This is the accepted manuscript made available via CHORUS.

The article has been

published as:

Evaluation of potential flow models for unsteady separated

flow with respect to experimental data
Field Manar and Anya R. Jones
Phys. Rev. Fluids 4, 034702 — Published 8 March 2019
DOI: 10.1103/PhysRevFluids.4.034702
 An evaluation of potential flow models for unsteady separated flow with
respect to experimental data

Field Manar∗ and Anya R. Jones†

University of Maryland, College Park
(Dated: February 15, 2019)
The wake characteristics and force production of an aspect ratio eight flat plate rectangular wing
starting from rest at high angle of attack (α = 45◦ ) were examined. The first portion of this paper
presents experimental measurements that focus on circulation production at the leading edge of the
wing. Circulation production is shown to be coupled to both the wing kinematics and the wake
state. The second portion of the paper reviews the implementation of several pertinent potential
flow models from the literature, and compares the results of these models to the experimental results.
Quasi-steady and fixed wake models were found to be poor predictors of both force and circulation.
A similarity solution based model was found to make accurate predictions only at very short times.
An unsteady strength two-vortex model predicted the circulation production during the acceleration
portion of the kinematics, but did not predict the force. A multiple discrete vortex model was found
to predict both the force and circulation production well, but had difficulty after the shedding of
the first leading-edge vortex.

I. INTRODUCTION AND BACKGROUND flight. Efforts to model leading-edge separated flows

are thus applicable to a broad range of problems and
Unsteady aerodynamics has been studied since the can be applied to insect flight, MAV gust encounters,
early days of the aerospace field, spurred largely by dynamic stall, and wind and tidal turbine flows.
the aeroelastic problems encountered in fixed wing In order to study the phenomena of leading-
flight. As a result, the theories of attached un- edge separation from a first-principles approach, the
steady flow are well developed and understood, for problem has been systematically reduced. The first
example, a comprehensive overview can be found in reduction is to use a rectangular flat plate wing as a
Bisplinghoff, Ashley, and Halfman [1]. The same representative geometry. The wing used here, and in
cannot be said of unsteady flows with leading-edge many other studies, is a thin flat plate with square
separation, which are still an open research topic edges; this is not too far removed from an insect
[2]. The original impetus for the current work wing and serves as a canonical geometry for experi-
stemmed from interest in reproducing insect-like mentation.
flapping wing flight in small unmanned “micro air The second reduction is to distill the flapping wing
vehicles” (MAVs). The realization of this dream has kinematics from oscillatory three degree-of-freedom
not yet come to pass in a meaningful way, due in flapping about a shoulder joint to a rectilinear surge
part to limited aerodynamic understanding and a at constant angle of attack. These simpler kinemat-
lack of applicable models of the unsteady separated ics are closest to the start and mid-stroke portions of
flows that such vehicles rely on. the full insect kinematics, but still differ in their dis-
Study of flapping wing aerodynamics has revealed regard of the rotational component of motion. Both
that, in general, the flows encountered in flapping of these simplifications are common to the field.
wing flight are similar to stalled transient flows. Fur- As a surrogate for the complete insect wing kine-
thermore, small fliers of any type must overcome matics, many studies have instead used translat-
the difficulties inherent to the low Reynolds num- ing wing motions to study the problem at hand
ber flight regime [3] and MAV applications natu- (e.g. Dickinson and Gotz [6]). This is generally
rally require operation in a Reynolds number range done to produce simpler, easier to understand flows.
of Re = O(102 − 105 ), much lower than traditional As such, rectilinear pitching and surging kinemat-
aerospace applications. This reduction in Reynolds ics have been particularly well documented in both
number leads to poor performance of traditional experimental [7–9] and computational studies [10–
thick airfoils [4, 5]. The combined effect of low 13]. Rotating wings, i.e. a wing revolving about
Reynolds numbers and the likelihood of encounter- its root, are another common surrogate for the in-
ing large angles of attack makes separation and dy- sect wingstroke [14–19]. These preserve the rota-
namic stall nearly a foregone conclusion for MAV tion of the entomological flapping wing in the stroke
plane, but neglect the out-of-plane motion. In gen-
eral, studies on simplified wing kinematics have pro-
∗ [email protected] duced a large body of data that serves as a quanti-
† [email protected] tative reference for the cases considered here [2], but
 2

the need for a computationally inexpensive predic- strength of the feeding shear layer. Wong et al. [29]
tive model remains. and Widmann et al. [30] use a similar philosophy to
Traditionally, those who seek to model unsteady estimate the size of the LEV.
separated flows and predict the forces from first prin- Vorticity flux measurements similar to those con-
ciples have used only the flow visualization and force ducted here have been carried out by Panah et
measurements from experimental studies (e.g. Ham- al. [31] and Wojcik and Buchholz [25]. These studies
mer et al. [20], or Wang and Eldredge [21]). How- were primarily concerned with establishing the im-
ever, for a model of even modest complexity, the portance of the secondary vorticity produced in the
force results are an amalgamation of all parts of the opposite-sign boundary layer below the LEV. The
model interacting, and using the force histories in analysis of Lighthill [32] indicated that the surface
an attempt to diagnose the success of a model leads pressure gradient is directly related to the vorticity
to vague answers at best. This state of affairs has production at a fluid/solid interface. To measure the
led to the continued use of the leading-edge Kutta secondary vorticity production, Panah et al.’s study
condition while simultaneously doubting its valid- included surface pressure measurements. Based on
ity. The caveat is usually then given that the Kutta these pressure measurements, both Panah et al. [31]
condition is inappropriate for the leading edge be- and Wojcik and Buchholz [25] concluded that the
cause flow separates here in the wing-normal direc- opposite-sign vorticity production on the surface of
tion rather than tangential to the leading edge. In the plate contributed approximately half the magni-
the current work, the form of Kutta condition used tude of circulation as the shear layer from the leading
is to enforce flow (and therefore the forming vortex edge to the circulation of the leading-edge vortex.
sheet) tangent to the plate. The separation point is Both of these studies, however, focused on kinemat-
also assumed to be fixed at the leading edge. Fur- ics dissimilar to the present surge case: Panah et
ther discussion of the Kutta condition is reserved for al. [31] used an oscillating plunging wing, and Woj-
section II C. cik and Buchholz [25] used a rotating wing. In both
The leading-edge vortex (LEV) is a prominent fea- of these cases the LEV is held nearer to the wing
ture in both dynamic stall events [22] and insect than is seen in pure translation, likely resulting in a
flight [23]. It forms when the wing motion is ag- stronger secondary boundary layer below the LEV.
gressive enough to cause the flow to separate from To remedy these gaps in the literature, the cur-
the leading edge of the wing and roll up into a vor- rent work focuses on a series of experiments con-
tex above the suction side of the wing [2]. The flow ducted to directly measure circulation production at
field is dominated by this vortex, which in turn dom- the leading edge of a wing starting from rest. Cur-
inates the force production of the wing [6, 24]. The rent low-order physics-based modeling techniques
LEV forms because the sharp leading edge forces the are then evaluated in the context of these measure-
boundary layer to separate there, which then rolls up ments. The goal of the model evaluation presented
into a single vortex. Several models of LEV growth here is to assess the strengths and weaknesses of the
have been developed based on the leading edge con- current methods and identify avenues for improve-
ditions [25–30]. Most of these studies are concerned ment. Close inspection of these models also aids in
with an oscillating wing (e.g. Buchholz et al. [26] understanding how the forces produced relate to the
and Widmann and Tropea [30]), so the scaling laws fluid processes.
they present are largely based on the parameters of
the oscillation. This produces valuable insights, but
can obfuscate the underlying mechanisms for the cir- II. POTENTIAL FLOW MODELING
culation growth. For example, including the ampli-
tude as a scaling factor, as in Buchholz et al. [26], Interest in flapping wing flight and unsteady aero-
does not make it clear if the vortex is larger because dynamics in general has produced a plethora of ap-
of the increased wing speed or because of the greater proaches to predicting the forces on a wing under-
distance covered during the wing motion. going unsteady motion. The extant models run
On the other hand, Sattari et al. [27] proposed the gamut from rigorous analytical analysis [33–37],
using the boundary layer exterior velocity to capture to modified versions of classical theories [38, 39],
the flux of vorticity in a feeding shear layer from the to vortex-based computational schemes [21, 40–44],
trailing edge of a waving plate. Kriegseis et al. [28] empirically based models [45–47], grid-based CFD
built on this idea and used the total flow velocity methods [48–50], and viscous vortex particle meth-
at the leading edge to successfully collapse the LEV ods [10].
circulations of wings plunging with different plunge The motivation behind the choice of models
amplitude. In their paper, the total flow velocity just presently being examined is to focus on those models
outside the shear layer is used as a surrogate for the suitable for engineering work, i.e. to trade accuracy
 3

for quick computation, while still including the req- steady separated flows, and is the basis for methods
uisite physics. Which physics are required is still such as Wang and Eldredge [21], Xia and Mohseni
something of an open question; at a minimum the [56], Ansari et al. [41, 57], and Minotti [58]. Con-
models should be able to predict the presence and formal mapping methods imply the use of a com-
bulk characteristics of the LEV. Ideally, the model plex potential formulation. The method works by
would also include a prediction of the shedding pro- stretching space so that a flat plate is mapped to a
cess. Preference is given to techniques that neatly cylinder. Arbitrary body shapes can be accommo-
categorize the various contributors to the force, as dated by extending the series of the mapping or by
this aids in developing intuition. These requirements using the Schwartz-Christoffel transform.
point the researcher towards potential flow theory. The no-penetration boundary condition can also
Potential flow has long been the bastion of pen and be also be fulfilled by solving for the strength of the
paper solutions, and also gave rise to the more ver- surface vortex sheet. When the body is infinitely
satile panel methods still in use today. Potential thin, a common approach is to parameterize the
flow, and thus panel methods, rely on a few key strength of that sheet with a truncated series expan-
simplifications, namely inviscid and incompressible sion. One must then solve for the coefficients of the
flow. These assumptions are largely appropriate for series expansion to enforce the no through-flow con-
the present class of problems, but it should be noted dition at a finite number of collocation points. The
that applying these models at the Reynolds numbers two most common basis for the series are a modified
considered here (O(104 )) stretches the limits of the Fourier series (usually attributed to Glauert [59]),
inviscid assumption. Nevertheless, the formation of and a Chebyshev series [40, 43, 60]. In either case,
the LEV from the shear layer presents itself as a the body is typically represented by a vortex sheet,
convection-dominated process, rather than diffusive although it is also possible to use a doublet sheet.
one. The strength of basis function methods lies in
Any unsteady potential flow model consists of their ability to capture either the singularities at
three components: a representation of the body, a the ends of the plate or naturally enforce the Kutta
representation of the wake, and a method for deter- condition, while still maintaining a tradeoff between
mining the circulation. For a general introduction to numerical and analytical computation. They also
potential flow theory, the reader is referred to any provide a sound analytical basis for the edge suc-
number of fluids texts [51–55], but an overview is ap- tion parameter, which relies on properly capturing
propriate here to understand where the models used the square-root singularity in sheet strength at the
in the present work come from. edge. The primary drawback of using basis functions
is that they do not extend to arbitrary geometry
(e.g. thick airfoil profiles) without explicit modifica-
A. Body Representation tion on the part of the user. It is still less clear how
to apply them for 3D problems.
An alternative method to solve for the vortex
The first consideration when modeling an exter-
sheet strength is to use a series of discrete panels,
nal flow problem is how to represent the body in
in much the same way as a finite element method.
question. The body comes into the equations as a
Instead of representing the singularity strength with
no-penetration boundary condition for the fluid at
a few complicated functions that span the domain, it
the body surface. Stated simply, it says that the
is now represented with many simple functions with
fluid velocity at the body surface must match that
limited support. Originally developed for non-lifting
of the body in their surface-normal components. To
flows by Hess and Smith [61, 62], panel codes have
enforce this, several methods have been developed
developed into a practical and general design tool
throughout history: conformal mapping, basis func-
for both 2D and 3D flows. A good overview of panel
tions for a vortex sheet, and panel methods. These
methods is given in Katz and Plotkin [52]. While
methods all converge on the same solution, and are
extremely flexible, it is difficult to capture the sin-
largely interchangeable outside of the details of their
gularities that naturally occur at the edges of bound
implementation.
vortex sheets with this method.
The oldest method for meeting the no through-
flow boundary condition is to use conformal map-
ping. It is the foundation for the basic unsteady flow
solutions that are known throughout the aerospace B. Wake Representation
field such as Wagner’s problem of an impulsively
started plate [33], or Theodorsen’s frequency re- In addition to computing the effect of the body,
sponse [35]. Despite its age, the conformal mapping potential models for unsteady flow require the in-
method still enjoys popularity in recent work on un- clusion of vorticity in the fluid. The job of the wake
 4

