Professional Documents
Culture Documents
schildhauer2014
schildhauer2014
1, 2014
Abstract: The numerical simulation of fluid-structure interaction requires a simulation tool that
can calculate both fluid flow and structure deformation and takes into account the results of one
code (i.e., forces and deformations) in the setup of the other (as motion and load). This article
shows a segregated (also: partitioned) approach that uses commercial, highly developed software
for each task: ANSYS CFX for fluid dynamics and ANSYS Mechanical for structural mechanics
with the ANSYS MFX solver to couple both programs. Two examples with large deformations
and strong influence on the flow field are shown: The Turek benchmark from Turek and Hron
(2006) consists of a laminar incompressible channel flow around an elastic object which results
in periodic oscillations of the structure. The second example is the discharge simulation of
compressed air through a rapidly moving reed valve in a piston compressor where opening and
closing contacts are important.
Keywords: fluid-structure interaction; reed valve; load transfer; computational fluid dynamics;
CFD; Turek benchmark; computational structural mechanics; CSM.
Reference to this paper should be made as follows: Schildhauer, M. and Spille-Kohoff, A. (2014)
‘Numerical simulation of fluid-structure interaction: Turek benchmark and kinetics of a reed
valve’, Progress in Computational Fluid Dynamics, Vol. 14, No. 1, pp.38–48.
Biographical notes: M. Schildhauer is an Engineer at MAN Diesel and Turbo SE, Berlin
(Germany). He received his MEng (2011) from Leipzig University of Applied Science
(Germany) and wrote his Masters thesis at CFX Berlin Software GmbH (Germany).
A. Spille-Kohoff is the Head of Research and Development at CFX Berlin Software GmbH
(Germany). He received his PhD in Theoretical Hydrodynamics (1999) from University
Oldenburg (Germany). His research interest is in the area of computational fluid dynamics,
electromagnetics and acoustics.
For transient FSI problems, the coupling might be block-structured meshes for the CFD simulation were
explicit, e.g., the CFD solution is updated to the new time generated with ANSYS ICEM CFD Hexa, Post-processing
step with deformations taken from CSM solver of last time was done with ANSYS CFD-Post. All ANSYS products
step or vice versa so that each solver is called only once were used in version 14.0.
each time step, or implicit, i.e., for one time step CFD and
CSM solutions are updated iteratively until a converged
solution is reached. The implicit coupling ensures consistent 2 Mesh deformation in CFD model
solutions and numerical stability; explicit coupling needs no
The deformations of the solid structure calculated by the
coupling iterations but often very small time step sizes.
CSM solver have to be imposed as boundary motion to the
The coupling algorithm has the following tasks: The
CFD model. Since the flow region around the solid is
forces acting on the fluid-structure interface have to be
discretised with a finite volume mesh, this mesh has to be
interpolated from CFD mesh to CSM mesh in a
deformed as well. This is a typical boundary value problem:
conservative form, the deformations back from CSM mesh
With given deformation values on the boundaries, calculate
to CFD mesh. To increase stability, under-relaxation of
a smooth relocation in the interior. ANSYS CFX uses
forces and deformations is often important. And finally the
Laplace equations for the three spatial components of the
algorithm has to decide when the iterations have converged,
deformation. But requirements on mesh quality are high in
i.e., when changes in forces and deformations are small
CFD: Boundary layers have to be resolved well, and bad
enough.
angles, high expansion factors, high length ratios or folded
In this contribution, ANSYS CFX is used for the CFD
volumes shall be avoided to get stable and converging
simulation and ANSYS Mechanical for the CSM
solutions.
simulation. The two-way implicit coupling is realised with
the ANSYS MFX Multi-field Solver, see Figure 1. The
Figure 1 Solution process of a FSI simulation with ANSYS CFX and ANSYS mechanical for CFD and CSM simulation, respectively,
and ANSYS MFX for coupling and interpolation (see online version for colours)
Figure 2 Mesh displacement at an upwards moving structure with constant stiffness (a) leading to bad orthogonality angles at lower
corner and folded elements at upper corner, and with hyperbolic stiffness (b) ensuring good mesh quality (see online version
for colours)
(a) (b)
40 M. Schildhauer and A. Spille-Kohoff
To fulfil these criteria, the user can choose a spatially and In this case the Reynolds number defined with the cylinder
temporally varying mesh stiffness ζ(x, y, z) for the Laplace diameter D is 200 (Re = cDρ /η) and a transient solution is
equations. Mesh stiffness can be increased near boundaries expected.
to prevent distortion in boundary layers, or near small
elements, or can be set arbitrarily. The authors chose 3.2 CFD meshing and setup
hyperbolic stiffness functions with poles at corners or edges
to avoid folded elements and to preserve mesh quality, see Since ANSYS CFX is a pure three-dimensional solver, 2D
Figure 2. problems have to be treated on a plane (extruded), one
element-thick layer with symmetry conditions on both base
areas. In view of an intense mesh motion during the FSI
3 Turek benchmark simulation it is necessary to generate a high quality regular
hexahedral mesh (done with ANSYS ICEM CFD Hexa).
