Author
others
View
2
Download
0
Embed Size (px)
V13 - 1
Use of Computational Fluid Dynamics in Civil Engineering
Prof. Dr.-Ing. Casimir Katz, SOFiSTiK AG, Oberschleißheim
ir. Henk Krüs, Cyclone Fluid Dynamics BV, Waalre
Zusammenfassung:
Der Einsatz von CFD im Bauwesen stellt nach wie vor eine Nischenanwendung dar. Die Gründe
dafür werden beleuchtet und aufgezeigt, warum sich das jetzt ändern wird.
Summary:
The use of CFD in the civil engineering community is still a rare event. The reasons for that will be
discussed and it will be pointed out why that will change now.
1 INTRODUCTION
1.1 Fluid dynamics and their applications
CFD is the acronym for „Computational Fluid Dynamics“. Fluids in civil engineering are mostly
air and water and the questions to be answered are the forces induced by fluid motion and the
transport of heat or particles within the fluid. Typical questions are
Wind loading on bluff bodies
Wind loading on moving bodies, especially bridges
Wind comfort and nuisance to cyclists and pedestrians
Wind energy
Heating and Ventilation (HVAC)
Fire safety engineering
Wave loadings
The difference to classical static or dynamic analysis is given by a different mathematical treatment
and thus different mathematical tools [1]. While structural mechanics use the Lagrangian approach
based on displacements, fluid mechanics prefer the Eulerian approach based on the velocity of the
fluid. Structural mechanics use the Finite Element Method, most fluid mechanics use the Finite
Volume Method. The basic principles of equilibrium and conservation of mass, energy and
momentum are common to both methods.
V13 - 2
1.2 Software for CFD
Software for CFD has been developed for a long time now, the most well known products today are
the big three FLUENT (now Ansys), STAR-CD/STAR-CCM (CD-Adapco) and CFX (now Ansys),
but there are many more like the CFDDRC (now ESI-Group), CFDesign (FEM by Blue Ridge
Numerics, now Autodesk) and FDS (by NIST). There are also different techniques available like
vortex particle and Lattice-Boltzmann methods.
The first impression for a structural engineer is, that everything is very complex, that there are
hundreds of features and parameters and that it will cost a fortune to start into this field. Over the
years the mantra of a technique much too complex for the common engineer and high licence costs
has established significant barriers. The same happened to structural FE-Software in the late
seventies. SOFiSTiK had success because we anticipated the wide spread of that technique on
personal computers in 1980.
But the CFD-market is changing now, cheaper versions of CFD-software enter the market, and the
open source software OpenFOAM has gained wide acceptance especially in the academic world.
However the structural beginner is still overwhelmed by a wide range of features. What is needed is
a robust, easy to use entry point for this journey.
SOFiSTiK has gained some experience with an academic multiphysics software PHYSICA and is
now supporting DOLFYN, an open source CFD-Solver used in practice in many engineering fields,
which has been fully integrated in the SOFiSTiK environment.
2 BASICS
2.1 Fluid dynamics and their solvers
Two important metrics are the Mach number (v/c) defining the ratio of the fluid velocity v to the
speed of the sound c and the Reynolds number derived from the fluid velocity u, the kinematic
viscosity n, and a characteristic dimension of the structure (like the diameter of a cylinder or the
height of a bridge deck):
kinematic viscosityRe
characteristicdimension
; .sec sec
2 26 6
d ud
m mAir 15 10 Water 1 31 10
V13 - 3
Fluid dynamics cover a broad range of fields, some of them are:
Potential Flow (Re = ∞ )
Creeping Flow (hardly any flow)
Laminar incompressible viscous flow
(Navier Stokes)
Turbulent Flow (Re large)
Compressible Flow (Ma > 0.3)
Supersonic Flow (Ma > 1.0)
Thermal Effects
Combustion & chemical reaction
Free Surfaces
Non-Newtonian fluids
Radiation
In civil engineering applications the subsonic incompressible turbulent flow is the most common
phenomenon. Compressible flow is needed for shocks and high temperature effects. The major
remaining problem is that a direct solution of the Navier-Stokes-equations is only possible for
Reynolds-numbers up to approximately 20 000, while practical examples are in the range of several
millions. Thus turbulence has to be modelled by RANS (Reynolds-Average-Navier-Stokes) models
allowing for very large Reynolds numbers but do not model all effects, that’s why LES (Large Eddy
Simulation) has gained some popularity, but requires still high computational effort.
2.2 Materialparameters
The selection of fluid material parameters is straight forward: There is a density [kg/m³], a dynamic
viscosity [Pa sec], a compressibility [Pa/m²] and some thermal properties.