representation is to provide a discrete representa- tion of higher order discretizations of the vorticity.
tion of that vorticity and to propagate the wake for- In two dimensions, besides point vortices (a zero-
ward in time by solving the vorticity evolution equa- dimensional representation) it is possible to have
tion. In two dimensions with the inviscid and incom- vortex lines [40, 43] and patches. Constant strength
pressible assumptions, vorticity evolution is simply vortex patch methods are referred to as contour
a statement that fluid particles that have vorticity dynamics methods [69]. In three dimensions, line
keep that vorticity, and so any vorticity must there- vortices are by far the most common element, and
fore convect with the fluid. Convecting the vorticity lead to the vortex lattice methods (see, once again,
with the fluid leads naturally to a Lagrangian for- Katz and Plotkin [52]). It is also possible to have
mulation, where propagating the wake in time boils sheets [70, 71], or volumes. These are attractive,
down to convecting the discrete representation with but adding this extra dimensionality also adds to
the flow. (To fully solve the vorticity form of Navier- the computational cost, and often requires ad hoc
Stokes also requires accounting for diffusion and, in methods to manage the connection between particles
three dimensions, vortex stretching.) An Eulerian when things become tangled. An excellent, if some-
approach is also possible, but these methods require what informal, review of vortex methods is given by
meshing the full fluid volume and are not as gen- Stock [72].
eralizable to low order models. They will not be
considered here.
The most convenient way to discretely represent C. Circulation Conditions
the vorticity field in potential models is to add point
vortices that convect with the fluid. This leads to There are three prevailing philosophies for deal-
the class of methods known as vortex methods, dis- ing with the creation of vorticity in potential meth-
cussed in detail in Cottet and Koumoutsakos [63] ods. One is to match the full no-slip surface condi-
and Saffman [64]. A problem with pure point vor- tion and to fully resolve the boundary layer. Meth-
tices is that their induced velocity tends to infinity ods using this approach account for separation by
the closer one gets to them. This causes difficulties capturing the boundary layer dynamics. Examples
in numerical simulations, where particles that are in of this class of methods can be found in Winckel-
too close proximity can garner unrealistically high mans and Leonard’s overview of early work in the
induced velocities and convect themselves out of the field [73], Eldredge’s VVPM method [10], the VRM
simulation domain. The same high velocities also method [74], or Kirchart and Obi’s recent work [75].
help propagate instabilities in vortex sheets, causing Resolving the boundary layer requires large num-
the interior structure of rolled-up sheets (as in the bers of particles, however, and it is not uncommon
LEV) to fall apart and become chaotic. To counter to see particle counts in the hundreds of thousands
this, the vortex can be regularized with a vortex core to millions of particles. With such a large particle
model that spreads the vorticity over a finite area count comes a commensurate increase in computa-
rather than a point. The first such methods were tion time, making these methods undesirable for the
proposed by Chorin, Krasny, and Bernard to study present goal of low-cost computations. A second op-
the problem of vortex sheet roll-up [65–67]. They tion is to use the integral boundary layer formula-
used a non-physical model of the vortex core that tion coupled with the potential solver, as presented
simply added a constant factor to the distance be- by Drela [76]. Unfortunately, outside of some pre-
tween the vortex and the interrogation point. Other liminary work, these methods are limited to steady
models, such as the more physical Rankine, Burgers, attached flows, and are of little use here. The third
or Lamb-Oseen vortex core can also be used, but philosophy is to allow for a slip velocity on the body
are often overlooked because of their computational surface, thus representing the boundary layer as an
cost. An approximation of the Lamb-Oseen vortex infinitely thin vortex sheet. When surface slip veloc-
core is proposed by Vatistas et al. [68], and has been ity is allowed, potential flow provides no mechanism
used for computations such as those by Ramesh et with which to generate circulation, and is mathe-
al. [44]. matically non-unique with regard to total circula-
The concept of a point vortex is applicable in tion. A surrogate circulation condition, such as the
both two and three dimensions, but the vorticity Kutta condition, must be applied at the separation
evolution equation has additional terms in three point and in the case of unsteady flows, additional
dimensions. Extra work is required to compute vortices are added at the separation point to main-
vortex stretching and keep the resulting vorticity tain both the total circulation and the specified edge
field divergence free. That extra work is made eas- condition. This approach keeps particle counts, and
ier when information about the arrangement of the therefore the computational cost, much lower than
point vortices is kept. This leads to the implementa- when resolving the full boundary layer. Note that
 5

the separation point must be specified from prior Cl = 2L/ρU 2 c = 2π sin(α) (for a freestream velocity
knowledge, or solved for with an additional layer U and angle of attack, α) and zero moment at the
of modeling. Fortuitously, the separation point in quarter chord are obtained. For the unsteady case,
attached flows is always the trailing edge and the the concept of fixing the ζ-plane velocity at the plate
shedding of vorticity is well described by the Kutta edges to zero at all times remains. This method is
condition. implemented reactively, in the sense that the simu-
The de-facto standard for specifying circulation in lation is updated with new circulation to remove the
potential flow models is the Kutta condition. There erroneous edge velocity at each time step. Thus it
are many ways to express the condition and a bewil- does not lend itself to an analytical expression of the
dering number of ways to implement it. As Sarp- circulation rate, dΓ/ dt.
kaya [77] puts it in his vortex method overview: Without resorting to conformal mapping, i.e.
“almost every paper, at least in part, represents a when using basis functions or a panel method, the
new method.” The most common implementation correct form of the condition, according to the work
is to enforce that the flow leave tangent to the shed- of Krasny [78, 79] and Jones [40], is to put the ul-
ding edge by adding a new wake vortex at each time timate focus on removing any infinite velocities in
step. At a minimum, this leaves the selection of the the flow. Doing this enforces that the pressure at
new vortex’s location as an open question. In prac- the edge of the plate remains finite, and that the
tice though, most methods achieve the same end re- flow leaves tangential to the edge, two criteria of-
sult albeit perhaps through a different implementa- ten given as the Kutta condition. This leads to the
tion. (A direct comparison of the various implemen- extremely simple result that
tations, while tedious to construct, could be very
dΓ
enlightening.) = γu (4)
dt
For the steady case, the Kutta condition is com-
monly implemented by enforcing zero bound vortex where dΓ/ dt is the rate of circulation being added
sheet strength at the edge of the plate, i.e. to the shed vortex sheet, γ is the bound vortex sheet
strength at the edge of the plate, and u is the tangent
γ(x̃ = x̃T E ) = 0, (1) velocity at the plate edge. This, like the conformal
mapping method, is difficult to use as the basis for
where γ is the vortex sheet strength in circulation an analytical expression of circulation production.
per distance, and x̃T E is the location of the trailing The first issue is that γ ties together everything in
edge in the plate frame of reference. When con- the flow field, including the motion of the plate and
formal mapping is used, the infinitely thin plate is the location and strength of all wake vorticity. Fur-
mapped to a cylinder. The complex plane with the ther, the true value of γ is generally unbounded at
cylinder has locations notated with ζ, and the cylin- the plate edge unless the Kutta condition is already
der is centered at the origin with radius one. The precisely met. Hence most implementations, as men-
expression of the Kutta condition specifies that the tioned previously, enforce the Kutta condition reac-
complex velocity, W = u − iv, at the edges of the tively.
plate in the ζ-plane are zero, i.e. Part of the reason that a plethora of methods ex-
ist for implementing the Kutta condition is that it
W (ζ = ±1) = 0. (2)
is not clear from the statement of the Kutta con-
This has the effect of canceling the singular fac- dition alone how exactly to add the new circulation
tor in the mapping from the cylinder back to the required to maintain the specified edge condition. In
plate plane, resulting in a finite edge velocity and a point vortex model, the question becomes where to
streamlines leaving smoothly from the edge. For place the new vortices. A popular answer given by
the steady translation case, the only velocity compo- Ansari et al. [41] is to place the new vortex one-third
nents present are the translation and bound vortex of the distance from the edge to the most recently
components. This leads to a simple expression for shed vortex. For finite-angle trailing edges (i.e. on
the bound vortex strength, thick airfoil shapes) Xia and Mohseni [80] recently
proposed a rigorous momentum analysis to theoret-
Γ = −πcṼ , (3) ically compute the angle and strength of a newly
created vortex sheet that was found to improve so-
where Γ is the circulation of the bound vortex, c lutions for rapid airfoil pitching.
is the chord, and Ṽ is the plate-normal compo- In the present problem, separation is fixed at the
nent of velocity at the midchord. When substituted leading and trailing edges of the flat plate wing. Un-
back into the expressions for force and moment, like the trailing edge, it is not clear a priori that
the usual results of a two-dimensional lift coefficient the Kutta condition is valid or useful at the leading
 6

edge. A general statement of the Kutta condition resentative of the common features in wing start-up
is that the flow must leave tangent to the separat- transients at high incidence. By studying this case in
ing edge. Based on flow visualization, the separated depth, the reader will be familiarized with the gen-
leading-edge shear layer appears to leave in an ap- eral picture of the real flow so that later comparisons
proximately plate-normal direction. This difference to model predictions will be put in context. The
of direction calls the validity of the Kutta condition experimental methods employed are documented in
into question. Nonetheless, it is a common choice. the appendix.
The leading edge suction parameter (LESP), as
proposed by Ramesh et al. [44], is a relaxation of
the Kutta condition to allow for the presumed abil- A. Flow Visualization
ity of finite thickness wing profiles to support a suc-
tion force at the leading edge [42]. This idea has
The vorticity fields for the baseline case described
been seen elsewhere, as in the philosophy behind
above were found via particle image velocimetry
the vortex shedding portion of Leishman-Beddoes
(PIV) measurements and are shown in fig. 1. Ar-
model [81] and Polhamus’s suction analogy [38]. In
tifacts of the laser shadow are visible in the images
their paper, Ramesh et al. argue that for a given
as the light red diagonal lines under the wing, but in
geometry and Reynolds number, the LESP provides
general, using vorticity as a flow visualization tool
a description of whether the leading-edge flow is at-
highlights the locations of shear and rotation in the
tached or separated that is independent of wing kine-
flow, e.g. the boundary layers, shear layers, and vor-
matics. The LESP also has the enviable property of
tices. As can be seen in the first image (top left,
reducing to the Kutta condition when the allowable
t∗ = 1), the vorticity in the flow is initially confined
suction is set to zero.
to boundary layers very close to the wing surface.
Having now identified the key components and At the edges of the wing, the leading-edge (LEV)
challenges of modeling separated flows using inviscid and trailing-edge (TEV) vortices have already be-
models, some experimental results are presented in gun to develop. As time progresses (t∗ = 2, 3, 4, 5),
the next section to provide an overview of the evo- flow leaves smoothly from the trailing edge of the
lution of the real flow field. Following this, in Sec- wing, forming a starting vortex made up of a series of
tion IV, these results are compared to those of the small-scale vortices centered at the original location
models described above to identify the strengths and of the wing’s trailing edge. Initially, the boundary
weaknesses of the various approaches and so identify layer at the leading edge of the wing is forced back
the dominant physical phenomena in these types of onto the wing, but it quickly rolls up into an LEV
flows. (see t∗ = 2, 3). As wing motion progresses, the LEV
moves slightly off of the wing surface but remains in
the vicinity of the body. While the LEV is near the
III. EXPERIMENTAL RESULTS wing, it continuously gathers circulation fed to it by
a shear layer emanating from the leading edge.
This section presents detailed results for a single Eventually, the LEV grows to a size where it
test case of an aspect ratio eight, 76.2 mm chord, can no longer be sheltered from the free stream be-
flat plate wing in rectilinear translation at a con- hind the wing, and it begins to convect downstream
stant angle of attack of 45◦ . The width of the towing (t∗ = 5, 6). This is when even the two-dimensional
tank was 1.5 m, approximately 2.5 times the span of picture of the flow starts to get complicated. A new
the wing. The plate velocity was piecewise linear in TEV rolls up behind the wing at t∗ = 6 and shortly
time, a constant acceleration over two chord-lengths thereafter another LEV forms as well (t∗ = 8, 9, 10),
of travel followed by a transition to a constant final followed by yet another TEV at t∗ = 12. Under
velocity, Uf , such that Re = 12, 500. The velocity the right conditions (two dimensional, and Reynolds
profile was slightly smoothed at the start and end number dependent), this alternating shedding pro-
of the acceleration to reduce vibrations. Because cess continues and the wake becomes a Kármán vor-
the wing mounting structure limited optical access tex street. Indeed, the details of vortex formation
at the center of the wing, flowfield measurements after the initial LEV depend on almost every aspect
were acquired one chord-length off of the centerline. of the flow and are extremely difficult to predict. In
This location on the aspect ratio eight wing was the present case, the wing is translating, has a finite
shown by comparison to other spanwise locations aspect ratio, and a moderately high Reynolds num-
and on shorter aspect ratio wings to have a nearly ber, so the flow devolves into a chaotic separated
two-dimensional flow. Results from this motion are wake at longer times (t∗ > 30) [51].
averaged over 5 runs and plotted with respect to Because the LEV remains close to the wing during
non-dimensional time, t∗ = tUf /c. This case is rep- its formation, it is expected to have a large effect on
 7

200
150 e f g h i j k l m
U [mm/s]
d
100
c
50 b
0

0 1 2 3 4 5 6 7 8 9 10 11 12 13
t*

(a) Time history of wing velocity, indicating the timing of the flowfield images below.

(b) t∗ = 1 (c) t∗ = 2 (d) t∗ = 3

(e) t∗ = 4 (f) t∗ = 5 (g) t∗ = 6

(h) t∗ = 7 (i) t∗ = 8 (j) t∗ = 9

(k) t∗ = 10 (l) t∗ = 11 (m) t∗ = 12

-20 0 +20
ωc/Uf

FIG. 1: Vorticity fields for a wing rapidly started in translation at 45◦ incidence, ensemble-averaged over 5
runs. The wing is moving from left to right, so the relative flow is from right to left. Red represents
counterclockwise rotation; blue represents clockwise rotation.
 8

the force and moment that the wing experiences; these local force maxima to the formation and shed-
this will be borne out in the force data discussed ding of LEVs above the wing. This process is partic-
in the next section. The development of the LEV ularly evident in the flowfield images at t∗ = 5 and
has a direct effect on the fluid dynamic force on the t∗ = 10, which show large LEVs and correspond
wing even though the LEV is not fully attached to to the peaks in forcing visible in fig. 2. The cyclic
the wing, but eventually sheds. For this reason, the shedding process observed through t∗ = 12 gradu-
state and dynamics of the LEV are the primary focus ally wears down, and the forces have nearly finished
of both measurements and modeling efforts. settling to a steady value by the time the wing has
traveled about 30 chord-lengths (t∗ = 30).

