Professional Documents
Culture Documents
September, 20 01__
I, __Ashutosh Gangwar ,
hereby submit this as part of the requirements for the
degree of:
Master of Science
in:
Aerospace Engineering
It is entitled:
Source Term Modeling of Rectangular Flow Cavities
________________________________________________
________________________________________________
________________________________________________
Approved by:
Orkwis P.
Jeng S.
Khosla P.
________________________
________________________
SOURCE TERM MODELING OF
by
Ashutosh Gangwar
University of Cincinnati
September 2000
Approved by ___________________________________________________
Chairperson of Supervisory Committee
Dr Orkwis P. ___________________________________________________
Dr Jeng S______________________________________________________
Dr Khosla P. ___________________________________________________
Program Authorized
to Offer Degree _________________________________________________
Date _________________________________________________________
ABSTRACT
The presence of small cavities has an effect on the primary fluid flow and should be
modeled properly. The goal of this research is to develop a source term module that
models the effect of these cavities for gas-path only simulations. In this approach the
unsteady effect of the cavity in the solution is modeled by adding source terms to the right
hand side of the Navier-Stokes equation without actually having to resolve the complex
cavity geometry. This idea increases considerably the numerical efficiency of the scheme by
avoiding the computation of small-scale fluid dynamic structures and complex geometric
details of the cavity.
Unsteady flow phenomena governing both subsonic and supersonic cavity oscillations
were studied. The exact cycle for both cases has been presented and compared with those
presented by other researchers. These unsteady solutions have been used to determine
deterministic source terms. Source terms were then inserted in steady solutions and used to
demonstrate that they can be used to model accurately cavity unsteadiness. Also two
methods to model these source terms were briefly reviewed.
ACKNOWLEDGMENTS
I wish to thank Dr. Paul Orkwis for the guidance and support through the entire research.
Also I wish to thank everyone in the Gas Turbine Research Laboratory for their helpful
comments.
I would like to thank Ohio Aerospace Institute, Core Collaborative Research Program OAI
CCRP 98-1-000, Dayton Area Graduate Studies Institute DAGSI PR-AFIT-99-07, US Air-
force Research Laboratory, Propulsion Directorate and General Electric Aircraft Engines
for financial and technical help.
ii
Table of Contents
Abstract....................................................................................................................................................................i
Acknowledgments ...................................................................................................................ii
1. Overview ............................................................................................................................5
Simplification to Rectangular Cavity........................................................................................................9
Parametric Variation.....................................................................................................................................10
Source Term Modeling................................................................................................................................11
Neural Network Model................................................................................................................................13
2. Numerical Algorithm........................................................................................................16
Computational Scheme................................................................................................................................16
Governing equations and Numerical Scheme.......................................................................................17
CFD++ Framework.....................................................................................................................................19
Computation Grids.......................................................................................................................................20
Computational Resources...........................................................................................................................22
Boundary and Initial Conditions...............................................................................................................22
Unsteady Calculations..................................................................................................................................23
Visualization Method...................................................................................................................................25
Combined Schlieren Simulation and Vorticity Contours................................................................25
3. Unsteady Flow Physics.....................................................................................................27
Earlier Research.............................................................................................................................................27
Unsteady Cavity Oscillations......................................................................................................................30
Rossiter's Empirical Formula.....................................................................................................................30
Heller and Bliss Cycle...................................................................................................................................31
Vortex interaction Mechanism ..................................................................................................................36
Results and Discussions...............................................................................................................................38
Supersonic Results.......................................................................................................................................38
Subsonic Results...........................................................................................................................................45
iii
Dominant Frequency..................................................................................................................................56
Parametric Studies .......................................................................................................................................56
4. Deterministic source terms...............................................................................................61
Lumped Deterministic Stress Technique................................................................................................61
Previous Research.........................................................................................................................................64
Modeling of Implicit Boundary Conditions...........................................................................................64
Modeling of Multistage Compressor Flowfields...................................................................................65
Hot Gas Segregation....................................................................................................................................66
Results and Discussions...............................................................................................................................67
Method of Calculation of Source Terms ..............................................................................................67
Supersonic Source Terms..........................................................................................................................67
Subsonic Source Terms..............................................................................................................................74
Source Term Modeling................................................................................................................................78
5. Directions for Future Work..............................................................................................83
6. Conclusion and Discussions.............................................................................................89
Direction for Future Work.........................................................................................................................93
References: ............................................................................................................................94
iv
1. OVERVIEW
Introduction
The aerodynamic design methodology has changed significantly in recent years. The last
decade or two have seen great advances in CFD techniques. Earlier designs were highly
simplified as compared to what is now possible. The design process was based on experience
rather than on analytical results. Ideas that were promising were typically included in the design
with their impact not known a priori. The designs were then tested and the experience was
used to update the existing knowledge base. Thus designs were based extensively on empirical
Due to the advent of high-speed computers and advanced computational methods, designers
can now use Computational Fluid Dynamics (CFD) to solve three-dimensional viscous flows
in a timely manner. This fundamentally changes the design process in two ways. First,
designers can compare two designs relatively quickly instead of relying strictly on past
experience or an upcoming test. Second, CFD allows designers to better visualize the detail of
The aerodynamic design of axial turbines too has undergone significant changes in the recent
years. In the last two decades the gaspath flow was modeled using computer programs that
While this model yielded successful designs, the total pressure loss levels of individual
5
bladerows were applied to the model a priori. As sufficient data was not available, this model
did not appropriately account various looses associated with the secondary flows. The
secondary flows are caused due various effects such as but not limited to the effects of the
boundary layers, un-shrouded tip clearance, effect of the film cooling and gaps between stator-
rotor at the hub. This neglecting of secondary flow resulted in designs that differed
considerably from experimental data. Also means to cut these loses were considered
empirically.
Although today’s designs include most of the above-mentioned secondary effects, the effect of
gap between the rotor stator at the hub is not appropriately included. These gaps are assumed
as flat surfaces running between the stator and rotor. In reality a gap is required between the
stationary and rotating bladerow at the hub. Usually an axial overlap - or seal - exists at the gap
to provide more aerodynamically efficient transition and to reduce gaspath flow from entering
the cavity. The gas that enters and leaves the gap is called wheelspace purge flow. The
wheelspace purge flow has a significant impact on the endwall boundary layer, which directly
The effect of these cavities is not limited to the gap region alone but is convected into the
main passage by means of two phenomena. First, pressure waves bouncing back and forth
between the cavity walls because of the self sustained oscillations within the cavity. Second the
shear layer that exists over the surface spans the cavity and is shed because of the oscillations
and is washed into the mainflow. The degree of secondary flow that forms is directly related to
upstream velocities and endwall boundary layer thickness of the flow that enters the bladerow.
Neglecting these cavities gives results that deviate considerable from the experimental results
6
Several attempts have been made, both experimental and computational, by various
researchers to understand the flow physics of the cavities notably by Heller and Bliss[1],
Rockwell and Naudasher [2] and Tam, et al. [3,4]. The complicated geometries of this purge
flow cavities and the unsteady flow phenomenon associated with them makes it very difficult
to include these small cavities directly in numerical simulations. A realistic picture of these
cavities is included here to help the reader appreciate the complexity of the geometry under
consideration. Figure 1.1 shows a single stage of a turbine, a single stator (region 2) followed
by the rotor (region 4) and the rotor-stator gap (regions 1& 3). The stator and rotor are
connected by means of a narrow cavity, which is seen at the bottom of the picture (region 5).
The cavity region (region 5) is zoomed out and shown separately so that the viewer can
appreciate the complex geometry of the cavity. Figure 1.2 shows a schematic of this wheel
space purge cavity. This is a highly simplified cartoon, which shows a single stage of a gas
1 3
5
2 4
7
1.2
Very fine grids are required with almost the same number of grid cells as the primary passage
to model these cavities even approximately. Also, lower velocities in the cavities lead to severe
restrictions on the time steps and hence considerable computational overhead. As a result
these cavities are often completely eliminated from the numerical simulations due to the
limited time frame available for design - often overnight. In order to study these cavities it
becomes necessary to resort to more simplified models that capture the essential flow physics
without incurring the computational costs associated with simulations of the purge cavities
directly. A rectangular cavity is one such simplification and will be discussed now.
