending on the face geometry.

3.3 Numerical scheme

Let’s assume we have a numerical solution for the displacement of structure nodes and velocities
of the fluid at time step n at time moment t=tn. To find the solution for the system of equations (3-
5) at time step n+1 (tn+1=tn+ t) FlowVision uses split algorithms solving Navier-Stokes equations
(Belotserkovsky, 1994) based on an accurate approximation scheme for convective impulse
transport (Aksenov et al, 1993). Navier-Stokes equations for cells with volumes changing in time
have the following notation:

V n +1 Vn v n +1 v n P n +1 1
(7) + ( V n +1 V ) = V n +1 + ( µ V n +1 ) + g
t v n +1

v n +1 v n
(8) V n +1 =
v n +1
here vn and vn+1 are the cell volume at time step tn and tn+1(Figure 2).

ABAQUS Users’ Conference 2004 Page4/9

 Structure FE
mesh

vn vn+1
Fluid FV
grid
n n+1
Figure 2. Changing volume of fluid cell from v to v in time step.

The first term in the right part of equation (7) is the reaction force from the structure on the fluid.
The term in the right side of equation (8) describes the changing mass inside the cell. If the cell
v n +1 v n
doesn’t contain a moving boundary, the term equals zero and (7-8) corresponds to the
v n +1
approximation of ordinary Navier-Stokes equations.
An explicit method is used for solving structure dynamics equations.
n
d 2u n
(9) = M 1 (P I)
dt 2
n +1 / 2 n 1/ 2 n
du du ( t n+1 + t n ) d 2 u
(10) = +
dt dt 2 dt 2
n +1 / 2
n +1 n +1 du
(11) u =u n 1
+ t
dt
FlowVision supplies the numerical solution for system (7,8), ABAQUS solves the system (9-11).

3.4 Selection of the time step t

The time step t is selected as minimum time steps of a stable solution in Abaqus and FlowVision.
The maximum time step is selected in Abaqus and equals

Lmin
(12) t structure = ,
cd

ABAQUS Users’ Conference 2004 Page5/9

where Lmin is the smallest element dimension in the FEA mesh and cd is the dilatational wave
speed.

+ 2µ
cd =

where and µ are Lame constants for structure material.

The maximum time step for fluid modeling is selected in FlowVision and equals

hmk
(13) t fluid = min( )
Vmk fluid structure int erface

where hmk distance between m-th and k-th cells and Vmk is velocity of the fluid between those cells.
Cells m and k are cells that lie near the fluid-structure interface.
One can see that both time steps are responsible for transferring the disturbance between the
nodes. The time step for integration of the system (7-11) can be selected as minimum between
tstructure and tfluid to get a stable solution

(14) t = min( tstructure, tfluid)

On another hand, the solution of the system (7-11) can lead to quite different time steps for the
“structure” part and the “fluid” part. The “limiting” system is the system requiring the solution
with the smallest time step. To avoid solving the system of equation with very small time step, the
time step is chosen as maximum ( tstructure, tfluid). In this case to satisfy the stability conditions,
the “limiting” system is solved iteratively with smaller time increments. During these iterations,
inside the time step, boundary conditions transferred from the “non-limiting” system stay constant
until the end of each time step.

3.5 Implementation

Link with ABAQUS is made using the user DLOAD function (VDLOAD for ABAQUS explicit)
for transfer of loads calculated in FlowVision. Node positions are obtained by FlowVision via
ODB utilities.
The following algorithm for calculating n-th time step is used:
1. Preliminary increment dT of n-th time step is calculated.
tn = tn-1+dT
2. ABAQUS calls DLOAD, pressure from n-1 time step transferred from FlowVision.
3. ABAQUS specifies its own stable time increment dt
4. ABAQUS makes increment, current time ABAQUS time t’ increased by dt

ABAQUS Users’ Conference 2004 Page6/9

 5. ABAQUS writes data to ODB file.
6. FlowVision seek last write in ODB file to get ABAQUS time t’.
7. if t’<tn, go to 2
8. FlowVision reads ODB file to get node positions on n-th time step and value of n-th time
step tn.
9. FlowVision makes simulations with time increment tn-tn-1
10. if it is not end of simulation, go to 1