2.3 Boundary Conditions
Inflow and outflow boundary conditions are in general not complex, but there is a major problem
with two aspects. There exists a boundary layer at every wall. At the wall itself there is no flow,
then we have a tiny laminar viscous sublayer, followed by the turbulent boundary layer. The
treatment of the wall boundary condition is quite difficult. Though there are possibilities for near
wall models, the common model is a logarithmic wall law describing the complete wall boundary
behaviour for a cell sufficiently far away from the boundary:
V13 - 4
However the atmospheric boundary layer has a height larger than one kilometre, thus all civil
engineering structures are completely encompassed by it. The design codes for wind loadings
describe the layer with a logarithmic or an exponential law.
It is not only the velocity but also the turbulence characteristics like the kinetic energy (turbulence
intensity) and the dissipation rate (Integral length scale) which need to be described. The correct
formulation requires not only to model the roughness of the ground with the correct value of z0, the
driving force at the top of the fluid domain, but also the treatment of the analytic solution of the
turbulence equations as initial conditions, which are fully implemented in DOLFYN:
*
*
*
( ) ln
( )
( )
ABL 0
0
2ABL
3ABL
0
u z zu zz
uk zC
uzz z
V13 - 5
3 MODELLING
3.1 Windtunnel or CFD ?
A wind tunnel is just a model of the reality, so is any CFD model. The wind tunnel needs scaling,
Reynolds number is not the same and there are cases where this does matter. There is a guideline
from the WTG (Windtechnologische Gesellschaft) describing in detail how to perform reliable
tests.
But for flexible structures, the measurement equipment may change the effects considerably. On the
other side there are known deficiencies of numerical analysis, which do not allow taking all results
for granted. “The purpose of computing is insight, not numbers.” said R. Hamming in 1962. So the
question is not which technique to be used but how to combine both methods to their best use.
3.2 Reality or design case ?
Is the requested result the mean values of the wind loading for a static analysis or the variation of
forces in time either to account for dynamic effects or to get reasonable loads at all. For example a
flat horizontal roof on columns will have a zero pressure as mean value, but the wind load is of
course not zero!
What is appropriate for the design of a building for wind loading, a solitaire in an empty
environment:
V13 - 6
Or the within the true environment:
Most design codes describe wind profiles based on the roughness of the terrain, for urban
environment the value of z0 may be over 2.0 m, which is not suitable as roughness for a CFD wall
boundary condition in general, although it is possible for an inlet wind profile. The main purpose is
to define loads just based on the velocity distribution, not to perform a fluid analysis. This can be
clearly seen by the fact that the velocity does not vanish at the ground.
For a wind tunnel test normally the whole environment is modelled. So to compare CFD and wind
tunnel, the model size of the wind tunnel (including the true geometry of the equipment) may be
analyzed but the analysis of the true scale of the natural model including the environment should be
more adequate.
And is our interest in a model matching reality as close as possible or can we agree on a model on
the safe side?
V13 - 7
4 EXAMPLE BUILDING
Gerhardt [3] has reported considerable deviations between a CFD-Analysis and measurements for a
hull of a building (Re = 4·107), given with a picture as follows:
As all other data was missing, personal enquiries yielded a profile exponent of 0.25 (wind tunnel)
and 0.27 (analysis), a rather sufficiently large air volume of 2000 x 2000 x 500 m and some
inhomogenities along the building and other neighboured buildings in the wind tunnel test. Some
tests showed however that the effect of those buildings should be neglectable allowing to
concentrate on a 2D section.
Thus the section was digitized from the picture, scaled to the real dimensions of 55 m width and 25
m height and an air volume with 500 m distance to the boundaries was used. The mesh density on
the hull was selected with 0.25 cm (approx. 30 y+) yielding a total of 12500 cells.
V13 - 8
The first outcome of the analysis showed, that the selection of the turbulence model had a
significant effect on the pressure distribution, the standard k- model created the distribution with
the large peaks (left picture) while the RNG-model showed better results (right picture):
A deeper look at the inflow parameters however showed some deficiencies. Input velocity and
turbulence intensities are well known, but the Eurocode specifies a value for the integral length
scale of more than 200 m in a height of 25 m, which is considerably larger than the common CFD
value of 0.4*H = 10 m. This contradiction can be overcome if the anisotropic structure of the
natural wind is considered. Further the used mesh does not allow for large roughness values. After
correcting those parameters, the following pressures have been obtained:
However the wind tunnel test was based on a very dense environment consisting not only of other
buildings but also of the roughness elements. So any further comparison of the CFD analysis and
the wind tunnel becomes useless.
1.36
1.361.28
1.25
1.14
1.13
1.03
0.963
0.895
0.839
0.821
0. 773
0. 740
0.7 1
2 0.689
0.67
2
0.65
0
0.64
60.
637
0.62
7
0.62
6
0.61
3
0.57
40.
497
0.487
0.42
20.
398
0.3910.391
0.391
-0.3900.389
0.389
0.3880.386
0.386
0.386
0.384
-0.383
0.382
-0.368
-0.362-0.36
1-0.359
-0.344
0.103
V13 - 9
Let’s use complex environments for the CFD as well:
V13 - 10
V13 - 11
5 EXAMPLE BRIDGE SECTIONS
5.1 Millau Bridge
A section of the Millau Bridge has been analyzed with CFD and compared to measurements in a
wind tunnel.
