Sie sind auf Seite 1von 28

AIAA 2016-4338

AIAA Aviation
13-17 June 2016, Washington, D.C.
46th AIAA Fluid Dynamics Conference

Toward Simulating a Pitch-Up Maneuver for a Small


Unmanned Aerial Vehicle
Joshua Blake∗, David Thompson†
Mississippi State University, Mississippi State, MS, 39762, USA

The heaving of a flat plate was investigated in order to determine computational require-

ments for simulating a pitch-up maneuver for a Small Unmanned Aerial Vehicle (SUAS). A
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

second-order, unstructured finite volume code was employed in conjunction with a Hybrid

RANS/LES turbulence model to study the flowfield structures. Results predicted using

the second-order method compared favorably to simulations by Visbal that employed a

higher-order, structured-grid LES method. Large-scale flow structures and phase-averaged

aerodynamic forces were accurately predicted. Greater spatial and temporal resolution is

needed to resolve smaller scale structures. The results shown here demonstrate that the

combined use of a second-order, unstructured method and a Hybrid RANS/LES turbu-

lence model is an appropriate strategy for simulating an SUAS vehicle undergoing a high

angle-of-attack maneuver.

I. Introduction

A. Background

The utility of SUAS (Small Unmanned Aerial Systems) has led to the development of many uniquely-shaped

vehicles. With a variety of arrangements of sensors, control surfaces, and propeller placements, each SUAS

has a unique and complex aerodynamic flowfield. Due to their small size, these unmanned vehicles operate at

lower Reynolds numbers than traditional, full size aircraft. These low Reynolds number flows are dominated

by several unsteady flow features, such as flow separation and complex vortex structures. The use of

non-streamlined body shapes only increases the likelihood of these flow features. In addition, many SUAS

are highly maneuverable, and many of these maneuvers produce highly non-linear and complex flowfields.

Yet still, the influence of proportionally-large propellers on the aerodynamics of the body adds another

element to the complex flowfield. Understanding the features of low Reynolds number flowfields is important

to the advancement of the current understanding of the aerodynamics of SUAS vehicles. The use of CFD
∗ Graduate Research Assistant, Center for Advanced Vehicular Systems, PO Box 5405, Student Member
† Airbus Helicopters Professor of Aerospace Engineering, Department of Aerospace Engineering, PO Box A, Associate Fellow.

1 of 28

American Institute of Aeronautics and Astronautics


Copyright © 2016 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
(Computational Fluid Dynamics) to simulate SUAS manuevers allows for detailed transient analysis of these

unsteady flow features, which will lead to improved understanding.

B. Objective

The objective of the research presented in this paper is to determine the grid and computational requirements

for simulating a pitch-up maneuver for a notional “pizza-box” SUAS. This first requires the selection of

an appropriate simulation strategy and its validation for a simplified, similar problem. The manuever of a

heaving flat plate was chosen as a representative problem to test the chosen simulation strategy. Results were

compared to simulation results from Visbal1 in order to develop confidence in the chosen methods.
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

C. Outline of Paper

Following the introduction, the paper proceeds with a short literature review. A method section details the

numerical approach taken to solve this problem. Results include the study of a flat plate at angle of attack

and a flat plate at angle of attack undergoing a heaving motion. Conclusions from the simulations and future

work are presented following the results.

II. Literature Review

To provide a proper understanding of the physics involved during a dynamic stall of a low aspect ratio flat

plate and a description of computational methods used to study this type of problem, this section includes a

discussion of low Reynolds number aerodynamics, flat plate simulations and dynamic stall, and turbulence

modeling.

A. Low Reynolds Number Aerodynamics

A typical SUAS vehicle operates in the low Reynolds number (below 1×105 ) flight regime.2 The characteristic

aerodynamics of low Reynolds number flows include a mixture of regions of attached and separated laminar

flow, transitional and turbulent mixing flow, and vortical flow.2 Taken together, these aerodynamic features

can lead to complex flow structures that are difficult to predict. However, Shyy2 reported on several

phenomena related to low aspect ratio SUAS vehicles and MAV’s (Micro Air Vehicles) that can be expected

in high angle-of-attack maneuvers at low Reynolds numbers.

One common flow feature is the laminar separation bubble, which can either lead to transition to turbulent

flow or simply to massively separated flow. The Reynolds number strongly affects this behavior, and Shyy2

suggests that a free-stream Reynolds number of 5×104 or higher is a general indicator that laminar separation

will transition to turbulent flow and reattach. Freestream turbulence intensity is also an important factor in

2 of 28

American Institute of Aeronautics and Astronautics


the transition of separated laminar flow.2

Vortex structures are another common feature. These structures are usually on the order of the vehicle size.

The interaction of wing-tip vortices and leading-edge vortices can often lead to unsteady forces and potential

vehicle stability issues.2 In flapping (or heaving) flight especially, the growth and shedding of a vortex (vortex

dynamics) is significant to the overall vehicle aerodynamics. Tip vortices can produce significant effects on

low aspect ratio wings, leading to reduced effective angle of attack and increased induced drag.3 In addition,

the low pressure region created by the comparatively large vortex core can provide additional lift2, 4 similar to

the manner in which lift is achieved on a delta wing at a high angle-of-attack and low speeds.2, 3 Leading-edge

vortices are common for maneuvers that lead to dynamic stall. These are vortices that form due to flow
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

separation at or near the leading edge, generally from a sharp leading edge, such as that found on a flat plate.

Leading-edge vortices often form at higher angles-of-attack and may enhance lift temporarily due to the

formation of a low-pressure core, especially during flapping flight.2

For smaller vehicles, one additional factor for consideration is gusts, which can be classified as freestream

flow unsteadiness. Gusts can change the effective Reynolds number and angle-of-attack of an SUAS vehicle,

thereby altering the forces and the flow structures significantly.

B. Flat Plate Simulations and Dynamic Stall

These expected flow structures, including leading-edge and wingtip vortices, have been confirmed by several

experiments and simulations that studied flat plate maneuvers. Experiments by Yilmaz and Rockwell5

investigated a flat plate with an aspect ratio of two at low Reynolds numbers undergoing a heaving motion

that initiated dynamic stall. Visbal1 studied this heaving motion with numerical simulations, which agreed

well with the experimental results from Yilmaz and Rockwell.5 Several low Reynolds number cases (1×103

to 2×104 ) were investigated. Visbal et al.6 combines results from these two papers and outlines a new

understanding of the evolution of flow structures for a plate undergoing dynamic stall.

