21 Month Review
21 Month Review
21 Month Review
Shakeel Ahmed
School of Engineering
August 2010
Executive Summary
List of Contents
Executive Summary............................................................................................... 2
List of Figures......................................................................................................... 5
List of Equations.................................................................................................... 8
List of Tables.......................................................................................................... 9
1.
Introduction..................................................................................................... 1
2.
3.
Literature Review............................................................................................ 3
3. 1. Aeroelasticity and Its Modelling.................................................................3
3.1.1.
Monolithic Procedure...........................................................................3
3.1.2.
Partitioned Procedure..........................................................................4
3. 2. Fluid-Structure Interaction........................................................................4
3. 3. Structural Analysis.................................................................................... 6
4.
3.3.1.
Static Analysis..................................................................................... 6
3.3.2.
3.3.3.
Validation of CFD............................................................................................. 9
4. 1. Validation of Clean Aerofoil.......................................................................9
4. 2. Validation of the Flap deployed configuration.........................................18
4. 3. Validation of The Droop Nose and Flap Deployed Configuration.............19
4.3.1.
Configuration................................................................................................. 19
4.3.2.
Configuration................................................................................................. 21
4. 4. Validation of The Droop Nose With The Flap Morphed Configuration......22
5.
Structural Model............................................................................................ 26
6.
Fluid-Structure Model.................................................................................... 29
7.
8.
Bibliography.................................................................................................. 38
List of Figures
Figure 1: Clean geometry profile for SADE (left), NACA 4412 aerofoil (right).........9
Figure 2: Clean aerofoil mesh (top) NACA 4412 aerofoil mesh (bottom)................9
Figure 3: Graph comparing the pressure profile of the NACA 4412 with the
experimental data and the conditions mentioned in [8]......................................11
Figure 4: Describes the locations for the speed velocity profile measurements[8].
............................................................................................................................ 12
Figure 5: Graphs comparing the velocity profiles of the calculated values tot the
experimental results. Black line represents the computational data and the red
asterisks are the experimental data points. (Top) graph comparing the profiles at
rake 2 from figure 4, (middle) graph comparing the profiles at rake 4 from figure
4, (bottom) graph comparing the profiles at rake 7 from figure 4.......................13
Figure 6: The y+ measured over the chord of the NACA 4412 aerofoil................14
Figure 7: Graphs comparing the velocity profiles of the calculated values tot the
experimental results. Black line represents the computational data and the red
asterisks are the experimental data points. (Top) graph comparing the profiles at
rake 2 from figure 4, (middle) graph comparing the profiles at rake 4 from figure
4, (bottom) graph comparing the profiles at rake 7 from figure 4.......................15
Figure 8: Comparison of the pressure distribution calculated using the
k SST
Figure 14: Comparing the SLE + flap deployed aerofoil in the landing and take-of
configurations...................................................................................................... 21
Figure 15: Graph comparing the results provided DLR (blue) to the computed
results (red) for the take-of configuration...........................................................22
Figure 16: Describes the SLE + flap morphed configuration................................22
Figure 17: Close up of the mesh for the SLE + Flap morphed configuration........23
Figure 18: Graph comparing the pressure distribution computed (solid green)
with the reference data (dashed red) supplied....................................................24
Figure 19: Graph comparing the pressure distribution computed using the
Gamma-Theta model (solid green) with the reference data (dashed red) supplied
and the previous results (dashed green).............................................................24
Figure 20: Graph comparing the pressure distribution computed using the
Gamma-Theta model (solid green) with the reference data (dashed red) supplied
and the previous results (dashed green); over the flap region (left) and leading
edge (right).......................................................................................................... 25
Figure 21: Structural model of the clean aerofoil (left) with a close up of the
leading edge mesh and load conditions (right)....................................................26
Figure 22: Results from the load case described above.......................................26
Figure 23: Structural model of the clean aerofoil with spars (left) with a close up
of the leading edge mesh and load conditions (right) the response from the
updated structural model (bottom).....................................................................27
Figure 24: Contour plot of the response due to an arbitrary applied load
resembling the correct direction vectors.............................................................28
Figure 25: Fluid-Structure model (left) shows the far field fluid model and (right)
shows the structural solid.................................................................................... 29
Figure 26: Fluid-Structure model (left) shows the far field fluid mesh and (right)
shows the structural solid mesh..........................................................................29
Figure 27: Figure describing the results obtained from the initial coupling model.