The Turek benchmark is described by Turek and Hron Hexahedral grids generally have numerical advantages over
(2006) to facilitate verification and comparison of FSI tetrahedral respectively prism meshes. Especially the former
solution algorithms on the basis of a numerically difficult are superior in case of large mesh deformations, where the
2D example. Turek and Hron present solutions with latter ones produce bad element angles without re-meshing.
reference values, calculated with a specialised monolithic The mesh has to take account of sufficiently high
ALE-FEM solver. resolution of flow gradients, particularly in boundary layers,
on the one side and a robust high-quality mesh deformation
3.1 Problem description on the other side. Clearly a third important aspect is the
minimisation of computational effort by optimising the
The setup is shown in Figure 3. A highly viscous
element distribution in order to reduce the number of nodes.
incompressible Newtonian fluid (η = 1 Pa·s) similar to
The first two objectives lead to a conflict especially at sharp
glycerine enters a channel with a parabolic velocity profile.
edges where the translational and rotational mesh
The laminar flow is expected to interact intensively with a
displacements are very large compared to the small element
solid body consisting of a rigid cylinder with a backwards
sizes necessary for boundary layer resolution. This often
mounted, highly flexible, hyperelastic structure which could
results in an unacceptably high mesh distortion which can
be made of an elastomer like polybutadiene. Fluid and
cause a solver termination in the worst case. The Turek
structure have the same density of 1,000 kg/m³. Through a
benchmark involves this problem at the tail of the structure
slightly asymmetric location of the body within the channel
which is already mentioned along its solution approach in
flow it is ensured that a lifting force develops in a steady-
Figure 2.
state flow and that laminar vortex shedding occurs at higher
The main part of the initial CFD mesh used for the FSI
flow rates. Based on variation of inlet velocity and structural
test is shown in Figure 4. It has 74,793 elements and a
stiffness the complete benchmark comprises different
minimum orthogonality angle of 41°. In pure CFD analyses
testing scenarios including pure CFD and CSM tests for
the mesh shows a transient periodic solution (vortex street)
preparation and verification purposes. For all scenarios, see
in very good accordance with reference values. For the
Schildhauer (2011); this article focuses on a single testing
ANSYS CFX setup refer to Figure 5. Since a mesh with half
scenario (FSI3 test):
the number of elements did not produce a transition from
• average inlet velocity c = 2 m/s stationary solutions to a transient one, the preferred
resolution can be regarded as suitable.
• shear modulus μ = 2 MPa; Poisson’s ratio ν = 0.4.
Figure 4 CFD mesh (main part) for Turek benchmark (see online version for colours)
Figure 5 CFD setup for Turek benchmark with boundary conditions in ANSYS CFX (see online version for colours)
Figure 6 CSM mesh and setup with boundary conditions in ANSYS mechanical (see online version for colours)
interface benefits from a higher or at least equally high of force transfer (factor < 0.1), but no displacement
resolution of the fluid mesh compared to the structural under-relaxation; strict coupling convergence criteria
resolution. (maximum residuals of force transfer < 0.001, for
displacement < 0.01), and a high maximum number of
coupling iterations (>≈ 30) per time step. With these settings
3.4 FSI setup, solution process and results
it is possible to gain a compromise between a stable FSI
The setups for CFD, CSM and mesh motion can be merged solution process with a sufficient accuracy on the one side
in the ANSYS workbench FSI setup. To enable the solver and a moderate computational effort on the other side.