In the simulations by Visbal,1 a leading-edge vortex structure was observed that evolved into an arch-vortex

structure that eventually convected downstream and was shed from the plate. This arch-vortex structure was

reported as newly-discovered.1, 6 Wingtip vortex formation and breakdown were also observed. Additionally,

small-scale transitional structures were reported, indicating that a turbulence model is appropriate for a

similar maneuver at these Reynolds numbers.

Visbal also studied a flat plate with an aspect ratio of two at low Reynolds numbers (1×103 to 4×104 )

undergoing canonical pitch-up and perching maneuvers.7 Pitch-up maneuvers revealed the same dynamic stall

behavior as the heaving simulations,1 including the arch-vortex, although the structure was not as compact.

The arch-vortex remained fairly stationary with an increased residence time and lead to a swirling flow

3 of 28

American Institute of Aeronautics and Astronautics


pattern and low pressure region on the plate. Perching maneuvers were performed at a Reynolds number of

1×104 . These maneuvers added freestream deceleration to the pitch-up motion, which is useful for simulating

a gust during a pitch-up maneuver. A leading-edge vortex system formed but collapsed with the freestream

deceleration. A swirl pattern was seen in one case. A trailing-edge vortex was seen near the end of the

maneuver. For both heaving plates and pitching/perching plates, Visbal showed that the arch-vortex and

swirl pattern are the main flow structures for a dynamic stall process for a low aspect ratio wing.

As part of the NATO Task Group AVT-202 investigation, Practical Extensions of MAV (Micro Air

Vehicle) Unsteady Aerodynamics, several experiments and computational efforts8–11 investigated a suite of

canonical flat plate maneuvers. The papers focused on high angle-of-attack flows for flat plates with low aspect
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

ratios (two to four). Pitching and surging (constant acceleration from rest) maneuvers were investigated. The

purpose for these studies was to investigate the time-scale evolution of leading-edge, trailing-edge, and wingtip

vortices since they are not well understood for unsteady pitching motions10 and to develop reduced-order

aerodynamic models. Lower Reynolds numbers (2×104 or less) were investigated. Vortex breakup was seen

for the accelerating (surging) flat plate cases.9 Leading-edge and trailing-edge vortices were shown to convect

downstream at differing speeds.9 One study by Bernal10 showed that the leading-edge vortex and trailing-edge

vortex merged downstream and a new leading-edge vortex was formed. Some arching of the leading-edge

vortex was observed for pitching motion by Jones et al.11

C. Turbulence Modeling

Flow separation can lead to a transition to turbulence even at lower Reynolds numbers. Therefore, any

simulation of low Reynolds number flows that also contains large regions of separated flow, such as a dynamic

stall maneuver, should consider modeling turbulent structures that the computational grid cannot resolve.

The two most common strategies for modeling turbulence are the Reynolds-Averaged Navier-Stokes

(RANS) and the Large Eddy Simulation (LES) approaches. RANS usually damps flow unsteadiness due

to larger values of the eddy viscosity12 but can accurately capture the effects of isotropic turbulence since

it is computing the mean flow. RANS models all flow scales. LES is often too expensive in the boundary

layer since it generally requires isotropic grids to be effective.13 LES methods solve a filtered version of the

Navier-Stokes equations rather than a mean flow equation and model only the smaller turbulent scales that

cannot be resolved by the grid. As a stand-alone model, neither of these is entirely appropriate for simulating

a dynamic stall problem for a realistic SUAS geometry.

Despite the usual expense associated with LES models, two previously mentioned papers by Visbal1, 7 and

a paper by Visbal et al.6 use an Implicit LES (ILES) method to simulate a flat plate undergoing dynamic stall.

This method does not use the typical SGS (sub-grid stress) model for modeling small unresolved turbulence

4 of 28

American Institute of Aeronautics and Astronautics


scales, but rather employs an eighth-order, low-pass filtering method that eliminates higher frequencies that

are not well resolved. The filter is applied to the conserved dependent variables, not to the Navier-Stokes

equations. The advantage of this model is that it can handle laminar, transitional, and turbulent flows.

Visbal1, 7 and Visbal et al.6 use this method in conjunction with a sixth-order Navier-Stokes solver on

structured grids. This method successfully matches the experimental data and resolves the small scales

well. However, it is apparent that the computational cost is significant, since a small timestep is required to

resolve the temporal turbulent scales. Additionally, the implementation of the ILES model in a higher-order

structured code likely limits this model’s applicability to simple geometries for which a structured grid might

easily be obtained. Therefore, it is unlikely that this scheme can be used on realistic, complex geometries,
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

which typically require unstructured grids.

The use of a Hybrid RANS/LES model12 has several advantages due to its combination of the best aspects

of RANS and LES models. In general, the smaller scales of turbulence are resolved with the isotropic SGS

model of the LES method, and RANS is only used to solve for the mean flowfield in the boundary layer.

This is advantageous because it removes the eddy viscosity damping of RANS12 that can occur outside the

boundary layer and removes the requirement for isotropic grids in the boundary layer for LES. Further,

it allows for the possibility of conserving computational resources by using RANS closer to the body to

calculate the small-scale energy cascade and dissipation of smaller turbulent scales with anisotropic grids

while using the Navier-Stokes equations to solve for larger turbulent eddies. The hybrid model generally is

more advantageous for higher Reynolds numbers12 where the range of turbulence scales is larger.14

Several examples of Hybrid RANS/LES models have been successfully implemented. Spalart et al.15

introduced the Detached-Eddy Simulation (DES) model, a widely-used model that switches between RANS

and LES based on the distance to a wall. Several modifications to the original DES method have been proposed

to eliminate observed shortcomings.16, 17 Another hybrid model is the multiscale (MS) model of Nichols and

Nelson.12 This model transitions from RANS to LES based on the comparison of a turbulent length scale to

the local grid scale. Nichols and Nelson performed simulations at Re = 8.0×106 and 1.95×106 with the MS

model.12 The results clearly demonstrate that the hybrid scheme is better at predicting the smaller scale

flow structures than a stand-alone RANS method since the hybrid model reduces the eddy viscosity that

usually damps out unsteady fluctuations. Walters et al.18 propose a Dynamic Hybrid RANS/LES model that

transitions from RANS to LES based on turbulence production. This method also shows better resolution of

small-scale structures in some problems than even other hybrid methods. Alam et al. modified the DHRL

model to make it sensitive to transition to turbulent flow.19

5 of 28

American Institute of Aeronautics and Astronautics


III. Methods

The methods used to simulate a heaving flat plate are described in this section.

