Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Summary
Flapping wing aerodynamics is the basis of flight for the large majority of airborne creatures – particularly
insects. It has been shown that flapping wing flow cannot be elucidated from the viewpoint of fixed wing
aerodynamics. In fact flapping wing flow involves a complex interaction of fluid vortices and wing kinematics
which enable things like insects perform otherwise difficult maneuvers such as rapid turning and hovering. If
the mechanics of the flapping wing can be understood thoroughly, the principles may be applied to aircraft
such as Micro Air Vehicles (MAVs) which are used by organizations such as the military.
Computational Fluid Dynamics (CFD) and Experimental flow visualization are the primary ways of investigating
these flows. Simulation is normally done in the low Reynolds number regime which is typical for insect flight
and under these conditions the viscous forces in the fluid are dominant, causing vortices to roll up from the
leading and trailing edges. Various aspects have been shown to affect this phenomena such as the flapping
frequency, flapping amplitude angle, wing geometry and forward flight speed. Literature suggests that insects
may use an optimal configuration of these parameters when flying.
This report presents the results of a CFD approach to the analysis of a two-dimensional flapping wing using
OpenFOAM, a computational tool using the Finite Volume Method to computationally solve the governing
equations of incompressible fluid flow. This is done by comparing numerical simulations to experimental
Particle Image Velocimetry (PIV) flow visualizations of flow over a quasi-elliptical wing section to establish the
validity of the solver used. Subsequently the parameters of motion are varied, such as the Reynolds number,
Strouhal number and the pitch angle amplitude. The way in which each of these parameters affects the
development of the flow is investigated in addition to the effect of the flow on the aerodynamic forces on the
wing.
Results have shown that the simulated predictions are in excellent agreement with PIV visualizations.
Qualitatively, all features of the flow identified in experiments have been observed in the computations.
Quantitatively, the vortex size and intensity is also predicted well with a clear match being shown between
simulation and experiment. The variation of Reynolds number, Strouhal number and pitch angle amplitude
were all shown to have an influence on the flow evolution. The Reynolds number elicited the least effect at
least in the range that was tested (500 to 1500), changing the aerodynamic wing forces only marginally.
Strouhal number had a marked effect on the flow and results indicate that higher Strouhal numbers lead to
higher peak values of lift and significantly more thrust production. Finally, varying the pitch angle amplitude
clearly demonstrated an optimal amplitude at which lift and drag production was maximized.
These findings have provided insight into the different aspects unsteady flows over a flapping quasi-elliptical
wing that require consideration in terms of providing a better understanding of insect flight and the design of
MAVs. Additionally, this work raises many questions for potential further study.
i
Monash University FYP 2010 Pavaman Bilgi 19334915
Nomenclature
𝑡𝑡 Time s
𝜎𝜎 Stress tensor
𝐈𝐈 Identity matrix
𝜈𝜈 Dynamic viscosity m2 ⁄s
𝑓𝑓 Frequency 1⁄s
𝐿𝐿 Reference length m
ii
Monash University FYP 2010 Pavaman Bilgi 19334915
𝑐𝑐 Chord length m
𝑆𝑆 Surface area m2
𝑊𝑊𝑀𝑀𝑖𝑖 𝑆𝑆𝑗𝑗 Weight factor of 𝑖𝑖 𝑡𝑡ℎ master patch face to 𝑗𝑗𝑡𝑡ℎ slave patch face
𝑆𝑆∩𝑀𝑀𝑀𝑀 Intersection areas of master patch face and slave patch face
iii
Monash University FYP 2010 Pavaman Bilgi 19334915
Table of Contents
Summary.................................................................................................................................................................... i
Nomenclature ........................................................................................................................................................... ii
List of figures and tables .......................................................................................................................................... vi
1. Introduction ...................................................................................................................................................... 1
1.1 Motivation ................................................................................................................................................ 1
1.2 Flapping wing flow .................................................................................................................................... 3
1.3 OpenFOAM ............................................................................................................................................... 4
1.4 Objectives and outline .............................................................................................................................. 5
2. Methodology .................................................................................................................................................... 6
2.1 Numerical Solution Method ..................................................................................................................... 6
2.1.1 The governing equations .................................................................................................................. 6
2.1.2 Discretisation .................................................................................................................................... 7
2.1.3 Generating a solution grid ................................................................................................................ 9
2.1.4 Solving the equations ..................................................................................................................... 10
2.2 Problem definition and set-up................................................................................................................ 11
2.3 Meshing and mesh movement ............................................................................................................... 14
2.3.1 Moving Mesh .................................................................................................................................. 14
2.3.2 2D meshing of a symmetric wing ................................................................................................... 17
2.4 Boundary conditions and validation ....................................................................................................... 20
2.4.1 Setup and boundary conditions ..................................................................................................... 20
2.4.2 Validation........................................................................................................................................ 22
3. Results and Discussion.................................................................................................................................... 24
3.1 Mesh resolution, temporal resolution and domain independence study ............................................. 24
3.2 Comparison of results with experiment ................................................................................................. 27
3.2.1 Qualitative comparison .................................................................................................................. 27
3.2.2 Quantitative comparison ................................................................................................................ 33
3.3 Effect of Reynolds number ..................................................................................................................... 35
3.4 Effect of Pitch angle amplitude .............................................................................................................. 36
3.5 Effect of Strouhal number ...................................................................................................................... 38
iv
Monash University FYP 2010 Pavaman Bilgi 19334915
v
Monash University FYP 2010 Pavaman Bilgi 19334915
Figure 26 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=500,
St=0.3 and 𝜽𝜽𝜽𝜽 = 𝟐𝟐𝟐𝟐° ............................................................................................................................................. 28
U
Figure 27 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=500,
St=0.3 and 𝜽𝜽𝜽𝜽 = 𝟐𝟐𝟐𝟐° ............................................................................................................................................. 29
U
Figure 28 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=1500,
St=0.2 and 𝜽𝜽𝜽𝜽 = 𝟐𝟐𝟐𝟐 ............................................................................................................................................... 31
U
Figure 29 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=1500,
St=0.2 and 𝜽𝜽𝜽𝜽 = 𝟐𝟐𝟐𝟐 ............................................................................................................................................... 32
U
Figure 30 Vortex movement comparison with experiment for 𝑹𝑹𝑹𝑹 = 𝟓𝟓𝟓𝟓𝟓𝟓, 𝑺𝑺𝑺𝑺 = 𝟎𝟎. 𝟑𝟑, 𝜽𝜽𝜽𝜽 = 𝟐𝟐𝟐𝟐° ....................... 33 U
vi
Monash University FYP 2010 Pavaman Bilgi 19334915
Figure 35 Drag coefficient as a function of pitch angle for different pitch angle amplitudes ............................... 37
Figure 36 Lift coefficient as a function of pitch angle for different pitch angle amplitudes .................................. 37
Figure 37 Lift coefficient as a function of pitch angle for different Strouhal numbers .......................................... 38
Figure 38 Drag coefficient as a function of pitch angle for different Strouhal numbers ....................................... 39
Figure 39 Complete mesh of the domain ............................................................................................................... 45
Figure 40 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝜽𝜽𝜽𝜽 for different 𝑹𝑹𝑹𝑹 (a) test number 1 (b) test number 4 and
(c) test number 6 from table 2 ............................................................................................................................... 45
Figure 41 Continuation of figure 41 ....................................................................................................................... 46
Figure 42 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝜽𝜽𝜽𝜽 for different 𝜽𝜽𝜽𝜽 (a) test number 2 (b) test number 8 and
(c) test number 9 from table 2 ............................................................................................................................... 47
Figure 43 Continuation of figure 42 ....................................................................................................................... 48
Figure 44 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝜽𝜽𝜽𝜽 for different 𝑺𝑺𝑺𝑺 (a) test number 1 (b) test number 2 and
(c) test number 3 from table 2 ............................................................................................................................... 49
Figure 45 Continuation of figure 44 ....................................................................................................................... 50
Figure 46 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝜽𝜽𝜽𝜽 (a) test number 5 and (b) test number 7 from table 2 .. 51
Figure 47 Continuation of figure 46 ....................................................................................................................... 52
vii
Monash University FYP 2010 Pavaman Bilgi 19334915
1. Introduction
1.1 Motivation
Although man-made flying vehicles are well known to have begun from the lauded achievement of the Wright
brothers in 1903, Nature’s own flying vehicles have arguably been in existence since the creation of the Earth.
While birds are assumed to be the evolutionary result of feathered dinosaurs, it is generally accepted that
insects took to flight far earlier still. Despite their ubiquity however, the aerodynamics of insect flight has
repeatedly continued to evade the efforts of conventional wing theory to explain the mechanisms of lift
generation due to the flapping motion of their wings. Until only recently however (due in most part to
advancements in experimental and numerical methods and technologies), the flow physics of flapping wings is
slowly being unraveled by continued experimental flow visualization and numerical computation (Sane, 2003).
While the endeavor to unveil the intricacies of insect flight is an interesting one, its salient value is in its
applicability to modern aircraft design. Though the mechanisms and aerodynamics of fixed wing aircraft are
generally thought to be well understood, the same cannot be said of their micro counterparts otherwise known
as Micro Air Vehicles (MAVs). The aerodynamics of MAVs (and small UAVs for that matter) is defined in a
different flow category to regular sized manned aircraft. Due to the combination of the relatively low velocities
and length scales involved, the flight regime of these aircraft in terms of the Reynolds number becomes quite
small. Refer to figure 1 (Mueller & DeLaurier, 2003) for a representation of the MAV flight regime relative to
mainstream aircraft. We can see from this figure that the flight regime of MAVs coincides with that of insect
flight in the low Reynolds number regime, implying that MAV design is quite amenable to the results obtained
by analyzing the mechanisms of insect flight – namely the flapping motion of wings.