B. Forces
C. Vortex Tracking
It is intuitive that large changes in flow structure
will correspond to an equally dynamic force history. Quantitatively measuring the wake behind the
The forces measured on the wing are shown in fig. 2 wing reveals further information about how the flow
as a function of t∗ = tUf /c. The figure shows four develops, and will turn out to be critical in the evalu-
different curves including the coefficients of lift, CL , ation of potential flow models. The quantities shown
and drag, CD , in the lab frame. The force coeffi- here are all derived from PIV measurements of the
cients in the wing-relative frame, the wing-normal, LEV, as this is the dominant flow feature early in
CN , and the wing-tangential, CT , coefficients are the motion of the wing. Quantification of the LEV
also shown. All of the force coefficients are defined helps to distinguish between cases that produce dif-
with respect to the final wing velocity, Uf . Although ferent forces on the wing but have visually similar
there are some high-frequency small-scale oscilla- LEV development.
tions in force visible in each of the curves due to Trajectories of the center of the LEV are shown in
the mechanical vibration of the towing carriage and fig. 3. These figures contain tracks computed from
wing mounting structure, the overall shape of the both the vorticity centroid and Γ1 criteria. (See Ap-
curves remains clear. pendix for details.) Note that the Γ1 data is re-
As shown in fig. 2, the lift and drag forces for the stricted to the duration of time that the first LEV
baseline case are essentially identical. This is a coin- is in the imaging frame. The Γ1 method tracks a
cidence due to the choice of angle of attack, α = 45◦ . rotation center and more closely aligns with the in-
A more universal observation for flat plates is that tuition of a vortex. Specifically, it tracks the initial
the majority of the force is concentrated in the wing- coherent dominant leading-edge vortex. The vortic-
normal component while the tangential component ity centroid method has been extended to account
is essentially zero. The lack of wing-tangential force for vorticity leaving the frame with a frozen wake hy-
in this case is a byproduct of the near complete pothesis. The locations are normalized by the wing
flow separation. Neglecting viscous drag, the wing- chord, and are relative to the leading edge. The
tangential component of force is exactly the force centroid method does not make distinctions between
associated with edge suction. Recall that the Kutta separate vortices and shear layers, and includes all
condition can be expressed as requiring zero suction the vorticity shed from the leading edge. This pro-
at the edge of a wing. As in the attached-flow case, vides a low order estimate for the net location of the
the Kutta condition is presumed to hold at the trail- wake vorticity rather than the specific location of a
ing edge so the lack of tangential force points to the single coherent vortex.
enforcement of the Kutta condition at the leading Looking at the vortex convection in the wing
edge of the wing as well. frame, shown in fig. 3a, one sees the path of the
Now focusing more on the shape of the curves, vortex as it leaves the wing, as well as the discrep-
there are several distinct features. The forces on the ancy between the centroid and Γ1 methods of vortex
flat plate wing start at zero in the quiescent flow, identification. Note that this figure does not indi-
and motion begins at t∗ = 0. At this time there is cate rate of convection, only location in space. Both
a rapid rise in force as the wing begins accelerating, methods indicate that the LEV leaves the suction
with both the rise in force and wing’s transition to surface of the wing from approximately a tenth of a
full acceleration occurring over t∗ = 0.25. This is chord behind the leading edge. The vortex center as
followed by a continued rise to a distinct peak that identified using the centroid measurement method
slightly lags the end of the acceleration. Following convects nearly straight aftwards, while that of the
that, forces slowly decay to steady state, though the Γ1 method convects slightly downwards. Results ob-
decay is punctuated by several peaks. Based on the tained using the centroid measurement also indicate
flow visualization given in fig. 1, one can attribute a distinct hump in the height of the vortex between
 9

3
CL CD CN CT
2.5

1.5
CF

0.5

-0.5
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
t*

FIG. 2: The forces on an AR = 8 wing undergoing surge at α = 45◦ . The shaded area corresponds to the
acceleration portion of the velocity profile.

0.2
0
-0.2
y/c

-0.4
-0.6 Centroid Γ1
-0.8
-6 -5 -4 -3 -2 -1 0
x/c

(a) Trajectory of the LEV in the wing frame.

0.2
0.0
-1.0
-2.0 0.0
-3.0
x/c

y/c

-4.0
-0.2
-5.0 dx/dt = (1/2)U
-6.0 Centroid
Γ1 Centroid Γ1
-7.0 -0.4
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
t* t*

(b) x-location of the LEV in time. (c) y-location of the LEV in time.

FIG. 3: Vortex location measurements on a surging wing. Data from five independent trials is overlaid for
both methods.

x/c = −2 and x/c = −1. Figure 3b indicates that fected by TEV formation than is the overall vorticity
the LEV is at x/c = −1 when t∗ = 6. Referring field and leading-edge shear layer.
back to fig. 1, note that the hump in the centroid-
The plots given in figures 3b and 3c show the
measurement results occurs at the same time that
x- and y-locations of the LEV returned by both
the trailing-edge vortex forms and pushes the initial
methods relative to wall clock time. Results from
LEV off the wing. It is interesting that this hump is
the centroid method have some noise at the begin-
not reflected in the Γ1 measurements. This suggests
ning of the run when the LEV has not yet gathered
that the center of rotation of the vortex is less af-
enough strength to achieve a good signal-to-noise
 10

ratio. During this time, the Γ1 method is unable one of the methods for evaluating models of un-
to conclusively identify an LEV. Figure 3b also in- steady flow. Vortex location and convection is a pri-
cludes a solid black line showing what the vortex mary factor in determining the resulting force on the
location would be if it had convected downstream wing, and thus a quantification of model success in
at a constant speed at half the freestream velocity. this regard helps point out some of the underlying
This would be the expected value if the circulation is causes for success and failure in predicting unsteady
produced at a constant rate, but the vortex initially loading. The other vortex characteristic of primary
convects slower than this. At early times (t∗ < 6) concern is its strength, which will be discussed next.
the path of the LEV does not appear to be linear,
but rather has a distinct curve indicative of the LEV
convection velocity increasing in time. The increase D. Vortex Circulation
in LEV convection picks up considerably at t∗ = 6,
which, as discussed previously, corresponds to the
formation of the TEV and the shedding of the first Because the amount of circulation produced at the
LEV. Figure 3b shows that at long times (t∗ ≥ 10), leading edge of the plate directly feeds the leading-
the centroid of the LEV convects at close to half of edge vortex and thereby affects the time history of
the freestream velocity without any obvious devia- the vortex strength, circulation production is one
tions. This leads to the hypothesis that the circu- of the most critical components in determining the
lation production at the leading edge of the wing overall success of a flow model. The measured cir-
reaches a relatively constant value, resulting in a culation flux, i.e., the time rate of vorticity passing
roughly constant-vorticity wake. through a boundary surrounding the leading edge,
is shown in fig. 4 for the baseline case. The cir-
The y-location, shown in fig. 3c, of the LEV is a
culation flux is computed using a square box that
bit more subtle in its development, staying nearer to
measures 0.08c on each side and is centered at the
zero (note smaller plot scale compared to x-location)
leading edge of the wing. Circulation is computed by
for the duration of the wing motion and only falling
summing the product of vorticity and velocity nor-
slightly at late times. As a first-order approximation
mal to the box. Further implementation details of
one could say that y/c = 0, especially up to t∗ =
the measurement are documented in the appendix.
6. Higher order trends are difficult to generalize,
The circulation flux as shown here relates directly
though the Γ1 method shows a downward trend that
to the rate of change of the total positive (i.e., coun-
is not reported by the centroid method.
terclockwise) circulation in the flow field. Vorticity
The first takeaway from the vortex tracking anal- generated at the leading edge of the wing is always
ysis presented here is that the leading-edge vortex is positive and is continuously produced, so the circu-
neither attached to the wing, nor is it swept away lation flux through the boundary is also always pos-
at the freestream velocity. From a modeling per- itive, and thus the circulation of the leading-edge
spective, this means that the motion of the vortex vortex monotonically increases. The overall shape
cannot be accounted for with a simple assumption of the circulation flux curve offers some interesting
of fixed location either in the wing or in the lab ref- insights into what is required for modeling the flux,
erence frame, but instead requires knowledge of the as well as the development of the leading-edge vortex
vortex convection speed. Thus, to model the flow, and the overall flow field.
an empirical rate of x-motion must be used, or the The circulation flux shown in fig. 4 increases
flow itself must by computed (e.g. with a potential nearly linearly from t∗ = 0.5 as the wing acceler-
flow model). ates, reaching a peak near t∗ = 3.5. The peak occurs
The second takeaway is that the exact trend in nearly in line with the end of acceleration, in con-
vortex location is difficult to determine from mea- trast to measurement of the forces, which showed a
surement. This stems from difficulty in precisely peak after the end of acceleration (fig. 2). After this
defining a vortex, especially in the present context of point, the wing moves at a constant velocity. The
a growing vortex near a wing, and as such, it is not circulation flux, however, falls off and continues to
clear which tracking method will produce the “cor- change, passing through more peaks and valleys as it
rect” vortex location. Furthermore, the first LEV approaches a steady state. These maxima and min-
can be tracked with reasonable success, but subse- ima correspond to the LEV formation and shedding
quent vortices are difficult to assess because of the process seen in the flow visualization. See fig. 1 at
turbulent nature of the flow. (Recall the flow im- t∗ = 4 for the first peak, t∗ = 6 for the subsequent
ages in fig. 1. That being said, the methods pre- minimum, and t∗ = 9, 10 for the second peak. These
sented here do appear to be consistent, at least up results share similarities with the work on vortex for-
to t∗ = 6. mation time, e.g. as reviewed by Dabiri [82]. Based
The vortex trajectories shown here will serve as on this case alone however, it is not clear that the
 11

1.8
1.6
1.4
1.2
(dΓ/dt)/Uf2

1
0.8
0.6
0.4
0.2
0
-0.2
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
t*

FIG. 4: Circulation production from the leading edge of the wing. The lines show filtered values from five
separate trials.

20 tion versus time curve is near linear overall, though

18 Direct Measurement it could be said to be more quadratic at very early
16 Integrated Flux times. This nonlinear variation at the start of mo-
tion is a direct consequence of the linear increase in
14
circulation flux seen in fig. 4. When that linear func-
12
Γ/(c Uf)

tion is integrated in time, a quadratic curve results.

10 The integrated circulation production matches with
8 the total circulation as directly measured, except
6 for the period between t∗ = 3 and 8. This is
4 in opposition to the results reported by Panah et
2 al. [31]. Their study found the leading edge to pro-
0 duce twice as much circulation as was measured di-
0 2 4 6 8 10 12 14 16 18 rectly. In that study, the difference was accounted
t*
for via their measurements of the secondary vortic-
ity produced under the LEV, which annihilated half
FIG. 5: Positive circulation nominally in the LEV of the leading-edge vorticity production. The dis-
for 5 independent measurements. crepancy between their study and the present work
likely lies in their use of reciprocating plunge kine-
matics, although the underlying reasons are not en-
tirely clear. This annihilation effect could also be
vortex formation ends due to the change in kinemat- the reason for the difference between measured and
ics or as a result of the vortex becoming saturated time-integrated circulations in the present study be-
with circulation. Note that the circulation produc- tween t∗ = 3 and 8, as this corresponds to a period
tion settles to a nearly constant value at later times, of very strong secondary vorticity generation. (Note
t∗ ≥ 12. the negative vorticity boundary layer at these times
The amount of total positive circulation measured in fig. 1.)
in the flow field, i.e. that nominally in the LEV and
including that in the free wake, is shown in fig. 5 as
a function of nondimensional time, t∗ . The figure
shows both the directly measured area integral of
circulation from each PIV frame and the time inte-
gral of the leading-edge circulation production. Cir- In summary, it has been shown that the produc-
culation increases monotonically with time as one tion and subsequent transport of vorticity are impor-
would expect based on the overall state of the flow tant factors in the force production of the wing, and
in section III A and the always-positive flux mea- therefore are also important factors in the prediction
surements. The vorticity generated at the leading of forces. In the following section, models from the
edge becomes a shear layer that feeds into the LEV, literature will be classified and evaluated on their
continuously increasing its strength. The circula- ability to predict the evolution of wake vorticity.
 12

IV. MODELING METHODS spectively:

1
This section evaluates the success of various mod- L = πρcU 2 sin(α) + πc2 ρ sin(α) cos(α)U̇ (6a)
els from the literature at predicting the forces expe- 4
1 2 2
rienced by a flat plate wing in the baseline case. In D = πc ρU̇ sin (α) (6b)
most previous work, model success has been judged 4
via comparisons of qualitative images of the wake 1
Mb = − πρc2 U 2 (2b − 1) sin(2α). (6c)
vorticity and the corresponding force production. 8
While these are certainly reasonable metrics, they
are both global flow measurements and so tend to The moment, M , is given about a pivot point, b,
obscure the root causes of any problems that may defined as 1 at the leading edge and −1 at the trail-
exist. In the current work, a closer look is taken ing edge. The pivot point is usually specified at
at quantitative comparisons. Crucially, model pre- the quarter chord, b = 1/2, because this causes the
dictions for the characteristics of the LEV, e.g. cir- moments that arise from wing translation and the
culation and circulation production, are compared bound vortex to cancel each other. Applying a nor-
to their experimentally measured counterparts. Cir- malization by dynamic pressure, q = 1/2ρUf2 c, as in
culation production in particular is a good method the experimental results and substituting b = 1/2
for isolating, insofar as it is possible, the critical results in expressions for the coefficients of lift, drag
leading-edge condition. While it is not possible to and pitching moment:
completely separate the effect of the leading-edge
condition from the rest of the flow, comparing cir-
culation production is as close as one can come to U2 c U̇
decoupling the system. This allows for an evaluation Cl = 2π sin(α) + π sin(α) cos(α) 2 (7a)
Uf2 2 Uf
of the Kutta condition as a leading-edge condition
independent of the global flow state. c U̇
Cd = π sin2 (α) 2 (7b)
2 Uf
Cm,c/4 = 0. (7c)