............................................................................................................................ 30
Figure 28: Updated structural model for the fluid-structure model......................30
Figure 29: Shows the mesh elements created in the dedicated structural solver
(left); (right) image shows the elements created by the fluid-structure solver....31
Figure 30: The converged solution of the failed aerofoil model...........................31
Figure 31: Structural model with contact applied................................................32
Figure 32: Image of the leading edge created using the fluid-structure software;
(left) close-up of the problematic region, (right) shows the overall structural
geometry.............................................................................................................. 32
Figure 33: (Top) shows the results of the coupled model at a time-step of 0.05
seconds; (bottom) shows the same model at a time-step of 0.35 seconds.........33
Figure 34: Structural model including the trailing edge section...........................34
Figure 35: Figure describing the mesh displacement and velocity contour of the
leading and trailing edge model..........................................................................34
Figure 36: (Top left) graph describing the residuals for the mass and momentum
equation, (top right) graph describing the convergence of the structural model,
(bottom left) graph describing the convergence across the interface boundary,
(bottom right) graph describing the displacement calculated at each iteration.. 35
List of Equation
List of Tables
Table 1: Table of aerodynamic parameters used in the individual analysis[8].....10
1. Introduction
The study of aircraft dynamics is troublesome enough with the creation of a
computational grid or modelling of turbulence. However this only gives a small
picture of the real physical world. It has been of great interest to many people to
be able to model the true world in its entirety; one step towards this would be to
combine the physical models which are already in existence. For example the
aeroelastic efects of a wing submersed in a moving fluid would usually require
the completion of two separate studies, combining the results later. However the
interactivity of these two wholly diferent fields is more like a continuum, where
one field can and does afect the other instantaneously. Furthermore, a solution
conforming to such true to life detail would be considered too computationally
expensive to be of any useful value, especially during the design process where
one requires quick analytical tools to assess the strength of an idea. Using this
analysis method it would also allow the practitioner to design and optimise for a
truer to world solution[1].
Moreover, reduction in emissions of aircrafts has become one of the upmost
important design factors, not just to save the environment but to also reduce the
draw on a depleting resource. As described in the previous interim report the
overall aim of the project is to tackle the issue of noise and air pollution around
airports.
developing the basic skills required to model the true aerofoils once they had
been made available. What is more a short introductory review of some coupling
procedures were studied to develop a better understanding for the in-depth
research carried out in this paper.
After which the fluid solution was combined with the appropriate
structural model using the MFX protocol to generate a more detailed automated
fluid-structure interaction model. Servo loads can then be applied to the skin of
the structure to generate the displacements prescribed by the consortium. The
author would then be in a position to convert the two dimensional models into
three dimensional models. Finally the positions of the spars would be optimised
to take advantage of the aeroelastic forces.
3. Literature Review
3. 1. Aeroelasticity and Its Modelling
Aeroelasticity is simple in context it is merely the interaction of inertial,
elastic and aerodynamic forces; however the interaction of these forces can
lead to very complex responses. Aeroelasticity can be broken down into two
sections, static and dynamic; the former ignores the inertial forces while the
latter includes efects such as vibrations. The aeroelastic tailoring of aircrafts
is becoming of increasing importance with the drive to reduce costs and
emissions. Before the days of such advanced computing power, structures
were stifened to the point at which the natural frequencies were far above
the dive speed of the aircraft. This required extra weight, bulkier structures
and left no room for deviation from the rigid structure. By incorporating the
aeroelastic bending and twist into the design it is hoped that the weight of
the aircraft can be reduced by removing or repositioning the support
structure.
Monolithic Procedure
The monolithic procedure strives to combine the two sets of
governing equation into a single system. This is harder than one would
first imagine due the combination of linear and non-linear terms, matrices
may be symmetrical and unsymmetrical.
expensive.
Partitioned Procedure
Partitioned procedures as the name suggests, solves the two
present independent solver codes such as CFX and Ansys can be used in
conjunction with an additional tool to handle data transfer.
Ansys is known as MFX.
This tool in
become apparent when one considers the interface layer; allowing the two
fields to be solved independently using whatever scheme and model suits
the situation.
including the issue of stability and judging whether the correct loads and
displacements have been transferred.
converged
value.
This
extra
convergence
also
means
the
3. 2. Fluid-Structure Interaction
To develop structures that can withstand the wear and tear of everyday
use but still be optimised for its particular duties is a troublesome task to say
the least.
cannot be avoided; others however are only due to the limitations of our
knowledge and our ability to deduce and manufacture the ideal solution. An
aircraft wing is a perfect example of compromise and optimisation. The
aerodynamic shape of the wing has been optimised for cruise. This is not a
very suitable shape for landing as the wings will inevitably stall at the high
angles of attack or not provide enough lift at the low speeds. However this is
4
overcome by attaching high lift devices to the wing to alleviate some of these
problems, but they themselves albeit necessary increase the complexity and
emissions of the aircraft. It was found earlier that the presence of the slot at
the leading edge of the wing accounted for a considerable proportion of the
noise of an approaching aircraft in its landing phase.
Due to the limitations of the computational resources a natural division
between the diferent physical behaviours grew.
It was computationally
easier to analyse the solid mechanics separately to the fluid dynamics to the
thermal dynamics.
The
transfer of parameters may be possible but as the solution progress and the
deformation increases these points will move further apart. In other words
the node transfer cannot be locked throughout the solution stage; thus a
5
coupling scheme is imperative to ensure that the correct nodes are receiving
the correct information at the appropriate time-step.