In addition to the above argument, there are other reasons for MAV design to adapt to the design rationale of
the insect. Using current fixed wing design elements, MAVs face serious limitations in flight compared to their
insect counterparts. Due to their small size, advanced and rapid control mechanisms must be employed to deal
with gust response and unsteady aerodynamic loads (Rojratsirikul, Wang, & Gursul, 2009). They are unable to
perform tight maneuvers as required when deployed by armed forces. Finally, they face difficulty in
maintaining position without the ability to hover in one position – all limitations which flapping wing design can
overcome.
What is so interesting and complicated about the flow over flapping wings? Flapping wings induce complex
vortical structures generated from the leading and trailing edges of the wings which then interact with other
features of the flow in ways that are not entirely understood. This affects the forces and performance
characteristics in both hovering and forward flight. This topic has been the subject matter of research efforts in
both the experimental side (Ellington, van den Berg, Willmott, & Thomas, 1996), (Weish-Fogh & Jensen, 1956)
and the computational side (Wang, 2008), (Sun & Tang, 2002) by leading investigators in fluid mechanics.
Both approaches have their advantages and disadvantages yet the experimental approach is limited in ways
that the numerical one easily overcomes. Apparatus can often be quite expensive if not to purchase then to
manufacture such as flow visualization equipment and water tunnels. More importantly however, due to the
moving wing, flow visualization techniques such as digital PIV are hampered due to shadow effects and the fact
that force field data (integration of surface pressures) is not easily obtained. In Computational Fluid Dynamics
all information of the flow field is immediately available and provided that the parameters of the simulation are
rigorously justified it can be a valuable and efficient tool in solving the problem of flapping wings.
Thus, for the above reasons, a general investigation on the simulation of a flapping wing was undertaken in the
low Reynolds number regime i.e. Ο(100~2000).
2
Monash University FYP 2010 Pavaman Bilgi 19334915
Vortex generation
Vortex formation is a common phenomenon in aerodynamics. They occur due to roll up of shear layers in the
flow characterized by large viscous forces (which is the case at the low Reynolds numbers considered) in
situations like flow at the wing tips of large aircraft or at the hub of a wind turbine. They can have quite
detrimental effects due to the localized changes in pressure that they carry with them which in turn generate
fluctuating forces which can cause vibration induced damage to the structure of the flow object. However this
has also been seen to be a beneficial attribute to vortex generating flows in the case of insects and birds – this
is primarily due to the formation of the leading edge vortex of their flapping wings.
The significance of vortex shedding over the surfaces of a wing has been highlighted in the past (Maxworthy,
1979), (Ellington, van den Berg, Willmott, & Thomas, 1996). Particularly, it is the presence of the vortex created
at the leading edge of the flapping wing that makes way for lift forces much higher than those predicted by
conventional wing theory. It has also been shown (in Ellington et. al.) that this vortex is essentially three-
dimensional and that its evolution is highly sensitive to the advance rate of the wing (Reynolds number) and
the especially the wing kinematics. Indeed, a direct relationship has been observed between the roll up of the
shear layers producing the leading edge vortex to the wing motion characteristics (Lentink & Dickinson, 2009)
and it has also been shown that an optimization can be achieved in terms of the thrust efficiency and force
coefficients of a flapping wing in relation to the kinematic parameters of the motion (Wang Z. J., 2000). This
emphasizes the need to further investigate the details of the influences of the wing kinematics to the flight
performance of vehicles that can be designed with such wings.
3
Monash University FYP 2010 Pavaman Bilgi 19334915
Various wing kinematic parameters have been tested in the literature and most of them have been shown to
have pointed effects on the forces generated. Harmonic translational motion has been researched numerically
with small and large amplitudes of oscillation (Lentink & Gerritsma, 2003) while frequency was varied also
(Wang Z. J., 2000). It was found that by selecting certain values, the lift forces can be maximized. More studies
in the same vein were performed on heaving wings (Lewin & Haj-Hariri, 2003). Again, optimal oscillatory
frequencies and pitch amplitudes were proven to exist and the variation of aerodynamic efficiency was
established. The geometry of the actual wing has also been investigated (Lentink & Gerritsma, 2003) while
fixing other motion parameters like frequency and pitch amplitude. The conclusions from these studies were
that the geometry did not significantly affect the lift forces generated when compared to the effect of minor
changes in the motion parameters. Rotational parameters have also been varied (Dickinson M. H., 1994) and
the results of this research have shown that the axis of rotation and the rotation speed profile are heavily
important in the lift generation of the wing.
Given the above broad area of literature pertaining to a wide spectrum of motion parameters of flapping wings,
the scope of this project has been chosen to be confined to a focused analysis of a few parameters including
the Reynolds number, geometry and pitch frequency. In this way it is hoped that this project will make a more
valuable contribution to the field than the ‘shotgun’ approach of detailing results for various independent
groups of parameter values.
An ongoing question posed by researchers in the problem of flapping wings is the accuracy of two-dimensional
simulations and experimental results compared to the three-dimensional ones. Additional issues introduced in
the three-dimensional problem are unsteady flow mechanisms, wing flexibility and wing tip vortex generation.
Studies have been made into the comparison between two and three dimensions (Wang Z. J., 2004) where
three dimensional experimental results of the Robofly (experimental robotic test-bed for flapping insect wings
at UC Berkely) to two dimensional results. The findings there and elsewhere (Dong, Mittal, Bozkurttas, & Najjar,
2005) (Blondeaux, Fornarelli, & Guglielmini, 2005) confirm that the two dimensional studies often over-predict
the lift and drag forces and do not account for the energy loss that occurs in the third dimension. Three
dimensional vortices were observed to exist in the wake of finite span wings.
Nevertheless, studies have been performed in recent times on two-dimensional flapping wing flow (Bos,
Lentink, & van Oudheusden, 2008) (Thaweewat, Bos, van Oudheusden, & Bijl, 2009). The justification used in
these cases is that while there are unresolved flow features in the 3D case, there is sufficient similarity between
the 3D and 2D cases (Wang Z. J., 2000b) to ensure that a parametric study in 2D can be quite useful. Given that
a 3D computation can be quite expensive, the scope of this project has until now been limited to the two
dimensional simulation of a solely pitching wing.
1.3 OpenFOAM
OpenFOAM is an open-source developed suite of finite volume solvers and utilities which for obvious reasons is
an attractive alternative to commercial codes. Furthermore it provides one with an object oriented
programming environment allowing the user far more flexibility than commercial codes do. OpenFOAM has
been validated in several PhD and MSc theses around the world and has been adapted for purposes such as
4
Monash University FYP 2010 Pavaman Bilgi 19334915
free surface flow, multi-phase flow, Direct Numerical Simulation, Large Eddy Simulation, turbulence modeling
and fluid-structure interaction (Jasak, 2010). Thus by demonstrating the capability of OpenFOAM in the
capacity of simulating moving boundary problems like flapping wings the present study aims to open the doors
to further work in this area using this tool.
Additionally, due to OpenFOAM being a relatively new software in CFD arena compared to mainstay packages
such as Fluent, CFX and StarCD, and more so due to its open-source nature, it would be of benefit to document
the method of setting up this type of problem in terms of solvers and utilities used.
To address these objectives, the next few sections deal with the methodology used to setup and numerically
solve the flow field around a pitching wing. This involves a description of the Finite Volume discretisation
method used by OpenFOAM and basic overview of the solution algorithm used. Further, the method used to
handle the moving boundary is covered followed by the rationale for the grid construction and boundary
conditions used. Finally, the motion parameters and important variables are defined and the experimental
results used for comparison are briefly covered.
Subsequently a presentation of the results is given. This entails a comparison of simulation data to
experimental data and a qualitative and quantitative discussion of the various points of similarity and
difference. The next part will detail the results of varying the motion and flow parameters. It will be seen that
these change in parameters affect the development and size of the leading edge vortex which consequently has
ramifications on the forces generated. Finally, the conclusions and recommendations for future work will
follow.
5
Monash University FYP 2010 Pavaman Bilgi 19334915
2. Methodology
𝜕𝜕𝜕𝜕
Mass cons. + 𝛁𝛁 ⋅ (𝜌𝜌𝐮𝐮) = 0 (2.1)
𝜕𝜕𝜕𝜕
Momentum 𝜕𝜕(𝜌𝜌𝐮𝐮)
+ 𝛁𝛁 ⋅ (𝜌𝜌𝐮𝐮𝐮𝐮) = 𝛁𝛁 ⋅ 𝛔𝛔 (2.2)
Cons. 𝜕𝜕𝜕𝜕
𝜕𝜕(𝜌𝜌𝜌𝜌)
Energy cons. + 𝛁𝛁 ⋅ (𝜌𝜌𝜌𝜌𝐮𝐮) = 𝛁𝛁 ⋅ (𝜎𝜎𝐮𝐮) − 𝛁𝛁 ⋅ 𝐪𝐪 + 𝜌𝜌𝜌𝜌 (2.3)
𝜕𝜕𝜕𝜕
2
Stress tensor 𝛔𝛔 = − �𝑝𝑝 + 3𝜇𝜇𝛁𝛁 ⋅ 𝐮𝐮� I + 𝜇𝜇(𝛁𝛁𝛁𝛁 + 𝛁𝛁𝐮𝐮𝑇𝑇 )
Here, the stress tensor is defined for a Newtonian fluid. Since the present work deals with flow around insect
wings, it is considered to be incompressible and more importantly, laminar (Williamson, 1995). Given these
conditions, we may make some simplifications to the above equations. When a fluid is incompressible, the flow
velocity is nowhere greater than 30% of the local speed of sound and thermal expansion effects may be
neglected. Under these conditions, the Navier-Stokes equations reduce to,
This system of equations is closed so there is no need for further relations such as the energy equation.