Here, U represents the instantaneous wing velocity

and Uf is the final, reference, velocity. The lift, Cl ,
A. Quasi-Steady and Data Driven Models equation has two components: the first is the cir-
culatory contribution from the bound vortex, and
the second is the non-circulatory contribution (some-
times called the “added mass” force). The drag, Cd ,
The simplest possible model of the flow over a flat
equation has only a non-circulatory component, and
plate wing is to assume that it is attached (perhaps
predicts zero drag in the steady state. The circula-
an overly bold assumption), and that the wing is
tory force component always acts in the vertical di-
undergoing steady translation at a constant angle of
rection (i.e. lift), while the non-circulatory force al-
attack. The start of the motion is assumed to have
ways acts in the plate-normal direction but is zero in
occurred long ago, and thus the starting vortex is
steady flow. The moment about the quarter chord,
considered irrelevant. The bound vortex strength
Cm,c/4 , is also predicted to be zero. Equations 7 are
(i.e., the bound circulation on the wing) is deter-
the classic results of thin-airfoil theory (see, e.g., An-
mined so as to satisfy the Kutta condition at the
derson [51]), slightly adapted to account for a large
trailing edge,
angle of attack, a time-varying wing velocity, and
Γ = πcU sin(α), (5) non-circulatory force.
The forces given by equation 7 can easily be com-
puted for the towing motion considered here and so
where c is the chord, U is the wing velocity, and α compared to the measured forces. In a quasi-steady
is the angle of attack. The quasi-steady modeling model, the flow state, and therefore the forces, can
approach neglects the wake entirely. The circulation only depend on the instantaneous wing kinematics
that would be shed into the wake to balance changes (i.e., wing location, angle of incidence, and their
in the bound circulation, as required by Kelvin’s first derivatives in time). This makes a quasi-steady
theorem, is ignored. This results in the following model extremely cheap to evaluate, but, as will be
expressions for lift, drag, and pitching moment, re- shown, makes it difficult to capture many of the gov-
 13

erning physics. fig. 7. Figure 7a compares the amount of bound cir-

Figure 6 shows a comparison of the measured culation predicted by the attached flow assumption
plate-normal (CN ) and plate-tangential (CT ) force to the measured circulation in the LEV. Figure 7b
coefficients to the predictions of the model. It also compares the time rate of change of the bound cir-
shows the two components of the model. The cir- culation to the measured circulation production at
culatory lift is shown as “CL circulatory,” and the the leading edge as computed from the vorticity flux
plate-normal added mass as “CN non-circulatory.” leaving a box around the leading edge. In these fig-
The steady-state force prediction gives the classic ures, one sees the underpinnings of the circulatory
result of Cl = 2π sin α and Cd = 0, which has been force prediction, as well as the reasons for its fail-
shown to compare favorably with experimental mea- ings. Up to t∗ = 1 the circulation and circulation
surements on thick airfoil profiles (see, for example, production matches reasonably well, but the predic-
Abbott and Von Doenhoff [83]), at least until stall. tions given by the quasi-steady model grow much
When the wing stalls, experimental values of the lift faster than the measured values through t∗ = 4,
force are significantly lower and the drag force is sig- and quickly outstrip the actual circulation in the
nificantly greater that this prediction. In the case of real (measured) case. The amount of circulation
the current experiments, the wing is clearly stalled predicted by the model becomes constant when the
and the normal force is over-predicted, particularly wing stops accelerating and reaches its final velocity
at long times. Additionally, focusing on normal and at t∗ = 4 because the bound circulation, given in
tangential force rather than lift and drag hides the equation (5), is a function of wing velocity only.
complete lack of a drag prediction in the classic re- Further, the lack of a wake causes the quasi-steady
sult, which fortuitously cancels the over-prediction model to miss the continual production of circulation
in lift to bring the predicted normal force closer to that occurs throughout the wing motion. Figure 7b
experiment. Particularly during the early part of the shows how the model stops “producing” any circu-
wing motion (t∗ = 0 to around 2), the combination lation after t∗ = 4, resulting in the fixed value of to-
of over- and under-prediction in the model gives a tal circulation after t∗ = 4. This fixed value causes
result that is surprisingly close to the measured nor- the initial over-prediction of circulation to eventually
mal force, but not the tangential force. become an under-prediction as the real flow contin-
The non-circulatory force, labeled “CN non- ues to generate vorticity at the leading edge. This
circulatory” in fig. 6, is quite difficult to isolate in ex- is indicative of a primary difference between steady
periment. Potential theory predicts a force per unit attached and stalled flows—stalled flows require the
depth of F = 0.25πc2 U̇ sin α in the plate-normal di- continued production of vorticity to feed the shear
rection. This is a two-dimensional calculation and layers at the edges of the plate, even in the fully de-
hence neglects any end effects, but simply multiply- veloped flow. Attached steady flow reaches a fully
ing by the span of the wing turns out to be essen- developed state with a certain amount of bound cir-
tially exact for the aspect ratios considered here, culation. Thereafter, attached flow does not produce
and is as confirmed in the review of added mass any further net circulation, except in the case of un-
by Brennen [84]. In the current problem, the non- steady motion, which will necessarily produce wake
circulatory force has a characteristic top-hat shape circulation to balance changes in the bound circula-
that corresponds exactly to the acceleration profile tion.
and matches the magnitude of the initial jump in Adding an empirical correction can mildly im-
forces. Puzzling, however, is the lack of a distinct prove the force prediction given by the quasi-steady
drop in the measured forces when the wing stops ac- model, and this approach is widely adopted in the
celerating, at t∗ = 4, when the non-circulatory force flapping wing controls community [45, 46, 85]. This
goes to zero. approach works best for rotating wing kinematics
Comparison of the quasi-steady model with the because the attached LEV that forms there causes
measured LEV data cannot be done in a direct fash- the force to rapidly achieve an LEV-enhanced steady
ion since the model does not include a wake of any state value, as seen in Manar et al. [86], making the
sort. However, a general comparison can be made quasi-steady assumption valid. Adding a bit of em-
between the bound circulation of the attached flow pirically derived foreknowledge to the quasi-steady
model and the strength of the LEV. This may not model can be done by simply using a lookup table
be all that far-fetched, as it has been previously for Cl,f and Cd,f based on the steady-state time-
argued by Pitt Ford and Babinsky [24] that a flat averaged coefficients. The coefficients for the single
plate wing at high angle of attack contains very lit- angle of attack considered are taken from the steady
tle bound circulation, and thus any circulation that state results in fig. 2, where CL,f = CD,f = 1.1.
would have been bound is present in the flow as an Since the current test case is fully three-dimensional,
LEV. The results of this comparison are shown in values of CL,f and CD,f (the finite wing form of
 14

4.5
CN measured
4 CT measured
3.5 CN model
3 CT model
2.5 CL circulatory
CN non-circulatory
2
CF

1.5
1
0.5
0
-0.5
0 2 4 6 8 10 12 14
t*

FIG. 6: The quasi-steady thin airfoil theory model force prediction compared to experimental data.

coefficients) from experiment have been used, and tains significant wake vorticity dynamics, in particu-
thus finite wing effects are empirically accounted lar the formation of a leading-edge vortex above the
for. The instantaneous values of the circulatory suction side of the wing. The presence of this vortex
forces are then CL (t) = U (t)2 /Uf2 CL,f and CD = near the wing produces a low-pressure region that
U (t)2 /Uf2 CD,f . The same equations as before are augments the force on the wing beyond what the
used for the non-circulatory component, resulting in quasi-steady model predicts. A quasi-steady model
the time history shown in fig. 8. The steady state has no hope of capturing the effect of the LEV be-
values for the coefficients are taken from fig. 2 at cause by definition all terms related to the wake and
t∗ = 30. The dependence on dynamic pressure gives time history have been neglected. In spite of this
the force transient a quadratic shape during wing ac- omission, the quasi-steady model is not a complete
celeration (proportional to U 2 ), followed by a fixed failure. The non-circulatory force, which does not
value after the wing reaches its final tow velocity. depend on the wake, captures the initial jump in
Note that the CD line has been omitted in this plot force production as the wing begins to move quite
since it is identical to the CL curve for a wing at well. This is a common theme throughout the mod-
45◦ . In the wing-fixed reference frame, using the els, as the theory behind the non-circulatory force
empirical data has given the correct result of zero does not require any of the approximations required
plate-tangential force. to capture the wake and circulatory forces.
In summary, the main attraction of the quasi-
To first order, the predicted and measured val-
steady model is its simplicity and ability to pro-
ues of CN , shown in fig. 8, are not all that different.
duce reasonable results when coupled with empiri-
Both have an initial sharp rise in force from the non-
cal data. Its weakness, however, lies in its lack of
circulatory component, as well as a subsequent rise
a wake model. The next models will take the first
in forces during the acceleration. The model gives
steps towards considering the full unsteady wake.
this as a quadratic function; the measured data has
a similar shape. The model also predicts peak force
near the end of acceleration at t∗ = 4. The em-
pirical quasi-steady model misses the magnitude of B. Fixed Wake Model
the peak by a fair margin, however, and does not
capture the subsequent relaxation to steady state.
For flows with unsteadiness, the change in bound
It does match the forces once they have settled to
circulation on the wing must be balanced by cir-
steady state after t∗ = 30 (not shown), but this is no
culation shed from the edges of the wing. In the
great feat considering the empirical correction used
fixed wake model, this change in circulation is cap-
in this data-driven approach.
tured, but only attached flow is considered to still
The model’s failure to capture the magnitude of allow for a pen and paper solution. The second as-
the peak forces and their subsequent relaxation is di- sumption taken here is that the wake sheet ema-
rectly related to its complete disregard of the wake. nating from the trailing edge of the wing does not
As was shown in section III A, the actual flow con- convect (in the lab-fixed reference frame) but is sim-
 15

18 unsteady motions that maintain attached flow, such

16 as avian flapping in Hall and Hall [93] and Hall et
14
al. [94]. Their work was then further extended by
Salehipour and Willis [95] to find optimum flapping
12
kinematics.
Γ/(c Uf)

10
8
The results of applying Wagner’s model to the
present case are shown in fig. 9. Mathematically,
6
this is accomplished by convolving the lift deficiency
4
measured function with the bound circulation result of fig. 7a.
2 model The lift deficiency function can be found in Wag-
0 ner’s original work [33], or many text books, e.g.
0 2 4 6 8 10 12 14 Bisplinghoff et al. [1]. Its basic shape is a curve
t* that begins at one half, and then gradually asymp-
totes to the final value, taking on the order of twelve
(a) Total Circulation chords of travel to reach ninety percent of steady
7 state. Thus the effect of convolving it with the cir-
measured culation curve is to delay the growth in lift force of
6
model the wing. The plot in fig. 9 gives the same com-
5 ponents as in the results of the quasi-steady model
4 shown in fig. 6 and fig. 8: the measured CN and CT
(dΓ/dt)/Uf2

3 curves alongside their predicted counterparts, and

the CL circulatory and CN non-circulatory model
2
components. As expected, Wagner’s model shows,
1 in fig. 9, many of the same features as the quasi-
0 steady model of section IV A. The obvious difference
-1 here is the delay in the buildup of the circulatory
0 2 4 6 8 10 12 14 force. Accounting for this delay causes the model to
t*
do a reasonable job of predicting the magnitude of
the normal force during acceleration (t∗ < 4, com-
(b) Circulation Production paring the red model results to the blue measured
results in fig. 9), and instead of over-predicting the
FIG. 7: A comparison of the circulation predicted peak force just before t∗ = 5, as in the quasi-steady
by pure thin airfoil theory and the measured case, it under predicts it. The steady state values
values. Note that technically the measured data is are, as before, vastly over predicted. The align-
that of the LEV in the wake, while the model ment of the normal forces at early times (t∗ < 4)
curves show bound circulation. is disingenuous however, as it neglects the plate-
tangent component of force. In keeping with the
Kutta-Joukowski lift theorem, Wagner’s fixed wake
ply left behind by the wing. Using this approach model predicts the presence of lift with no drag. In
produces the well-known lift deficiency function that the wing-normal and wing-tangential axes, this cor-
describes the delay of circulatory force production, responds to equal plate-normal and plate-tangential
e.g. Wagner’s model [33, 87]. The fixed trailing components at α = 45◦ . It is thus only a twist of fate
edge wake is the core principle behind the models of that the normal force has nearly the correct magni-
Wagner [33] and Theodorsen [35], which cover im- tude. The plate-tangential force, which was mea-
pulsive and oscillatory wing motions, respectively. sured at close to zero, is modeled to have the same
Several authors have since extended on the basic magnitude as the normal force. Thus, in reality, the
principle, such as Greenberg’s model of an oscil- total force is predicted by this model to be much
lating free stream [88], Leishman’s model account- larger than the measured total force.
ing for compressibility effects [89], von Kármán and To see why the lift is over-predicted, fig. 10 shows
Sears’s impulse derivation [87], Kussner’s model for a comparison of the bound circulation in the fixed
a sharp-edged gust response [90], and Sears’ oscil- wake model to the measured LEV circulation, as well
lating gust model [91]. These approaches have each as the predicted rate of change of circulation to the
been experimentally validated (e.g., in Ref. [92]) for measured vorticity flux at the leading edge. These
small amplitude attached flow disturbances and en- results show the same over-prediction of total circu-
joy widespread use. Numerical implementations of lation and circulation production that was seen in
these ideas have also been used for large amplitude quasi-steady model during the acceleration portion
 16

3
CN measured
2.5 CT measured
CN model
2 CT model
CL circulatory
1.5 CN non-circulatory
CF

0.5

-0.5
0 2 4 6 8 10 12 14
t*

FIG. 8: The quasi-steady model augmented with empirical force coefficients and compared to experimental
data.