A. Loci/CHEM

For this study, Navier-Stokes simulations were performed using the Computational Fluid Dynamics (CFD)

code Loci/CHEM.20, 21 Loci/CHEM was implemented in Loci, a framework for developing multi-physics

simulation tools,22, 23 as a full-featured flow solver that has been extensively verified and validated.24–26

This cell-centered, finite volume code employs a second-order, upwind spatial discretization to solve the

Navier-Stokes equations on unstructured, generalized grids.27 The Beam and Warming28 implicit time-
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

integration scheme is used, and a solution is obtained through a combination of Newton and linear Symmetric

Gauss-Seidel (SGS)22 iterative methods. The flow solver uses pre-conditioning for low Mach number flows and

can be extended to high Mach number flows through the incorporation of models for finite-rate chemistry.21

The code employs over-set techniques for grid motion.

B. Hybrid RANS/LES Turbulence Model

From the discussion of Hybrid RANS/LES models in the Literature Review section, it is clear that a hybrid

model has several advantages over pure RANS or LES models, including the possibility of reducing grid

resolution requirements usually imposed by LES schemes. The flexible framework of Loci/CHEM allows

for easy implementation of different turbulence models, including the multiscale (MS) Hybrid RANS/LES

model of Nichols and Nelson.12 It was desirable to test the Nichols-Nelson model in the low Reynolds number

regime and evaluate its effectiveness.

The Nichols-Nelson model can incorporate different RANS models.12 In the simulations performed with

Loci/CHEM, the Nichols-Nelson model was implemented with Menter’s SST (Shear Stress Transport) RANS

model,29, 30 which is Menter’s BSL30 (baseline) model modified to include the transport of the principal

shear stress.31 It is a two-equation eddy-viscosity model that uses Wilcox’s k − ω model32 near the wall and

transitions to the k −  model away from the wall.

Loci/CHEM implements the Nichols-Nelson multiscale model with one modification. Equation 1 shows

the definition of the turbulent length scale (lT ), a quantity associated with two-equation RANS models, given

by Nichols-Nelson.12 The definition used in Loci/CHEM is given in Equation 2, which is the definition used

by Wilcox in the 1998 k − ω model.14 In the equations, k is turbulent kinetic energy,  is the dissipation, and

ω is the specific dissipation rate.

3/2
lT = kRAN S /RAN S (1)

6 of 28

American Institute of Aeronautics and Astronautics


3/2
lT = β ∗ kRAN S /RAN S = k 1/2 /ω (2)

IV. Results

Results from the numerical simulations performed in this study are presented below, including a comparison

to a simulation from Visbal1 of a flat plate in a freestream flow, comparisons to a simulation from Visbal1 of

a heaving flat plate, and a grid resolution study for the heaving flat plate simulations. Finally, limitations of

the employed simulation strategy are presented.


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

A. Flat Plate Simulation Setup

Simulations by Visbal1 investigate a flat plate undergoing a heaving motion that initiates a dynamic stall.

Visbal’s simulations were performed at several Reynolds numbers, with both a flat plate at 8◦ angle of

attack (α) and a flat plate at α = 8◦ undergoing a vertical heaving motion.1 These results agreed well with

experimental results from Yilmaz and Rockwell.5, 6 Visbal presents a thorough understanding of the evolution

of flow structures for a plate undergoing dynamic stall and shows a newly-discovered arch-vortex structure.1

Therefore, in the present study, simulations were performed and compared with the computational results

from Visbal1 in order to determine if the numerical methodology used in this paper correctly predicts the

same flowfield features for the flat plate.

The flat plate was rectangular, with an aspect ratio (span/chord) of two and a uniform thickness

(thickness/chord = 0.016). The edges of the plate were sharp and not rounded. Pointwise33 was used to

generate a grid comprised of mostly isotropic tetrahedra. Anisotropic triangular prisms were used in the

viscous boundary layer region near the plate to save on the grid’s cell count near the wall. The wake and

areas below and above the plate were further resolved by the use of an additional refinement box. A cut

down the mid-plane of the rotated grid is shown in Figure 1, with the plate highlighted in blue. The grid

used for the heaving simulations had 93 million cells and is the referred to as the fine grid. The boundary

conditions for the grid were as follows: inlet plane and sides were farfield (pressure, temperature, and velocity

specified), the plate surface was a viscous wall (no-slip condition), and the outlet plane was an outlet (pressure

specified). The simulation did not make use of a symmetry plane as was done in the simulations by Visbal.1

To simulate the heaving motion, the grid was translated vertically. However, the oncoming freestream velocity

was achieved by imposing a uniform flow condition at the farfield boundary and not by translating the grid.

7 of 28

American Institute of Aeronautics and Astronautics


(a) Overview of grid (b) Zooming in around the flat plate

Figure 1: Computational grid cut at the flat plate centerline.


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

B. Comparison with Results from Visbal1 for a Flat Plate at Re = 10,000

A transient simulation was performed for a flat plate at an angle of attack of 8◦ relative to the oncoming flow.

A freestream velocity was specified that produced a Reynolds number of 10,000 based on the chord. The

simulation was run until the lift force reached an oscillatory steady-state value. Instantaneous and mean flow

results are presented in this section. All variables have been non-dimensionalized by freestream conditions to

match the results from Visbal1 (pressure was normalized by ρu2∞ ). This case will be referred to hereafter as

Simulation A. As a note, the grid for this case is the medium grid (56 million cells). Since the simulation

produced acceptable results, the simulation was not repeated with the fine grid.

The average CL and CD for Simulation A are shown in Table 1. The table also includes values from

Visbal.1 The values for both lift and drag agree quite well and show that the simulations predict the mean

forces well.
Coefficient Simulation A results Visbal1 results % difference
CLmean 0.431 0.428 0.70
CDmean 0.094 0.095 1.05

Table 1: Comparison of average lift and drag coefficient values.

Figure 2 (a) shows an isosurface of vorticity from the top rear of the plate colored by Cp. Figure 2 (b)

shows the corresponding image from Visbal.1 The images reveal the unsteady nature of the flow, and show

a large separation region and several distinct vortices as well as small-scale transitional structures at the

plate’s trailing edge.1 The results from Simulation A show coarser resolution of the flowfield than the results

from Visbal.1 This is expected because of inadequate grid resolution in Simulation A. Additionally, Visbal1

used a higher-order numerical method which is expected to provide better resolution of the flowfield than

second-order spatial schemes.

8 of 28

American Institute of Aeronautics and Astronautics


(a) (b)
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

Figure 2: Comparison of isosurfaces of instantaneous vorticity from Simulation A and from Visbal1 for the
flat plate.

Figure 3 (a) shows the time-averaged Cp values on the top surface of the flat plate in Simulation A, and

Figure 3 (b) shows the corresponding image from Visbal.1 The two images have similar mean flowfields on

the surface of the plate, although Simulation A predicts a slightly lower pressure than the Visbal simulation.