coupling the FSI interface has to be defined at a fluid Using a transient solution of a pure CFD analysis as
domain boundary (wall) and the corresponding structural initial condition for the FSI solution process can be
domain surface. The numerical setup of the coupling convenient to reduce calculating time. Because of the
procedure (ANSYS MFX solver) is done within the ANSYS initially periodically acting forces the structure immediately
CFX-Pre environment and includes settings like coupling starts to oscillate and heavily interacts with the flow. After
time step size and convergence criteria, under-relaxation already eight oscillations a periodic state is reached, see
factors or the maximum number of coupling iterations. Figure 7. A time step size of 0.002 s yields a temporal
The Turek benchmark is characterised by the necessity resolution of about 92 time steps per period. Together with a
of making an extensive use of those coupling parameters to CFD residual target of 0.001 (max) and maximum number
gain a stable solution with a partitioned approach. The of ten coefficient loops per coupling step a well converging
partitioned approach involves, unlike the monolithic solution process is achieved. The convergence of the
approach, an inherent instability effect which is frequently structural solution part is relatively unproblematic and not
called artificial added mass effect. time-critical. A computing time of 4.43 hours for one
Each acceleration of a solid body within a fluid causes a oscillation was detected on an Intel Xeon processor (64-bit,
pressure force in the opposite direction and can therefore be 2.8 GHz) with four cores in parallel. In fact, the possible
interpreted as an inertial force caused by an added mass. number of 300 coefficient loops per time step (30 coupling
Under certain circumstances this opposing force causes a iterations with each ten coefficient loops) is never reached,
sudden reversal of acceleration which is followed by a new but is still at a quite high number of 36 in average. About
estimation of the force transfer in the opposite direction and 80% of the computing time is used by the CFD solver only.
an even greater acceleration response of the structure during The accuracy of the simulation can be evaluated by the
the next coupling step and so on. The forces and the comparison with reference values (mean value, amplitude
corresponding displacements can grow until a solver error and frequency) for the force acting on the solid body
occurs. In contrast to other instability effects with (drag and lift component) and the displacement of the tail at
time-stepping schemes this instability cannot be reduced by point A (x- and y-direction), see Table 1. Turek and Hron
decreasing the time step size, but precisely the opposite is used a time step size of 0.001 s and a mesh consisting of
true: A smaller time step size amplifies the acceleration 15,872 biquadratic elements to obtain these reference
response and makes the situation worse if the given problem values. The values could be reproduced well. In fact, a
is not generally stable. further improvement is achieved by reducing the time step
The artificial added mass effect is caused or promoted size from 0.002 s to 0.001 s. The higher temporal resolution
by the following physical influences in descending order, accelerates the convergence speed of the CFD solution and
see Schildhauer (2011) and Förster et al. (2006): the average number of coefficient loops per time step drops
from 36 to 27, although some additional coupling steps are
• high fluid density in relation to structural density (esp.: required. In consequence the computing time rises only
density ratio ≥1) 60%.
• movement with heavy fluid mass displacement
Table 1 Test results and reference values
• low structural stiffness
ANSYS/CFX Reference
• high viscosity
DispX at A (mm) –2.73 ± 2.59 –2.69 ± 2.53
• incompressibility or low compressibility of the fluid.
DispY at A (mm) 1.53 ± 33.40 1.48 ± 34.38
The opposite of those attributes indicates stable behaviour. Drag (N/m) 459.7 ± 25.2 457.3 ± 22.66
But obviously the Turek benchmark exhibits all instability
Lift (N/m) 1.85 ± 142.45 2.22 ± 149.78
causes to a high extent and as a result shows numerically
very difficult conditions. Frequency (Hz) 5.5 5.3
Therefore special parameters for the implicit Source: Turek and Hron (2006)
coupling procedure are necessary: Strong under-relaxation
Numerical simulation of fluid-structure interaction 43
Figure7 Turek benchmark (FSI3 test) (see online version for colours)
Note: Sequence of states over one period (streamlines and pressure field).
until the pressure in the compression chamber rises above The latter means that the total fluid domain is everywhere
the pressure in the discharge chamber; then, the reed is continuously connected. Even when the reed collides with
abruptly bended up and collides with the reed catcher that is the valve plate or the catcher, a very thin layer of fluid
supposed to limit the maximum bending stress. After the elements has to be maintained in between. In other words,
rebound the reed starts to oscillate within the emanating gas only continuous deformations of domains are allowed, no
flow. This is the so called valve flutter. It can delay the cutting, if the fluid-structure simulation shall be done in a
sealing when the piston arrives at the top dead-centre single solver run without stopping, remeshing, restarting
position. Together with the abrupt opening behaviour the and interpolation.
flutter heavily affects the service life, the discharge/intake
process and thereby the compressor efficiency and 4.2 CFD meshing and setup
construction.