4.5
CN measured
4 CT measured
3.5 CN model
3 CT model
2.5 CL circulatory
CN non-circulatory
2
CN

1.5
1
0.5
0
-0.5
0 2 4 6 8 10 12 14
t*

FIG. 9: The results of Wagner’s fixed wake model compared to experimental results.

of the wing motion (t∗ < 4), followed by an under cation of that wake model.
prediction at steady state. The Wagner model, be-
ing an attached flow model, also shares the same
deficiency as the quasi-steady model in predicting C. Similarity Solution
long term force behavior. Attached flow models pre-
dict a fixed total amount of circulation, while the To further improve on the solution offered by the
actual stalled case continues to create circulation ad- fixed wake model, the convection of the vortex sheets
infinitum. due to both the wing’s velocity field, and the effect of
However, the early portion of the circulation pro- the wake on itself, needs to be considered. A general
duction (t∗ < 6) and total circulation (t∗ < 3) has solution is analytically intractable, but for simple
been significantly improved in this model. Circula- cases a similarity solution to the shape and strength
tion production matches experimental results until of the vortex sheets shed from a sharp edge can be
t∗ = 2, and the decay of production after the end found. This is the idea behind the work of Pullin
of acceleration (t∗ = 4 to 6) at least has the cor- [96], who proposed a truncated series solution for a
rect shape. This is a good sign that the addition of vortex sheet shedding from the end of a semi-infinite
a wake to the flow model has improved the overall plate in cross-flow. Pullin and Wang [36] applied
quality of the prediction despite the gross simplifi- that solution to a finite plate, and showed that to a
 17

18 the acceleration portion of the test (i.e. t∗ <= 4),

16 measured as it cannot accommodate the change to constant
model velocity.
14
12
Γ/(c Uf)

10
The equations for force on a flat plate wing start-
8
ing from rest given in Pullin and Wang [36] break
6 down into two components: a non-circulatory and a
4 circulatory component. Both force components have
2 only a plate-normal term, as meeting the Kutta con-
0 dition at both plate edges removes any possibility of
0 2 4 6 8 10 12 14 plate tangential force (i.e., edge suction). The no-
t* tation from their paper has been modified to fit the
notation used here. The non-circulatory component
(a) Total Circulation is purely plate-normal, and is given as:
4.5 1 2
4 measured N= πc ρ sin(α)U̇ (9)
model 4
3.5
3 where N is the plate-normal force, and all other
(dΓ/dt)/Uf2

2.5 quantities have the same meanings as before. This

2 matches the non-circulatory force used previously in
1.5 equation (6), albeit written as a plate-normal force
1 rather than decomposed into lift and drag.
0.5
0
-0.5 The circulatory component, also purely plate-
0 2 4 6 8 10 12 14
normal, is the heart of the model. Using the sim-
t* plified velocity profile given in equation (8), the
evolution of the shed vortex sheet can be repre-
(b) Circulation Production
sented by an infinite-series similarity solution based
on Kaden’s spiral. The leading term of the series is
FIG. 10: A comparison of the circulation predicted
then solved for to give the shape of the rolled-up vor-
by Wagner’s model and the measured values. Note
tex sheet. In the course of the derivation, given in
that technically the measured data is that of the
Ref. 36, it is also argued that the effects of the plate-
LEV in the wake, while the model curves show
tangent velocity component only enter the problem
bound circulation.
as higher order terms of the series and are thus ne-
glected. The resulting equation for the circulatory
force is given as
first-order approximation, the leading- and trailing-
edge vortex sheets do not influence one another. 2 5(1+m)
n
1/2
o
N= (5m + 2)Kρc1/2 J0 a5/3 t 3 −2 < ω0
To make an analytical solution of the wake sheet 3
tractable, Pullin’s similarity models assume a wing (10a)
velocity profile with a simple polynomial dependence a = c1/2 B sin(α) (10b)
on time:  2/3
3
K= (10c)
U = Btm (8) 4(1 + m)

where B is a scaling factor and m determines if the where N is once again the normal force, and a and K
profile is of constant speed, constant acceleration, are convenient scaling factors. The values of ω0 and
etc. The flow is assumed to be quiescent at t = 0. J0 represent the similarity solution shape and cir-
For comparison to the present work, a constant ac- culation, respectively, of the rolled-up vortex sheet
celeration profile of m = 1 was used. The value of B and are non-analytical functions of m. Their values
was set to match the velocity profile modeled here are taken, as in Pullin and Wang [36], from the nu-
to that of the experiments, resulting in B = 0.25. merical solution in Pullin’s previous work [96], and
Choosing a constant acceleration profile immedi- are ω0 = −0.17 + 0.33i and J0 = 2.185. A series
ately limits the applicability of this model to only expansion for circulation of the LEV is also given in
 18

6 5
CN measured 4.5 measured
5 CT measured 4 model
CN model 3.5
4 CT model 3

Γ/(c Uf)
3 2.5
CF

2
2 1.5
1
1
0.5
0 0
0 2 4 6
0 2 4 6
t*
t*
(a) Total Circulation
FIG. 11: The similarity solution model compared 1.8
to experimental data. 1.6 measured
1.4 model
1.2

(dΓ/dt)/Uf2
Ref. 36, which to leading order is 1
0.8
δ02 (t) 0.6
Γ= J0 (11a)
K 3/2 t 0.4
δ0 (t) = Ka2/3 t2(1+m)/3 (11b) 0.2
0
where δ0 is another intermediate scaling factor. The -0.2
rate of circulation produced by the accelerating wing 0 2 4 6
is the time derivative of the expression for Γ. t*
These equations for force are based on solutions
for both the shape and circulation of the wake sheet (b) Circulation Production
that satisfy only the leading term in an infinite se-
ries, and thus accuracy is expected to be limited to FIG. 12: A comparison of the circulation predicted
short times. In addition, it is worth noting that by the similarity solution model and the measured
making the solution analytically tractable required values.
Pullin and Wang to ignore the effect of the LEV on
the TEV and vice versa, although this was shown to
be a second-order effect for the short times of inter- a very simple exponential velocity profile definition.)
est. At longer times (t∗ > 0.25) the mismatch is likely
Results from implementing Pullin and Wang’s due to neglecting the higher order terms, particu-
similarity model are given in fig. 11, where they show larly those that deal with the velocity induced by
agreement with the experimentally measured results the LEV on TEV and vice versa.
through t∗ = 2, after which time they predict a In spite of its shortcomings, the similarity solution
smaller plate-normal force than was measured. This model presents a closed form solution to the sepa-
is the expected result based on conclusions in Pullin rated flow that correctly predicts the lack of plate-
and Wang’s paper [36]. A reason for the under- tangential force. In doing this, the model provides
prediction of the forces can be found in the circula- valuable insight into how the flow behaves at early
tion prediction given in fig. 12. The total circulation times. This model also serves to illustrate the limits
is predicted reasonably well for t∗ < 1.5, but does of what a purely analytical solution is capable of cap-
not scale properly with time, and does not adhere to turing. While the assumption of no LEV-TEV inter-
the linear growth of the experimental measurements. action in the Pullin and Wang model produces viable
This results in under-predicted circulation and force results for short times, even at moderate times the
values at longer times. At short times, the differ- interaction must be captured in order to accurately
ence in shape between the measured and predicted model the resulting force production. Furthermore,
circulation productions is likely due to the model’s the similarity solution model is not capable of deal-
ignorance of the smoothing used for the experimen- ing with anything except single-term polynomial ve-
tal velocity profile. (Recall that the model requires locity profiles. In order to implement a more general
 19

solution method, Pullin admits, and this author con- vortex model to experimental results stemmed from
curs, that a numerical procedure must be used [96]. enforcement of the Kutta condition at the leading
While the similarity solution model’s stand-alone use edge. Either the Kutta condition was an incorrect,
may be limited, it is quite useful in its capacity as overly aggressive, choice for the leading-edge bound-
a method for kick-starting more general numerical ary condition, or using a point vortex far from the
methods (e.g., those in Refs. 21, 40) by providing a shedding edge lead to overzealous addition of circula-
solution to the early, nearly singular, vortex behav- tion. The effect of a point vortex falls off with 1/r, so
ior. to maintain the same condition at the edge, the vor-
tex strength must increase faster for lone shed point
vortices than if a shear layer was included. However,
D. Two-Vortex and Impulse-Matching Models the measured circulation data in fig. 14 shows that
the strength of the LEV is captured by the model.
The expected effect of the point vortex receding from
To combat the growth in computational cost and
the wing is seen in the increasing overshoot of the
bridge the gap between numerical and analytical
modeled circulation production after t∗ = 3. Bet-
models, Wang and Eldredge [21] extended the work
ter, in fact, than the force data, which rapidly di-
of Cortelezzi and Leonard [97] on modeling unsteady
verges from the measurement for t∗ > 1. Based on
separated flows with point vortices that have time-
the prediction of LEV circulation, one would expect
varying strengths. In this approach, the wake is rep-
reasonable force predictions through at least t∗ = 2.
resented by a very small number of point vortices
(i.e., one for each shedding edge) whose strengths Note that there is some noise in the circulation
change in time to maintain the Kutta condition as flux just after the start of the wing motion. This
the vortices convect downstream. This constitutes is associated with the leading-edge vortex not being
a wake model that captures the effect of flow sep- placed in quite the correct location by the initial con-
aration from both edges while keeping the degrees ditions and the time step being too large to handle
of freedom to an absolute minimum, making it an to nearly singular velocity close to the plate edges in
extremely computationally efficient modeling tech- the circle plane. The noise quickly settles out as the
nique. vortex moves away from the singularity, and the ini-
The results of applying this two-vortex model to tial placement has only a small effect on the overall
the present case are shown in fig. 13. The figure outcome.
shows the measured forces in the wing-relative ref- Since circulation appears to be predicted reason-
erence frame, CN and CT , compared with predic- ably well, the error in force prediction must come
tions given by a two-vortex model. It also shows the from elsewhere. The answer lies in the location of
model’s circulatory and non-circulatory components the vortex. The model predictions for LEV location
independently. The two-vortex model correctly cap- are shown in fig. 15. The two subfigures show the x/c
tures the lack of plate-tangential force, but over- and y/c values obtained from the two-vortex model,
predicts the strength of the plate-normal force by al- as well as experimental measurements in which the
most a factor of two. The non-circulatory forces are vortex is identified using both the vorticity centroid
captured in exactly the same manner as in the pre- and Γ1 criteria. While the modeled x/c location of
vious models, with similarly successful results. The the LEV is on the right track with respect to the
discussion will thus once again focus on the circula- experimental results, the two-vortex model predicts
tory force contribution from the wake. a lower value of horizontal displacement than the
The force history in fig. 13 does not extend past measurements, indicating that the model represents
t∗ = 5 because the simulation becomes unstable. In a LEV that is closer to the wing than observed in
certain configurations, it happens that the vortices experiments. The trend in y/c location as predicted
have little ability to enforce the Kutta condition at by the model is backwards, indicating an LEV that
the edges. This leads to large changes in circula- rises above the leading edge of the wing, rather than
tion for small changes in vortex location. When one that sinks below it. Incorrectly predicting the
coupled with the Brown-Michael convection scheme location of the LEV has a direct impact on the force
[98], which alters the vortex velocity to account for experienced by the wing.
circulation change, these two effects feed back into Note that the location of the vortices also im-
each other and cause a divergence of vortex strength pacts their circulation as the vortex strengths are
and location. The initial stages of divergence can be updated each time step to continue enforcement of
seen in the squiggles of rapidly increasing magnitude the Kutta condition. As the vortex moves further
at the end of force curves in fig. 13. away from the plate its image approaches the cen-
It was initially assumed that the over-prediction ter of the cylinder, and thus is never more than a
of force observed in the comparison of the two- cylinder radius away from the edge. Therefore the
 20

6
CN measured
5 CT measured
CN model
4 CT model
CL circulatory
3 CN non-circulatory
CF

CN impulse match
2

0 2 4 6
t*

FIG. 13: The Wang and Eldredge point vortex model.