This could indicate that Simulation A predicts a slightly different flow structure. Other dissimilarities could

arise due to difficulties replicating the color map from Visbal.1 It is clear from this analysis that, although the

mean flowfields are similar, the flowfield in Simulation A is not as well resolved as in the Visbal1 simulation.

(a) Simulation A (b) Visbal1

Figure 3: Comparison of time-averaged Cp on the top of the flat plate from Simulation A and from Visbal.1

9 of 28

American Institute of Aeronautics and Astronautics


C. Comparison with Results from Visbal1 for a Heaving Flat Plate at Re = 10,000

Visbal1 studied a heaving flat plate at Re = 10,000 and the resulting induced dynamic stall flow structures.

This simulation is of note because of the newly-observed1 arch-structure seen during the dynamic stall

process and the observed breakdown and regrowth of the wingtip vortices. The heaving motion consisted

of a sinusoidal path, divided into eight evenly-spaced critical points. The height (non-dimensionalized by

chord) described as a function of the critical points as shown in Figure 4. The non-dimensional plunging

amplitude, ho , is 0.25 and the reduced frequency, k = π · f · chord/U∞ , is 1.0, which is fairly high as noted

by Visbal.1 The heaving motion puts the flat plate (initially at α = 8◦ ) into an effective angle of attack

range −18.6◦ ≤ αef f ≤ 34.6◦ . Equations to describe the motion, specified in the boundary conditions in the
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

Visbal1 simulations, are given below:

(a) Height vs. Phase1

0.2
Non-Dimensional Y Height

0.1

-0.1

-0.2

1 2 3 4 5 6 7 8
Step Number
(b) Height vs. Critical Step Number

Figure 4: Heaving motion

10 of 28

American Institute of Aeronautics and Astronautics


us = −ḣ · sin(α0 ), vs = ḣ · cos(α0 ), ws = 0 (3)

h(t) = h0 · sin(2 · k · F (t) · t) (4)

F (t) = 1 − e−at , a = 4.6/t0 , t0 = 0.25 (5)

In the heaving simulations, Visbal1 employed a multi-zoned stretched Cartesian grid with 30 million grid

points, although symmetry was assumed at center span of the plate. The numerical methods were sixth-order

schemes and used an Implicit LES (ILES) turbulence model with an eighth-order low pass spatial filter.

A small non-dimensional timestep of ∆t · U∞ /c = 2π × 10−5 was used, which equates to 50,000 timesteps
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

per plunging cycle. The heaving motion lasted for fourteen heaving cycles to guarantee time-asymptotic

simulations. The results from Visbal1 are used as a comparison to test the accuracy of the chosen numerical

methods.

In the simulations presented in this paper, the flat plate underwent the same heaving motion. The grid

was rotated 8◦ prior to the simulation and a freestream velocity was specified that produced a Reynolds

number of 10,000. After the initial start-up transient, the grid was translated in the y direction for the

heaving motion, using Equations 3, 4, and 5 to prescribe that motion. These results will hereafter be referred

to as Simulation B. Due to limited computational resources, a timestep was chosen that equated to 1032

timesteps per heaving cycle (versus the 50,000 of Visbal1 ), which allowed the eight critical points to fall at

regular intervals of 129 steps. The simulation was run for twenty heaving cycles plus one extra “buffer” cycle

to minimize the effects of a transient related to start-up or the beginning of the heaving motion.

Figure 5 (a) shows results from Visbal1 and Figure 5 (b) shows results from Simulation B. The images

show isosurfaces of vorticity. These results are phase-averaged, meaning that the flow variables were averaged

over all twenty cycles (fourteen for results from Visbal1 ) at only one specified location in the heaving motion.

This averaging was done for all eight critical points. The numerical values on or below each image correspond

to the numbered positions in Figure 4. The isosurfaces in the images from Simulation B are colored by Cp.

The images show motion starting from the top of the plunging motion. As the plate plunges, a leading-edge

separation region appears. This leading edge vortex (LEV) grows until the bottom of the cycle (step 5), at

which point the LEV evolves into an arch-like or Λ-shaped structure. This arch-vortex continues to move

toward the trailing edge of the plate until it eventually separates off the plate (steps 2 and 3), and the legs of

the arch connect to produce a ring-shaped structure that convects downstream (this is shown more clearly in

Figure 6). The wingtip vortices also grow with the initial downward motion (step 2) of the pitching cycle,

but vortex breakdown is seen as the plate reaches the bottom of the heaving motion (steps 3-5). The tip

vortex has disappeared by the middle of the upward part of the heaving motion (step 7). Tip vortices are

11 of 28

American Institute of Aeronautics and Astronautics


seen to form on the bottom of the plate (step 8) as the plate moves back upward (this is also more clearly

seen in Figure 6). This is the reported behavior of dynamic stall as observed by Visbal.1

In general, the results from Simulation B show similar large-scale flow structures to those observed by

Visbal.1 The developing leading-edge and arch vortex structures are clearly seen, as are the tip vortex

structures, including vortex breakdown. In addition, some of the small-scale structures are seen, including

the vortex structures shed off the trailing edge (step 4) and those near the center of the arch vortex (step 5).

Differences may exist due to difficulties in reproducing the exact color map or value of isosurface used by

Visbal,1 though most of the differences in the structures are assumed to relate to inadequate spatial resolution,

temporal resolution, or the difference in numerical methods. Despite slight discrepancies, Figure 5 indicates
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

that the simulation methodology can predict the larger-scale structures of the flowfield during heaving.

In addition to vorticity, isosurfaces of total pressure were plotted, non-dimensionalized as shown in

Equation 6.
1 ρ∗V2
Cptot = Cp + ∗ 2
(6)
2 ρ∞ ∗ V ∞

Plotting total pressure essentially filters out some of the smaller-scale features that are weaker1 and reveals

the core of the larger fluid structures. Figure 6 (a) shows these isosurfaces from Visbal1 and Figure 6 (b)

shows similar results from Simulation B. In Figure 6, it is clearly seen that the arch vortex forms into a

ring structure at the start of the next heaving cycle (step 1) and then is shed off the plate (steps 2 and 3).

Additionally, tip vortices are seen below the plate (step 8) near the end of the upward motion. In general, the

same structures are seen for both sets of figures, and it is apparent that the numerical methods are correctly

predicting the large-scale structures. However, some noticeable difference do exist. A value of total pressure

for the isosurface was chosen so that the ring-structures in steps 1, 2, and 3 would be visualized. This resulted

in structures that appear larger than the images from Visbal1 in some frames. The differences in shape and

size indicate that slight differences in the pressure field exist, and this is likely due to inadequate resolution in

the region above the flat plate. Also, the missing shed arch-structure in step 3 and the missing tip vortices in

step 1 also imply that either the pressure fields do not agree completely or that these structures are dissipated

due to improper grid resolution. This is confirmed further by the apparent difference in the colors of Cp as

shown in the isosurfaces from Simulation B. Of all the images, steps 4, 5, and 6 compare the best.