Many coupling procedures have been devised, ranging from the simple
manual transfer of data[4] to the more complex algorithms that account for
the energies within the boundary layer. Manual transfers require running two
separate analyses and transferring the loads between them through pressure
and displacement tables. It is a very laborious and long winded process with
the least accuracy of all the methods, but can prove useful when one is
carrying out a structural analysis which incorporates significant monotonous
aero-loads. A method described in [5] known as geometry smoothing method
was developed for handling complex geometries in which the mesh is kept
the same but the geometry is updated. Moreover, this method is useful when
the structural complexity may cause instability in the fluid solver. However,
the mesh density around the moving boundary would have to be resolute
enough to capture the important features. This technique may lead to a fluid
domain which is of poor or even unacceptable quality.
The
loss of conservation of forces across the interface. It was found in [8] that the
time step and level of flexibility of the structure play an important role in
maintaining stability of the numerical solution. In essence the more flexible
a structure and the shorter the time-step the greater the numerical instability
and the harder it will be to model.
3. 3. Structural Analysis
To produce an accurate fluid structure interaction model both the fluid and
structural model must be of a certain degree of accuracy; if either one of the
models is inaccurate in its response then this inaccuracy will be systemic in
the entire model. It is therefore necessary to gain a basic understanding of
the principle of a structural analysis so one can judge the correctness of the
solutions. There are many types of structural analysis available all of which
have been used and optimised for their specific duties for many years. These
techniques can be broken down into six main types; static, modal, harmonic,
transient dynamic, spectrum or a buckling analysis. For the purposes of this
study only the static and transient dynamic analysis will be looked at closer.
3.3.1.
Static Analysis
A static analysis is similar to that in an aeroelastic analysis, where
the inertial and damping terms are neglected. A static analysis assumes
that the variations are slow with respect to time. This is fairly obvious as a
static analysis can be assumed to have a very large time step leading to a
final solution.
3.3.2.
of the structure as the load is applied (or removed). Whereas the static
analysis ignored the inertia and damping terms by only including the
stifness matrix in
the
governing
equation;
the
dynamic analysis
( M ) {u }+ ( C ) {u }+ ( K ) {u }={F (t )
Equation 1: Governing equation for a dynamic analysis.
Where
M ,C
respectively.
and
u , u
and
displacement, respectively.
F (t )
run using the Newmark time integration technique which assumes that
there is a linear change in the acceleration over the time-step.
The
method efectively creates two additional equations which are then used
to solve the equations.
Both these methods have the added ability to be in linear or a non-linear
fashion. A linear analysis as the name suggests is best suited when there
is little change from the datum, any large deviations or abrupt changes in
the stress level. This may be true when steel is still undergoing elastic
deformation or the deviation of a flat plat with a moderate weight applied;
however past the yield strength or for large flexes and non-linear
behaviour begins to exhibit itself.
non-linear system:
never change and in others the force orientation may stay the same with
respect to the structure. Therefore one could say a sense of foresight or
initial modelling is required when selecting and applying the loads.
3.3.3.
before one could reasonably assume their model was adequate for an
analysis. The use of a full 3D element when modelling thin surfaces would
lead to elements of high aspect ratios along the surface; further the
matrices would become cumbersome and inefficient. It was found in [9] in
order to use these full three-dimensional elements one had to ensure a
difering order across the axes. For example, in [9] along the so called
chord the polynomial order was set to five whereas the order was set at
8
above case it would not suitable to use this technique. The use of shell
elements helps to alleviate this problem by modelling the thickness
numerically rather than having a physical node at each corner. All shell
elements have interpolation points through the thickness, which can be
altered to increase the accuracy depending on the loading and stress
conditions. According to the Ansys Structural User Guide an integration
step of 2 is required for elastic behaviour. To resolve nonlinear and plastic
strain regions more integration points are required through the thickness.
The shape function of the element predicts a constant shear strains and
stresses through the thickness. What is more, the normal stresses are set
to the negative value of the applied pressure on the element surface and
are linearly interpolated to the bottom surface.
4. Validation of CFD
4. 1.
Figure 1: Clean geometry profile for SADE (left), NACA 4412 aerofoil (right).
mesh was easily done by replacing the aerofoil in the existing mesh; altering
areas of dissimilarity and adapting the mesh for the flow condition.
comparison the two meshes have been displayed below:
10
For
Figure 2: Clean aerofoil mesh (top) NACA 4412 aerofoil mesh (bottom).
The aerofoils themselves are not similar by any means as can be seen
from the previous page; however this analysis should provide a sense of the
accuracy of the clean SADE aerofoil. For added confidence the wind tunnel
results provide both pressure distributions as well as boundary layer velocity
measurements; by comparing both sets of data it will be possible to evaluate
both the results generated close to the surface as well as the more general
pressure drop calculation. From the previously mentioned documentation the
corrected initial and boundary conditions for this particular case were found
and have been tabulated along with the details of the SADE wing below:
Parameter
SADE wing
NACA 4412
Chord (m)
Mach
0.15
0.18
Reynolds Number
20 106
4.17 106
12
12.49
Angle of Attack
( )
11
Again the similarities are few but the high incidence provides a good
reference to whether stalled flow was being captured.