6 image vortex, rather than the vortex itself, accounts

measured for the majority of the induced velocity at each edge.
5 model Furthermore, since the image moves relatively little,
the location of the vortex actually has a diminishing
4
impact on its own circulation as the vortex recedes
Γ/(c Uf)

3 from the plate. The force produced by the vortex

pair however, is independent of vortex location.
2 Wang and Eldredge [21] were aware of the incor-
rect convection, and the second thrust of their pa-
1
per was to propose a new model for convection that
0 moves the vortex so as to cancel the effect its mo-
0 2 4 6 tion has on the total force. As an added benefit,
t* the impulse-matching model allows arbitrary choice
of vortex shedding time. The results of switching
(a) Total Circulation to their new “impulse matching” convection scheme
are shown in fig. 13. Applying the impulse matching
2.5
measured convection improved the force, but did not entirely
2 model eliminate the problems already noted.
The two-vortex wake model is extremely attrac-
1.5 tive from a conceptual and computational cost
(dΓ/dt)/Uf2

standpoint, but is unfortunately hamstrung by its

1
over-simplified wake representation. The key omis-
0.5 sion appears to be the effect of the shear layers that
form at the leading and trailing edges of a wing in
0 viscous flow. These shear layers are where the roll-
up of new vortices occurs; without capturing their
-0.5
0 2 4 6
dynamics either an additional ad-hoc vortex shed-
ding relationship must be added or coherent vor-
t* tex shedding forgone altogether. These deficiencies
(b) Circulation Production would merely limit the model to short times if it
were not for its gross overestimation of the forces.
FIG. 14: A comparison of the circulation predicted This shortcoming has been recognized by the El-
by Wang and Eldredge’s model and the measured dredge group, and an extension of the present model
values. that includes the shear layer has been proposed [99].
In summary, the two-vortex model presented here
shows great promise for relatively short times when
the flow is dominated by the initially shed LEV and
 21

0.2 0.2
panels
0 0.1 vortices
-0.2
0

y
-0.4
-0.1
-0.6
x/c

-0.8 -0.2
-0.4 -0.2 0 0.2 0.4
-1 centroid x
Γ1
-1.2
model FIG. 16: A representative panel method mesh with
-1.4 N = 12. The actual computations used a finer
0 2 4 6 mesh where N = 64.
t*

(a) LEV x/c

The method implemented for the present work is
0.1 similar to the one outlined by Katz and Plotkin [52].
0 The wing is represented by N = 64 point vortex pan-
-0.1 els, with cosine spacing across the chord, depicted
-0.2 in fig. 16. Each panel contains a point vortex at
-0.3 the panel center; collocation points are at the panel
-0.4 edges. This setup places collocation points exactly
y/c

-0.5 at the plate edges and thus implicitly satisfies the

-0.6 Kutta condition at both edges. The addition of two
centroid
-0.7 Γ1 new vortices each time step models the shedding of
-0.8 model circulation. For the first time step, these vortices
-0.9 were placed a distance of 2% of the chord away from
0 2 4 6 the shedding edge. The exact location of the ini-
t* tially shed vortex was found to have little impact on
the results. In subsequent time steps, new vortices
(b) LEV y/c were placed at a third of the distance from the shed-
ding edge to the previously shed vortex, as per the
FIG. 15: A comparison of the location predicted by method in Ansari et al. [41]. The time step size was
Wang and Eldredge’s model [21] and the measured selected to be t∗ = 0.015 as in Xia in Mohseni [56].
values. The net result at each time step is N + 2 vor-
tices to solve for no through-flow at N + 1 colloca-
tion points. The system is closed by specifying zero
TEV. For the flapping wing applications, for exam- total circulation. This method enforces the Kutta
ple, this may be all that is required before the wing condition by meeting the plate-tangent shedding ve-
reverses course. locity at the edges. Vortex locations were evolved in
time with an explicit Euler scheme. A vortex core
model was not used so that an off-the-shelf fast mul-
E. Multi-Vortex Model tipole solver, FMMLIB2D [100], could be employed.
Forces were computed with the impulse method of
Lagrangian multi-vortex methods present them- Wu [101]. As a final consideration, the model used
selves as an attractive option to solve the problem a linear trapezoidal velocity profile rather than the
of wake convection in a more holistic way. Allowing smoothed version used in experiment.
the wake to convect with the flow leads to the roll-up The downside of using a panel method for body
of the shear layers at the leading and trailing edges representation is that it becomes difficult to sepa-
and the formation of the leading- and trailing-edge rate the resulting force into the circulatory and non-
vortices. This natural inclusion of the wake evolu- circulatory components. This could be remedied by
tion has made Lagrangian vortex models a popular instead representing the plate via conformal map-
choice, and many examples exist in the literature ping as in Xia and Mohseni [56], though the point
including Katz’s thick airfoil method [42], Xia and vortex panel method was selected in this case for
Mohseni’s conformal mapping method [56], Ansari its extreme ease in calculating the forces from the
et al. ’s strip theory approach for 3D wings [41, 57], vortex impulse.
and Hammer et al. ’s work [20], to name but a few. By continuously shedding vorticity from the wing,
 22

2
ues outside of a slight over-prediction between t∗ = 5
1 and 6. A salient point to make here is that the Kutta
0 condition performs well at the leading edge in spite
y

-1 of earlier misgivings. It appears that the Kutta con-

-2 dition is both valid and useful for determination of
-3
leading-edge shedding for thin wings.
0 1 2 3 4 5 6 7 8 9 The question that now arises is how the circu-
x
lation production can be well-predicted throughout
(a) Multiple-vortex model the wing motion, but the forces less so. This is per-
haps because the truly two-dimensional model is be-
ing compared with three-dimensional experimental
results that were obtained on a finite aspect ratio
wing. When two-dimensional measured force data is
available, two-dimensional multi-vortex models have
(b) Vorticity field from PIV with frozen wake been shown to accurately reproduce the force histo-
hypothesis ries [20, 44, 102].
If some method for determining when shedding
FIG. 17: Wake comparison at t∗ = 10.0 should or should not occur from the leading edge
were included, the present calculations could also be
improved to account for more general kinematics,
the shear layers at both edges of the wing are more such as low angles of attack, pitching, and oscilla-
properly captured. Furthermore, allowing the vor- tion. In a low order model such as this, such a de-
ticity to convect naturally results in the roll-up termination is generally left to ad-hoc methods such
and shedding of large-scale vortices. As seen in as the leading edge suction parameter (LESP) [44],
fig. 17, the resulting wake compares favorably with a flow angle limit [102], or a simple force limit [81].
the experimental frozen wake result. Using a large The necessary leap to make the discrete vortex
(O(103 )) number of vortices also captures the ef- methods considered in this section a complete an-
fect of the spatial extent of shed vorticity. In sum- swer to the full flapping wing problem is to extend
mary, Lagrangian multi-vortex models make for a these methods into the third dimension, but this
relatively well resolved wake compared to the pre- seems to exist only rarely in practice (e.g. Roccia
viously discussed methods, albeit at the expense of et al. [71]). It appears that fluids researchers who
mild additional computational cost, which may be seek fundamental understanding of separated flows
more expensive than some models but is still signif- tend to overlook discrete vortex models because they
icantly cheaper than full CFD. stray too far from pen and paper models into numer-
The force results from the multiple-vortex model ical methods, and thus offer little insight into the
are shown in fig. 18. The first feature of note is origin of forces. Nevertheless, the results presented
that the model correctly predicts the lack of plate- here, and others in the literature, show that compu-
tangent force. The second important feature is that tational methods like the present two-dimensional
the shape of the measured force histories is cap- discrete vortex model can be excellent tools for de-
tured fairly well. The model correctly predicts the termining the force on a two-dimensional profile un-
peak force connected with shedding of the first LEV der arbitrary motion.
around t∗ = 5, as well as the second force peak due
to the formation of a second LEV around and af-
ter t∗ = 8. The magnitude of the force is, however, V. CONCLUSIONS
not perfect. In particular, the minimum at t∗ = 7
is under-predicted and the model over-predicts after The first portion of this paper has examined ex-
t∗ = 8. perimental measurements of the bulk wake charac-
The total circulation and leading-edge circulation teristics of a flat plate wing at high angle of attack
production as predicted by the multi-vortex model starting from rest. These measurements focused on
are shown in fig. 19. The model does a remarkable the process of circulation generation at the leading
job of reproducing the leading-edge circulation pro- edge of the wing. The second portion of the paper
duction, especially through t∗ = 6. The minimum presented the implementation of several of the per-
at t∗ = 6 and second maximum at t∗ = 9 are not tinent models from the literature and compared the
as accurately resolved. The success in matching the results of these models to the experimental results.
circulation production carries over to the total circu- Section III examined a single experimental test
lation which falls well in line with the measured val- case in which an aspect ratio eight wing was held
 23

6
CN measured
5 CT measured
CN model
4 CT model

3
CN

-1
0 2 4 6 8 10 12 14
t*

FIG. 18: The point vortex model compared to experimental data.

at a 45◦ angle of attack and accelerated from rest useful at a conceptual level for intuitively predict-
to a Reynolds number of 12,500 over a distance ing the response of the wing. The non-circulatory
of two chord-lengths. The experimental results force component of all of the models shown here was
included two-dimensional flow visualization, time- able to successfully predict the initial acceleration-
resolved force measurement, vortex tracking and cir- dependent jump in forces at the start of the motion.
culation measurement, and time-resolved leading- This led to the conclusion that forces due to plate
edge circulation measurement. The analysis pre- motion can be considered superimposable on the cir-
sented here showed how even this relatively sim- culatory forces regardless of the wake state. Further,
ple case exhibits complex and interconnected wake these forces are well-captured in present models de-
dynamics. It also exposed leading-edge circulation scribed in the literature. The circulatory compo-
production as the critical element of the wake sys- nents, however, account for the action of the wake.
tem, and thus highlighted the need to understand These require a case-by-case analysis, as each model
the wake dynamics in order to understand the forces uses a different method of representing the wake.
on the wing. The attached flow models examined here (i.e. the
The next section, section IV, presented an analy- “quasi steady“ models and Wagner’s “fixed wake”
sis of the strengths and weaknesses of various phys- model in fig. 20) do not include adequate wake mod-
ically based models to test their ability to predict els, and were unable to predict the unsteady forces
the forces experienced by the rapidly accelerated produced by the rapidly accelerated wing on all but
wing. These models included (in increasing order of the most rudimentary levels. The excess of pre-
wake fidelity) a quasi-steady attached flow model, dicted force was found to be due to additional cir-
a data-driven quasi-steady model, Wagner’s fixed culation erroneously present in the flow. The sur-
wake model, Pullin’s similarity solution for sepa- plus in bound circulation stemmed from enforcing
rated plate flow, a two-vortex convected wake model, attached flow on the plate where, in reality, the flow
and a multiple vortex convected wake model. A com- separates from the leading edge. The attached-flow
parison of the plate-normal forces predicted by these assumption also leads to an incorrectly predicted
models, as well as the measured force history, was plate-tangent force of magnitude equal to the normal
given in fig. 20. Of the models selected, only the force. These problems could be alleviated by using
last three account for shedding at both plate edges. empirical values of the lift and drag coefficients, as in
The force predicted by the models studied here the data-driven model, but even this approach still
can be split into non-circulatory (sometimes called neglects the extra forces produced by the formation
“added mass”) and circulatory components, though and shedding of the initial LEV.
whenever invoking such a conceptual divide, it is im- Pullin’s similarity solution model, labeled “simi-
portant to remember that it is a purely conceptual larity” in fig. 20, presents an analytical method for
one. In an actual experiment, such a distinction does obtaining the force on the plate, including the effect
not exist and there is only pressure and skin friction of leading-edge separation. The addition of vorticity
due to the flow as a whole. The concept of a non- shedding from the leading edge brought the direction
circulatory force arises purely as a separation of the of the predicted force in line with the measurements,
terms in the potential flow equations. It is, however, i.e. this model predicted only the plate-normal force
 24

18 excellent agreement with the measured leading-edge

16 measured circulation production. The magnitude of the nor-
model mal force was over-predicted with respect to experi-
14
ments, especially after the first LEV sheds (t∗ > 8).
12
The fully convected wake included here allows this
Γ/(c Uf)

10
model to accurately capture the timing of LEV shed-
8 ding and reformation, something that was not seen
6 in any of the other models. Given the success of
4 the circulation predictions, the failures in predicting
2 forces must be due to other factors such as vortex
0 convection or wake distribution.
0 2 4 6 8 10 12 14 In addition to evaluating each of these models in-
t* dependently and with respect to each other, another
purpose of this work was to assess the validity of
(a) Total Circulation applying the Kutta condition at the leading edge.
Earlier work and intuition have cast doubt on the
2
1.8
validity of the Kutta condition at the leading edge
1.6 based the observed lack of flow leaving tangent to
1.4 the flat plate wing. The models examined here that
1.2 included leading-edge separation all used the Kutta
(dΓ/dt)/Uf2

1 condition at the edge, and were all, within limits,

0.8 successful at predicting the circulation in the LEV.
0.6 The present analysis thus leads to the conclusion
0.4 that the Kutta condition, despite earlier misgivings,
0.2 measured is both valid and useful at the leading edge for high
0 model angle attack flows (α > 30◦ ). Errors in the model
-0.2
0 2 4 6 8 10 12 14
force predictions appear to come from other sources,
such as the discretization or the convection of the
t* wake.
(b) Circulation Production

FIG. 19: A comparison of the circulation predicted ACKNOWLEDGEMENTS

by the point vortex model and the measured values.
This material is based upon work supported by the
Air Force Office of Scientific Research under award
to be nonzero. Ignoring higher order terms limits number FA9550-16-1-0508 and the National Science
the model’s applicability to a short time window af- Foundation under grant no. 1510962.
ter the start of the motion.
Wang and Eldredge’s two-vortex wake model [21],
labeled “two vortex” in fig. 20, uses point vortices of Appendix: Experimental Methods
changing strengths to enforce the Kutta condition at
both plate edges. It was found to predict the produc- This section gives a description of the experimen-
tion of circulation at the leading edge well, at least tal methods and subsequent data analysis used to
until the end of the wing’s acceleration, but over- produce the results in the present experiments. The
predicted the forces on the plate by a factor of two experimental methods very closely followed those of
or more. The error in force prediction was traced to Manar and Jones [103], and further details can be
the model’s incorrect convection of the LEV when found there.
compared to experiment, and the instantaneous mo- The experimental facility used was the 1 m deep,
tion of newly produced circulation from the shedding 1.5 m wide, 7 m long towing tank at the University
edge to the vortex. of Maryland. In brief, the tow motion is provided
The multiple vortex model includes shedding from by a carriage above the tank, which connects to the
both plate edges, accomplished by adding a vortex wing with vertical plunge rods. The force balance
to the wake at each time step. The wake vortices are and subsequent wing mount were located on the
then convected with the background flow while keep- end of a horizontal sting to provide a separation of
ing its strength fixed. This model enforces the Kutta two chord lengths between the wing the plunge rods.
condition at both plate edges, and was found to give The wings used for these experiments were flat alu-
 25

6
Measured
Quasi-steady
5
Data-driven Quasi-steady
Fixed Wake
4
Similarity
Two Vortex
3
CN

Multi-Vortex
2

0
0 2 4 6 8 10 12 14
t*

FIG. 20: The normal force predictions from all models.