8
Simplification to Rectangular Cavity
In order to understand the flow physics of the cavity a simplified geometry consisting of a
rectangular cavity and passage have been used. The rectangular cavity serves as a good starting
point for modeling the actual geometry. Although a rectangular cavity leads to an over
simplification of the purge flow cavity geometry, it nevertheless helps us understand some
important aspects of the flow physics of the purge flow cavity. However, there is one very
important difference between the purge cavities and the rectangular cavity - the stator and
rotor of purge cavities are connected by means of a narrow passage. Chokran [5]
experimentally and Jeng [6] computationally have shown that the unsteady oscillations existing
in a rectangular cavity may be considerably damped in case of the purge cavity because of the
fluid being injected or removed from the cavity. It must be pointed out that this connecting
path has a very narrow constriction as seen in figure 1.1 and causes very little fluid to escape
from the cavity. Also the front and back wall of the trench cavity rotate in the circumferential
direction while the cavity under consideration has stationary walls.. Also the trench cavities are
axis-symmetric while the cavity under consideration is rectangular. Another important reason
for using a rectangular cavity in the current research is that various researchers have used
rectangular cavities to understand the unsteady effects associated with the cavity and a good
9
Figure 1.3: Rectangular cavity grids
In order to model these cavities appropriately, the flow phenomenon associated with these
cavities needs to be understood properly. Also of interest to us is the variation of this flow
physics with cavity geometry and flow parameters. Heller and Bliss [1] based on wave
interaction mechanism and Rockwell and Naudascher [2] based on vortex interaction
mechanism based on computations have put forward a theory to explain the cavity oscillation
cycle mechanism. Also Tam, et al. [3] provides a theory of this oscillation cycle mechanism.
Although all of them agree that there is an oscillating shear layer, which sheds and impinges on
the aft wall of the cavity, differences exist between the researchers regarding the exact
mechanism cause of the pressure oscillations. These oscillation cycles are presented in detail in
the next chapter. A study of flow physics of a 2-D rectangular cavity was performed, and the
oscillation cycle mechanism has been compared with those suggested by other researchers.
Parametric Variation
Although the cavity oscillation mechanism has been more or less understood the variation of
this mechanism with geometry of the cavity or flow parameters has not been studied. Cavities
10
are classified as open, closed or transitional based on the length to depth ratio (L/D). Zhang
and Edwards [6] have studied in detail the variation of the oscillation cycle mechanism with the
L/D ratio of the cavity. They reported significant differences in the oscillation mechanism for
different cavities. Their studies suggest that although there is a single dominant frequency for
lower L/D ratios, for larger L/D ratios there is more then one dominant frequency. For very
large ratios one observes no dominant frequency. These differences are pointed out in detail in
Parametric studies with variation in flow parameters have been scarcely seen in literature.
Zhang and Edwards [7] have obtained results for M= 1.5 and M= 2.0. The unsteadiness
present in the cavity depends on the cavity geometry, thickness of the boundary layer and inlet
injection velocities. To study this parametric variation with Mach number, simulations were
performed for a range of Mach number from subsonic to supersonic (0.3-2.0). The cavity
oscillation behavior has been observed using unsteady numerical simulation. Variation of the
cavity flow physics with Mach number has been extensively studied and described in detail in
the following chapters. This study has also formed a background for further investigation of
Understanding the unsteady flow physics of the cavity helps in two ways; one it helps to
develop flow control methods to suppress the oscillations and two, to develop models that
include the effect of these cavities in numerical simulations. Most of the existing stuies have
been aimed at suppressing the cavity oscillations. Both active and passive flow control devices
have been tested to suppress the cavity oscillation. Sarno [8] studied the effect of pulsating
flow injection at the leading edge of the cavity. Jeng [1] computed the effect of porous walls at
11
front bulkhead and rear bulkhead to suppress the cavity oscillations. The attempts to model
these cavities have been relatively scarce. An important aspect of these cavities is that there is
dissipation of energy inside the cavity, which makes it difficult to model the cavity interface
using a suitable boundary condition. There is flow being extracted and/or injected inside the
cavity. This dissipation of momentum and energy can be accounted well by using source
terms. Source terms are terms that are added to right hand side of the Navier-Stokes equations.
Although the use of source terms is not new to the field of computational fluid dynamics, the
manner in which they are utilized here is sparingly seen in the literature. These source terms
are both very effective in capturing the unsteady effects and are very easy to incorporate into
the solver. The strength and distribution of source terms however needs to be determined.
Hunter, et al. [9] have used strip source terms lined along the cavity interface. A functional for
is identified and the coefficients of the functional form are determined from experiments. An
approach using strip source terms placed along the interface of the cavity only applies the
boundary condition correctly. It does not capture the unsteady flow features such as moving
shock waves and vortex shedding phenomenon. Thus their model captured only the boundary
effects and not the field effect of the unsteadiness. An attempt to capture the field
phenomenon requires the source terms to be distributed inside the cavity and in the region
The current research focuses on field source terms rather then point source terms. Field
source terms are distributed in the entire region as opposed to being lined up at the cavity-
main flow interface. The strength of these source terms has been calculated directly from the
unsteady calculations. As a result they are exact and represent the flow physics more accurately.
12
To demonstrate the effectiveness of the approach solutions obtained from steady simulations
with source terms inserted were compared with time average of unsteady simulations.
It must be noted however, that in calculating the source terms from unsteady solutions one is
post-dicting the effect of unsteadiness as opposed to predicting the unsteadiness. Thus the
source terms calculated in this manner can not be directly used for modeling. These source
terms are however used for two purposes; first, they are used to demonstrate the feasibility of
using the source terms to model the unsteadiness. This has been demonstrated by comparing
the solutions obtained from steady simulations with source terms inserted with time average of
unsteady simulations, Second these source terms obtained over a range of parameters form a
database that is used for training a neural network. This neural network can then in turn
predict the source terms for an arbitrary range of flow parameters. Each of the aspect of
rectangular cavity flow unsteadiness modeling will be discussed in the following chapters.
The next chapter describes the numerical algorithm used. The numerical code CFD++ used
for current simulations is discusses in detail. Each of the aspect of the code such as governing
equations, numerical scheme, computational grids, initial and boundary conditions are
discusses in detail. The chapter ends with a small note on the Schlieren pictures, which are
used for visualization of unsteady pressure waves. In this research these pictures were obtained
Chapter 3 starts with a literature survey discussing in brief the current research on unsteady
supersonic cavity. Cavity oscillations mechanism suggested by other researchers are presented.
The cycle observed in current study is then presented and compared to those suggested by
13
other researchers. Both supersonic and subsonic results are presented. In the end a parametric
Chapter 4 discusses a method of modeling the cavity unsteadiness using source terms. It also
presents an elegant technique of calculating the source terms from unsteady solutions called
source terms in other applications. A method to calculate the source terms and there variation
with Mach number for both subsonic and supersonic cases is discussed. A parametric study of
variation of source terms with Mach number is included. Solutions obtained using source
terms and those obtained by time averaging unsteady solutions are then compared to
demonstrate the accuracy of source terms to model time averaged flow unsteadiness.
Chapter 5 suggests directions for future work in this research. The unsteady effect of the purge
cavity is shown by means of some preliminary results. A strip term source model used by
Chapter 6 concludes the report by summarizing the important findings of the research.