Comparison of the Pressure Distribution Calculated Using the NACA 4412 Aerofoil With Experimental Data
-7
-6
-5
-4
Cp
-3
-2
Experimental
Computational Data
-1
1
2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
Chord
Figure 3: Graph comparing the pressure profile of the NACA 4412 with the experimental
data and the conditions mentioned in [10].
There is a clear similarity with the two graphs; however there also exists a
clear dissimilarity as the CFD has overestimated the pressure distribution
across the under surface of the aerofoil.
CL
CL
values.
CL
The
slightly higher percentage error of 8.8%, but still within the tolerances of this
study.
The separation zone was not predicted well, with the CFD results
most likely due to the large irregularities that would arise from the large angle
of attack, which the turbulence model did not capture. What is more, this
CL
and
CD
Figure 4: Describes the locations for the speed velocity profile measurements[10].
As the velocity profiles are created due to the viscous forces close to the
surface, a close match of these results should verify whether or not the
correct turbulence model was chosen.
that the results were a close match, as can be seen in the group of figures
below.
The top graph from figure 5 describes the fully turbulent boundary that
one would expect to develop so far downstream of the flow. The computed
velocity ratio, here shown as the solid black lines, was under predicted
leading one to think that the turbulence is being over estimated. This may
explain the delayed separation experienced in the calculated data; the
separation point was found to occur close to line 4.
However, the
experimental data showed severe stall by this point of the flow. This should
be expected as the turbulence model was efectively run as a fully turbulent
flow, which would not be the case in the experiment. Thus in an efort to
increase the accuracy a transition model was applied in conjunction with the
turbulence model.
13
14
0.01
Experimental
Computational Data
-0.01
0.00
0.20
0.40
0.60
0.80
1.00
1.20
-0.10
0.10
0.30
0.50
0.70
0.90
1.10
U/U*
d/c
-0
.0 0.0 0.0 0.1 0.1
1
3
7
1
5
Experimental
Computational Data
0.00
0.20
0.40
0.60
0.80
1.00
1.20
-0.10
0.10
0.30
0.50
0.70
0.90
1.10
U/U*
0.05 Experimental
Computational Data
0.00
-0.05
-0.10
0.10
0.30
0.50
0.70
0.90
1.10
-0.20
0.00
0.20
0.40
0.60
0.80
1.00
U/U*
15
Figure 5: Graphs comparing the velocity profiles of the calculated values tot the
experimental results. Black line represents the computational data and the red asterisks
are the experimental data points. (Top) graph comparing the profiles at rake 2 from figure
4, (middle) graph comparing the profiles at rake 4 from figure 4, (bottom) graph comparing
the profiles at rake 7 from figure 4.
Two transitions models were available for calculating this position, namely
the Gamma and the Gamma-Theta model.
additional equation for the intermittency of the flow but requires the user to
input the appropriate transition Reynolds number. This would be possible to
estimate with the following equation:
within 0.001< y
<5.
If the
+
y
+
y were to go below 0.001, the transition
point artificially moves downstream; the opposite is true for when the
+
y is
greater than 8. Any value above 25 and the transition zone would be too far
+
y
of the above
16
+
y along the chord
1.
1
1.
4
1.
7
-0
.1
0.
2
0.
5
0.
8
y+
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
Chord
Figure 6: The y+ measured over the chord of the NACA 4412 aerofoil.
+
y does remain within the specific range to
adequately resolve the transition region. The leading edge showed a large
spike in comparison to the remainder of the chart; however the peak only had
an amplitude of 1.9 which can be considered within the bounds of the
equation. Furthermore, this peak was expected as this section would contain
the thinnest fraction of the boundary layer.
+
y dropping to 0.02.
This area is of
greater concern as the bottom limit of the Gamma model was almost
reached.
reviewed earlier, namely line 2, 4 and 7, where once again overlaid on the
experimental data.
17
d/c
-0
.0 0.0 0.0 0.0 0.0
1
1
3
5
7
d/c
-0
.0 0.0 0.0 0.1 0.1
1
3
7
1
5
0.00
0.20
0.40
0.60
0.80
1.00
Without Gamma-Theta Model
-0.10
0.10
0.30
0.50
0.70
0.90
1.10
U/U*
d/c
-0
.0 0.0 0.1 0.2
5
5
5
5
Figure 7: Graphs comparing the velocity profiles of the calculated values tot the
experimental results. Black line represents the computational data and the red asterisks
are the experimental data points. (Top) graph comparing the profiles at rake 2 from figure
4, (middle) graph comparing the profiles at rake 4 from figure 4, (bottom) graph comparing
the profiles at rake 7 from figure 4.
18
The diferences are not clear at first glance; whence the previous results are
super-positioned it becomes clear that the Gamma-Theta model does improve
the correlation along the turbulent section of all the measurement lines.
However, along the trailing edge of the aerofoil where the flow is separated
the Gamma-Theta model seems to under-predicted the velocity ration.
Nevertheless the most important parameter to match was the pressure
distribution as this would be the information being transferred, thus a
comparison of the new pressure distribution can be seen below.