6
Monash University FYP 2010 Pavaman Bilgi 19334915
For the purposes of comparison of different solutions of the incompressible NS equations it is preferred that
the dimensionless variables be used. These are defined below:
We may substitute equation (2.6) into (2.4) and (2.5) to obtain the following non-dimensional form of the
continuity and momentum equations:
2.1.2 Discretisation
To solve the equations (2.7) and (2.8) in a numerical fashion, the solution domain needs to be discretised. In
this particular case this means dividing the domain into a set of discrete control volumes that are joined by
faces and do not overlap. Each field variable is the solved at the center of each control volume. Logically, the
smaller the volumes become, the more accurate the solution will be and as the volume approaches zero, the
error of the solution also approaches zero.
The center of each control volume, 𝐱𝐱 𝐏𝐏 is defined by the equation, ∫𝑉𝑉 (𝐱𝐱 − 𝐱𝐱 𝐏𝐏 )𝑑𝑑𝑑𝑑 = 0. Each of these control
𝑃𝑃
volumes (tetrahedral, hexahedral or polyhedral) then has a certain number of faces and points associated with
it and the governing equations are to be approximated over these cells. This is done by assuming a linear
variation of any field variable across any two cells centers. Thus, using the notation in figure 4 where 𝑓𝑓 and 𝑆𝑆𝑓𝑓
7
Monash University FYP 2010 Pavaman Bilgi 19334915
denote the center of and normal vector of each face respectively, we can use a Taylor expansion on any scalar
field variable, 𝜙𝜙 about the cell center of any volume to obtain,
𝜕𝜕𝐮𝐮 𝛁𝛁𝑝𝑝
� + � 𝛁𝛁 ⋅ (𝐮𝐮𝐮𝐮) − � 𝛁𝛁 ⋅ (𝜈𝜈𝛁𝛁𝛁𝛁)𝑑𝑑𝑑𝑑 = � (2.11)
𝑉𝑉𝐶𝐶𝐶𝐶 𝜕𝜕𝜕𝜕 𝑉𝑉𝐶𝐶𝐶𝐶 𝑉𝑉𝐶𝐶𝐶𝐶 𝑉𝑉𝐶𝐶𝐶𝐶 𝜌𝜌
Before discretising each term we establish the discretisation of a general volume, surface, divergence and
gradient integrals of a scalar field variable, 𝜙𝜙 (this may represent a velocity component for example) and
vector, 𝐚𝐚 (perhaps a velocity vector).
• ∫𝑉𝑉 𝜙𝜙(𝑥𝑥)𝑑𝑑𝑑𝑑 = ∫𝑉𝑉 𝜙𝜙𝑃𝑃 + (𝐱𝐱 − 𝐱𝐱 𝐏𝐏 )(𝛁𝛁𝜙𝜙)𝑃𝑃 𝑑𝑑𝑑𝑑 = 𝜙𝜙𝑃𝑃 ∫𝑉𝑉 𝑑𝑑𝑑𝑑 + �∫𝑉𝑉 (𝐱𝐱 − 𝐱𝐱 𝐏𝐏 )𝑑𝑑𝑑𝑑 � ⋅ (𝛁𝛁𝜙𝜙𝑃𝑃 ) ≈ 𝜙𝜙𝑃𝑃 𝑉𝑉𝑃𝑃
𝑃𝑃 𝑃𝑃 𝑃𝑃 𝑃𝑃
• ∫𝑉𝑉 𝛁𝛁 ⋅ 𝐚𝐚𝑑𝑑𝑑𝑑 = ∮𝑆𝑆 𝐝𝐝𝐝𝐝 ⋅ 𝐚𝐚 = ∑𝑓𝑓 ∫𝑆𝑆 𝐝𝐝𝐝𝐝 ⋅ 𝐚𝐚 ≈ ∑𝑓𝑓 𝐝𝐝𝐒𝐒𝐟𝐟 ⋅ 𝐚𝐚𝐟𝐟
𝑃𝑃 𝐶𝐶𝐶𝐶 𝑓𝑓
In the above, Gauss’ formula is utilized which allows one to express a volume integral as a surface integral on
the volume. The integral is then evaluated using the finite number of faces on the volume in question.
• ∫𝑉𝑉 𝛁𝛁𝜙𝜙𝜙𝜙𝜙𝜙 = ∮𝑆𝑆 𝐝𝐝𝐝𝐝𝜙𝜙 = ∑𝑓𝑓 ∫𝑆𝑆 𝐝𝐝𝐝𝐝𝜙𝜙 ≈ ∑𝑓𝑓 𝐝𝐝𝐒𝐒𝐟𝐟 𝜙𝜙𝑓𝑓
𝑃𝑃 𝐶𝐶𝐶𝐶 𝑓𝑓
Finally, the terms evaluated at the surface, 𝑆𝑆𝑓𝑓 must be interpolated using a scheme of the users’ choice. The
surface interpolation schemes used for this project were the linear, the Koren and Van Leer limiter schemes.
The details of these schemes are not within the scope of this project although it was found that the Van Leer
scheme provided the best results.
Convection term
Diffusion term
8
Monash University FYP 2010 Pavaman Bilgi 19334915
Temporal term
Here, a number of different time differencing schemes can be adopted. For this project, the backward
difference second order implicit scheme was used (Hirsch, 1988).
𝜕𝜕𝜕𝜕 1 3 1
= � 𝜙𝜙 𝑛𝑛+1 − 𝜙𝜙 𝑛𝑛 + 𝜙𝜙 𝑛𝑛−1 �
𝜕𝜕𝜕𝜕 Δ𝑡𝑡 2 2
In this formulation, 𝜙𝜙 𝑛𝑛 denotes the state of the field variable 𝜙𝜙 at time 𝑛𝑛. As an aside, from stability and error
propagation analysis, the Courant number defined as,
𝑢𝑢 ⋅ Δ𝑡𝑡
𝐶𝐶𝐶𝐶 =
Δ𝑥𝑥
must be less than unity for the solution to be stable over time.
The limitation of the structured grid arises in the difficulty of creating it around complicated geometries which
is a case for using an unstructured grid. The cell shape and number of faces may be varied in an unstructured
grid which permits one to maintain high degrees of cell orthogonality even around relatively complex
boundaries. Compared to a structured grid however, the degree of cell orthogonality achievable of the
unstructured grid is not as high.
Cell orthogonality or more generally, cell ‘quality’ is crucial to the accuracy of a numerical solution. This is
because it affects the interpolation of the field variables between cell-centers at the cell surfaces. A
representation of non-orthogonality and skewness is given in figure 6. Here, it is the angle 𝛼𝛼𝑁𝑁 that is the
measure of orthogonality and this must be as low as possible for an accurate flow solution and particularly, to
9
Monash University FYP 2010 Pavaman Bilgi 19334915
minimize the error of the diffusion term. Skewness is increased when the line connecting two cell centers
moves away from the center of the interface of those two cells. The quantitative measure of cell skewness is
given by 𝜓𝜓 = |𝐦𝐦|⁄|𝐝𝐝|.
For this project it was decided to use structured grids wherever possible in the vicinity of the wing and the near
wake region and use unstructured grids a sufficient distance away from the wing. This is so that the important
flow features near the wing can be accurately resolved while the grid cell size can easily be inflated away from
the wing where comparatively no flow structures exist.
The second difficulty is that the equations are coupled – the velocity field depends on the pressure and vice-
versa. To this end, the PISO scheme was used (Issa, 1986). While the detailed explanation of this scheme is not
within the scope of this report, it will suffice to say that it essentially consists of three steps. First, the
momentum equation is solved for velocity using the old pressure field. Then, the pressure field is updated using
an equation combining the continuity and momentum equations. Finally, the velocity is also corrected using
the updated pressure field using a modified version of the momentum equation.
10
Monash University FYP 2010 Pavaman Bilgi 19334915
In the summer of 2005, two dimensional Digital PIV (DPIV) visualizations were carried out on the flow over a
symmetric aerofoil pitching about the mid-chord in the context of insect aerodynamics (Green, Parker, & Soria,
2005). This was performed in a water tunnel at the Laboratory of Turbulence Research for Aerospace and
Combustion (LTRA&C). The working section of the water tunnel measures 1m long with a 0.25m2 cross section.
Furthermore, turbulence intensity levels in the core region are purportedly less than 0.35%. Measurements
were taken of flow visualized in a single 2D plane at the mid-span of the wing. While the flow is intrinsically 3D,
for reasons outlined in section 1.2 these flow features were not pursued.
11
Monash University FYP 2010 Pavaman Bilgi 19334915
A quasi-elliptical cross-section aerofoil was used with a chord length of 31mm and an aspect ratio of 3.3 where
the aspect ratio is defined as the square of the span divided by the chord length of the section; 𝐴𝐴𝐴𝐴 = 𝑏𝑏 2 ⁄𝑐𝑐 . The
diagram below illustrates this.
The pitching displacement profile was a sinusoidal one enforced by stepper motors which rotate the wing
about the mid-chord point. The motors were programmable so that different pitch frequencies and amplitudes
could be tested, allowing for different displacement profiles. The diagram below illustrates the basic scheme of
the setup for a particular pitch amplitude of 20 degrees.
The flow was characterized by the Reynolds number and the Strouhal number introduced earlier. The Reynolds
number is a representation of the ratio of convection time to diffusion time in a flow. Alternatively, it is the
ratio of inertial to viscous forces in a fluid. Its magnitude will determine what kind of flow structures will be
responsible for the forces developed on a flapping wing – for example at low Reynolds numbers (which are
considered in the present study), the viscous term is dominant, leading to viscous phenomena such as shear
layers and the creation of vortices. The Strouhal number is normally used in the context of vortex shedding
flows and is quite relevant to flapping wing flows. It is defined by the flapping frequency multiplied by the
reference length of the flapping motion divided by the reference flow velocity.