Figure 7 shows contour plots of phase-averaged Cp on the upper surface of the flat plate during the heaving

maneuver for each of the eight critical points. The images show low pressure regions at the leading edge

during the downward motion (steps 1-4) of the heaving cycle, confirming the presence of a leading-edge vortex

attached to the plate. Step 4 shows that the leading-edge vortex has lifted off the plate at the centerline.

During the upward motion (steps 5-8), regions of low pressure show the attached legs of the arch vortex as

it moves downstream and detaches from the plate. The effects of tip vortices are seen in steps 2-4 as well,

12 of 28

American Institute of Aeronautics and Astronautics


although step 4 confirms that the vortex has started to break down. In general, the results of Simulation B

match the results of Visbal1 quite well, though some slight differences exist. These differences are likely due

to difficulties in replicating the exact color map used by Visbal.1 One major difference is that the images

from Simulation B are clearly non-symmetric about the centerline of the plate. This is particularly evident in

steps 1, 5, 6, 7, and 8 when considering the locations and size of pressure in the attached legs of the arch

vortex. In results from Visbal,1 symmetry is enforced at the plate centerline to reduce computational costs.

However, Simulation B does not employ a symmetry plane. Even though asymmetries could exist in plots of

instantaneous Cp, asymmetries in plots that have been phase-averaged over twenty cycles indicates a source

of asymmetry in each heaving motion. Though not clearly visible in Figures 5 and 6, the asymmetry also is
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

seen during the upward motion of the heaving cycle. Figure 8 shows isosurfaces of phase-averaged vorticity

from the front of the plate in steps 5, 6, and 7. The images clearly show asymmetric flow structures. While

the exact reason for the asymmetry is unknown, it is most likely due to asymmetries in the wake region of

the grid since the grid is fairly coarse at that location (roughly 15 cells across the core of the lambda vortex).

Although Figures 5, 6, and 7 confirm that the large-scale features are predicted well, it is also of interest

to investigate the ability of the chosen methods to predict smaller structures, especially the transitional and

turbulent structures. For this reason, instantaneous isosurfaces of Q-criterion34, 35 (a method for identifying

vortices in the flowfield) are shown in Figure 9. Figure 9 (a) shows images from Visbal,1 which reveal several

smaller-scale transitional structures being shed of the trailing edge. Images from Simulation B (Figure 9 (b))

show some of the same fine scale structures. Additionally, a few of the smaller scale structures are seen in

front of the arch vortex in steps 5-7, although it is clear that these are fewer and not as well resolved in

Simulation B as in Visbal.1 Steps 2-4 show a fairly significant deviation from the results of Visbal1 in the

arch structure and at the rear of the plate. It is likely that some of the noted differences are related to the

chosen value of Q-criterion for the isosurface. Another reason for the dissimilarities could be that the region

near the trailing edge of the plate is not as well resolved in Simulation B as in Visbal.1

One final comparison is made for the phase-averaged lift and drag forces. Phase-averaging was performed

for each of the 1032 timesteps in the heaving cycle, over the twenty completed cycles for Simulation B.

Figures 10 (a) and 10 (b) show images from Visbal1 and Figures 10 (c) and 10 (d) show corresponding

images from Simulation B with the critical points displayed on the graphs. The lift and drag forces appear

to agree quite well, except for a small portion of the CL and CD graphs between steps 2 and 3 near the

maximum positive value of α. Despite these differences, the similarity between the phase-averaged forces

further indicates, along with Figures 5, 6, and 7, that the large-scale structures are predicted correctly.

13 of 28

American Institute of Aeronautics and Astronautics


D. Grid Convergence Study

A grid convergence study was performed to determine if the results were converged numerically. The solutions

on three computational grids of successive refinement were investigated: fine (grid 1), medium (grid 2), and

coarse (grid 3), as suggested by Roache.36–38 Table 2 shows the number of cells and nodes in each grid, as

well as a grid coarsening factor. To generate successively coarser grids, all specified point spacings for the fine

plate grid, including the distance to the first point off the flat plate, the spacing on the flat plate, the spacing

on the refinement box, and the spacing on the outer boundaries, were multiplied by the grid coarsening

factor. The initial spacing of the first point off the flat plate (d) is also given in Table 2, non-dimensionalized

by the chord (c). All three simulations used the same timestep size, which corresponded to 1032 timesteps
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

per heaving cycle. The simulations took a lengthy time to finish running on 400 processors. For a total of

22,962 timesteps, the coarse simulation required around 144,000 processor-hours, the medium simulation took

around 240,000 processor-hours, and the fine grid took around 393,600 processor-hours.

Table 2: Sizes of the computational grids.

Grid Number of Cells Number of Nodes Grid Coarsening Factor Initial Spacing (d/c)
1 (fine) 98,009,172 23,440,877 1 9.33×10−4
2 (medium) 56,211,346 14,848,229 1.5 1.40×10−3
3 (coarse) 32,782,044 8,795,335 1.875 1.75×10−3

A typical method of calculating the discretization error is the Grid Convergence Index (GCI), usually

calculated on a flow-based quantity obtained from the solution on the finest grid. In essence, the GCI of a fine

grid estimates the percentage that some quantity on the fine grid differs from that quantity measured on an

assumed “converged” numerical solution.36–39 This method is usually applied to structured or geometrically

similar grids,38, 40 but the method can be used on unstructured grids with some special considerations.41–43

When applied to three grids, the GCI method assumes that quantities, such as CL and CD , lie within the

asymptotic range and that they monotonically increase or decrease toward the numerically converged solution

with grid resolution. If the data is not monotonic, the GCI method cannot be reliably applied.44, 45

It was intended to apply the GCI to the three grids listed in Table 2 at each of the eight critical points

from the heaving motion in the previous section. However, the CL and CD data were not monotonic at each

of these eight points. Therefore the three solutions were not in the asymptotic range and the GCI method

was not applied.