Comparison of the Pressure Distribution Calculated Using the NACA 4412 Aerofoil With Experimental Data
-7
-6
-5
-4
Cp
-3
-2
Experimental
Computational Data
-1
1
2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
Chord
k SST
model with the gamma-theta model active (dashed) of (solid) and also from experimental
work (red asterisk).
Once again the diference was minimal, with no increase of accuracy in the
trailing edge region.
Gamma-Theta model on the inlet viscosity ratio. It may also be because the
simple geometrical flow is being resolved to the best of the turbulence
models accuracy will allow. A final check for grid independence was carried
out to ensure this was the most accurate data available.
19
Graph Comparing The Experimental Pressure Distribution With The Low, Medium, High Density Grids
-7
-6
-5
Experimental Data
-4
-3
Cp
-2
-1
0
1
-0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95
Chord
Figure 9: Graph comparing the pressure distribution of the diferent mesh refinement
steps.
The low density mesh expectedly produced the worst results, with the
medium
and
high
mesh
density
grids
producing
similar
pressure
through each
20
paring The Pressure Distribution of The Clean Aerofoil Configuration With The Gamma-Theta Model Activated A
-10
-8
-6
Cp
-4
Gamma-Theta Enabled
Gamma-Theta Disabled
-2
0
2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Chord
-10
-8
-6
Cp
-4
-2
0
2
0
0.02
0.04
0.06
0.08
0.1
Chord
Figure 10: Graph comparing the pressure distribution calculated using the Gamma-Theta
model enabled (dashed) and disabled (solid).
As with the NACA 4412 investigation little diference was seen over the
majority of the aerofoil, with slight deviations near the leading edge of the
aerofoil. Therefore no firm decision was taken to discontinue the use of the
Gamma-Theta model as it may prove useful in the more complex models that
are to follow.
21
4. 2.
DLR Results
Computed Results
-5
0
0
1.1 1.2
Chord
Figure 11: Graph comparing the pressure distribution calculated with the results supplied
by the consortium.
The pressure peak could not be evaluated properly as the original data was
provided on a graph with the y axis limited to just -13. However the pressure
distribution showed some good agreement with the datum.
showed excellent correlation while the upper surface and leading edge were less
accurate. The author tried changing many parameters, such as the increase of
the grid density and the addition of the Gamma-Theta model, with no avail.
Furthermore, due to the limited data of the results it was almost impossible to
tell whether the peak zone was being modelled correctly.
22
4. 3.
Configuration
4.3.1. Validation of The Droop Nose and Flap Deployed in Landing
Configuration
Particular
attention
was
paid
to
meshing
the
droop
nose
Figure 12: Close up of mesh for the SLE + flap deployed configuration.
The pressure distribution was exported and compared to the data provided
by the SADE consortium which can be seen below:
DLR's Results
Computed Results
-1
Cp
-5
-9
-1
3
Graph Comparing The Pressure Distributions of The Computed Results and That Provided By SADE
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
Chord
Figure 13: Graph comparing the results provided DLR (blue) to the computed results (red)
for the landing configuration.
23
The addition of the Gamma-Theta model improved the results greatly; the
region over the flap was predicted with very little errors. However there
was a reduction in the pressure peak over the leading edge resulting in
larger errors. This reduction may be due to the inaccuracy in finding the
stagnation point at the leading edge. A close up of the leading edge and
the flap region is figured below:
Graph Comparing The Pressure Distributions
Graph
Over
Comparing
the Flap Region
The Pressure Distributions Over Leading Edge
-1
3.
00
-2.00
-1.00
Computed Results
Cp DLR's Results
-5
.0
0
Cp DLR's
0.00 Results
Computed Results
1.00
0.70
0.80
0.90
1.00
1.10
0.00
Chord
0.10
0.20
Chord
Figure 14: Close up of the flap section (left) and the leading edge peak (right).
Figure 15: Comparing the SLE + flap deployed aerofoil in the landing and take-of
configurations.
24
DLR Results
Computed Results
-2
Cp
-5
-8
-1
1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.1
Chord
Figure 16: Graph comparing the results provided DLR (blue) to the computed results (red)
for the take-of configuration.
Once again the correlation was excellent with very few deviations from the
datum graph; what is more the inaccuracy seen over leading edge was not
as prevalent in the above case.
prediction of the pressure has not been found and further studies are
continuing. This may be an issue as the proposed design solution would
have a flexible leading edge; hence the efect of the increased pressure
will inevitably have an efect. The level of influence will of course depend
on the flexibility of the structure and will be analysed at a later date. It is
the authors belief that the results obtained are of reasonable accuracy to
use at least towards the initial application of the fluid-structure interaction
phase.
4. 4.
Configuration
The final configuration released was the droop nose with the flap morphed
configuration; the flap geometry was optimised for the deflection sort by the
SADE consortium. The author noticed an unusual kink in the flap section of
25
the aerofoil; however as this data was supplied one would have to assume the
data is correct. The aerofoil is figured below:
The mesh was created similar to the previous mesh as once again the leading
edge remained unchanged; the flap region required more attention due to the
irregular geometry. The flap would need to be modelled accurately as the
flow over this region would greatly afect the flow over the leading edge. A
close up of the mesh can be seen below:
Figure 18: Close up of the mesh for the SLE + Flap morphed configuration.