200 where Uf is the final wing velocity, c is the wing

chord, and ν is the kinematic viscosity of the fluid
(water in this case). The level of acceleration is spec-
150
ified by the distance traveled during the acceleration
interval: 2c.
U [mm/s]

100
Time-resolved loads were measured on the
wing at 1000 Hz using a submersible ATI Mini40
50 force/torque transducer. The resulting signals were
smoothed with a Savitzky-Golay filter with a width
0
of 0.25 seconds and then ensemble-averaged over five
trials.
0 1 2 3 4 5 6 7 8 9 10 In order to make quantitative statements about
t* the evolution of the wake behind the wing, parti-
cle image velocimetry (PIV) was used to measure
FIG. 21: The velocity profile of the motion. the time-resolved velocities of the flow. The PIV
flow fields were analyzed to track three LEV-related
quantities: the vortex location, the vortex circula-
minum plates. They had a chord, c, of 76.2 mm (3 in) tion, and the circulation production at the leading
and thickness, t, of 3.175 mm (1/8 in) for a thickness- edge of the wing.
to-chord ratio of t/c = 0.042 and an aspect ratio of The planar PIV tests were performed using a
AR = 8. double-pulsed Nd:YLF laser (Litron LDY304, 30
The experimental work discussed here makes use mJ/pulse, 10 kHz max), with the laser sheet ori-
of a piecewise linear towing velocity profile consist- ented in the chordwise direction. Soda-lime glass
ing of a constant acceleration portion followed by a spheres with an average diameter of 34 µm were used
constant velocity portion. The corners (start and as the tracer particles. Images were acquired using
end of acceleration) of the plate’s velocity versus a Phantom v641 camera (4 MP CMOS sensor, up
time profile were smoothed slightly to minimize un- to 3.2 kHz at max resolution) placed orthogonal to
necessary vibrations in the model from rapid changes the laser sheet and tank wall. After a sliding back-
in the driving force applied. Smoothing was ac- ground subtraction to increase the signal-to-noise ra-
complished with a 7th order polynomial in veloc- tio, correlation was performed in DaVis v8.1 using
ity, so that velocity, acceleration, and jerk could be multi-pass interrogation with 50% overlap. A me-
matched at either end of the smoothing segment. dian filter was then applied on 3 × 3 pixel regions,
The smoothing segment was centered at the corner replacing vectors whose peak ratio (the ratio of the
and was set to last 20% of the acceleration time. The highest correlation peak to the second-highest cor-
velocity profile is shown in fig. 21. relation peak) was less than 2 with an interpolated
The motion is set to start at t1 = 0 for conve- velocity vector. The measurements were taken at a
nience. The final velocity of the wing is specified us- chord-wise plane one chord length from the center of
ing the chord-based Reynolds number, Re = Uf c/ν, the wing span and were ensemble-averaged over five
 26

FIG. 22: The results of applying the frozen wake hypothesis. The extent of the actual PIV frame is shown
to the right of the vertical black line.

The vortex location was tracked using two meth-

ods: the Γ1 criteria and via the single-sign vor-
ticity centroid. The Γ1 criteria is the same from
Graftieaux et al. [104], with a threshold value of 0.6.
To further improve robustness, the centroid of the
first region leaving the leading of the wing and sat-
isfying this threshold was computed and this point
was taken as the location of the LEV. For the vor-
ticity centroid, the vorticity field was split into pos-
itive and negative regions along the same lines as
the circulation measurement and the first moment
of vorticity was calculated.
One of the limitations of the PIV setup is that the
camera moves with the wing, and thus the wake is
FIG. 23: The location of flux measurement marked continually convected out of the frame. To alleviate
by the black box around the leading edge. this, the vorticity leaving the frame is assumed to be
frozen in place, forming a first order approximation
to the wake as a whole. The results of this approach
trials. can be seen in fig. 22. It allows for a reasonable
To assess whether the PIV recordings had suffi- estimate of the total circulation and centroid of vor-
cient spatial resolution to measure the vorticity flux ticity throughout the entire test run in spite of the
across the relatively small shear layer, measurements wake leaving the frame. Figure 22 also shows the
with two different fields of view (FOV) were taken alternating LEV and TEV shedding that occurs.
of several different repetitions of the same case. The The frozen-wake approximation is implemented
tight field of view was processed with a 24-pixel win- by adding a small strip of vorticity at the exit of the
dow and 50% overlap, resulting in a vector spacing of PIV frame onto a stationary background grid, which
0.84% of chord. The wide field of view was processed maintains a running average of the contribution from
with a 32-pixel window and 50% overlap, resulting all frames. Obviously, this approximation is far from
in a vector spacing of 1.90% chord. Thus the tight perfect. The vorticity in the wake that would nor-
field of view has about half the vector spacing as mally convect does not, leading to erroneous loca-
the wide field of view (the chord is 76.2 mm for both tions of the shed vortices, and any vorticity that
cases). leaves through the top and bottom of the frame is ne-
The LEV circulation was directly computed as glected. Furthermore, vortices that slowly leave the
the area sum of the positive vorticity over the en- frame become stretched in the fixed-wake represen-
tire PIV frame. In order to still maintain a dis- tation. Despite the disadvantages, making the fixed-
tinction between the LEV and TEV, the vorticity wake approximation is a vast improvement over sim-
was split into positive and negative regions and inte- ply neglecting the wake that has left the frame.
grated separately. This required the use of a thresh- The production of circulation at the leading edge
old to remove the background noise. To ensure that was measured via the flux of vorticity out of a control
a proper threshold was used, a sensitivity study was volume around the leading edge of the wing. The ex-
conducted that that showed that if a large enough tent of the control volume is depicted in fig. 23, and
threshold was used, then the measured quantities is 0.08c on a side centered at the upper corner of the
were nearly independent of the threshold. A thresh- leading edge. Along the edge of the box, the vortic-
old value of normalized vorticity ωthresh = ωc/Uf = ity flux was computed with the standard integral
2.5 was selected as a conservative result from the sen- Z
sitivity study, and was 5 standard deviations above dΓ
= ω~u · n̂ ds (A.1)
the background noise level for all cases. dt s
 27

2
1.8 Wide FOV Tight FOV
1.6
1.4
1.2
(dΓ/dt)/Uf2

1
0.8
0.6
0.4
0.2
0
-0.2
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
t*

FIG. 24: Flux measurements for the two different fields of view. Note that each field of view is represented
by five independent trials.

where dΓ/ dt is the vorticity flux (expressed as the to add noise to the measurements with the wide
rate of circulation), ω is the vorticity, and ~u · n̂ is field of view, and secondarily to impart a slightly
the velocity component normal to the boundary, s. higher magnitude for the wide FOV. The wide field
The vorticity is computed with a three-point cen- of view also has a slight delay in the start of the mea-
tral differencing scheme in each direction. The inte- surement compared to the tight field of view. This
gral is discretized with the PIV vector spacing and was due to the measurement plane being displaced
computed numerically with the trapezoid rule. This slightly further behind the leading edge in the wide
method measures the circulation flux slightly behind field of view to ensure that it captured the whole
the leading edge. Doing so was a conscious choice shear layer. The wide field of view also reports a
to avoid using vectors whose correlation might be slightly higher initial peak in the flux measurement.
contaminated with stationary wing pixels. Overall, the two methods report the same trends
and magnitude of circulation production. The fac-
The impact of the two different fields of view on tor of two difference in their resolution verifies that
the flux measurement is shown in fig. 24. The effect the shear layer is sufficiently resolved to capture the
of halving the measurement resolution is primarily true value of circulation production.

[1] R. L. Bisplinghoff, H. Ashley, and R. L. Halfman, ditions,” Journal of Aircraft, vol. 47, no. 1, pp. 80–
Aeroelasticity. Dover Corporation, 1996. 86, 2010.
[2] J. D. Eldredge and A. R. Jones, “Leading-edge vor- [8] T. Yilmaz, M. Ol, and D. Rockwell, “Scaling of
tices: Mechanics and modeling,” Annual Review of flow separation on a pitching low aspect ratio
Fluid Mechanics, vol. 51, no. 1, pp. 75–104, 2019. plate,” Journal of Fluids and Structures, vol. 26,
[3] D. J. Pines and F. Bohorquez, “Challenges facing no. 6, pp. 1034–1041, 2010.
future micro-air-vehicle development,” Journal of [9] K. O. Granlund, M. V. Ol, and L. P. Bernal, “Un-
Aircraft, vol. 43, no. 2, pp. 290–305, 2006. steady pitching flat plates,” Journal of Fluid Me-
[4] J. McMasters and M. Henderson, “Low-speed chanics, vol. 733, p. R5, 2013.
single-element airfoil synthesis,” Technical Soaring, [10] J. D. Eldredge, “Numerical simulation of the fluid
vol. 6, no. 2, pp. 1–21, 1980. dynamics of 2d rigid body motion with the vor-
[5] P. Lissaman, “Low-Reynolds-number airfoils,” An- tex particle method,” Journal of Computational
nual Review of Fluid Mechanics, vol. 15, no. 1, Physics, vol. 221, no. 2, pp. 626–648, 2007.
pp. 223–239, 1983. [11] M. Ol, J. Eldredge, and C. Wang, “High-amplitude
[6] M. H. Dickinson and K. G. Gotz, “Unsteady pitch of a flat plate: an abstraction of perching
aerodynamic performance of model wings at low and flapping,” International Journal of Micro Air
Reynolds numbers,” Journal of Experimental Bi- Vehicles, vol. 1, September 2009.
ology, vol. 174, no. 1, pp. 45–64, 1993. [12] M. R. Visbal, R. E. Gordnier, and M. C. Galbraith,
[7] D. Rival and C. Tropea, “Characteristics of pitch- “High-fidelity simulations of moving and flexible
ing and plunging airfoils under dynamic-stall con- airfoils at low Reynolds numbers,” Experiments in
 28

Fluids, vol. 46, no. 5, pp. 903–922, 2009. two-dimensional plunging and flapping plates with
[13] D. J. Garmann and M. R. Visbal, “Numerical in- varying sweep,” Journal of Fluids and Structures,
vestigation of transitional flow over a rapidly pitch- vol. 43, pp. 231–243, 2013.
ing plate,” Physics of Fluids, vol. 23, no. 9, 2011. [30] A. Widmann and C. Tropea, “Parameters influ-
[14] A. Jones, C. Pitt Ford, and H. Babinsky, “Three- encing vortex growth and detachment on unsteady
dimensional effects on sliding and waving wings,” aerodynamic profiles,” Journal of Fluid Mechanics,
Journal of Aircraft, vol. 48, no. 2, pp. 633–644, vol. 773, pp. 432–459, 2015.
2011. [31] A. E. Panah, J. M. Akkala, and J. H. Buchholz,
[15] S. K. Venkata and A. R. Jones, “Leading-edge vor- “Vorticity transport and the leading-edge vortex of
tex structure over multiple revolutions of a rotating a plunging airfoil,” Experiments in Fluids, vol. 56,
wing,” Journal of Aircraft, vol. 50, no. 4, pp. 1312– no. 8, pp. 1–15, 2015.
1316, 2013. [32] M. Lighthill, Laminar Boundary Layers. Oxford
[16] F. Manar, A. Medina, and A. R. Jones, “Tip vor- University Press, 1963.
tex structure and aerodynamic loading on rotating [33] H. Wagner, “Über die Entstehung des dynamis-
wings in confined spaces,” Experiments in Fluids, chen Auftriebes von Tragflügeln,” ZAMM-Journal
vol. 55, no. 9, pp. 1–18, 2014. of Applied Mathematics and Mechanics/Zeitschrift
[17] M. Bross and D. Rockwell, “Flow structure on a si- für Angewandte Mathematik und Mechanik, vol. 5,
multaneously pitching and rotating wing,” Journal no. 1, pp. 17–35, 1925.
of Fluid Mechanics, vol. 756, pp. 354–383, 2014. [34] H. Glauert, The force and moment on an oscillat-
[18] Z. R. Carr, A. C. DeVoria, and M. J. Ringuette, ing aerofoil. HM Stationery Office, 1929.
“Aspect-ratio effects on rotating wings: circulation [35] T. Theodorsen, “General theory of aerodynamic
and forces,” Journal of Fluid Mechanics, vol. 767, instability and the mechanism of flutter,” NACA
pp. 497–525, 2015. Report, no. 496, 1935.
[19] D. J. Garmann, M. R. Visbal, and P. D. Ork- [36] D. Pullin and Z. J. Wang, “Unsteady forces on
wis, “Three-dimensional flow structure and aero- an accelerating plate and application to hovering
dynamic loading on a revolving wing,” Physics of insect flight,” Journal of Fluid Mechanics, vol. 509,
Fluids, vol. 25, no. 3, p. 034101, 2013. pp. 1–21, 2004.
[20] P. Hammer, A. Altman, and F. Eastep, “Valida- [37] M. N. J. Moore, “Torsional spring is the optimal
tion of a discrete vortex method for low Reynolds flexibility arrangement for thrust production of a
number unsteady flows,” AIAA Journal, 2014. flapping wing,” Physics of Fluids, vol. 27, no. 9,
[21] C. Wang and J. D. Eldredge, “Low-order phe- 2015.
nomenological modeling of leading-edge vortex for- [38] E. C. Polhamus, “Predictions of vortex-lift charac-
mation,” Theoretical and Computational Fluid Dy- teristics by a leading-edge suction analogy,” Jour-
namics, vol. 27, no. 5, pp. 577–598, 2013. nal of aircraft, vol. 8, no. 4, pp. 193–199, 1971.
[22] G. J. Leishman, Principles of Helicopter Aerody- [39] Y. S. Baik, L. P. Bernal, K. Granlund, and M. V.
namics. Cambridge University Press, 2006. Ol, “Unsteady force generation and vortex dynam-
[23] C. Ellington, “The aerodynamics of hovering insect ics of pitching and plunging aerofoils,” Journal of
flight. V. a vortex theory,” Philosophical Transac- Fluid Mechanics, vol. 709, pp. 37–68, 2012.
tions of the Royal Society B: Biological Sciences, [40] M. A. Jones, “The separated flow of an inviscid
vol. 305, no. 1122, pp. 115–144, 1984. fluid around a moving flat plate,” Journal of Fluid
[24] C. Pitt Ford and H. Babinsky, “Lift and the Mechanics, vol. 496, pp. 405–441, 2003.
leading-edge vortex,” Journal of Fluid Mechanics, [41] S. Ansari, R. Żbikowski, and K. Knowles, “Non-
vol. 720, pp. 280–313, 2013. linear unsteady aerodynamic model for insect-like
[25] C. J. Wojcik and J. H. Buchholz, “Vorticity flapping wings in the hover. Part 1: Methodology
transport in the leading-edge vortex on a rotat- and analysis,” Proceedings of the Institution of Me-
ing blade,” Journal of Fluid Mechanics, vol. 743, chanical Engineers, Part G: Journal of Aerospace
pp. 249–261, 2014. Engineering, vol. 220, no. 2, pp. 61–83, 2006.
[26] J. H. Buchholz, M. A. Green, and A. J. Smits, [42] J. Katz, “A discrete vortex method for the non-
“Scaling the circulation shed by a pitching panel,” steady separated flow over an airfoil,” Journal of
Journal of Fluid Mechanics, vol. 688, pp. 591–601, Fluid Mechanics, vol. 102, pp. 315–328, 1981.
2011. [43] R. K. Shukla and J. D. Eldredge, “An inviscid
[27] P. Sattari, D. E. Rival, R. J. Martinuzzi, and model for vortex shedding from a deforming body,”
C. Tropea, “Growth and separation of a start- Theoretical and Computational Fluid Dynamics,
up vortex from a two-dimensional shear layer,” vol. 21, no. 5, pp. 343–368, 2007.
Physics of Fluids, vol. 24, no. 10, p. 107102, 2012. [44] K. Ramesh, A. Gopalarathnam, J. R. Edwards,
[28] J. Kriegseis, M. Kinzel, and D. E. Rival, “On the M. V. Ol, and K. Granlund, “An unsteady air-
persistence of memory: Do initial conditions im- foil theory applied to pitching motions validated
pact vortex formation?,” Journal of Fluid Mechan- against experiment and computation,” Theoretical
ics, vol. 736, pp. 91–106, 2013. and Computational Fluid Dynamics, vol. 27, no. 6,
[29] J. G. Wong, J. Kriegseis, and D. E. Rival, “An in- pp. 843–864, 2013.
vestigation into vortex growth and stabilization for [45] S. P. Sane and M. H. Dickinson, “The aerodynamic
 29