14
15
2. NUMERICAL ALGORITHM
Computational Scheme
The commercial CFD package CFD++ developed by Metacomp Technologies was used for
the numerical simulations. CFD++ is a highly versatile CFD solver. It has been used to solve a
large variety of problems such as: high speed reacting/non-reacting flows, low speed
preconditioning, multiphase flows - Eulerian with finite size distribution, conjugate heat
CFD++ is based on unified grid, unified physics and unified computing framework. A multi-
numerical oscillations in the computed flow-fields. These polynomials are exact fits of multi-
dimensional linear data. Approximate Riemann solvers are used to define upwind fluxes, with
preconditioned variants available for low-speed flows. CFD++ also features non-linear
variants, which automatically enforce entropy and positivity constraints. CFD++ features:
3. Finite-Volume Framework
16
5. Fast computation of unsteady flows
The conservation form of the governing differential equations currently supported by CFD++
∂q ∂ ( f i − f v ) ∂( g t − g v ) ∂( ht − hv )
+ + + =S
∂t ∂x ∂y ∂z
Where the vector q represents the dependent conservation variables; f, g and h represent the
fluxes in the three spatial directions; and S represent the source terms. The subscripts i and v
are the inviscid ad viscous terms respectively. For the Reynolds averaged Navier-Stokes
equations the dependent quantities and the inviscid fluxes can be written as follows
where e is the total energy; ρ is the density; u, v, w are x, y and z direction velocities
respectively. p is the pressure and σ’s represent the color tracer or turbulence transport
quantities such as turbulence kinetic energy and “undamped” eddy viscosity in the point wise
turbulence models. The infrastructure for inclusion of multiple species exists in the code. The
17
first five rows represent the standard Euler equations with the first being the energy equation
followed by the continuity and the three momentum equations. The equation of state couples
18
Where the Stokes theorem is assumed to hold true, thus relating the second coefficient of
CFD++ Framework
Unsteady compressible and incompressible Navier Stokes equations with turbulence modeling
including hybrids (unified grid). Unification of treatment of various cell shapes including
hexahedral, tetrahedral and triangular prism cells (3-d), quadrilateral and triangular cells (2-d)
and linear elements (1-d). 4) Treatment of multi-block patched aligned (nodally connected),
patched aligned (nodally connected, patched-nonaligned and overset grids (unified grid). 5)
19
versions for preconditioned forms of the governing equations. 7) Consistent and accurate
Pointwise turbulence models that do not require knowledge of distance to walls. 9) Versatile
types for the various sets of equations. 10) Implementation on Massively Parallel Computers
MPI, PVM etc. (unified computing). For the details of implementation in the software
program the reader is referred to Chakravarthy and Peroomian [11,12] and the CFD++
Software Manual.
Computation Grids
The problem setup consists of a flat plate with a rectangular cavity. The plate is 1 meter in
length. The region above the plate is extended to a depth of 0.25 meters. The cavity has a
length of 0.5 meters and depth of 0.25 meters. Thus the length to depth (L/D) ratio of the
cavity is fixed at 2. The region above the plate has been modeled as a rectangular structured
grid with (100x100) grid points. The cavity is attached to the passage grid and has (50x50)
points. The two blocks have been combined and the interface points deleted to form a single
unstructured block. The grids are clustered in the region above the cavity. The grids have been
produced using an algebric method. A C code has been written to create binary files for the
above-mentioned grid in a format suitable for CFD++. The flat plate and the walls of the
cavity are modeled as adiabatic viscous walls. The upper wall of the passage has been modeled
as a reflecting wall. The listing for the code are provided in Appendix II. Figure 2.1 shows a
schematic of the problem and the actual grids used are shown in figure 2.2.
20
U Upper wall
I O
n u
l t
e l
t e
X t
21
Computational Resources
To facilitate enough computational resources to carry out full-scale unsteady simulations a new
Beowulf cluster was setup. The cluster consists of 4 dual processor nodes with a total of 8 PIII
-450 Mhz Intel processors. Each node consists of 256 Mb of RAM and 4 GB hard disk. The
nodes are connected by means of a 100 Mbps fast Ethernet switch by means of a dedicated
network. Another machine with PII -450 MHz processor and 512 MB of RAM serves as the
master and the gateway. All systems run SuSE Linux and MPICH- an implementaion of MPI
message passing library. MPI is used for inter process communication by CFD++.
Boundary conditions for the supersonic computations were prescribed variables at inlet, no-
slip adiabatic viscous wall function on the solid surfaces, all variables extrapolated at the exit
plane. The upstream profile is uniform stream and the boundary layer is allowed to develop
before the flow enters the cavity. At the upper wall a symmetric boundary condition has been
used which causes the reflection of shock/expansion waves. For subsonic cases three variables
were specified and the fourth variable extrapolated. Backpressure was specified at the outlet
plane.
The initial conditions for steady computations were free stream velocities everywhere. Density
was initialized to a non-dimensional value of one. Energy was calculated from temperature.
For the unsteady run the solutions obtained from the subsonic simulations were used as initial
conditions. For steady simulation without the cavity the following was done to determine the
velocity at the cavity interface. Velocities were extracted from the time-averaged solutions
obtained from the unsteady simulations at the cavity interface. These velocities were then used
22
as initial condition at the interface of the cavity. Frozen wall boundary condition was then used
at this interface, i.e., the velocities were kept fixed at the interface of the cavity. This assures
that correct velocities are obtained at the cavity interface even when the cavity is not physically
Unsteady Calculations
Roe's scheme was used to provide proper propagation of signals. The scheme is second order
accurate in space and uses a Total Vanishing Diminishing TVD discretization to ensure
stability. Both explicit and point-implicit methods with relaxation were tested. The results
presented are for implicit calculations. The code was run viscous laminar although turbulent
In order to ascertain the convergence of the code to a steady cyclic behavior a pressure probe
was done at the hind wall of the cavity. Data was sampled after every iteration and a time
history of the pressure was obtained. A Fast Fourier Transfer (FFT) of this pressure signal was
then performed and the dominant frequencies identified. The code was run for 25- 30
characteristic time steps to allow the solutions to become periodic, which was also ascertained
from the FFT. Figure 2.3 shows the residual plot for an unsteady run. All four residuals are
seen to converge to a cyclic behavior. Figure 2.4 shows the FFT of the corresponding pressure
signal. One single frequency is seen to dominate the spectrum. The initial part of the pressure
signal during which the pressure converges to cyclic behavior is removed to avoid the initial
noise.
23
Figure 2.3: Residual plot for M=1.5
24
The time step (dt) is chosen carefully so that it is large enough to facilitate faster convergence
and small enough to be inside stability limits. The run starts with a smaller value of delta t and
is gradually ramped to larger values to facilitate faster convergence. One needs to be careful
because if the time step is large then the unsteadiness is not observed. This could be because as
the time step is increased it increases the damping of the scheme which suppresses the
unsteady oscillations.
In order to determine the sampling rate first the period of oscillation was determined from the
FFT of the pressure signal. The sampling rate was then chosen in such a way so that there
were about 20 data points over a complete cycle. Data was accumulated over approximately 4
cycles for each simulation. These results were then converted to a format suitable for
visualization.
Visualization Method
A commercial visualization package Tecplot was used for visualization of results. For
visualization of unsteady data successive images over a period of a cycle were converted into
animation. Density, pressure and vorticity contours were used for visualization. In order to
Schlieren images are obtained traditionally from experiment and represent the density gradient
in the direction perpendicular to the knife-edge. Schlieren pictures are very helpful in
recognizing the various unsteady compression and expansion waves. Experimental Schlieren
pictures thus have a preferred direction, i.e., They are either ∂ρ/∂x or ∂ρ/∂x depending on the
25
obtaining the absolute value of both the x and y density gradients. Figure 2.5 shows a
comparison between the computational simulated Schlieren and those obtained experimentally
by Zhang and Edwards [7]. More physics seems to be captured in simulated Schlieren pictures.
Figure 2.5: Comparison between simulated Schlieren obtained with those obtained
2 2
∂ρ ∂ρ
∆ ρ = +
∂x ∂y
These simulated Schlierens thus do not represent the same thing as the Schlierens obtained
from experiment. In certain aspects these simulated Schlierens are better as they represent
density gradient in both directions. Some results using above visualization technique are
presented in the next chapter, which aims at explaining the cavity resonance cycle.
26
3. UNSTEADY FLOW PHYSICS
Earlier Research
This chapter discusses the unsteady flow phenomenon associated with a subsonic/supersonic
two-dimensional rectangular cavity. Cavities are classified as one of the three types based on
their physical dimensions; they are open, closed or transitional. Closed cavities are typically
shallow with length to depth ratio (L/D) greater then 13. In this case the shear layer that
impinges the lower cavity wall produces two re-circulation zones essentially closing the
connection between the two cavity regions. Open cavities are short with L/D ratio less then
10. Open cavities are typical of those found in an aircraft. The flow phenomenon associated
with closed cavities is far more complicated because of the various interactions between the
pressure waves and shear layer and shed vortices. Because of the highly unsteady flow
phenomenon associated with these cavities, they are easier to visualize using numerical
time histories of the entire flow field with relatively less computational effort, which enabling
the better understanding of the physical mechanisms causing the cavity oscillations.