Once again the pressure data was supplied by the partner responsible for the
creation and optimisation of this case; moreover for validity a short
comparison study of the computed results and the results given by CIRA was
started.
same parameters as those used in section 4.3.1; this was because the overall
geometric space had not been significantly changed.
The pressure
26
-4.00
CIRA Results
Low density
-2.00
0.00
2.00
0.00
0.20
0.40
0.60
0.80
1.00
1.20
Chord
Figure 19: Graph comparing the pressure distribution computed (solid green) with the
reference data (dashed red) supplied.
The first 21% showed great dissimilarity to the reference data; however the
remaining 79% showed excellent agreement. The erroneous region indicated
there may have meant an improper resolution was chosen for the leading
edge. As with the previous results the same mesh was run using the GammaTheta model enabled; it was the authors belief that the addition of the
Gamma-Theta model may assist in finding the stagnation point more
accurately. The graph below shows the updated pressure distribution:
Graph Comparing The CIRA Results With Those Computed
-10.00
-8.00
-6.00
Cp
Computed Data
CIRA Results
-4.00
-2.00
0.00
Low density
2.00
0.00
0.20
0.40
0.60
0.80
1.00
1.20
Chord
Figure 20: Graph comparing the pressure distribution computed using the Gamma-Theta
model (solid green) with the reference data (dashed red) supplied and the previous results
(dashed green).
27
From the above one can see the Gamma-Theta model had a great influence
on the pressure distribution improving the correlation over the flap region and
predicting a larger pressure spike.
leading edge has been expanded below to better compare the results.
Graph Comparing The CIRA Results With
Those
ComputedThe CIRA Results With Those Computed
Graph
Comparing
Cp
-4
-10
-3
-8
-2
-6
-1
Cp
0
1
2
0.80
-4
-2
Computed Data
CIRA Results
0
2
0.00
1.00
Chord
Low density
0.10
0.20
Chord
Figure 21: Graph comparing the pressure distribution computed using the Gamma-Theta
model (solid green) with the reference data (dashed red) supplied and the previous results
(dashed green); over the flap region (left) and leading edge (right).
It is clear that the Gamma-Theta assisted in the prediction of the flap region
showing very little deviation from the reference data up until the extreme
curvature of this section. However, the leading edge continued to show large
disagreement; as a final check for grid independence a convergence study
was completed for the aerofoil.
28
Medium Density
Low Density
-4.00
-2.00
High Density
0.00
2.00
0.00
0.20
0.40
0.60
0.80
1.00
1.20
Chord
Figure 22: Comparison of pressure distribution for the grid convergence study along with
the datum results.
5. Structural Model
For the fluid structure model to be truthful both the fluid model and the
structural model have to be of a similar accuracy; as the least accurate model
will determine the overall correctness of the results. Therefore the author felt it
best to create several structural models to assist with the initial modelling. It
was decided that the clean aerofoil would be modelled initially due to the
relatively simple geometry. The first model to be created was a shell model of
the aerofoil; the interior of the model was left hollow so not to complicate the
solution. A simple quad mesh was created around the aerofoil while trying to
maintain regularity over the leading edge. The figures below picture the above
case.
29
Figure 23: Structural model of the clean aerofoil (left) with a close up of the leading edge
mesh and load conditions (right).
It is clear that the solution is on nonsensical value but it does show that
restraining the model at the edges is an incorrect approach. What is more, the
S shaped contortions signify the force loading is incorrect with the structures
leading edge moving down rather than rotating. Therefore, to provide an area
for supporting the structure without hindering the analysis a simple wing box
structure was created.
30
Figure 25: Structural model of the clean aerofoil with spars (left) with a close up of the
leading edge mesh and load conditions (right) the response from the updated structural
model (bottom).
The response is once again not what is sort after; however by altering the
loading magnitudes directions it was possible to generate a shape of reasonable
approximation. The loading represents the force applied by an actuation system
to maintain the desired shape. An arbitrary loading pattern was applied to mimic
the actuation system and produce the bending. Once the coupling procedure
had been finalised the correct internal loading pattern would be applied; as
application of complex loading early on in the process will only hinder the
progress.
31
Figure 26: Contour plot of the response due to an arbitrary applied load resembling the
correct direction vectors.
The image above was created using this arbitrary load pattern and the response
is not unreasonable however not accurate either. It would be foolish to come to
the conclusion that a very fine mesh is required as displacement vectors can
easily be calculated, to a reasonable accuracy, with a very course. Nonetheless,
one cannot say a course mesh will suffice as the fluid elements close to the
interface may require more elements than a standard structural analysis to help
maintain the smoothness of the fluid surface. It was therefore the authors belief
that the structural model could be improved to the necessary level once the
coupling procedure had been resolved and the level of accuracy required
understood further.