12
Monash University FYP 2010 Pavaman Bilgi 19334915
In the DPIV experiments the Reynolds number and Strouhal number were varied. These are summarized in the
table below.
The sinusoidal motion may be parameterized in terms of the Reynolds number and Strouhal number as follows.
The displacement follows a function of the form,
𝐿𝐿𝑟𝑟𝑟𝑟𝑟𝑟 = 𝑐𝑐 ⋅ tan(𝜃𝜃𝑎𝑎 )
Further, the free stream velocity, 𝑈𝑈∞ can be represented in terms of the Reynolds number, 𝑅𝑅𝑅𝑅 as,
𝑅𝑅𝑅𝑅 ⋅ 𝜈𝜈 ⋅ 𝑆𝑆𝑆𝑆
𝜃𝜃(𝑡𝑡) = 𝜃𝜃𝑎𝑎 sin �2𝜋𝜋 ⋅ ⋅ 𝑡𝑡� radian (2.13)
𝑐𝑐 2 tan 𝜃𝜃𝑎𝑎
where 𝜈𝜈 = 1 × 10−6 m2 ⁄s which is the value for water at the temperature used in the experiment.
𝑐𝑐 2 tan 𝜃𝜃𝑎𝑎
𝑇𝑇 = seconds (2.14)
𝑅𝑅𝑅𝑅 ⋅ 𝜈𝜈 ⋅ 𝑆𝑆𝑆𝑆
In order to extend the findings of the experimental results, further values of 𝑆𝑆𝑆𝑆 and 𝑅𝑅𝑒𝑒 were tested in this study
in addition to varying 𝜃𝜃𝑎𝑎 , the pitch angle amplitude. This is summarized in the table below.
Test # 1 2 3 4 5 6 7 8 9
𝑹𝑹𝑹𝑹 500 500 500 1000 1000 1500 1500 500 500
𝑺𝑺𝑺𝑺 0.2 0.3 0.4 0.2 0.4 0.2 0.4 0.3 0.3
𝜽𝜽𝒂𝒂 (°) 20 20 20 20 20 20 20 30 40
13
Monash University FYP 2010 Pavaman Bilgi 19334915
Finally, this study also aims to extend the results of the experiment by considering the lift and drag forces
generated on the wing over time and analyse the effect of the vortex evolution on these forces. This will be
looked at in terms of the lift and drag coefficient defined as below.
2 (2.15)
𝐶𝐶𝐿𝐿 = 𝐿𝐿⁄(0.5 ⋅ 𝜌𝜌𝑈𝑈∞ 𝑆𝑆)
2 (2.16)
𝐶𝐶𝐷𝐷 = 𝐷𝐷⁄(0.5 ⋅ 𝜌𝜌𝑈𝑈∞ 𝑆𝑆)
One way to address the problem is to impose the boundary condition of a moving boundary at different nodes
for each time-step. This allows one to have as structured a grid as one wants. However this method introduces
challenges since the boundary does not always adhere to the grid nodes at all times and thus the term,
‘Immersed Boundary’ (Mittal & Iaccarino, 2005). Hence, interpolation algorithms are required in order to find
the appropriate boundary condition to impose at the nearest node locations. This can necessitate dense
gridding in the vicinity of the boundary movement and the interpolation algorithms need to be of a high order
– both of which can be computationally expensive.
Figure 10 Example mesh setup for immersed boundary method for blades in a mixer
14
Monash University FYP 2010 Pavaman Bilgi 19334915
Another approach is to define a new reference frame in which the boundary does not move. In the case of a
pitching wing for example, this would be a non-inertial reference frame that rotates with the wing. In this
reference frame, since the wing does not move, the boundary condition at the wing then becomes a no-slip
condition, while the Coriolis and centrifugal acceleration terms must be added to equation (2.5). These
additional terms add complexity to the solving method however, and the enforcing of the far-field boundary
condition also becomes complicated. This is because technically at the far-field, in a non-inertial reference
frame rotating with the body, the velocity is infinite by definition of the far-field (Sun & Tang, 2001).
Dynamic Meshing
The dynamic meshing method involves explicit topological changes in the mesh. When the individual nodes of
the boundary are moved at each time, the solver includes a routine which computes the redistribution of the
nodes of the mesh of the previous time-step. This method of treating moving boundaries is required by fluid-
structure interaction problems since the movement of the boundary is not prescribed. The mesh movement is
done in two ways – the first is to treat the connections between nodes as springs and to use for example,
Laplacian smoothing to redistribute the nodes. The second is to move each node individually, otherwise known
as point-by-point methods. All dynamic mesh methods require a separate mesh motion solver which can come
at a heavy computational expense.
The GGI method was developed for the purposes of turbo-machinery simulations where separate meshes are
normally created for different components of the system such as a rotor and a stator. The main advantage is
that it removes the necessity of topologically altering the mesh all the time while still letting the mesh move. In
the GGI method, it is the interface between two meshes which is treated at the matrix level in order to balance
the mass flow through the interface. Due to the simplicity of implementation and the computational efficiency
of the method, this method was chosen as the best way to approach the pitching wing problem.
15
Monash University FYP 2010 Pavaman Bilgi 19334915
Because the GGI method operates on the interface of two meshes, another advantage is that the resolutions
of the two meeting meshes need not be the same. The field variables of the flow are controlled over the
interface by weighting factors which are used in the interpolation between the meshes (Beaudoin & Jasak,
2000).
A GGI is represented by a meeting of a master and shadow patch. The master patch is composed of 𝑛𝑛 faces
while the shadow patch is composed of 𝑚𝑚 faces. During rotation, these patches do not meet seamlessly and
may intersect each other. Thus the field variables on a particular patch are to be calculated using the values on
the corresponding patch as necessary.
Equations (2.17) and (2.18) dictate that a field variable value on a particular face of a patch is calculated as a
weighted sum of the values on ‘neighboring’ faces. These neighboring faces are defined as those faces on the
opposing patch which intersect the face in question. Further constraints are imposed on these weighting
factors in order for the interpolation to be conservative.
16
Monash University FYP 2010 Pavaman Bilgi 19334915
�𝑆𝑆∩𝑀𝑀𝑆𝑆𝑖𝑖 �
𝑊𝑊𝑀𝑀𝑆𝑆𝑖𝑖 = (2.22)
�𝑆𝑆𝑀𝑀𝑛𝑛 �
�𝑆𝑆∩𝑆𝑆𝑀𝑀𝑗𝑗 �
𝑊𝑊𝑆𝑆𝑀𝑀𝑗𝑗 = (2.23)
�𝑆𝑆𝑆𝑆𝑚𝑚 �
Given this formulation for the weighting factors, it remains only to calculate the intersection areas of faces and
neighboring faces which is a geometric consideration. This involves the determination of which are the
neighborhood faces for each face on the master or slave patches and efficient algorithms are implemented in
the mesh motion solver to take care of this.
Inner mesh
On the inside of the GGI region, the inner part of the mesh was a circular region containing the wing. Two
general symmetric wing shapes were considered during this project – blunt edges and sharp edges. In the case
of blunt edges the wing can be described analytically as an ellipse with a semi-major and minor axis. The grid
constructed for the blunt edge wing shape is shown in figure 13.
Since the circle is simply a specific case of an ellipse it is not difficult to gradually release the elliptical plan-form
of a wing into a circle while maintaining a high degree of cell quality near the wing. When using sharp leading
and trailing edges however the wing profile is no longer well defined and is discontinuous. It was decided to
release the ‘sharp-edged’ wing into an ellipse which is then released into the outer circle.
The top and bottom surfaces of the sharp-edged wing were layered with highly structured quadrilateral
(hexahedral in 3D) mesh elements whereas the regions on the ends of the wing were filled with high density
high quality triangular (tetrahedral in 3D) mesh elements. This arrangement conveniently resulted in an
intermediate elliptical profile which was then released in a conformal fashion to the outer circle as shown in
figure 15.
17
Monash University FYP 2010 Pavaman Bilgi 19334915
18
Monash University FYP 2010 Pavaman Bilgi 19334915
Outer mesh
The outer mesh interfaces with the inner mesh at the GGI and also extends to the far-field boundaries where
the flow is expected to approach the free stream velocity profile. The nominal cell size in each area of the outer
mesh must be such that while the flow features in that area are adequately resolved, but also that excessive
computational time is not wasted in those regions. Important areas to consider in the outer region are the
forward and wake regions of the wing and top and bottom far-field boundary areas of the flow. The outer mesh
used is pictured in figure 16.
It was decided that high density quadrilateral meshing would be used for the near wake region while triangular
meshing with inflation would be applied to the rest of the domain. As will be shown in later sections, this was
also for the convenience of performing the grid resolution study where it became easy to adjust the resolution
of the parameterized wake region while keeping the rest of the mesh unchanged. It is also important to note
the choice of the ‘OH’ type grid. This is a standard approach when considering flow over a wing and is done so
that the inlet conditions imposed on the left boundary do not create unwanted reflections at the corners of a
normal ‘H’ type grid.
19
Monash University FYP 2010 Pavaman Bilgi 19334915
20
Monash University FYP 2010 Pavaman Bilgi 19334915
The constant directory contains full information of the mesh and is also where the dynamic mesh method is
specified. The system directory contains information on the discretisation schemes used and the solution
algorithms chosen to solve the equations for the solver to be used (icoDyMFoam in this case, which is the
standard solver for the incompressible NS equations in OpenFOAM). Finally the time directories are
destinations for solution output with the exception being the 0 time directory where initial and boundary
conditions are specified. Refer to Appendix B for the full details of the files.