Despite the good agreement of the measurements with the analysis, some questions remain also in
this case. The turbulence parameters in the wind tunnel have not been fully specified. And nobody
really knew the wind conditions at the site of the bridge. There is a deep valley and fundamental
CFD analysis had been undertaken to get the wind conditions in the nature. But to perform a
dynamic analysis of a moving bridge in the wind field, the drag coefficients may be used.
V13 - 12
5.2 Bridge in the wind tunnel
The second example is the analysis of a bridge section in a wind tunnel [4]. Here the CFD model
matches exactly the geometry of the wind tunnel. The inflow parameters are well known (I=3 %,
L=0.03 m)
The lift and moment coefficients are quite closely matched, but for the horizontal drag, the
measurements deviate and the simulation is closer to the pressure measurements.
V13 - 13
When analyzing this deck the results deviate to a certain extend. First the flow field is considerably
different for the standard k- and the RNG variant:
The pressure distribution is very sensitive to the mesh definition, but the principal distribution is:
Location / Value Experiment Simulation
openFOAM DOLFYN
k- DOLFYN
RNG top spt 01 (tap 31) 0,494 0,650 0,723 0,689
top spt 05 (tap 35) -0,413 -1,090 -0,676 -0,662
top spt 06 (tap 36) -0,902 -1,230 -1,307 -1,181
top spt 18 (tap 8) -0,202 -0,270 -0,216 -0,149
top spt 19 (tap 9) -0,264 -0,260 -0,199 -0,135
top spt 21 (tap 11) -0,270 -0,250 -0,133 -0,096
dwn spt 40 (tap 30) 0,496 0,670 0,779 0,799
dwn spt 34 (tap 24) -0,708 -1,110 -0,730 -0,677
dwn spt 33 (tap 23) -1,083 -1,250 -1,180 -1,065
dwn spt 26 (tap 16) -0,295 -0,340 -0,335 -0,261
dwn spt 25 (tap 15) -0,305 -0,300 -0,297 -0,218
dwn spt 22 (tap 12) -0,303 -0,250 -0,133 -0,109
Drag coefficient (range of values)
0,088 0,075
0,073 0,093 0,081
0,082 0,069
-1.80
-1.70
-1.68
1.15
1.15
-1.06
-0.842
0.77
9
0.722
0.60
6
-0.589
-0.586
-0.4900.48
2
0.481
-0.448
-0.424
-0.413
-0.413
-0.3
91
0.37
5
-0.352
-0.315
-0.290
-0.280
-0.272
0.27
0
-0.260
-0.260
0.252
-0.250
-0.239
-0.235
-0.2
25
-0.223
-0.217
-0.213
-0.213
-0.2
08
-0.204
-0.202
-0.202
-0.1
99
-0.192
-0.183
-0.182-0.169
-0.168
0.16
0
-0.158
-0.1
58
-0.153
-0.152
-0.146-0.143
-0.139
-0.138
-0.136
-0.127
-0.123 -0.121
-0.116
-0.114
-0.1
13
-0.109
-0.106
-0.102
-0.100
-0.0977
-0.0958
-0.0955
-0.0948
-0.0932
-0.0921
0.03
84
V13 - 14
6 CONCLUSION
The critical definitions for a CFD analysis are the selection of the mesh and the inflow conditions.
Handled with greater success are bluff bodies with sharp edges, slender structures (e.g. airfoils)
need more attention.
So there is no free lunch by just buying a CFD software, which is also valid for any other type of
complex simulation software (like structural analysis Finite Element). Each journey starts with the
first step. It’s time to take that first step.
7 DOLFYN
DOLFYN [1] is an open source collocated face based Finite Volume software to solve
incompressible fluid dynamic problems in 3D. The key features are:
Standard k- and RNG turbulence models
Stable numerical procedures
Temperature / scalars / particles included
Postprocessing with ParaView / VisIt (VTK-files)
The implementation in the SOFiSTiK environment gives additional
Mesh generation with SOFIMSHA / SOFIMSHC
Postprocessing with WINGRAF
Full CADINP support including formulas for boundary and initial conditions
Easy wind definition (atmospheric boundary layer) directly or via SOFiLOAD
Direct generation of resulting wind loading in the data base
Possible (planned) extensions are:
Compressible subsonic flow
Free surfaces (VOF)
Conjugate heat transfer (heat transfer with different materials)
8 LITERATURE
[1] www.dolfyn.net
[2] J.H.Ferziger, M.Peric, Numerische Strömungsmechanik, Springer, 2008
[3] H.J. Gerhardt, Experimentelle und numerische Verfahren bei der Bauwerks-Bemessung, Der
Prüfingenieur Vol. 24, April 2004
[4] A. Sarkic, C. Neuhaus, R. Höffer, Numerical and experimental determination of
aerodynamic forces at a long span bridge girder, Eurodyn 2011, Leuven