effects of wing rotation and a revised quasi-steady versity Press, 2000.

model of flapping flight,” Journal of Experimental [64] P. G. Saffman, Vortex Dynamics. Cambridge Uni-
Biology, vol. 205, no. 8, pp. 1087–1096, 2002. versity Press, 1992.
[46] H. E. Taha, M. R. Hajj, and P. S. Beran, “State- [65] A. J. Chorin, “Numerical study of slightly viscous
space representation of the unsteady aerodynamics flow,” Journal of Fluid Mechanics, vol. 57, no. 4,
of flapping flight,” Aerospace Science and Technol- pp. 785–796, 1973.
ogy, vol. 34, pp. 1–11, 2014. [66] A. J. Chorin and P. S. Bernard, “Discretization of a
[47] H. Babinsky, P. Stevens, A. R. Jones, L. P. Bernal, vortex sheet, with an example of roll-up,” Journal
and M. V. Ol, “Low order modeling of lift forces of Computational Physics, vol. 13, no. 3, pp. 423–
for unsteady pitching and surging wings,” in 54th 429, 1973.
AIAA Aerospace Sciences Meeting, 2016. [67] R. Krasny, “Desingularization of periodic vortex
[48] F. M. Bos, D. Lentink, B. Van Oudheusden, and sheet roll-up,” Journal of Computational Physics,
H. Bijl, “Influence of wing kinematics on aerody- vol. 65, no. 2, pp. 292–313, 1986.
namic performance in hovering insect flight,” Jour- [68] G. H. Vatistas, V. Kozel, and W. Mih, “A simpler
nal of Fluid Mechanics, vol. 594, pp. 341–368, 2008. model for concentrated vortices,” Experiments in
[49] P. Trizila, C.-K. Kang, H. Aono, W. Shyy, and Fluids, vol. 11, no. 1, pp. 73–76, 1991.
M. Visbal, “Low-Reynolds-number aerodynamics [69] N. J. Zabusky, M. Hughes, and K. Roberts, “Con-
of a flapping rigid flat plate,” AIAA Journal, tour dynamics for the Euler equations in two
vol. 49, no. 4, pp. 806–823, 2011. dimensions,” Journal of Computational Physics,
[50] J. Lankford, D. Mayo, and I. Chopra, “Computa- vol. 30, no. 1, pp. 96–106, 1979.
tional investigation of insect-based flapping wings [70] M. J. Stock, W. J. Dahm, and G. Tryggvason, “Im-
for micro air vehicle applications,” International pact of a vortex ring on a density interface using a
Journal of Micro Air Vehicles, 2016. regularized inviscid vortex sheet method,” Jour-
[51] J. D. Anderson Jr, Fundamentals of Aerodynamics. nal of Computational Physics, vol. 227, no. 21,
McGraw-Hill Education, 2010. pp. 9021–9043, 2008.
[52] J. Katz and A. Plotkin, Low-speed aerodynamics. [71] B. A. Roccia, S. Preidikman, J. C. Massa, and
Cambridge University Press, 2nd ed., 2001. D. T. Mook, “Modified unsteady vortex-lattice
[53] R. L. Panton, Incompressible Flow. John Wiley & method to study flapping wings in hover flight,”
Sons, 3rd ed., 2006. AIAA Journal, vol. 51, no. 11, pp. 2628–2642,
[54] L. M. Milne-Thomson, Theoretical Hydrodynamics. 2013.
Dover Publications, fifth ed., 1968. [72] M. Stock, “Summary of vortex method literature.”
[55] G. K. Batchelor, An Introduction to Fluid Dynam- via http://markjstock.org/research/, downloaded
ics. Cambridge University Press, 1967. 12/24/2017, 2006.
[56] X. Xia and K. Mohseni, “Lift evaluation of a two- [73] G. Winckelmans and A. Leonard, “Contributions
dimensional pitching flat plate,” Physics of Fluids, to vortex particle methods for the computation of
vol. 25, no. 9, p. 091901, 2013. three-dimensional incompressible unsteady flows,”
[57] S. Ansari, R. Żbikowski, and K. Knowles, “Non- Journal of Computational Physics, vol. 109, no. 2,
linear unsteady aerodynamic model for insect-like pp. 247–273, 1993.
flapping wings in the hover. Part 2: Implemen- [74] S. Shankar and L. Van Dommelen, “A new dif-
tation and validation,” Proceedings of the Institu- fusion procedure for vortex methods,” Journal of
tion of Mechanical Engineers, Part G: Journal of Computational Physics, vol. 127, no. 1, pp. 88–109,
Aerospace Engineering, vol. 220, no. 3, pp. 169– 1996.
186, 2006. [75] M. Kirchhart and S. Obi, “A splitting-free vorticity
[58] F. Minotti, “Unsteady two-dimensional theory of redistribution method,” Journal of Computational
a flapping wing,” Physical Review E, vol. 66, no. 5, Physics, vol. 330, pp. 282–295, 2017.
2002. [76] M. Drela, “XFOIL: An analysis and design system
[59] H. Glauert, The Elements of Aerofoil and Airscrew for low Reynolds number airfoils,” in Low Reynolds
Theory. Cambridge University Press, 2nd ed., number aerodynamics, pp. 1–12, Springer, 1989.
1947. [77] T. Sarpkaya, “Computational methods with vor-
[60] M. A. Jones and M. J. Shelley, “Falling cards,” tices,” ASME, Transactions, Journal of Fluids En-
Journal of Fluid Mechanics, vol. 540, pp. 393–425, gineering, vol. 111, pp. 5–52, 1989.
2005. [78] R. Krasny, “Vortex sheet computations: roll-up,
[61] J. L. Hess and A. Smith, “Calculation of non-lifting wakes, separation,” Lectures in Applied Mathemat-
potential flow about arbitrary three-dimensional ics, vol. 28, no. 1, pp. 385–401, 1991.
bodies,” tech. rep., Douglas Aircraft Co Long [79] R. Krasny and M. Nitsche, “The onset of chaos
Beach CA, 1962. in vortex sheet flow,” Journal of Fluid Mechanics,
[62] J. Hess, “Panel methods in computational fluid vol. 454, pp. 47–69, 2002.
dynamics,” Annual Review of Fluid Mechanics, [80] X. Xia and K. Mohseni, “Unsteady aerodynamics
vol. 22, no. 1, pp. 255–274, 1990. and vortex-sheet formation of a two-dimensional
[63] G.-H. Cottet and P. D. Koumoutsakos, Vortex airfoil,” Journal of Fluid Mechanics, vol. 830,
Methods: Theory and Practice. Cambridge Uni- pp. 439–478, 2017.
 30

[81] J. G. Leishman and T. Beddoes, “A semi-empirical [94] K. C. Hall, S. A. Pigott, and S. R. Hall, “Power
model for dynamic stall,” Journal of the American requirements for large-amplitude flapping flight,”
Helicopter society, vol. 34, no. 3, pp. 3–17, 1989. Journal of Aircraft, vol. 35, no. 3, pp. 352–361,
[82] J. O. Dabiri, “Optimal vortex formation as a unify- 1998.
ing principle in biological propulsion,” Annual re- [95] H. Salehipour and D. J. Willis, “A coupled
view of fluid mechanics, vol. 41, pp. 17–33, 2009. kinematics–energetics model for predicting energy
[83] I. H. Abbott and A. E. Von Doenhoff, Theory of efficient flapping flight,” Journal of Theoretical Bi-
Wing Sections, Including a Summary of Airfoil ology, vol. 318, pp. 173–196, 2013.
Data. Courier Corporation, 1959. [96] D. Pullin, “The large-scale structure of unsteady
[84] C. Brennen, “A review of added mass and fluid self-similar rolled-up vortex sheets,” Journal of
inertial forces.,” tech. rep., Naval Civil Engineering Fluid Mechanics, vol. 88, no. 3, pp. 401–430, 1978.
Laboratory, 1982. [97] L. Cortelezzi and A. Leonard, “Point vortex model
[85] G. J. Berman and Z. J. Wang, “Energy-minimizing of the unsteady separated flow past a semi-infinite
kinematics in hovering insect flight,” Journal of plate with transverse motion,” Fluid Dynamics Re-
Fluid Mechanics, vol. 582, pp. 153–168, 2007. search, vol. 11, no. 6, p. 263, 1993.
[86] F. Manar, P. Mancini, D. Mayo, and A. R. [98] C. Brown and W. Michael, “Effect of leading-edge
Jones, “Comparison of rotating and translating separation on the lift of a delta wing,” Journal of
wings: force production and vortex characteris- the Aeronautical Sciences, vol. 21, no. 10, 1954.
tics,” AIAA Journal, 2016. [99] D. Darakananda, J. Eldredge, T. Colonius, and
[87] T. VonKarman and W. Sears, “Airfoil theory for D. R. Williams, “A vortex sheet/point vortex dy-
non-uniform motion,” Journal of the Aeronautical namical model for unsteady separated flows,” in
Sciences, vol. 5, no. 10, pp. 379–390, 1938. 54th AIAA Aerospace Sciences Meeting, pp. 4–8,
[88] J. M. Greenberg, “Airfoil in sinusoidal motion in a 2016.
pulsating stream,” Tech. Rep. 1326, NACA, June [100] L. Greengard, “The FMMLIB2D suite.”
1947. https://cims.nyu.edu/cmcl/fmm2dlib/fmm2dlib.html.
[89] J. Leishman and K. Nguyen, “State-space rep- [101] J. Wu, “Theory for aerodynamic force and moment
resentation of unsteady airfoil behavior.,” AIAA in viscous flows,” AIAA Journal, vol. 19, no. 4,
Journal, vol. 28, no. 5, pp. 836–844, 1990. pp. 432–441, 1981.
[90] H. G. Küssner, “Stresses produced in airplane [102] C. Chabalko, R. D. Snyder, P. S. Beran, and
wings by gusts,” NACA-TM-654, 1932. G. Parker, “The physics of an optimized flapping
[91] W. R. Sears, “Some aspects of non-stationary air- wing micro air vehicle,” in 47th AIAA Aerospace
foil theory and its practical application,” Journal Sciences Meeting, pp. 2009–801, 2009.
of the Aeronautical Sciences, vol. 8, no. 3, 1941. [103] F. Manar and A. R. Jones, “Transient response of
[92] J. Leishman, “Validation of approximate indicial a single degree-of-freedom wing at high angle-of-
aerodynamic functions for two-dimensional sub- attack,” AIAA Journal, pp. 1–12, 2017.
sonic flow,” Journal of Aircraft, vol. 25, no. 10, [104] L. Graftieaux, M. Michard, and N. Grosjean,
pp. 914–922, 1988. “Combining PIV, POD and vortex identification
[93] K. C. Hall and S. R. Hall, “Minimum induced algorithms for the study of unsteady turbulent
power requirements for flapping flight,” Journal of swirling flows,” Measurement Science and Technol-
Fluid Mechanics, vol. 323, pp. 285–315, 1996. ogy, vol. 12, no. 9, p. 1422, 2001.