32
6. Fluid-Structure Model
Using the above models it would be theoretically simple to couple them;
however due to the complexity in setting up the process a step-by-step process
was implemented.
One tutorial file existed for this feature with very little
Figure 27: Fluid-Structure model (left) shows the far field fluid model and (right) shows
the structural solid.
Both the structural and fluid space can be seen in the above figure; Ansys
Workbench was used as the initial coupling program.
structural model and fluid model be meshed separately; however the model can
be considered as a single simulation. A basic mesh was created for both fields
and can bee seen below:
Figure 28: Fluid-Structure model (left) shows the far field fluid mesh and (right) shows the
structural solid mesh.
The mesh was of a very course nature for both the structure but especially for
the fluid field. Furthermore, an unstructured grid was used for the fluid mesh
33
which required additional attention to the meshing process and the results from
the completed simulation.
case; the only real value was to develop the modelling technique to further the
research. The results from the case can be seen below:
Figure 29: Figure describing the results obtained from the initial coupling model.
The above diagram shows the results generated by the above model
configuration, no force data was applied to the aerofoil as the exchange of data
was the topic of study. The structural model was held in place at the trailing
edge; obviously this would be considered an incorrect form of modelling however
for initial purposes the author felt allowing the aerofoil to move was more
important than realism.
green line highlighted in the figure, the deflection were of the order one would
expect from motion due to aerodynamic forces.
This however was only an exercise of principles as the structural model was
clearly of no useful value. The structural model was improved by modelling the
skin as a surface; in doing so the same problem of supporting the structure faced
in the previous section had to be tackled.
34
The model however failed to solve due to what the error file described as a read
error. The cause of the error was unknown therefore a check of the two fields
had to be run to deduce some correct measures.
rechecked for inaccuracies; upon which the author discovered the fatal flaw in
modelling the aerofoil as described. It was found that in the dedicated solver the
T-joint at which the spar connects to the skin was resolved as a single structure;
whereas in the fluid-structure model the program misinterpreted the geometry
file. The figure below shows the diferences in the models. The left image shows
the mesh elements created in the dedicated structural solver; while the right
image shows the elements created by the fluid-structure solver.
Figure 31: Shows the mesh elements created in the dedicated structural solver (left);
(right) image shows the elements created by the fluid-structure solver.
From the images it can be seen the fluid-structure model does not model the Tjoint correctly.
35
Without a shadow of a doubt one can say the connections at each individual
planes edge had not been modelled correctly. Furthermore the contact between
the spar plane and the leading edge had not been resolved, thus allowing the
edge to travel through the structure. Upon closer inspection one can see the left
corner (facing the aerofoil) of the leading had remained attached while the
opposite corner remained attached to the spar. Furthermore the bottom edge
experienced no deviation from its datum shape signifying motion with no
contact. Contact was applied to the appropriate locations to efectively glue the
structure together.
However the model once again failed due to a negative volume error; in other
words elements within the model either fluid or structural collapsed in on
themselves causing a catastrophic failure of the solver. Furthermore, by adding
an artificial connection one may have inadvertently altered the load flux over the
surface by representing the structure with a non-physical model.
A further
complication arises in the choice of the contact model several choice were given
to the author; each having their advantages for their respective application. In
the above case a fixed body-body contact was used which restricts the
movement of the contact zones in all degrees of freedom.
Once again a test case was run for the author to familiarise himself with the
parameters; the structural model was updated by excluding everything behind of
the first spar. The spar position was left in place to support the structure and
avoid rigid body movement. The model can be seen below:
36
Figure 34: Image of the leading edge created using the fluid-structure software; (left)
close-up of the problematic region, (right) shows the overall structural geometry.
The above model is similar to the model mentioned earlier in which the manual
static aeroelastic load transfer was applied to. It was the authors belief that as
the leading edge would be the most difficult to model accurately therefore it
should be the first tackled. The skills learnt from completing the above case will
hopefully be easily extrapolated to the full model. The above model description
was taken from another PhD student currently modelling the leading edge of the
above aerofoil[4].
implement as one would have hoped as the interface region had to be modified
to account for the correct load. According to [7] if the region flagged to send
data does not match the receiving side the discrepant region is ignored;
furthermore, the receiving side be will automatically set the unmapped zone to
zero.
recreated the fluid mesh to allow one to select the leading edge as the interface.
37
Figure 35: (Top) shows the results of the coupled model at a time-step of 0.05 seconds;
(bottom) shows the same model at a time-step of 0.35 seconds.
The two images above describe the mesh displacement and velocity field of the
coupled modelled at varying time-steps. The structural model can be seen to
update through each of the time-steps however the fluid model looked to have
remained fairly steady state. This may be due to relatively small deviations from
the original shape, with the largest displacement being a mere 0.003m.
The
above diagram also shows that the fluid mesh is of poor quality, in fact severely
inadequate if this were anything other than a test case. The mesh quality was
left intentionally very course so as to prevent the early problem described,
namely the collapse of elements.