To begin solving the equations intialisation of the flow is necessary. This was done using potentialFoam, the
potential flow solver in OpenFOAM to get a sensible starting distribution of velocity and pressure. The resultant
flow solution from this solver is then used as initialization to the icoFoam solver. However additionally,
boundary conditions must be specified in order to have a closed system of equations after discretisation
described in section 2.1.2. The boundary conditions specified are summarized in table 3.
𝑼𝑼 𝒑𝒑
Inlet 𝑈𝑈∞ zeroGradient
Outlet zeroGradient 0
Top/Bottom symmetryPlane symmetryPlane
Wing surface movingWallVelocity zeroGradient
The values of U and p at the boundaries must be such that their effect on the flow in the vicinity of the wing is
negligible. To this end, on the top and bottom of the domain (above and below) the wing, a symmetry
boundary condition is used for both U and p. This condition specifies the boundary to act like a mirror,
effectively setting the normal component of the solution at that boundary to zero.
The inlet and outlet boundaries are given a combination of Neumann and Dirichlet boundary conditions. At a
boundary the Dirichlet type condition sets the value of a variable while a Neumann condition sets the value of
the gradient of a variable. Thus the Neumann boundary condition allows for one dimensional variation on the
21
Monash University FYP 2010 Pavaman Bilgi 19334915
boundary. Since pressure perturbations propagate further upstream than downstream, pressure at the inlet is
given a Neumann condition while the velocity is set to free-stream (Dirichlet). Similarly, since velocity
perturbations propagate much further downstream than upstream, the velocity at the outlet is given a
Neumann condition while the pressure is set to free-stream (Dirichlet).
Finally, the condition imposed at the boundary of the wing itself was that of a moving wall for obvious reasons.
This means that this boundary condition is one that changes in time and OpenFOAM provides a routine to be
able to calculate the velocity of a moving boundary at all times. Should the velocity imposed at the wing
boundary not match the velocity of the boundary we should expect mass flux through the walls of the wing. An
excerpt from the flux field at the wing boundary shows that the moving wall velocity calculator is working very
well since we achieve (close to) zero mass flux through the wing wall boundary during motion:
2.4.2 Validation
Before beginning simulations a validation study was performed on the GGI method to ensure that the results
using this method were satisfactory. This was done by using the case of flow over a cylinder where the vortex
shedding is reminiscent of the wake of a flapping wing. Starting from flow over a stationary cylinder at a
Reynolds number of 1 × 105 , a circular GGI was placed in the wake of the cylinder with the inner mesh region
rotating sinusoidally. The solution was then compared to the case with no GGI and if the operation at the GGI is
working as expected, the differences between the solutions should be minimal.
22
Monash University FYP 2010 Pavaman Bilgi 19334915
In figure 18 we can see a schematic of the validation test. Furthermore in figure 19 we can see that the
subtraction of the solution with the GGI and that without gives a maximum error of less than 1%. This was done
with a sinusoidal rotation of the inner mesh at a Strouhal number of 0.3 where the length scale used was the
diameter of the cylinder. While this is a rather crude method of establishing the validity of the GGI method, due
to time constraints it would have to do. It should be noted that results may be different for different rotation
frequencies and Reynolds numbers. Further testing is required to establish these effects.
Figure 19 Subtraction of solution with GGI from that with no GGI. Maximum error is no more than 1%
The second validation study concerned the appropriate discretisation scheme for the convection term in
equation (2.11). This is because it was seen that the evolution of the generated vortices was dependent on the
amount of diffusion the convection scheme was producing. To do this, a set of discretisation schemes
commonly used for the convection term were used and the results were measured in terms of the vortex
preservation of the scheme i.e. the intensity of vortices after a specified amount of time. As can be seen in
figure 20, the vanLeerV scheme provided the best vortex energy preserving properties over time as the vortices
convect down-stream. This scheme was used throughout the present study. Refer to Appendix B for a full
account of all discretisation schemes used in the fvSchemes file.
1.6
Peak vorticity vs time
vanLeerV
Vortex Intensity
Koren
Linear
0
0 Time 7
Figure 20 Peak vorticity decay over time of a single shed vortex using different discretisation schemes for the convection term
23
Monash University FYP 2010 Pavaman Bilgi 19334915
24
Monash University FYP 2010 Pavaman Bilgi 19334915
The inner and immediate wake regions of the mesh are where the mesh resolution is critical. Hence these areas
were the focus of the mesh resolution study. Due to the implementation of the GGI method described in
section 2.3.1, the mesh resolutions of the inner and wake regions can be altered independently since the inner
and outer regions of the mesh are topologically disconnected. Thus, a separate resolution study was conducted
for each of these regions. The resolution of the inner mesh was ascertained before that of the wake region
since flow moves from the inner region to the wake region.
The resolution in the vicinity of the wing and the wake region was increased by decreasing the width of the
average cell in that region. Each time the resolution was increased, the vorticity solution with the new mesh
was interpolated onto the old mesh and subtracted from the solution of the old mesh. If this was greater than
1% anywhere in the domain, the resolution was further increased.
|𝜔𝜔𝑜𝑜𝑜𝑜𝑜𝑜 − 𝜔𝜔𝑛𝑛𝑛𝑛𝑛𝑛 |
< 0.01 for all 𝑥𝑥, 𝑦𝑦 in the inner mesh
|𝜔𝜔𝑜𝑜𝑜𝑜𝑜𝑜 |
A similar process was used for determining the appropriate temporal resolution of the simulation i.e. finding
the right time step size to use. A further limitation in the size of the time-step used is in the use of the GGI
method. Since a large time-step corresponds to a greater movement (rotation) of the inner mesh, this may
affect the quality of the interpolation at the GGI. This resulted in a requirement of 1500 time-steps per period
of pitching motion of the wing. For a Reynolds number of 500 and a Strouhal number of 0.3, this corresponds to
a time-step of 1.5 milliseconds.
Domain independence
Since the boundary conditions are derived at the ‘far-field’ boundaries (at a distance infinitely far away from
the wing), the actual distance of these boundaries needs to be established to obtain domain independence.
Due to the structure of the domain used, two parameters need to be determined; the radius of the inlet
boundary, 𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 and the 𝑥𝑥-coordinate of the outlet boundary, 𝑑𝑑. As 𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 and 𝑑𝑑 tend to infinity, we expect that
25
Monash University FYP 2010 Pavaman Bilgi 19334915
the solution (if adequately resolved by the mesh) to approach the exact solution. As before for the mesh
resolution, the domain size was increased until a solution residual of less than 1% was observed.
In the case of domain independence however, it was the residuals of lift force that were observed.
Test 𝒓𝒓𝒊𝒊𝒊𝒊𝒊𝒊𝒊𝒊𝒊𝒊 𝒅𝒅
1 5𝑐𝑐 10𝑐𝑐
2 10𝑐𝑐 15𝑐𝑐
3 15𝑐𝑐 20𝑐𝑐
4 20𝑐𝑐 25𝑐𝑐
4.05E-05
Peak lift vs test #
Lift/Unit span (N/m^2)
3.90E-05
3.75E-05
1 2 Test # 3 4
Figure 23 Peak lift over time for each domain independence test number
Domain size independence is clearly achieved at test number 3. This is represented in figure 24.
where 𝜙𝜙 represents any flow field variable. We see the flow reaching periodic steady state conditions in figure
25 of the lift and drag forces produced on the wing for test number 2 of table 2 – this is clearly around 4 to 5
periods of motion from initialization.
4.00E-06
0.00E+00
-25 -15 -5 5 15 25
Lift
-4.00E-06
Drag
-8.00E-06
Pitch angle (degrees)
Figure 25 Variation of lift and drag over time for 𝑹𝑹𝑹𝑹 = 𝟓𝟓𝟓𝟓𝟓𝟓, 𝑺𝑺𝑺𝑺 = 𝟎𝟎. 𝟑𝟑 and 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐°
The comparisons indicate good accuracy for the numerical simulations and most of the features of the flow
present in the PIV visualizations are also present in the simulations. In figure 26 (a) we can clearly see the roll
up of the LEV in the early stages of the pitching motion and at or around -10 degrees displacement the LEV
detaches from the leading edge. Shortly after this, the TEV also detaches from the trailing edge and begins to
convect downstream. Meanwhile the shear layer created at the trailing edge during the birth of the TEV begins
to move up the surface of the wing as indicated in figure 26 (d).
27
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
(d)
Figure 26 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=500, St=0.3 and 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐°
28
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
(d)
Figure 27 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=500, St=0.3 and 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐°
29
Monash University FYP 2010 Pavaman Bilgi 19334915
Figure 27 (a) then shows the LEV being absorbed into the advancing shear layer as the wing then begins to
pitch upward and move into the LEV. Since the shear layer and the LEV are of the same sign vorticity, they are
able to merge constructively. Subsequently, as seen in figure 27 (c) at 20 degrees positive displacement of the
wing, the shear layer has reached the leading edge of the wing which is about to pitch down. This shear layer
contributes to the roll-up development of the next LEV (with vorticity of direction into the page) which explains
why the LEV is created stronger than the TEV.
It should be noted that since we are dealing with symmetrical motion, the flow is symmetrical about the 𝑦𝑦 = 0
line. Thus although the flow on the bottom of the wing is not captured in the experimental visualizations due to
the laser shadow cast by the wing, this region of the flow is simply the reflection of the top side about 𝑦𝑦 = 0
with a phase lag of half a period of motion, or 𝑇𝑇/2.