The simulation of the discharge process uses the model The hexahedral mesh shown in Figure 9 is mainly adapted
geometry shown in Figure 8. Only a half model is to the needs of a stable quality preserving mesh motion
considered with assumption of a symmetry plane. The which is the most critical part of this simulation. Basically
structural model consists of three solid bodies: reed, catcher the methods mentioned above are applied and extended to
and valve plate, whereas the latter two are primarily ensure a well operating mesh motion. Explaining the many
meaningful for the contact surfaces. The geometry of the diverse measures that have been taken would go far beyond
fluid domain roughly represents the conditions within a real the scope of this report, for details see Schildhauer (2011).
compressor. Only the gas flow through the valve opening The element distribution is characterised by the
and around the reed is of main interest. So the suction intention to obtain solution convergence in all areas
chamber with corresponding valves is neglected, and the of the mesh at a relatively low spatial resolution
geometry of the discharge chamber can be strongly (about 360,000 elements). In order to keep the
simplified by defining a cylindrical boundary with a large computational effort moderate the chosen boundary element
circular opening on the top. The piston stroke is modelled as heights permit no boundary layer resolution. Between the
a time-dependent moving ground plane in the cylindrical reed and the valve plate there is a tiny gap of 0.1 mm height;
compression chamber to obtain realistic and stable flow about 20 element layers across this gap width ensure a
conditions at the valve opening. It is convenient to start the sufficient vertical resolution when the gap is extremely
simulation at a piston position where a zero pressure enlarged due to the bending up of the reed.
difference can be assumed between the compression and Figure 10 contains the most essential information about
discharge chamber. the CFD setup. Since the air is modelled as ideal gas, the
Another important issue has to be regarded: Narrow temperature is introduced as additional field variable
intersection zones of fluid and solid domains are necessary calculated by the total energy model. Turbulence is
to allow changing contact states between the solid bodies modelled by the SAS SST turbulence model.
without causing topological changes of the fluid domain.
Figure 8 Model geometry of a reed valve in a piston compressor (see online version for colours)
Figure 9 CFD hexahedral mesh (generated with ANSYS ICEM CFD hexa) for piston compressor (see online version for colours)
Figure 10 ANSYS CFX setup for piston compressor with initial conditions (see online version for colours)
A uniform pressure distribution of 20 bar (relative to produced in the worst case. An overview about a well
1 atm reference pressure) and zero velocity are assumed at performing mesh and setup delivers Figure 11.
the simulation start. The piston motion is defined through A reliable solution convergence is achieved at a time
the kinematics of a crankshaft. It starts at a certain upwards step size of 0.01 milliseconds together with frictionless
velocity in a lower position and reaches the top dead-centre contact definitions according to the augmented Lagrange
position after approximately five milliseconds. formulation. Furthermore a manual definition of the contact
stiffness (factor ≈ 1) and the contact search radius (8 mm) is
4.3 CSM meshing and setup necessary. The chosen contact stiffness features a
compromise between good solution convergence and low
Plenty of setup and meshing variants had to be tested until a penetration.
satisfying answer to the difficult contact problem was For simplicity reasons a FSI interface has been
found. Important requirements are a reliable solution considered for the reed surface only and not for the catcher.
convergence, realistic boundary conditions for the reed and In consequence the catcher has to be fully constrained.
finally but equally important an excessive penetration of the Otherwise the oscillation of the catcher would cause serious
bodies has to be avoided. It rather should be limited to a problems with the CFD mesh. A constant displacement of
very small extent. Otherwise the topology of the CFD mesh the catcher secures the required clamping force on the reed.
is violated within the small gaps and a solver termination is
46 M. Schildhauer and A. Spille-Kohoff
Figure 11 ANSYS Mechanical setup of reed valve with meshing (see online version for colours)
Finally three important findings regarding the simulation improved through reducing the time step size in favour of
effort shall be mentioned. Firstly, both examples suggest the fewer coupling iterations insofar the stability is provided.
fact that FSI simulations are mainly dominated by the setup And latterly, there is no general good coupling parameter
of a stable moving mesh within the fluid domain. Secondly, set, but it depends on the special requirements of each
the cost-benefit ratio of the computational effort can be particular FSI case.
Figure 12 Kinetics of the reed valve in the piston compressor (see online version for colours)
Note: Sequence of states over the complete discharge process (streamlines, pressure field and structural stress).
48 M. Schildhauer and A. Spille-Kohoff
Figure 13 (a) Piston position h, position of end and centre point of reed valve uA and uB, (b) maximum von-Mises stress σ over time
(see online version for colours)
(a) (b)