Researchers seem to be in some agreement that an oscillating shear layer exists and that the
shear layer drives the primary and secondary vortices residing in the cavity. The pressure
oscillations exist which cause the shear layer to shed. Coherent shed vortices, unsteady shock
and pressure waves, and interaction between the shed vortex and the vortex that resides inside
the cavity are present. Flow field characteristic appear to depend primarily upon the geometry
of the cavity, inlet Mach number and the thickness of the boundary layer.
27
Apart from computing the unsteady cavity flow field, a good amount of effort has been
directed in developing suppression devices to damp the oscillation of the cavity. Periara et al.
measured the time averaged velocity characteristics of different cavity trailing edge geometries
(sharp, nose shaped and round) using laser Doppler anemometry. Sarno [8] studied the effect
of pulsating flow injection at the leading edge of the cavity. Jeng [6] computed the effect of
porous walls at front bulkhead and rear bulkhead to suppress the cavity oscillations. All this
phenomenon.
Zhang and Edwards [7,13,14] have studied in detail various aspects of unsteady supersonic
flow over a cavity. A Reiman solver with k-omega turbulence model was used for their
studies. Inlet boundary condition consisted of boundary layer profile obtained from
experiment. Part of their work involved capturing major feature of the cavity using Adaptive
Mesh Refinement (AMR). They obtained results for length to depth ratio (L/D) for the cavity
varying from 1 to 5. They found the unsteady motion to be random or regular based on the
L/D ratio of the cavity. At short lengths (L/D =1-3) the measured pressure fluctuation was
highly regular and repetitive. For larger value of L/D >3 they found the measured pressure
fluctuation became increasingly irregular. Though the oscillations were dominated by a single
frequency, other harmonics of frequency started to appear with increase in L/D ratio of the
cavity. Also they reported mode switching with laminar boundary layer with same boundary
conditions as the earlier case. The mode switch is caused by shear layer instability and is
Tam et al. [4] have also performed extensive experimental and computational analysis aimed at
understanding the unsteady physical mechanism of the cavity flow oscillations They used
28
double thin layer Navier-Stokes (DTLNS) equations for their simulations. The upstream
profile was determined from flat plate computations from which the proper profile was
chosen by finding the x-location with same momentum boundary layer thickness. A complete
description of the mechanism can be found in the section on unsteady oscillation mechanism.
Reynolds averaged Navier-Stokes equations with an algebraic turbulence model have been
used in most analysis. More advanced model such as one- [12] and two- [13] equation model
have also been attempted, although with no better result. The general results indicate that a
wide variety of solutions can be obtained with various turbulence models, but that the large-
scale features are similar for all the models. Good results are typically obtained for time
averaged properties like surface pressure and shear stress. However the unsteady properties
like sound pressure levels and cavity resonance frequencies are not computed consistently.
Although a large amount of research effort has been directed toward understanding the cavity
flow phenomenon the parametric variation of these results with inlet parameters has not been
reported. Zhang and Edward report that there at certain length and time scale cavity
oscillations may be amplified by feedback from the shear layer and vice versa be damped at
certain other frequencies. In light of this finding parametric studies conducted in this research
aim at understanding the variation of oscillation cycle with inlet parameters. Another goal is to
establish the cavity oscillation mechanism identified with Tam et al. for the supersonic cavity.
This would both establish the results obtained earlier and build confidence in the accuracy of
29
Unsteady Cavity Oscillations
In order to explain the unsteady flow physics of the cavity various researchers have suggested
mechanisms for the cavity oscillations. Notable among them are efforts by Rossiter [15], Heller
and Bliss [1], Rockwell and Naudascher [2] and Tam et al. al. [3].
Rossiter derived and empirical formula for cavity resonance frequencies using dimensional
U m−γ
f =
L 1 +M
K
Where:
L is length ,
M is Mach number,
Rossiters formula predicts the dominant frequency and has been verified by various
researchers. Rossieter's model was derived using edge tone analogy and the assumption that
acoustic radiation is due to shed vortices impinging on the aft cavity wall. His experiments
visualized the shed vortices and the pressure waves external to the cavity. His work laid
foundation for more comprehensive theories regarding cavity resonance put forth by Heller
30
Equation 2 was later modified by Heller et al. to account for the recovery factor measured to
be close to unity rather then zero assumed by Rossiter. Thus, Rossiter’s formula was
improved for the higher Mach number range by assuming that the cavity sound speed is equal
to the free stream stagnation sound speed. The modified Rossiter’s formula becomes
U m −γ
f =
L +
1 M
K 2 1/ 2
M
1 + (γ − 1)
2
The mechanism proposed by Heller and Bliss [1] and cited by several other researchers,
defines the shear layer oscillation mechanism on the basis of steady compression waves. The
upstream traveling compression wave reaches the front wall of the cavity and reflects. The
resulting wave pattern in cavity causes unsteadiness in the shear layer. In turn, the shear layer
lotion is responsible for aft wall mass addition and removal that initially generated the cavity
internal wave structure. Figure 3.1 shows the process of feedback loop in the cavity.
1. The pressure wave from previous trailing edge disturbance reaches the front wall of
the cavity and reflects. Another wave, which has already reflected of the front wall,
approaches the aft wall of the cavity. At the same time the shear layer is deflected
above the aft wall and the fluid leaves the cavity.
31
2. A new compression wave begins to for at the aft wall as the flow impinges on the
trailing edge causing the fluid to enter the cavity. The compression wave reflected from
3. The shear layer which is now below the trailing edge of the aft wall, forma an upstream
traveling compression wave. The reflected wave from the front wall continues to move
4. The upstream and downstream compression interacts near the cavity center.
5. After interaction the waves continue to move in their respective directions. The
external part of the upstream traveling wave moves into the supersonic flow, thus
causing it to be more tipped then the external flow mach angle. The downstream wave
moves in the same direction as the external flow and travels at supersonic speed.
6. The shear layer is now at the trailing edge. The wave generated at the trailing edge
approaches the front wall of the cavity, and the downstream traveling wave approaches
the aft wall of the cavity. The oscillation cycle then repeats.
32
Figure 3.1: Heller and Bliss Cycle
33
Heller and Bliss using water table visualization technique by Shapiro [16], reported the pressure
fluctuation in the cavity. Tam et al. [1] used simulated Schlieren pictures to explain the cyclic
behavior. They reported important differences from the Heller and Bliss cycle. Their cycle is
1. Compression wave formed at the aft corner starts not from the previous cycle wave
but from the wave just formed below the shed vortex.
2. The downstream traveling compression wave that has reflected off the front wall
dissipates near the center of the cavity and does not strike the aft wall.
3. The upstream traveling compression wave does not have a component outside the
4. Although the shed vortex and the vortex impingement are observed little evidence
exists to suggest that the torn vortex is responsible for the pressure oscillations.
Further it seems to acts as a forcing function thus fixing the period of oscillation.
34
Figure 3.2. Schlieren images, Reference -Tam, et al. [1]
35
Vortex interaction Mechanism
Rockwell and Naudascher [2] provided an explanation of the cavity resonance cycle, which
stipulates that the shear layer oscillation is driven primarily by transient vortex motions within
the cavity. Both water channel visualizations and laser Doppler anemometry were used to
study the nature of this feedback mechanism. The flow visualization was achieved using
hydrogen bubble technique in the water channel. It should be noted that this visualization
technique is entirely different from the water-table visualization used by Heller and Bliss [1] A
provide different flow features. For example a water table is used to see the pressure waves but
not the shed vortices, and vice versa is true for water channel. In this Rockewell and
Naudascher were able to detect the shed vortices that occurred with thin neither the shear
layer nor the pressure waves. They proposed that the vortices which for with in the oscillation
shear layer sometimes impinge upon the aft wall and produce upstream propagating
disturbances within the cavity. These fluctuations travel forward and eventually displace the
shear layer at the leading edge. This produces and external excitation of the shear layer that
initiates and “locks in” the shedding of another vortex and closes the feedback loop show in
figure 3.3.