Even without a GCI analysis, it was important to investigate whether or not all three grids predicted the

same large-scale flow features. Figure 11 shows isosurfaces of phase-averaged Q-criterion at step 5 for the

fine, medium, and coarse grids. All three grids show the same large-scale features, including the arch vortex

and the wingtip-vortex breakdown. It is clear that the coarser grids do not show some of the structures

14 of 28

American Institute of Aeronautics and Astronautics


seen on the fine grid, such as the smaller-scale vortices extending upward in front of the arch vortex. It is

also clear that the tip vortex itself is not resolved as well on the coarse and medium grids, especially at the

tips. Some slight differences can be observed in the thickness and shape of the arch-vortex in the fine grid as

compared to the other two grids, which indicates that the solution may not be fully converged. This also

suggests that the grid resolution in this region could be causing the asymmetric results observed in Figures 7

and 8. Despite these differences, all three grids do adequately predict the large-scale structures. Additional

smaller-scale structures are introduced through grid refinement as expected.

The phase-averaged lift and drag for each case are plotted in Figure 12 with each of the eight critical

points overlayed onto the plots. From the figure, the only noticeable differences in CL from the fine grid in
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

Figure 12 (a) are seen when α is negative, which is during the upward motion near steps 6 and 8. The only

noticeable differences in CD from the fine grid in Figure 12 (b) are seen when α is positive, which is during

the upstroke near steps 1 and 4. Figure 12 (c) plots the log of the percent difference from the medium grid to

the fine grid. For the majority of the points in the heaving motion, the fine and medium grids agree within

2%. However, just after steps 4 and 7, the percentage difference for both CL and CD spikes. The explanation

for this is that a temporal offset occurs between the two results as the CL and CD curves pass through zero

and leads to large percentage differences. Despite these large percentage differences, it is clear that even the

medium and coarse grids predict the correct phase-averaged forces.

15 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

(a) Visbal1

1 2 3

4 5 6

7 8
(b) Simulation B

Figure 5: Isosurfaces of phase-averaged vorticity magnitude colored by Cp from Visbal1 and Simulation B.

16 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

(a) Visbal1

1 2 3

4 5 6

7 8
(b) Simulation B

Figure 6: Isosurfaces of phase-averaged total pressure colored by Cp from Visbal1 and Simulation B.

17 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

(a) Visbal1

1 2 3 4

5 6 7 8
(b) Simulation B

Figure 7: Contours of Cp on upper surface of flat plate from Visbal1 and Simulation B.
18 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

(a) step 5

(b) step 6

(c) step 7

Figure 8: Flow asymmetries revealed in isosurfaces of phase-averaged vorticity colored by Cp from Simulation B

19 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

(a) Visbal1

1 2 3

4 5 6

7 8
(b) Simulation B

Figure 9: Isosurfaces of instantaneous Q-Criterion colored by Cp from Visbal1 and Simulation B.

20 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

Visbal1

2
3 0.3
2 2

1 1
3
0.2

1
4
CD

0.1
CL

8 4
8

0 0

5 5
7 -0.1
7
-1 6 6

-0.2
-20 -10 0 10 20 30 -20 -10 0 10 20 30
α eff α eff

(c) (d)
Simulation B

Figure 10: Phase-averaged forces from Visbal1 and Simulation B.

21 of 28

American Institute of Aeronautics and Astronautics


Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

(a) Fine Grid

(b) Medium Grid

(c) Coarse Grid

Figure 11: Comparison of isosurfaces of phase-averaged Q-Criterion colored by Cp at step 5 of Simulation B


for three grids.

22 of 28

American Institute of Aeronautics and Astronautics


2 Medium
Baseline
Coarse 3 0.3 Coarse
2 Fine 2
Fine
1 1
3
0.2

1
4

CD
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

0.1
CL

8 4
8

0 0

5 5
7 -0.1
7
-1 6 6

-0.2
-20 -10 0 10 20 30 -20 -10 0 10 20 30
α eff α eff

(a) CL (b) CD

CL
100 CD
Log of % Difference

0.01

2 4 6 8
Step Number
(c) Percent difference from fine to medium grid

Figure 12: Phase-averaged forces from Simulation B for three grids.

23 of 28

American Institute of Aeronautics and Astronautics


E. Summary of Simulation Results and Limitations of Methods

The transient evolution of dynamic stall reported by Visbal1 and Yilmaz and Rockwell5 was investigated

using an alternative computational method. It was expected that we would not be able to reproduce as

finely resolved results as Visbal1 due to the use of lower-order methods with inadequate spatial and temporal

resolution. However, the results of Simulations A and B show that the large scale structures, the phase-

averaged forces, and some small-scale structures are correctly predicted. Still, several small-scale structures

are not resolved, indicating that further grid resolution is needed. Further refinement should also allow

for calculating a GCI in the asymptotic range to demonstrate grid convergence. Since Visbal1 employed

a sixth-order method on a multi-zonal stretched Cartesian grid, it was expected that this strategy would
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

require fewer computational cells than our second-order scheme. The grid for Simulation B used 93 million

cells, almost 160% more cells than the grid used by Visbal1 if Visbal’s grid size was doubled to account for

the use of a symmetry plane. Therefore, as the grid from Simulation B is refined further, a large cell count is

expected, which is a potential limitation of this computational strategy. However, while a stretched Cartesian

grid is acceptable for a flat plate, an unstructured grid would be preferred for more geometrically complex

configurations.

Temporal resolution also differs greatly between Simulation B (1,032 timesteps per cycle) and the results

of Visbal (50,000 timesteps per cycle).1 Although it is assumed that spatial grid resolution is the main cause

for differences between the results from Simulation B and the results of Visbal,1 it is recognized that temporal

resolution could have an effect on the flow structures as well. Since the same timestep size was used for three

successively coarser grids, differences due to temporal resolution are not included in the comparisons shown

and should be investigated further.

Additionally, Simulation B made use of a Hybrid RANS/LES turbulence model, as compared to the

Implicit LES model with an eighth-order low-pass filter used by Visbal.1 The hybrid model was able to

predict the larger scale structures. The use of the Hybrid RANS/LES model allows for anisotropic cells in

the boundary layer, which reduces the required grid spacing as compared to LES, which usually requires

isotropic cells in order to be effective.13 It appears that the Hybrid RANS/LES model, though mostly

used on simulations with higher Reynolds numbers, is still applicable to these low-to-moderate Reynolds

numbers simulations at least for the current grids. This turbulence model is a potential limitation for the

methods presented in this paper since it involves the assumption that the entire flowfield is turbulent. At

this low-to-moderate Reynolds number, it is clear that the flow over the plate is not initially turbulent. This

is confirmed by Visbal,1 who states that the ILES turbulence model is especially appropriate for a flow

with “mixed laminar, transitional, and turbulent regions.” The claim of the ILES method is that it uses the

unfiltered Navier-Stokes equations without additional SGS (sub-grid stress) terms, and therefore can be used

24 of 28

American Institute of Aeronautics and Astronautics


in any flow regime. The low-pass filter is applied only to the conserved variables, not to the equations as in a

typical LES framework. In this manner, the ILES method has a distinct advantage. However, it is possible to

investigate the addition of a transition model to the Hybrid RANS/LES scheme.19 Even considering a fully

turbulent flowfield, it is unclear if the model, especially the RANS part, will allow resolution of smaller-scale

structures with grid refinement.