4. Simulating sloshing water in a flexible tank

The described simulation approach of the fluid structure interaction problem is performed for
calculating water motion inside a flexible tank (Figure 3, a). The bottom of the tank is oscillating
with speed 0.2sin(1.8 t) m/sec, where t is time in sec. The walls of the tank have the following
properties – density 3.000 kg/m2, Young modulus 9x105 N/m2, Poisson coefficient 0.3.
Dimensions of the tank and finite element mesh (Figure 3, a). The tank is filled with water with
height 2 meter (Figure 3, b)

a) b)

Figure 3. Flexible tank. a) finite-element mesh and dimension, b) initial water state

The problem is solved using ABAQUS-Explicit v6.4 and FlowVision 2003 solving Navier-Stokes
equations for incompressible fluid with k- turbulence model. The water-air interface is tracked by
VOF (Volume of Fluid) -like method (Hirt , Nichols, 1981).

ABAQUS Users’ Conference 2004 Page7/9

The number of cells in the finite-volume grid is 16.000, the finite-element (tank) mesh includes
5.100 nodes and 3.300 elements. The time step is 0.03 sec.

Figure 4. Pressure (Pa) and water surface distribution at time 3 sec.

The water pressure distribution is shown in Figure 4 by color map, time moment is 3 sec. The
water surface is shown by contour. The computational domain is cut by plane to visualize details
inside the tank.

5. Conclusion

An efficient realization of the fluid-structure interaction is possible due to the unique Sub-Grid
Resolution Method providing natural link between CFD and FEM meshes. The simulation of
sloshing water in a flexible tank with oscillating bottom is just one example. The described
approach has been successfully used for other simulations involving complex geometry, multi-
phase media, moving bodies and large structural deformations.
Known limitation of the approach: the fluid boundary is defined by the FEM mesh, which in some
cases is not accurate enough to describe the boundary for the fluid flow calculation. An illustrative
example is bending of a turbine blade caused by the passing gas flow. To describe accurately the
flow separation on the blade surface the fluid grid has to be very fine. Thus, the finite element
mesh must also be very fine. This is not practical as the resulting FEM models become huge and
unmanageable. To overcome it we will use the original CAD geometry. The Sub-Grid Resolution
Method constructs the fluid grid by intersecting the orthogonal fluid cells with the CAD surface
exactly describing the fluid-structure border. The elements on the fluid-structure border can now
be matched with the much coarser FEM mesh. This is ultimately solving all limitations of the
described above fluid-structure interaction approach and will be implemented in the next
FlowVision release.

ABAQUS Users’ Conference 2004 Page8/9

6. References

1. Aksenov A, Dyadkin A, Pokhilko V. Overcoming of Barrier between CAD and CFD by

Modified Finite Volume Method, Proc. 1998 ASME Pressure Vessels and Piping Division
Conference, San Diego, ASME PVP-Vol. 377-1., 1998
2. Aksenov A., Dyadkin A., Gudzovsky A., 1996, “Numerical Simulation of Car Tire
Aquaplaning”. Computational Fluid Dynamics ’96, J.-A. Desideri, C.Hirsch, P.Le Tallec,
M.Pandolfi, J.Periaux edts, John Wiley&Sons, pp. 815-820.
3. Aksenov A., Gudzovsky A., Serebrov A., 1993, “Electrohydrodynamic Instability of Fluid Jet
in Microgravity”, 19-24, in Proc. of 5th Int. Symposium on Computational Fluid Dynamics
(ISCFD), Aug. 31 - Sept. 3, 1993, Sendai, Japan, Japan Society of Computational Fluid
Dynamics, Vol.1, 1993.
4. Belotserkovsky, 1994 Numerical Methods in Continuum Mechanics, Moscow, Fizmatlit, 2nd
edition, p.441.
5. Hirt C., Nichols B. 1981, “Volume of Fluid (VOF) method for the dynamics of free
boundaries”, J.Comput. Phys. v. 39, pp 201-225.

ABAQUS Users’ Conference 2004 Page9/9