The above model was then expanded to include the trailing edge section with
the spar; however still ignoring the wing box section of the aerofoil.
The
geometry and mesh of the structural section has been highlighted below for
clarity.
As can be seen the two parts are completely separate from one another, this
may be valid for an analysis in which the local displacements and stresses are of
38
primary concern.
Figure 37: Figure describing the mesh displacement and velocity contour of the leading
and trailing edge model.
The above model produced similar if not the same results for the leading edge as
the previous model which only incorporated the named section.
The model
39
Figure 38: (Top left) graph describing the residuals for the mass and momentum equation, (top
right) graph describing the convergence of the structural model, (bottom left) graph describing the
convergence across the interface boundary, (bottom right) graph describing the displacement
calculated at each iteration.
According to the manual convergence for the two domains is attained in the
normal fashion; however at each time-step convergence is restarted therefore
one should expect the results to converge before spiking and settling to a new
asymptote level.
leading edge; the curved section shows the convergence of this displacement.
Convergence over the interface is judged diferently, the manual states once the
variables have settled to a value below zero the interface can be considered
converged. The fluid model showed a sudden jump in the residuals around the
40th iteration step; this step corresponded to the contour plot shown in bottom
image of figure 33.
40
1 10
conducted to generated pressure data; this data was exported onto a structural
model to generate a basic aeroelastic model.
The report consisted of a literature review of aeroelastic modelling and the
complications of modelling fluid-structure interaction were completed.
Special
attention was paid to modelling thin structures and the use of shell elements in
formulating a response. Validation of the work presented in the previous review
was completed to judge the accuracy and reliability of the models.
Close
attention had to be paid during the validation process as the aerofoil of interest
did not have publically available reference data. The clean aerofoil configuration
was the first to be validated; this was achieved by using a similar aerofoil, i.e. the
NACA 4412, with the same discretisation grid. It was found that the pressure
distribution was predicted very well; however the velocity profiles adjacent to the
aerofoil surface were not so accurate.
activated and altered, such as the Gamma-Theta and the grid density, to improve
the results.
correlation of the results to the datum; while the Gamma-Theta model showed no
significant improvement.
The droop nose with the flap deployed configuration was supplied in subconfigurations, landing and take-of; the only diference being the position and
inclination of the flap. Both configurations were validated via a grid convergence
study as well as comparing the pressure distribution supplied by a partner
company.
The
aerofoil in the landing configuration showed the some deviation; with the
greatest inaccuracies arising over the leading edge. The addition of the GammaTheta model increased the processing time greatly, however the model showed a
significant increase in accuracy of the predictions.
showed greater correlation than the landing case with an almost exact fit of the
two graphs.
41
The Droop nose with the flap morphed was the last configuration to be
modelled and validated. Once again the only datum for comparison was another
set of computational data supplied by a partner company. Careful attention was
given to leading edge as this region has shown to be a problematic area. It was
found that the leading edge once again showed poor agreement with the data
supplied by the partner; strangely however the trailing edge was modelled with
great precision. With the Gamma-Theta model activated the trailing edge results
were predicted almost identically as those provided.
Furthermore, a grid
modelling process when creating a structural model. The first model was of just
the skin; it was found that the constraints applied to the aerofoil were unrealistic.
The model was updated a skin structure with some internal supports placed in
their appropriate positions. The author was then able to constrain the model in a
more realistic fashion however the loading pattern used were incorrect. It was
the authors belief that the addition of complex loading patterns would hinder
the coupling process due to resulting complex motion.
The first fluid-structure model was created to develop an understanding of
setting up the coupling process. The fluid model was left similar to one created
and validated above however for ease of modelling the author decided to use a
solid aerofoil. The structural model was then modified to resemble the structural
model created earlier. However the modifications may have been too large of a
step in the modelling process as the model failed to solve. It was found the error
was due to an improper use of the importing function, which left a panel
detached from its host.
over the coupling issue. The fluid model was modified to allow the selection of
particular sections; this was utilised when the structure was split into the leading
edge and trailing edge sections.
The author intends to extend the current by including the wing box in the
structural model; the author proposes to achieve this by improving the structural
modelling techniques implemented.
experience to begin modelling the flap deployed configuration from which all the
other configurations will arise. The above models do not include any servo force
applied to the skin to cause the deflections.
pattern required to created the required deformation has been completed[4] and
can be used within the current model. In fact due to the use of the reference
geometry the loading patterns and techniques can be directly imported into the
structural model; giving an insight into the aeroservoelastic response of the
aerofoil. The concept is hoped to be expanded to a three dimensional model so
as the twist and bending experienced by the wing can also be incorporated into
the analysis of the high lift device.
43
8. Bibliography
1.
2.
3.
4.
5.
6.
de Boer, A., A.H. van Zuijlen, and H. Bijl, Review of Coupling Methods for
Non-Matching Meshes. Computer Methods in Applied Mechanics and
Engineering, 2007. 196(8): p. 1515-1525.
7.
8.
9.
Scholz, D., et al., Thin Solids for Fluid-Structure Interaction. 2006. p. 294335.
10.
44