Figure 28 and figure 29 depict comparisons of vorticity distributions for test number 6 in table 2. The velocity in
this case was 𝑈𝑈∞ = 0.05 m/s and the period, 𝑇𝑇 = 1.09 seconds which corresponds to a pitch frequency of
𝑓𝑓 = 0.92 Hz. Scaling was done as before and the contour colors range from red to blue and the vorticity from
−15 1/s to 15 1/s. Once again the comparisons show that the simulations are quite agreeable to the
experimental PIV results with all the flow features observed in the experimental results being present in the
simulations.
We can see in figure 28 (a) that the LEV has already been created in the first half of the pitch stroke while the
TEV is only beginning to form. Then, in figure 28 (c) we see that at 𝜃𝜃 = −20° the TEV has formed and detached
while the LEV is proceeding down-stream across the wing surface. As before in test number 2, we see that the
shear layer is advancing across the top surface of the wing – however we also see that while it comes into
contact with the LEV, they do not merge and the LEV proceeds down-stream and absorbs the rear of the shear
layer. Also worth noting is that in figure 29 (a) we can see that the LEV on the upper surface of the wing is
paired with a counter rotating vortex of much smaller proportions which is evident in the experimental results
also.
Thus, since the LEV is not absorbed, it is released at the trailing edge to join the TEV formed at time 𝑡𝑡 = 𝑇𝑇⁄2
from when the LEV was formed. This produces an oblique wake in which counter rotating vortex pairs travel
down-stream. The bottom part of the wake consists of pairs where a clockwise rotating vortex leads the pair
while the top part of the wake consists of pairs where the anti-clockwise cortex leads the pairs. These pairs
produce jets of fluid moving away from the wing in either direction (up and down).
Some differences arise between the simulations and experiment however, most notably in the distribution of
vorticity around each vortex. It seems that the vortices generated in simulation are markedly more coherent
and diffuse less when convecting down-stream. This is evident in figure 28 (a) where the right most TEV is very
diffuse in the experiment while still retaining coherence in the simulation. These differences are better seen in
the quantitative comparisons in section 3.2.2.
30
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
(d)
Figure 28 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=1500, St=0.2 and 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐
31
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
(d)
Figure 29 Numerical (left) and DPIV Experimental Data (right) sequential vorticity comparison for Re=1500, St=0.2 and 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐
32
Monash University FYP 2010 Pavaman Bilgi 19334915
We first consider the position of the LEV and TEV over time from the time of birth for test number 2 in table 2
(the flow solution is pictured in figures 26 and 27). Figure 30 shows a favorable comparsion of the movement of
the LEV and TEV over time – the simulations accurately predict the time of detachment of the vortices and their
speed as they travel downstream.
2
x-coordinate (position)
LEV
1.5
TEV
ExpLEV
1
ExpTEV
0.5
0
0 0.5 1 1.5 2 2.5
Time (s)
Figure 30 Vortex movement comparison with experiment for 𝑹𝑹𝑹𝑹 = 𝟓𝟓𝟓𝟓𝟓𝟓, 𝑺𝑺𝑺𝑺 = 𝟎𝟎. 𝟑𝟑, 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐°
Furthermore, in figure 31, the variation of the peak vorticity of the LEV and TEV are shown as compared to that
of the experiment. Again, the comparison is quite favorable with the initial rapid decay of both vortices being
correctly predicted and a slowing of the decay being observed later on. The quicker decay of the LEV can be
understood to be due to the interaction with the shear layer on the surface of the wing which is of oppositely
signed vorticity as seen in figure 26 (c).
Figures 32 and 33 show the peak vorticity comparisons for test number 6 in table 2. Naturally since a greater
amount of activity is present in these test conditions (due to a much higher Reynolds number) the comparisons
are not as favorable as for test number 2. Nonetheless patent agreement is found between simulation and
experiment in the behavior of the vortices.
33
Monash University FYP 2010 Pavaman Bilgi 19334915
30
25
Peak vorticity (1/s)
LEV
20
TEV
15 ExpLEV
ExpTEV
10
0
0 0.5 1 1.5 2 2.5
x-coordinate
Figure 32 clearly shows the accurate prediction of the vortex speed over time. The speed-up of the TEV is
evident in both simulation and experiment and is due to the separation of the vortex from the shear layer. This
can also be observed in visualizations of figure 29 (c) where the TEV is about to detach from the trailing edge.
The LEV on the other hand can be seen to be speeding up at around 0.5 seconds which corresponds
approximately to figure 29 (a) where the LEV is halfway across the upper surface of the wing. This is where the
3
Vortex position vs Time
2.5
Position (x-coordinate)
LEV
1.5
TEV
ExpLEV
1
ExpTEV
0.5
0
0 0.2 0.4 0.6 0.8 1 1.2
Time (s)
Figure 32 Vortex movement comparison with experiment for 𝑹𝑹𝑹𝑹 = 𝟏𝟏𝟏𝟏𝟏𝟏𝟏𝟏, 𝑺𝑺𝑺𝑺 = 𝟎𝟎. 𝟐𝟐, 𝜽𝜽𝒂𝒂 = 𝟐𝟐𝟐𝟐°
34
Monash University FYP 2010 Pavaman Bilgi 19334915
LEV splits the advancing shear layer into two parts and consumes the rear of the shear layer while proceeding
down the remainder of the wing. Finally, when the LEV departs from the trailing edge the speed of the vortex
begins to drop again. All of the above mentioned phenomena are observed in the computations and the
experimental results are a good indication.
As before in figure 30, figure 33 shows the variation of vortex intensity with vortex position for test number 6
of table 2. We can see that while the large changes in the rate of decay of the vortex are correctly modeled by
the simulations, smaller changes effected by complex phenomena are seemingly not represented. The LEV
begins a strong rate of decay from birth but then slows almost simultaneously with the increase in velocity
observed in figure 32. This indicates a possible correlation between the vortex velocity and the rate of decay of
the strength of the vortex.
According to the experiment however, the TEV is still building up strength where the simulations indicate that it
is decaying. Thus, differences arise concerning the formation and detaching of the TEV but agreement resumes
once the TEV is released from the wing at around 1.6 seconds corresponding to figure 29 (a).
40
35
Peak vorticity (1/s)
30
LEV
25
TEV
20 ExpLEV
15 ExpTEV
10
0
0 0.5 1 1.5 2 2.5 3
Vortex position (x-coordinate)
The results showed that the effect of Reynolds number was minimal. Looking at the flow solutions, at
𝑅𝑅𝑅𝑅 = 500, the LEV was seen to be consumed by the advancing shear layer. Increasing 𝑅𝑅𝑅𝑅 to 1000 results in the
creation of a stronger LEV which manages to survive the journey over the surface of the wing and travel into
the wake. Increasing 𝑅𝑅𝑅𝑅 further to 1500 magnified this effect and creates a more sparse wake structure.
35
Monash University FYP 2010 Pavaman Bilgi 19334915
Meanwhile the TEV remains unaffected by the change in Reynolds number. Refer to figure 40 and 41 in
Appendix A for vorticity solutions of these tests.
The behavior of 𝐶𝐶𝐿𝐿 and 𝐶𝐶𝐷𝐷 shown in figure 34 confirm the above observations. We can see that increasing the
Reynolds number marginally increases 𝐶𝐶𝐿𝐿 over all pitch angles and also reduces 𝐶𝐶𝐷𝐷 . One interesting observation
however is the thrust production (indicated by negative 𝐶𝐶𝐷𝐷 ) increase with increasing 𝑅𝑅𝑅𝑅. For 𝑅𝑅𝑅𝑅 = 500 thrust is
produced at 5° < |𝜃𝜃| < 14°. However at 𝑅𝑅𝑅𝑅 = 1500 the range of thrust production expands to 1° < |𝜃𝜃| <
16°. In addition, the thrust coefficient is almost tripled as the Reynolds number goes from 500 to 1500. This is
most likely due to the survival of the LEV over the rear surface of the wing at the higher Reynolds numbers.
1.5
1
Lift/Drag coefficient
0.5
0
-25 -20 -15 -10 -5 0 5 10 15 20 25
-0.5
Cd500
-1
Cl500
-1.5 Cd1000
Cl1000
-2 Cd1500
Cl1500
-2.5
Pitch angle (degrees)
Figure 34 Lift and drag coefficients as a function of pitch angle for a single period of motion
The vorticity plot comparisons shown in figures 42 and 43 in Appendix A show the differences in the flow at
intervals of Δ𝑡𝑡 = 0.125𝑇𝑇 or 1⁄8 𝑡𝑡ℎ of the period of pitching motion. We see that as 𝜃𝜃𝑎𝑎 is increased, while the
formation and travel of the TEV is relatively unaffected but that of the LEV undergoes interesting changes. The
strength of the LEV is ostensibly much greater as the pitch angle amplitude is increased however it is its
36
Monash University FYP 2010 Pavaman Bilgi 19334915
movement that plays a critical role in the generation of forces on the wing. In the case of 20° and 30° the LEV
remains close to the wing surface and since vortices are regions of low pressure, the LEV is responsible for the
generation of lift and thrust forces as it travels over the wing. In the 40° case however we notice that the LEV is
generated quite early and also separates from the leading edge early. Subsequently it is deflected by the rear of
the wing during the latter part of the stroke and moves away from the wing surface.
The lift and drag coefficient plots (figures 35 and 36) confirm the above observations by showing that lift and
thrust generation are maximized at 30° while it begins to drop after this point (for 𝜃𝜃𝑎𝑎 = 40°) due to the moving
away of the LEV at higher 𝜃𝜃𝑎𝑎 . We also notice that the lift generated for the pitch angle amplitude of 40° is
generally less than that produced at 𝜃𝜃𝑎𝑎 = 30°.