In summary, we have confidence in the computational methods employed in this paper to predict the

dynamic large-scale flow structures and aerodynamic forces in low-to-moderate Reynolds number maneuvers

involving large angles of attack. We anticipate that refinement of the spatial and temporal scales will lead to

a better resolution of these flowfield features. More investigations will be required as to the effectiveness of
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

the Hybrid RANS/LES model for these flow conditions.

V. Conclusion

In conclusion, simulation results produced by second-order, unstructured methods were compared to results

from simulations produced by sixth-order structured methods.1 The simulations compared favorably, and

the lower-order method accurately predicted the large-scale flow structures and phase-averaged lift and drag

forces. These methods will be applied to the simulation of a pitch-up, perch, or otherwise non-linear maneuver

with a notional “pizza-box” SUAS vehicle with actuating propellers. The grid generation and computational

cost associated with this case would be significant with a structured, higher-order method. Therefore the

lower-order, unstructured methods presented in this paper appear desirable for these simulations.

There are several concerns associated with this problem, and future work is planned to address these

concerns. Once concern is the use of a turbulence model in transitional flow regions. With the current

implementation of the Nichols-Nelson multiscale Hybrid RANS/LES turbulence model, it is assumed that

the entire boundary layer is turbulent since no transition model was used. This is obviously not physically

correct for most low-to-moderate Reynolds number flows. Therefore, investigations will be made into the

effects of incorporating transition into the RANS turbulence model in a manner similar to Alam et al.19

Additionally, other RANS models, such as Menter’s BSL model30 and Wilcox’s k − ω model,32 will be

investigated with the Hybrid RANS/LES model. Furthermore, different levels of freestream turbulence

intensity will be investigated. Additional work will include refinement of the grid and simulation timestep

size to investigate grid convergence. Investigations will be made to identify the cause of the asymmetries

observed in the flow structures during the upward motion of the plate.

25 of 28

American Institute of Aeronautics and Astronautics


Acknowledgments

Effort sponsored by the Engineering Research & Development Center under Cooperative Agreement

number W912HZ-15-2-0004. The views and conclusions contained herein are those of the authors and should

not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied,

of the Engineering Research & Development Center or the U.S. Government.

References
1 Visbal, M., “Three-Dimensional Flow Structure on a Heaving Low-Aspect-Ratio Wing,” AIAA, No. AIAA 2011-2019 in
49th AIAA Aerospace Sciences Meeting, 2011.
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

2 Shyy, W., Lian, Y., Tang, J., Viieru, D., Liu, H., Shyy, W., Lian, Y., Tang, J., Viieru, D., and Liu, H., Aerodynamics of
low Reynolds number flyers, Vol. 22 of Cambridge Aerospace Series, Cambridge University Press, 2008.
3 Anderson Jr, J. D., Fundamentals of aerodynamics, McGraw-Hill, New York, NY, 4th ed., 2007.
4 Mueller, T. J. and DeLaurier, J. D., “Aerodynamics of small vehicles,” Annual Review of Fluid Mechanics, Vol. 35, 2003,
pp. 89–111.
5 Yilmaz, T. O. and Rockwell, D., “Three-dimensional flow structure on a maneuvering wing,” Experiments in Fluids,
Vol. 48, No. 3, 2010, pp. 539–544.
6 Visbal, M., Yilmaz, T. O., and Rockwell, D., “Three-dimensional vortex formation on a heaving low-aspect-ratio wing:
Computations and experiments,” Journal of Fluids and Structures, Vol. 38, 2013, pp. 58 – 76.
7 Visbal, M. R., “Flow structure and unsteady loading over a pitching and perching low-aspect-ratio wing,” AIAA, New
Orleans, Louisiana, No. AIAA 2012-3279 in 42nd AIAA Fluid Dynamics Conference and Exhibit, 2012.
8 Ol, M. V. and Babinsky, H., “Unsteady Flat Plates: a Cursory Review of AVT-202 Research (Invited Paper),” AIAA, San
Diego, California, No. AIAA 2016-0285 in 54th AIAA Aerospace Sciences Meeting, 2016.
9 Stevens, P. R. R. J., Babinsky, H., Manar, F., Mancini, P., Jones, A. R., Granlund, K. O., Ol, M. V., Nakata, T., Phillips,
N., Bomphrey, R. J., et al., “Low Reynolds Number Acceleration of Flat Plate Wings at High Incidence (Invited Paper),” AIAA,
San Diego, California, No. AIAA 2016-0286 in 54th AIAA Aerospace Sciences Meeting, 2016.
10 Bernal, L. P., “Unsteady Aerodynamics of Pitching Low Aspect Ratio Wings: A review of AVT 202 panel results (Invited
Paper),” AIAA, San Diego, California, No. AIAA 2016-0287 in 54th AIAA Aerospace Sciences Meeting, 2016.
11 Jones, A. R., Manar, F., Phillips, N., Nakata, T., Bomphrey, R., Ringuette, M., Percin, M., van Oudheusden, B., and
Palmer, J., “Leading Edge Vortex Evolution and Lift Production on Rotating Wings,” AIAA, San Diego, California, No. AIAA
2016-0288 in 54th AIAA Aerospace Sciences Meeting, 2016.
12 Nichols, R. and Nelson, C., “Applications of RANS/LES Turbulence Models,” AIAA, Reno, Navada, No. AIAA 2003-0083
in 41st Aerospace Sciences Meetings, 2003.
13 Spalart, P. R. and Streett, C., “Young-person’s guide to detached-eddy simulation grids,” Technical Report CR-2001-211032,
NASA, 2001.
14 Wilcox, D. C., Turbulence Modeling for CFD, DCW Industries, La Canada, CA, 2nd ed., 1998.
15 Spalart, P. R., Jou, W. H., Strelets, M., and Allmaras, S. R., “Comments on the feasibility of LES for wings, and on a
hybrid RANS/LES approach,” in Liu, C. and Liu, Z., eds., “First AFOSR International Conference on DNS/LES,” Greyden
Press, Columbus, OH, 1997, pp. 137–147.