36
Figure 3.3. Vortex Interaction Mechanism
37
Results and Discussions
Supersonic Results
Simulated schlieren pictures were produced to visualize the unsteady cavity oscillation
mechanism. The test parameters are M=2.0, Re=3.69e+4, L/D=2. The complete cycle has
been divided into eight pictures and is presented in Figure 3.4. Visible in the picture is the
complex nature of wave interaction and propagation about the cavity. The important flow
features are a moving compression wave at the leading edge of the cavity, a bow shock wave at
the trailing edge of the cavity, an oscillating shear layer and vortex shedding. Velocity vectors
shown in Figure 3.5 were used in conjunction with the Schlieren pictures Figures 3.4 to
determine the exact mechanism of the cycle. The wave patterns observed are closely associated
with the vortex motion of the flow in the cavity. The appearance of these complex features is
believed to be closely associated with the amplification and damping of small disturbances in
the unsteady shear layer. The oscillating shear layer causes a compression wave to be formed at
the leading edge of the cavity. As the shear layer oscillates the shock waves forms and
disappears. As the shear layer rolls up it forms a vortex, which grows in strength as the fluid
enters the cavity. The entering fluid causes a pressure wave to form inside the cavity below the
vortex. The pressure wave reflects from the bottom aft corner and interacts with the vortex
causing it to shed. The shed vortex then travels downstream and impinges the aft lip of the
cavity and breaks into two. At the same time the fluid escapes the cavity completing the cycle.
Contours of temperature have also been plotted in Figure 3.6. The exact details of oscillation
38
Observed Oscillation Cycle
The results in Figure 3.4, 3.5 and 3.6 shows time histories of simulated Schlieren images,
vorticity contours and Temperature contours for one oscillations cycle respectively. The
complete cycle has been divided into eight parts and numbered 1-8 in the figure. The Schlieren
1. A newly formed vortex (v) is seen at the leading edge of the cavity. A compression
wave (c) is has fomed at the leading edge of the cavity. Another compression wave (b)
is formed at the aft lip of the cavity. This compression wave remains more or less
2. The vortex (v) becomes stronger and moves downstream. The compression wave (a')
from previous cycle has interacted with the vortex (v) and now dissipates. This wave
3. The shear layer impinges on the aft lip of the cavity. At this point the fluid can be seen
entering into the cavity in Figure 3.5. The entering fluid causes a compression wave (a)
to be formed just below the vortex and can be seen in both Schlieren pictures and the
temperature contours.
4. The fluid continues to enter the cavity and the compression wave (a) becomes
stronger. As the fluid enters the cavity the compression wave (c), which has been more
or less stationary, now disappears at the front wall of the cavity and moves upwards.
39
5. The vortex (v) impinges on the hind wall of the cavity and splits into two. Part of it is
washed downstream above the cavity while other half enters the cavity. A pressure
wave (p) is formed below the vortex that enters into the cavity. This wave now moves
downwards. A new compression wave (c’) is formed at the leading edge of the cavity.
Both the waves can be observed in the Schlieren pictures and the temperature
contours.
6. The pressure wave (a) is reflected from the bottom aft of the cavity and starts moving
upstream as (a"). The shed vortex is washed further downstream of the cavity.
7. The upstream moving wave (a”) starts interacting with the vortex at the leading edge
of the cavity. The pressure wave (p) reflects from the bottom wall and moves upwards.
8. Wave (a”) causes the vortex (v) to shed. The fluid moves out of the cavity causing the
shear layer to roll up and form a new vortex (v'), seen at the leading edge of the cavity.
The essential features of the observed cycle mechanism are the following,
1. It is seen that the compression wave seen below the vortex is formed because of
the fluid rushing into the cavity. This is an essential difference from the Heller and
Bliss cycle and is also observed by Tam, et al. in their cycle mechanism.
2. This wave then hits the bottom aft corner of the cavity reflects and moves
upstream until it hits the vortex causing the shear layer to shed. This is slightly
different from that observed by Tam, et al. who suggest that the pressure wave
40
formed below the vortex reflects from the aft wall, moves upstream and reflects
from the front wall at which time it interacts with the vortex causing it to shed.
This could be because the upper wall is represented as a reflecting wall, which
causes a reflecting compression wave to strike at the aft lip of the cavity. This is
3. This shed vortex then impinges on the aft lip of the cavity and breaks down and
moves partially into the cavity while the other half gets washed into the primary
flow.
MOVIE001 MOVIE002
41
MOVIE003 MOVIE004
MOVIE005 MOVIE006
MOVIE007 MOVIE008
42
MOVIE001 MOVIE002
MOVIE003 MOVIE004
MOVIE005 MOVIE006
43
MOVIE007 MOVIE008
TEMP000 TEMP002
TEMP003 TEMP004
44
TEMP005 TEMP006
TEMP007 TEMP008
Subsonic Results
For subsonic flow the vortex motion dominates the flow physics. The shear layer developed
over the boundary floats over the cavity. This shear layer rolls into a vortex, detaches and
moves towards the aft wall of the cavity. The vortex after impinging on the aft wall of the
cavity, breaks into two vortices, part of which enters the cavity and moves towards the lower
wall while other half gets washed into the passage. A time history of the oscillation cycle is
presented using series of Schlieren images in Figure 3.7, over the period of oscillation.
45
Although the formation and consequent shedding of the vortex can be observed in the
Schlieren pictures, the oscillations cycle is better observed by following the behavior of the
shed and resident vortices in the cavity. The vorticity contours are plotted over the period of
oscillation. In order to visualize the behavior of the vortex, the streamlines are plotted on top
of the vorticity contours. Figure 3.8 shows the vorticity and streamlines plotted on top of each
other using a set of pictures not essentially equally placed in time over the cycle. A cartoon has
been created and presented in Figure 3.9 to easily follow the vortices through the cycle.
Temperature contours are also presented in Figure 3.10 for completeness. A step-by-step
1. Two resident vortices are seen in the cavity. Vortex (a) stays more or less stationary
2. A new vortex (c) is created because of the large component of the Y velocity in the
cavity because of the resident vortex (b). It causes the vortex (b) to move towards the
3. Vortex (b) and (c) interact and forma new vortex now called (bc).
4. A new vortex (d) is formed which pushes the vortex (bc) towards the aft wall
5. Vortex (bc) splits after hitting the aft lip and breaks down into two vortices (b”) and
(c”). While vortex (b”) becomes resident in the cavity vortex (c”) gets washed into the
46
6. The resident vortex (c”) combines with the newly formed vortex (d) forming a larger
vortex (c”d).
8. Vortex (e) combines with vortex (c”d) forming a new vortex (b’), which is the resident
vortex (b) of the precious cycle. Vortex (a) is not seen at this stage.
Important feature observed in the cycle is that the resident vortex after combining with the
newly formed vortex may or may not escape the cavity. If the resident vortex is a fragment
of the split vortex then it just absorbs the newly formed vortex and stays inside the cavity.