1.2
Drag coefficient
Cd20
0.8
Cd30
0.4 Cd40
0
-50 -40 -30 -20 -10 0 10 20 30 40 50
-0.4
Pitch angle
Figure 35 Drag coefficient as a function of pitch angle for different pitch angle amplitudes
1
Lift coefficient
0
-50 -40 -30 -20 -10 0 10 20 30 40 50
-1
-2 Cl20
Cl30
-3
Cl40
-4
Pitch angle
Figure 36 Lift coefficient as a function of pitch angle for different pitch angle amplitudes
37
Monash University FYP 2010 Pavaman Bilgi 19334915
Figure 37 and 38 show the lift and drag coefficient variation with pitch angle. It is clearly evident that at the
higher Strouhal number, larger, more intense vortices are created at the leading edge (stronger LEVs) in
addition to stronger TEVs. This is obviously the cause of the greater production of lift and thrust than at the
lower strouhal numbers (comparing 𝑆𝑆𝑆𝑆 = 0.3 to 0.2 for example). We notice something different happen when
the Strouhal number moves up to 0.4 however. The lift profile changes such that much lower lift coefficients
are registered at the lower pitch angles whereas much higher lift coefficients are found at the higher pitch
angles. Furthermore, looking at the drag coefficient curves we see that thrust production is markedly increased
at 𝑆𝑆𝑆𝑆 = 0.4 with the range of pitch angles at which thrust is generated being significantly increased also.
2
Lift coefficient
0
-25 -20 -15 -10 -5 0 5 10 15 20 25
-2
Cl0.2
-4 Cl0.3
Cl0.4
-6
Pitch angle
Figure 37 Lift coefficient as a function of pitch angle for different Strouhal numbers
Figures 44 and 45 in appendix A reveal the reasons behind the observations noted above. While indeed we do
see the stronger LEVs associated with the higher pitch frequencies (higher strouhal numbers), at 𝑆𝑆𝑆𝑆 = 0.4 the
frequency is so high as to not allow the LEV to convect downstream before it is destroyed in the return stroke
of the wing. This means that the LEV (a region of very low pressure) remains at the leading edge at all times
thus inducing the generation of thrust. Furthermore this allows the advancing shear layer created at the rear of
the wing to advance over the surface of the wing unimpeded by the LEV – this leads to very high lift coefficients
being generated.
38
Monash University FYP 2010 Pavaman Bilgi 19334915
1.2 Cd0.2
Cd0.3
Drag coefficient
0.8
Cd0.4
0.4
0
-25 -20 -15 -10 -5 0 5 10 15 20 25
-0.4
-0.8
Pitch angle
Figure 38 Drag coefficient as a function of pitch angle for different Strouhal numbers
3.6 Limitations
Before concluding it is necessary to outline the most prominent limitations that were faced in obtaining the
aforementioned results in this section. These should be addressed in any future work continuing the present
investigation.
It is important to note firstly that the analysis done in section 3.1 concerning mesh and temporal resolution was
only done for certain flow and motion conditions and that other conditions may yield different results. Ideally
the conditions which impose the most stringent conditions on the mesh and time step should have been tested
but due to time constraints, this was not able to be done. Because of this a single mesh was used throughout
the study which may not always have resolved the flow accurately in all conditions (especially at the higher
Reynolds numbers and higher pitch frequencies).
Another limitation due to time constraints was the limitation to two dimensional simulations. Section 1.2
established the three dimensionality of flapping wing vortical flows. However due to the sheer computational
expense of three dimensional simulations, this was not pursued. Consequently, the stretching of the vortices
and energy dissipation in the third dimension (𝑧𝑧-axis) will not have been captured which will affect the vorticity
distribution in the 𝑥𝑥𝑥𝑥-plane. There is also a limitation from the experimental point of view due to the finite
span of the wing. This will create unwanted wing tip vortices that will affect the flow along the span of the wing
to some extent.
Finally a glaring limitation also brought on by time constraints was the limited number of tests that could be
performed. Although for example it was found that Reynolds number had a relatively minimal effect on the
results the case could have been different for different pitch frequencies and pitch angle amplitudes. Ideally
the parametric study would be performed on a much wider and more comprehensive range of parameter
values.
39
Monash University FYP 2010 Pavaman Bilgi 19334915
40
Monash University FYP 2010 Pavaman Bilgi 19334915
The dynamic mesh method used was the General Grid Interface method which proved to be an effective
technique to approach the problem both from viewpoint of computational efficiency and ease of
implementation. A key advantage of this method was its ability to handle different rotational motion
parameters without any modification which other mesh motion methods require. Vadidation tests and
comparison with experimental data show that questions must still be posed as to the accuracy of the method
under different rotation speeds and different mesh resolutions at the interface.
Domain and mesh resolution studies showed that a mesh of at least 500,000 elements was required for the
problem and a domain of at least 15 chord lengths in both the 𝑥𝑥 and 𝑦𝑦 directions was required for the effects of
the boundary conditions imposed to be negligible. Simulations at 𝑅𝑅𝑅𝑅 = 500, 𝑆𝑆𝑆𝑆 = 0.3 and 𝑅𝑅𝑅𝑅 = 1500,
𝑆𝑆𝑆𝑆 = 0.2 with a pitch angle amplitude, 𝜃𝜃𝑎𝑎 = 20° were compared to experimental flow visualizations using
Digital PIV equipment at the LTRA&C. Vorticity fields showed that the 2D simulations accurately predicted
creation of both the leading and trailing edge vorticies including the evolution of those vortices with time. Thus,
although the flow is in principle three dimensional, the present study shows that two dimensional simulations
can potentially deliver accurate results. The degree of accuracy seemed to vary between test conditions,
however and this is a point requiring further experimentation and study.
The results of the PIV experiments were extended to include the effects of Reynolds number, Strouhal number
and pitch amplitude angle. The variation of the Strouhal number and pitch angle amplitude induced interesting
effects on the development and evolution of the leading edge vortex while the Reynolds number had little
effect (in the range that was tested). It was found that when the Strouhal number was increased for a fixed
Reynolds number and pitch amplitude, the strength of the LEV was significantly increased leading to increasing
thrust and lift generation. Increasing the pitch amplitude angle for fixed Reynolds and Strouhal numbers on the
other hand also increased vortex strength but also altered vortex movement resulting in the observation of an
optimal pitch angle amplitude in terms of lift and drag.
The above conclusions are a summary of an introductory study into flapping wing flows using OpenFOAM®.
Many complexities of the flow have been identified and no doubt will attract further study in this area. Based
on the findings the following questions for further analysis arise:
1. Increasing the distance of the GGI from the wing reduces the error incurred at the GGI but reduces the
maximum allowable time-step for computation. Thus, what is the optimum distance of the GGI from
the wing?
2. The present study has been performed for a quasi-elliptical cross section with sharp edges. How do the
conclusions change (if at all) for differing wing geometries?
41
Monash University FYP 2010 Pavaman Bilgi 19334915
Acknowledgements
The author would like to acknowledge the guidance and supervision of Prof. Julio Soria and Brendon Anderson
and the ongoing advice and support from fellow students and friends. In particular, the efforts of Dr. Melissa
Green (currently working at the Department of Energy in the USA) in recovering and explaining the
experimental data of the companion PIV experiments are to be recognized as a major contribution to the
present work. Thanks are also extended to the various Monash Mechanical Engineering postgraduates who
provided useful insight into understanding CFD and the Finite Volume Method. Finally the author also
acknowledges the computational facility provided by the National Computing Infrastructure without which the
present work would not have been possible.
42
Monash University FYP 2010 Pavaman Bilgi 19334915
5. References
Anderson, J. (1991). Fundamentals of Aerodynamics, second edn. McGraw-Hill Inc.
Beaudoin, M., & Jasak, H. (2000). Turbomachinery section. Retrieved 2010, from University of Zagreb, Power
Engineering Department: http://powerlab.fsb.hr/ped/kturbo/OpenFOAM/Berlin2008/SessionIV/
Blondeaux, P., Fornarelli, F., & Guglielmini, L. (2005). Numerical Experiments on flapping foils mimicking fish
like locomotion. Physics of Fluids , 113601-12.
Bos, F. M., Lentink, D., & van Oudheusden, B. W. (2008). Influence of wing kinematics on aerodynamic
performance in hovering insect flight. Journal of Fluid Mech. , 341-368.
Dickinson, M. H. (1994). The effects of wing rotation on unsteady aerodynamic performance at low Reynolds
number. The Journal of Experimental Biology , 179-206.
Dickinson, M. H., & Gotz, K. G. (1993). Unseady aerodynamic performance of model wings at low Reynolds
numbers. The Journal of Experimental Biology , 45-64.
Dickinson, M. (1999). Wing rotation and the aerodynamic basis of insect flight. Science , 1954-1960.
Dickson, W. B., & Dickinson, M. H. (2004). The effect of advance ratio of the aerodynamics of revolving wings.
Journal of Experimental Biology , 4269-4281.
Dong, H., Mittal, R., Bozkurttas, M., & Najjar, F. (2005). Wake structure and performance of finite aspect-ratio
flapping foils . 43rd AIAA Aerospace Sciences Meeting and Exhibit, Reno .
Ellington, C. P., van den Berg, C., Willmott, A. P., & Thomas, A. (1996). Leading-edge vortices in insect flight.
Nature , 626-630.
Green, M., Parker, K., & Soria, J. (2005). 2D DPIV of a Pitching Aerofoil. Fourth Australian Conference on Laser
Diagnostics in Fluid Mechanics and Combustion .
Hirsch, C. (1988). Numerical Computation of internal and External Flows, Volume I: Fundamentals of Numerical
Discretisation. John Wiley & Sons.
Issa, R. I. (1986). Solution of the implicitly discretised fluid flow equation by operator splitting. Journal of
Computational Physics , 2053-2059.