26 of 28

American Institute of Aeronautics and Astronautics


16 Spalart, P. R., Deck, S., Shur, M. L., Squires, K. D., Strelets, M. K., and Travin, A., “A new version of detached-eddy
simulation, resistant to ambiguous grid densities,” Theoretical and computational fluid dynamics, Vol. 20, No. 3, 2006, pp.
181–195.
17 Shur, M. L., Spalart, P. R., Strelets, M. K., and Travin, A. K., “A hybrid RANS-LES approach with delayed-DES and
wall-modelled LES capabilities,” International Journal of Heat and Fluid Flow, Vol. 29, No. 6, 2008, pp. 1638–1649.
18 Walters, D., Bhushan, S., Alam, M., and Thompson, D., “Investigation of a dynamic hybrid RANS/LES modelling
methodology for finite-volume CFD simulations,” Flow, turbulence and combustion, Vol. 91, No. 3, 2013, pp. 643–667.
19 Alam, M. F., Walters, K., and Thompson, D., “A Transition-Sensitive Hybrid RANS/LES Modeling Methodology for
CFD Applications,” AIAA, Grapevine (Dallas/Ft. Worth Region), Texas, No. AIAA 2013-0995 in 51st AIAA Aerospace Sciences
Meeting including the New Horizons Forum and Aerospace Exposition, 2013.
20 Luke, E. A., Tong, X.-L., Wu, J., Tang, L., and Cinnella, P., “A Chemically Reacting Flow Solver for Generalized Grids,”
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

AIAA Journal.
21 Luke, E. and Cinnella, P., “Numerical simulations of mixtures of fluids using upwind algorithms,” Computers and Fluids,
Vol. 36, No. 10, 2007, pp. 1547–1566.
22 Luke, E. A., A rule-based specification system for computational fluid dynamics, Ph.D. thesis, Mississippi State University,
1999.
23 Luke, E. A. and George, T., “Loci: a rule-based framework for parallel multi-disciplinary simulation synthesis,” Journal
of Functional Programming, Vol. 15, No. 03, 2005, pp. 477–502.
24 Veluri, S. P., Roy, C. J., and Luke, E. A., “Comprehensive code verification for an unstructured finite volume CFD code,”
AIAA, Orlando, Florida, No. AIAA 2010-127 in 48th AIAA Aerospace Sciences Meeting, 2010.
25 Veluri, S. P., Roy, C. J., and Luke, E. A., “Comprehensive code verification techniques for finite volume CFD codes,”
Computers and Fluids, Vol. 70, 2012, pp. 59–72.
26 Roy, C. J., Tendean, E., Veluri, S. P., Rifki, R., Luke, E. A., and Hebert, S., “Verification of rans turbulence models
in Loci-CHEM using the method of manufactured solutions,” AIAA, Miami, Florida, No. AIAA 2007-4203 in 18th AIAA
Computational Fluid Dynamics Conference, 2007.
27 Koomullil, R. and Soni, B., “Flow simulation using generalized static and dynamic grids,” AIAA Journal, Vol. 37, No. 12,
1999, pp. 1551–1557. Dec.
28 Beam, R. M. and Warming, R. F., “An Implicit Factored Scheme for the Compressible Navier-Stokes Equations,” AIAA
Journal, Vol. 16, No. 4, 1978, pp. 393–402.
29 Menter, F. R., “Influence of freestream values on k-omega turbulence model predictions,” AIAA Journal, Vol. 30, No. 6,
1992, pp. 1657–1659.
30 Menter, F. R., “Two-equation eddy-viscosity turbulence models for engineering applications,” AIAA Journal, Vol. 32,
No. 8, 1994, pp. 1598–1605.
31 Johnson, D. A. and King, L., “A mathematically simple turbulence closure model for attached and separated turbulent
boundary layers,” AIAA Journal, Vol. 23, No. 11, 1985, pp. 1684–1692.
32 Wilcox, D. C., “Reassessment of the scale-determining equation for advanced turbulence models,” AIAA Journal, Vol. 26,
No. 11, 1988, pp. 1299–1310.
33 Pointwise, Inc., Pointwise User Manual, 2014.
34 Hunt, J. C., Wray, A., and Moin, P., “Eddies, streams, and convergence zones in turbulent flows,” Tech. Rep. Report
CTR-S88, Center for Turbulence Research, 1988.
35 Jeong, J. and Hussain, F., “On the identification of a vortex,” Journal of fluid mechanics, Vol. 285, 1995, pp. 69–94.

27 of 28

American Institute of Aeronautics and Astronautics


36 Roache, P. J., Verification and Validation in Computational Sciences and Engineering, Hermosa Publishers, Albuquerque,
New Mexico, 1998.
37 Slater, J. W., “Examining Spatial (Grid) Convergence,” Website, http://www.grc.nasa.gov/WWW/wind/valid/tutorial/spatconv.html,
2008. Accessesd: 2016-03-18.
38 Roache, P. J., “Perspective: a method for uniform reporting of grid refinement studies,” Journal of Fluids Engineering,
Vol. 116, No. 3, 1994, pp. 405–413. 10.1115/1.2910291.
39 Schwer, L. E., “Is your mesh refined enough? Estimating discretization error using GCI.” in “LSDYNA User Forum,
Bamberg, Germany,” DYNAmore GmbH, Bamberg, Germany, Validation / Verification, 2008.
40 Roy, C. J., “Review of code and solution verification procedures for computational simulation,” Journal of Computational
Physics, Vol. 205, 2005, pp. 131–156.
41 Pelletier, D. and Ignat, L., “On the accuracy of the grid convergence index and the Zhu-Zienkiewicz error estimator,”
Downloaded by UNIVERSITY OF CAMBRIDGE on June 15, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-4338

ASME-Publications-FED, Vol. 213, 1995, pp. 31–36.


42 Roache, P. J., “Quantification of uncertainty in computational fluid dynamics,” Annual Review of Fluid Mechanics,
Vol. 29, 1997, pp. 123–160.
43 Eça, L., Hoekstra, M., Beja Pedro, J., and Falcão de Campos, J., “On the characterization of grid density in grid refinement
studies for discretization error estimation,” International Journal for Numerical Methods in Fluids, Vol. 72, 2013, pp. 119–134.
44 Cadafalch, J., Pérez-Segarra, C., Consul, R., and Oliva, A., “Verification of finite volume computations on steady-state
fluid flow and heat transfer,” Journal of Fluids Engineering, Vol. 124, 2002, pp. 11–21.
45 Eça, L. and Hoekstra, M., “A verification exercise for two 2-D steady incompressible turbulent flows,” in “Proceedings of
the ECCOMAS,” Jyväskylä, Finland, 2004.

28 of 28

American Institute of Aeronautics and Astronautics

Das könnte Ihnen auch gefallen