Else if the resident vortex has already combined with a newly formed vortex earlier after
being split then it grows and splits forming two vortices again. Thus the vortex splits once
47
48
49
Figure 3.7: Schlieren pictures for subsonic cavity oscillations
50
MOVIE001 MOVIE002
MOVIE004 MOVIE005
MOVIE007 MOVIE009
51
MOVIE010 MOVIE011
MOVIE014 MOVIE015
MOVIE016 MOVIE017
Figure 3.8: Vorticity Contours and streamlines for subsonic cavity oscillations
52
a b b’
c e
a b
a dc
bc
a a dc
b”
d
d
bc a
a c“
53
Figure 3.9: Cartoon to explain the interaction of various vortices in the subsonic cavity
TEMP000 TEMP003
TEMP006 TEMP009
TEMP012 TEMP015
54
TEMP018 TEMP021
TEMP024 TEMP027
TEMP030 TEMP033
55
Dominant Frequency
The time history of pressure was taken along the wall surface at the trailing edge of the cavity
with data being taken at every iteration. There are a total of 200,000 sample data points. The
data is taken after the solutions have become cyclic so as to avoid initial transition. A Fast
Fourier Transform (FFT) was taken of the pressure signal. This dominant frequency has been
shown to be same throughout the cavity by Tam et al. [3]. These periods of oscillations were
Parametric Studies
The unsteady calculations were repeated for complete range of Mach numbers starting from
subsonic (0.3) to supersonic (1.9). For subsonic Mach numbers only single frequency seems to
dominate the oscillation mechanism. Figure 3.11 shows the FFT for M=0.3. A single
frequency is seen to dominating the pressure signal. As the Mach number is increased and
nears M=1.0 the pressure signal becomes less regular and more frequencies start appearing in
the FFT. An FFT of pressure signal for M=1.4 is seen in Figure 3.12. At supersonic Mach
numbers above 1.4 the oscillations are dominated by a single frequency. As the Mach number
is further increased, at M = 2.0 the spectra again starts showing high frequency component. A
plot of the pressure cycle is shown at M=2.0 in Figure 3.13. Although a single frequency is
seen to dominate the pressure oscillations a large amount of noise is observed in the later half
of the cycle. An FFT of the pressure signal shows the dominant frequency with large amount
56
Figure 3.11. FFT of pressure signal at aft wall, M=0.3
57
Figure 3.13: Pressure signal, M=2.0
58
Largest fluctuations were found for Mach number close to 1.5, which seems consistent with
Zhang et al. conclusions that certain length scales tend to amplify the disturbances more then
other. For larger Mach numbers smaller disturbances were observed. However the mechanism
59
60
4. DETERMINISTIC SOURCE TERMS
Deterministic stresses are analogous to turbulent stresses in that they are correlations between
the fluctuating flow quantities. The time scales of the “deterministic” motion are much larger
then the time scale of the “stochastic” (turbulent) motion. The time-average of the solutions
from an unsteady simulation will in general differ from the solution from a steady-state
simulation. These differences can be attributed to the effect of the deterministic stresses. If the
deterministic stress field is known, it can be used to compute source for the Navier-Stokes
equations. A steady state simulation now with source terms included will then yield the same
Lumped Deterministic Stress Terms (LDST) are easily calculated directly from time average
unsteady simulations. In this method, the values of individual deterministic stress components
are not computed; only there sum is deduced at each grid point from the time average of
unsteady simulations, hence the name “lumped deterministic stress”. Lumped deterministic
stress fields are computed from unsteady, viscous simulations of the flow for a variety of flow
conditions. Both inviscid and viscous effects contribute to the unsteadiness in the flowfield.
Since deterministic stresses are analogous to turbulent stresses, decomposing velocities into
mean and fluctuating components and substituting the decomposed velocities into Navier-
Stokes equations is a used as a starting point. In conventional Reynolds stress modeling, the
velocities are decomposed into mean and fluctuating components, and the stress represents
the molecular stress and turbulent stress. In the deterministic stress, the velocity fluctuations
61
are considered to have a random (turbulent) component and a deterministic component. The
deterministic fluctuations occur on a larger space and time scale than the random fluctuations.
The flow variables are first decomposed into a “deterministic velocity" and stochastic
~
uj = u j + uj'
fluctuation u’,
The deterministic velocity is further decomposed into a mean value and a deterministic
_
~ ~
u j = u j +uj"
fluctuation.
These decompositions are then substituted in the governing equations and then time averaged
,first over the time scale of the fluctuations of the turbulence and then over the frequency of
the deterministic fluctuations. This averaging procedure yields two additional stress terms,
Where the first term on the right is the conventional Reynolds stress and the second term on
Rewriting the energy equations for example, consider the general form of the unsteady energy
∂ Et ∂ [( Et + p )u] ∂ [( Et + p )v ]
+ + =0
∂t ∂x ∂y
62
Unsteady solution techniques solve this expression by moving the spatial derivatives to the
RHS, i.e.,
∂ Et ∂ [( E t + p )u] ∂ [( E t + p) v ]
= − +
∂t ∂x ∂y
This approach is also taken when solving for the steady state solution, where Equation (2) is
If we split the variables into time mean and fluctuating quantities, i.e.,
On time averaging the above equation the time average of the perturbed quantities become
zero. Therefore the second bracket on the right reduces to zero. Also because of equation (1)
the first bracket is equal to the remaining LHS. Which means that upon time averaging a flow
63
∂ [( E t + p)u ] ∂ [( Et + p )v ] ∂ [( Et' + p ' ) u ' ] ∂ [( E t' + p' ) v ' ]
0= − + − +
∂x ∂y ∂x ∂y
Bracket one is the usual residual solved by relaxation methods for the steady state (see
Equation (2). Bracket two represents the source terms that must be added to the steady state
equations to include the effect of unsteadiness. Note the following: The solutions obtained
from Equations (2) and (5) are different since they respectively represent the steady state and
time average solution variables. The source term can be found by taking the time mean of the
unsteady solution variables. These variables are then inserted into Equation (5) to solve for the
lumped deterministic stresses that are the second bracket terms. Once these source terms are
calculated they can be inserted into the unsteady equations to include the unsteady effects due
to the cavity.
Previous Research
Use of source terms is not new to Computational Fluid Dynamics but only a few applications
exist in which they are used in a manner suggested here. Some of the applications will be
Orkwis, et al. [17] have used source terms to impose boundary conditions. This is sometime
useful when developing implicit schemes in which the structure of Jacobian matrix changes
because of inclusion of boundary conditions. The new matrix is considerably more difficult to
solve using already existing numerical algorithms. Suitable alterations are then needed to solve
the new system of equation compromising efficiency of the code. By implementing boundary
64
conditions as source terms only explicit side of the scheme is altered leaving the implicit side
unaltered. This unaltered matrix has a simplified matrix structure, which can be solved more
efficiently without requiring modification to the numerical algorithm used to solve the system
of equations.
An even more clever application exists in turbomachinery for the problems with coupled
stator-rotor blade rows. Traditional multistage design is based on mixing plane approach which
represents the simplest formulation. The mixing plane approach employs an arbitrary interface
plane located roughly midway between the neighboring blade rows across which the spanwise
information is transferred from one blade domain to another. Flow variables on both sides are
circumferentially averaged. The accuracy of this method is clearly in question because of loss
of spatial information resulting from circumferential averaging. Adamczyk et al [18] and later
Hall [19,20] have shown that it is possible to correct this difficulty by means of a time averaged
calculation by including the effect of moving (non-moving) blade rows as a source term in the
non (moving) bladerow. Hall’s deterministic stress model is designed to operate within the
framework of mixing-plane approach and uses unsteady flow physics to model the source
terms. The model assumes that the important time-dependent flow features manifest
themselves as two primary mechanisms. First, wake shed from upstream row and propagating
through the downstream rows. Second, unsteady velocity fluctuations caused due to potential
The solutions obtained using the deterministic stress field incorporated with compared with
experimental results and time average of the unsteady results. Spanwise profiles for mixing-
plane deterministic stress calculations were found to be in good agreement with the test data.
65
There was no significant increase (less then 5%) in CPU time due to addition of the
A more recent application of source term has been in simulation of gas turbine combustor hot
streak.. Experimental data taken from gas turbine combustors indicates that the flow existing
the flow exiting the combustor has both circumferential and radial temperature gradients.
These temperature gradients have a significant effect in the wall temperature of the first stage
rotor. A combustion hot streak which has temperatures typically twice the free stream
temperature, has a greater stream wise velocity and than the surrounding fluid and therefore a
larger positive incidence angle to the rotor as compared to the free stream. Due to this
incidence variation through the hot streak and slow convection speed on the pressure side of
the rotor the hot streak typically accumulates at the pressure surface on the suction side. For
steady state computation the tangential component of the hot streak at the exit of the stator
are flux averaged and only the radial variation in the rotor frame are retained. It has been
shown earlier Orkwis, et al. [21, 22] that the tangential variation in the hot streak is of prime
importance in establishing the hot streak migration path. By mixing out the tangential variation
at the rotor inlet, the steady state computations do not model the temperature segregation in
the blade passage nor produce correct temperature distribution on the blade surface.
Busby et. al. [23] and Orkwis et. al. [21] have used lumped deterministic stress (LDS) to model
these effects in a time averaged manner in steady simulations. The LDS model obtained by
Busby used an inviscid model on a single stage, three-dimensional turbine and vane blade with
hot streak entering at mid-pitch and mid-span. The LDS model reproduces the time-averaged
temperature distribution to within 2.3% while the steady viscous has an error of 8,4 percent.