Jasak, H. (2010). List of Papers: Hrvoje Jasak. Retrieved July 1st, 2010, from Hrv's Homepage:
http://www.h.jasak.dial.pipex.com/
Lentink, D., & Dickinson, M. H. (2009). Rotational accelerations stabilise leading edge vortices on revolving fly
wings. Journal of Experimental Biology , 2705-2719.
43
Monash University FYP 2010 Pavaman Bilgi 19334915
Lentink, M., & Gerritsma, M. H. (2003). 43rd AIAA Fluid Dynamics Conference and Exhibit. Orlando .
Lewin, G. C., & Haj-Hariri, H. (2003). Modelling thrust generation of a two-dimensional heaving airfoil in a
viscous flow. Journal of Fluid Mech. , 339-362.
Maxworthy, T. (1979). Experiments on the weis-fogh mechanism of lift generation by insects in hovering flight.
part 1. dynamics of the 'fling'. Journal of Fluid Mechanics , 47-63.
Mittal, R., & Iaccarino, G. (2005). Immersed Boundary Methods. Ann. Rev. Fluid Mech. , 239-261.
Mueller, T. J., & DeLaurier, J. D. (2003). Aerodynamics of Small Vehicles. Ann. Rev. Fluid Mech. , 89-111.
Pedro, G., Suleman, A., & Djilali, N. (2003). A Numerical study of the propulsive efficiency of a flapping
hydrofoil. International Journal for Numerical Methods in Fluids , 493-526.
Rojratsirikul, P., Wang, Z., & Gursul, I. (2009). Unsteady fluid-structure interations of membrane airfoils at low
Reynolds numbers. Experiments in Fluids , 859-872.
Sane, S. P. (2003). The aerodynamics of insect flight. The Journal of Experimental Biology , 4191-4208.
Sun, M., & Tang, J. (2001). Unsteady aerodynamic force generation by a model fruit fly wing in flapping motion.
The Journal of Experimental Biology , 55-70.
Sun, M., & Tang, J. (2002). Unsteady aerodynamic force generation by a model fruit fly wing in flapping motion.
The Journal of Experimental Biology , 55-70.
Thaweewat, N., Bos, F. M., van Oudheusden, B. W., & Bijl, H. (2009). Numerical study of vortex-wake
interactions and performance of a two-dimensional flapping foil. 47th AIAA Aerospace Sciences Meeting,
Orlando .
Wang, Z. J. (2008). Aerodynamic efficiency of flapping flight: analysis of a two-stroke model. Journal of
Experimental Biology , 234-238.
Wang, Z. J. (2000b). Two dimensional mechanism for insect hovering. Physical Review Letters , 2216-2219.
Wang, Z. J. (2004). Unsteady forces and flows in low Reynolds number hovering flight:two-dimensional
computations vs robotic wing experiments. The Journal of Experimental Biology , 449-460.
Wang, Z. J. (2000). Vortex shedding and frequency selection in flapping flight. Journal of Fluid Mech. , 323-341.
Weish-Fogh, T., & Jensen, M. (1956). Biology and physics of locust flight. i. basic principles of insect flight: a
critical review. Phil. Trans. Royal Society London , 415-458.
Williamson, C. (1995). Fluid Vortices (Vol. Vol. 30). Kluwer Academic Publishers.
44
Monash University FYP 2010 Pavaman Bilgi 19334915
Re St 𝜽𝜽𝒂𝒂 (deg) 𝑻𝑻(secs) 𝚫𝚫𝒕𝒕 𝑼𝑼∞ (m/s) Write interval (s) Frequency (Hz)
500 0.3 20 2.184 1.456E-03 0.0167 0.091 0.46
500 0.4 20 1.638 1.092E-03 0.0167 0.068 0.61
500 0.2 20 3.276 2.184E-03 0.0167 0.136 0.31
500 0.3 30 3.464 2.309E-03 0.0167 0.144 0.29
500 0.3 40 5.035 3.356E-03 0.0167 0.210 0.20
1000 0.2 20 1.638 1.092E-03 0.0333 0.068 0.61
1000 0.4 20 0.819 5.460E-04 0.0333 0.034 1.22
1500 0.2 20 1.092 7.279E-04 0.0500 0.045 0.92
1500 0.4 20 0.546 3.640E-04 0.0500 0.023 1.83
(a)
(b)
(c)
Figure 40 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝒂𝒂 ⁄𝟐𝟐 for different 𝑹𝑹𝑹𝑹 (a) test number 1 (b) test number 4 and (c) test number 6 from table 2
45
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
46
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
Figure 42 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝒂𝒂 ⁄𝟐𝟐 for different 𝜽𝜽𝒂𝒂 (a) test number 2 (b) test number 8 and (c) test number 9 from table 2
47
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
48
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
Figure 44 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝒂𝒂 ⁄𝟐𝟐 for different 𝑺𝑺𝑺𝑺 (a) test number 1 (b) test number 2 and (c) test number 3 from table 2
49
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
(c)
50
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
Figure 46 Vorticity solutions at intervals of 𝚫𝚫𝜽𝜽 = 𝜽𝜽𝒂𝒂 ⁄𝟐𝟐 (a) test number 5 and (b) test number 7 from table 2
51
Monash University FYP 2010 Pavaman Bilgi 19334915
(a)
(b)
52
Monash University FYP 2010 Pavaman Bilgi 19334915
fvSchemes
/*--------------------------------*- C++ -*----------------------------------*\
| ========= | |
| \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
| \\ / O peration | Version: 1.3 |
| \\ / A nd | Web: http://www.openfoam.org |
| \\/ M anipulation | |
\*---------------------------------------------------------------------------*/
FoamFile
{
version 2.0;
format ascii;
class dictionary;
object fvSchemes;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //
ddtSchemes
{
default CrankNicholson 0.5;
}
gradSchemes
{
default Gauss linear;
grad(p) Gauss linear;
}
divSchemes
{
default none;
div(phi,U) Gauss vanLeerV;
}
laplacianSchemes
{
default none;
laplacian(nu,U) Gauss linear corrected;
laplacian(rAU,pcorr) Gauss linear corrected;
laplacian(rAU,p) Gauss linear corrected;
53
Monash University FYP 2010 Pavaman Bilgi 19334915
interpolationSchemes
{
default linear;
interpolate(HbyA) linear;
interpolate(1|A) linear;
}
snGradSchemes
{
default corrected;
}
fluxRequired
{
default no;
pcorr;
p;
}
// ************************************************************************* //
fvSolution
/*--------------------------------*- C++ -*----------------------------------*\
| ========= | |
| \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
| \\ / O peration | Version: 1.3 |
| \\ / A nd | Web: http://www.openfoam.org |
| \\/ M anipulation | |
\*---------------------------------------------------------------------------*/
FoamFile
{
version 2.0;
format ascii;
class dictionary;
object fvSolution;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //
solvers
{
pcorr PCG
{
preconditioner DIC;
tolerance 1e-07;
relTol 0;
};
p PCG
{
preconditioner DIC;
54
Monash University FYP 2010 Pavaman Bilgi 19334915
tolerance 1e-06;
relTol 0.05;
};
pFinal PCG
{
preconditioner DIC;
tolerance 1e-07;
relTol 0;
};
U PBiCG
{
preconditioner DILU;
tolerance 1e-05;
relTol 0;
};
}
PISO
{
nCorrectors 4;
nNonOrthogonalCorrectors 1;
pRefCell 0;
pRefValue 0;
}
// ************************************************************************* //
Notes:
• PCG: Pre-conditioned Conjugate Gradient solver. It is the iterative solver for solving the pressure equation.
• PBiCG: Pre-conditioned Bi-stab conjugate Gradient solver. It is the solver used to solve the pressure-
velocity coupling.
• DIC: Diagonal Incomplete Choleski (a pre-conditioner)
• DILU: Diagonal Incomplete LU decomposition pre-conditioner
mixerGgiFvMesh
The new variables introduced are declared like so in the mixerGgiFvMesh.C file:
Pi_ =
3.141592653589793238462643383279502884197169399375105820974944592307816406286208998628
03482534211706798214808651328230664709384460955058223172535940812848111745028410270193
85211055596446229489549303819644288109756659334461284756482337867831652712019091456485
66923460348610454326648213393607260249141273724587006606315588174881520920962829254091
71536436789259036001133053054882046652138414695194151160943305727036575959;
Re_ = readScalar(dict_.lookup("Re"));
c_ = readScalar(dict_.lookup("c"));
theta_ = readScalar(dict_.lookup("theta"));
St_ = readScalar(dict_.lookup("St"));
nu_ = readScalar(dict_.lookup("nu"));
55
Monash University FYP 2010 Pavaman Bilgi 19334915
theta_ = theta_*Pi_/180;
Info<< "Mixer mesh :" << nl
<< " origin : " << cs().origin() << nl
<< " axis : " << cs().axis() << nl
<< " rpm : " << rpm_ << nl
<< " Re : " << Re_ << nl
<< " c : " << c_ << nl
<< " theta : " << theta_ << nl
<< " St : " << St_ << nl
<< " nu : " << nu_ << endl;
Since the GGI method was originally developed for rotating geometries with constant rotational speed, the
definition of the RPM variable had to be changed to accommodate the sinusoidal rotation of the wing. This is
shown below and is simply the time derivative of equation (2.13).
rpm_
=(60.0*theta_*Re_*nu_*St_)/(c_*c_*tan(theta_))*cos((2.0*Pi_*Re_*nu_*St_*time().value()
)/(c_*c_*tan(theta_)));
movePoints
(
csPtr_->globalPosition
(
csPtr_->localPosition(allPoints())
+ vector(0, rpm_*360.0*time().deltaT().value()/60.0, 0)
*movingPointsMask()
)
);
56