66
The solution using the LDS model is obtained at a cost in CPU time that is 26 percent of that
cavity. The solutions are allowed to become cyclic, which is ascertained by probing a solution
variable. A Fast Fourier Transform of the pressure signal is then carried to determine the
period of the oscillation. The solution variables are sampled at discrete locations over the
period of the cycle. To determine the time average of the unsteady simulations, the solution
variables are time-averaged over a known period of oscillations. This time-average solution is
then put back into the governing equations to determine the residuals. These residuals as
These terms are useful in understanding the nature of the density, X, Y momentum and
energy. As seen in the figure, the red terms are the positive and the blue terms are the negative
source terms. When these terms are added to the right hand side of the respective Navier-
Stokes equations, the sign of the terms is reversed. Thus a red source term would mean that
particular quantity is being removed from that location and put elsewhere where the source
terms are negative. Thus a red energy source term above the cavity and a blue energy source
term below the cavity indicate that the energy is being removed from the region above the
cavity and added to the cavity. As was discussed earlier the source term represent the
difference between the time average unsteady solutions and the steady solutions. They
67
represent the effect of the unsteady flow features that are not seen in a steady solution. It is
possible to observe most of the unsteady flow phenomenon that are observed in Schlieren
pictures of unsteady cavity cycle presented in the previous chapter. All the unsteady waves
seen in the Schlieren image – the compression wave seen at the leading edge of the cavity, the
bow shock seen at the aft edge of the cavity and the oscillating shear layer – are observed in
The compression wave appears as positive source terms along the shock. As a stationary wave
would not appear in a source term plot, the presence of the compression wave in the source
term plot verifies that the wave is unsteady. Also note that the waves are seen in all four-source
terms, demonstrating all four quantities change as the shock moves. The presence of small
magnitude energy source term for the unsteady shock implies that small amount of energy
dissipates through the moving shock wave. This is probably because of the shock wave
curvature, which creates the additional vorticity. The vortex formed because of the oscillating
shear layer is seen as concentrated positive and negative source terms juxtaposed. The
impinging vortex causes a pressure wave to move downwards at the aft wall of the cavity. The
shed vortex is best seen in the energy source terms because of the energy dissipation in the
vortex region. The path of the vortex is not seen as clearly as it is seen for a subsonic case
The unsteady calculations are repeated for a range of Mach numbers from (0.3-2.0). For Mach
number (0.9-1.4) the source terms are not obtained, as there are a large number of frequencies
present, which make it impossible to identify a frequency for time averaging. All four source
terms have been plotted for the entire range of Mach number in figure 4.2-4.5. The same
general pattern is observed with regard to source term location although the strength of the
68
source terms changes with Mach number. The shock waves are not seen in the energy terms
indicating that there is no energy redistribution in the cavity. The pressure wave formed above
and below the cavity and responsible for the shedding of the vortex can be best seen in the Y-
Momentum plot.
Density X-Momentum
Y-Momentum Energy
69
1.5 1.6
1.7 1.8
1.9
70
source2.1.5 source2.1.6
source2.1.7 source2.1.8
source2.1.9
71
source3.1.5 source3.1.6
source3.1.7 source3.1.8
source3.1.9
72
source4.1.5 source4.1.6
source4.1.7 source4.1.8
source4.1.9
73
Subsonic Source Terms
The source terms obtained for the subsonic case are different from the supersonic case since
the flow physics governing them are different. The appearance of expansion and compression
waves complicates the source terms for the supersonic Mach numbers. The supersonic source
terms are dominated by the unsteady compression waves while the subsonic source terms are
dominated by the vortex shedding phenomenon. The unstable behavior of the oscillating shear
layer and consequent shedding of the vortex govern the unsteady subsonic cavity oscillation.
All the four source terms, namely, density, X,Y-momentum and energy are plotted in Figure
4.6. As seen in the Figure 4.6, density and X-momentum are very small as compare to other
two. Small density source terms imply that mass is not entering or leaving the cavity. This is
different from what is observed for the supersonic cavity for which the fluid moves in and out
of the cavity every cycle. It can be observed in velocity vector plots presented in the previous
chapter. This is expected as the subsonic flow is governed by the vortex shedding
phenomenon as opposed to the supersonic case where shock waves dominate. This causes the
fluid to swirl inside the cavity without any effective mass transfer - or resulting density source
terms. The vortex motion was also observed in the streamline plot in Figure 3.11 of the
previous chapter. The Y-Momentum source terms are small compared to the energy source
terms. The energy source terms appear as a cloud of positive source term above the cavity and
negative source terms below the cavity. This is expected as the moving vortex causes the
energy to re-distribute within the cavity. The energy is removed from the main flow and added
into the cavity. This addition provides energy for the cavity flow oscillations.
The trail of the moving vortex can also be clearly seen in the energy source term plot. The
vortex impinges on the trailing edge of the cavity and then breaks into two. A part of the
vortex is seen to enter into the cavity while the other half gets washed with the flow above the
74
cavity. Also there are concentrated positive and negative source terms juxtaposed and placed at
The calculations are repeated for Mach number 0.3-0.8. The Figure 4.7 shows the Y-
Momentum source terms and Figure 4.8 shows the energy source terms. The density and X
momentum source terms are not plotted, as they are almost zero. As seen in the Figure 4.7 and
4.8 the strength of source terms increases with the Mach number. This is expected since the
pressure waves are stronger at higher Mach numbers because of increased compressibility. The
path followed by the vortex, as seen in the energy source term plot, changes with Mach
number. For higher Mach numbers the vortex hits the aft lip of the cavity and breaks into two
parts. Part of the broken vortex moves into the cavity and the other half gets washed into the
main flow. This is better seen at higher Mach numbers because downward moving pressure
wave created at the aft lip of the cavity becomes stronger and pushes the vortex down towards
75
1 2
3 4
76
source3.0.3 source3.0.4
source3.0.5 source3.0.6
source3.0.7 source.3.0.8
0.3 0.4
77
0.5 0.6
0.7 0.8
Figure shows a comparison between the unsteady time average solutions obtained by including
the cavity in the simulation and the results of steady simulations obtained by including the
source terms without the cavity. This is done for the complete range of Mach number form
0.3 to 0.8 and 1.5 to 1.9. The match obtained is exact. This is observed for all subsonic and
supersonic cases tested and are shown figure 4.9 for subsonic cases and figure 4.10 for
supersonic cases. If the region of the cavity is covered with a sheet of paper and the flow in the
region above the cavity in the following figures it is easy to see that the two flows are identical.
A 2-D plot of flow quantities at the cavity interface shows that the two plots lie on top of each
other. Therefore a comparison of 2-D flow quantities is not represented in the current
78
discussion. This is expected as the source terms used are obtained from unsteady simulation
and are exact. This clearly demonstrates that the source terms obtained using Lumped
Deterministic Stress Technique are exact. Also it demonstrates that source terms can be used
to model cavity unsteadiness and that a potential exists for using “modeled” to represent the
cavity.
79
density.0.5 Density with sources.0.5
80
density.0.8 Density with sources.0.8
density_M=1.5 density_M=1.5_with_sources
density_M=1.6 density_M=1.6_with_sources
81
density_M=1.7 density_M=1.7_with_sources
density_M=1.8 density_M=1.8_with_sources
Density_M=1.9 density_M=1.9_with_sources
82
5. DIRECTIONS FOR FUTURE WORK
It was demonstrated in the previous chapter that the source terms obtained using the lumped
deterministic source term technique can be used to exactly include the unsteady effect in
steady simulations without any significant computational overhead. These source terms were
however calculated using unsteady solutions which in the first place is what we are trying to
model. Thus without a method to model these source terms the current approach is just an
academic exercise. The next step at this stage is to develop a technique to model these source
terms. A general source term prediction would require a parametric scaling to account for the
appropriate flow physics variations. The exact functional form of the source term is not clear
in a general sense at this stage, although insight into the variation of source terms with Mach
number can be obtained by observing the source terms for a simplified geometry.
These source terms can be modeled to a varying degree of complexity. A more elaborate
model would no doubt be very accurate, but a lower order model may be more practical to
implement and applicable to a wider range of problems. A building block approach needs to
be used to determine the level of complexity required for modeling the source terms correctly.
The simplest model consists of a strip of source terms put at the interface of the cavity. Fitting
a curve through the experimentally obtained data identifies a functional form of the source
term. The coefficients of the parametric form are then calculated from the experiments.
Hunter, et al. [20] used strip source terms calculated from detailed
experimental/computational analysis to capture wheel space purge flow seal effects in low
pressure turbines. The source terms are introduced directly into the layer of cells above the
83
cavity-gaspath interface. The results from detailed analysis of the bladerow with the hub cavity
are utilized to define the source terms at the interface. The source term model is conceptually
straightforward since the mass, energy and three components of momentum are defined
explicitly at the gap location. These quantities can be extracted over the period of oscillation
and averaged to obtain time mean quantities. These terms can then be used as source terms for
steady analysis.
Hunter et al. [20] used two levels of modeling with different level of complexity; Uniform
Source Term Model and the Distributed Source Term Model. The first level of source term
modeling applies a uniform distribution of mass, momentum and energy sources at the cavity-
gaspath interface. The cavity entrance and the exit gap locations are specified and the uniform
sources are applied to the first layer of cells. The profiles of mass flux, the three components
of velocity and turbulence quantities at the entrance and exit interface are calculated. The
profiles were then averaged in the axial direction to obtain the value applied to the uniform
source term at the cavity entrance. Also they used a distributed source term model in which
the source terms were calculated based on the quantities at the interface of the passage and
cavity. No time averaging was done in the axial direction for this case. They obtained results
that matched more closely with experimental value when they used distributed source term
As found in the figures shown in the last chapter the actual source terms even for a simplified
cavity are distributed inside the cavity and in the region above the cavity. An approach using
strip source terms placed along the interface of the cavity only applies the boundary condition
correctly. It does not capture the unsteady flow features such as moving shock waves and
vortex shedding phenomenon. An attempt to capture this phenomenon requires the source
84
terms to be distributed inside the cavity and in the region above the cavity. A problem with
using field source terms is that they require more elaborate modeling procedure.
An approach currently in investigation by Lukovic et al. [21] approximates the field source
terms using Neural Networks. This approach does not rely on experiments to determine the
value of coefficients for the functional form of the source terms. Instead unsteady solutions
are used to calculate deterministic source terms for a wide range of inlet parameters. The
density, three momentum and energy source terms are calculated using the Lumped
Deterministic Stress Technique. A neural network is trained using the source terms calculated
in this way. Once the neural network is trained, it can be used to approximate the source term
for any arbitrary set of parameters. A slightly better approach is to identify a parametric
functional form by observing the behavior over a range of parameters. A neural network can
then be employed to determine coefficients that describe the source terms. The neural network
can be trained to create the desired general form provided a reasonably large set of test cases is
available. Starting from a small number of neurons and then increasing the number until they
are sufficient determine the number of neurons and the amount of training required. This is
determined by inserting the source terms generated using the neural network and performing
steady simulation that can be compared to unsteady time averaged solutions to determine
accuracy.
Although it has been successfully demonstrated in the current research that the deterministic
source terms can capture the flow unsteadiness exactly, the current research and the following
research to model these source terms is only an academic exercise. The work assumes
importance in light of the fact that these rectangular cavities are simplifications of the purge
85
To understand the difference between the unsteady rectangular cavity and a turbine purge
cavity flow unsteadiness both steady and unsteady simulations were carried out for a realistic
purge flow cavity. The geometry consists of a single stage rotor and stator connected by a
purge cavity. This cavity exists because of the clearance required between the stationary and
moving parts – the stator and the rotor respectively. The grid used to model the geometry is
multi-block unstructured and consists of approximately 80,000 grid cells. A commercial code
Star-CD was used for the simulations. The code was run viscous with wall functions to model
the wall boundary layer. Second order discretizations were used for both time and space
derivatives. Cyclic boundary conditions were used to simulate the blade row. The code was run
on an SGI Origin 2000 machine at the Ohio Supercomputing Center. Both 3-D steady and
unsteady simulations were performed. Solutions form steady solutions were used as initial
Figure 6.1 shows the velocity contours in two planes; one in the plane of the blade and the
other perpendicular to it. The results presented are still preliminary results and calculation of
source terms for this case requires considerable more effort along this line. The preliminary
results suggest that the unsteady effect of this cavity is not limited to the cavity region alone
but is also felt in the primary flow path. This is where the current research becomes of
significant importance. It has already been demonstrated that the deterministic source terms
can model the unsteadiness exactly for a rectangular cavity. If source terms can be
calculated/modeled for the purge flow cavity they will provide a convenient and cost effective
86
Figure 5.1: Velocity Vectors in Plane of the Cavity
87
88
Figure 5.3. Velocity magnitude contours, zoomed in cavity region
89
6. CONCLUSION AND DISCUSSIONS
Unsteady calculations are performed for flow over a 2-D subsonic/supersonic rectangular
cavity. Simulated Schlieren pictures are used for visualization of unsteady flow. The
calculations are repeated for Mach numbers (0.3-0.8) in subsonic and (1.5-1.9) in the
supersonic range.
The flow mechanism governing the unsteady pressure oscillation cycle of the rectangular cavity
are compared with those suggested by other researchers. The research verifies the oscillation
cycle mechanism established by Tam, et al. for a supersonic cavity. The essential features of the
1. It is seen that the compression wave found below the vortex is formed because of the
fluid rushing into the cavity. This is an essential difference from the Heller and Bliss
2. This wave then hits the bottom aft corner of the cavity and continues to move
upstream until it hits the vortex causing the shear layer to shed.
3. This shed vortex then impinges on the aft lip of the cavity and breaks down and
moves partially into the cavity while the other half gets washed into the primary flow.
4. As the shear layer breaks down the fluid rushes out of the cavity.
5. The driving mechanism of the oscillation is the upstream moving compression wave,
90
6. It was observed that the compression wave seen below the vortex also observed by
Tam, et al., appears to be formed because of the fluid rushing into the cavity.
The cause for oscillation for a subsonic cavity is slightly different from the supersonic cavity.
For subsonic cavity the oscillation cycle is explained better by following the resident and shed
1. The changing behavior of the resident vortex after interacting with the newly formed
vortex causes the oscillation to “lock in” and cause the cavity to oscillate.
2. A resident vortex may or may not get squashed against the aft lip and break into two
3. If the resident vortex is a fragment of the split vortex then it combines with the newly
formed vortex and stays resident inside the cavity. Else if the resident vortex has
already combined with a newly formed vortex earlier after being split then it grows and
splits again.
4. Thus the vortex splits once every two times it combines with a newly formed vortex.
A parametric study was performed to study the variation of pressure oscillation cycle with
1. The flow physics governing the subsonic cavity is very different from flow physics of a
supersonic cavity
91
2. Subsonic cavity oscillations are primarily caused because of vortex shedding and its
consequent breaking up into two parts after hitting the back wall of the cavity.
3. For supersonic Mach number the oscillation cycle mechanism does not change
4. For Mach number near sonic 0.9 to 1.3 there is no clear dominant frequency observed
The unsteady solutions were time averaged and used to calculate the deterministic source
terms
1. These source terms represent the difference between the steady solutions and the
unsteady time averaged solution. These source terms represent the same unsteady flow
2. Deterministic source terms when used in steady simulations without the cavity capture
the effect of the cavity. This was demonstrated for the entire range of Mach number
studied by comparing the unsteady time averaged solutions with the steady solutions
Term Model used by Hunter, et al. [] to model trench cavities and 2. Neural Network
92
Direction for Future Work
1. The feasibility of source term modeling needs to be further demonstrated by using the
source terms to model more realistic turbine purge flow cavity geometry.
93
REFERENCES:
1. Heller, H., Bliss, D., 1975, “ The Physical Mechanism of Flow-induced Pressure
Fluctuation in Cavities and concepts and Concepts for their Supression”, AIAA
Paper 75-491.
2. Rockwell, D., Naudasher, E., 1979, “Self Sustained oscillations of Impinging Free
Shear Layers”, Annual Review of Fluid Mechanics, Vol. 11, pp.- 67-94.
3. Tam C. J., Orkwis P.D., Disimle P.J., 1995, “A Comparison of Several Standard
Turbulence Models for 2-D Open Cavity Flow Field Computations”, AIAA 95-0361
4. Tam, C.-J., Orkwis, P.D. and Disimile, P.J., “Supersonic Open Cavity Flow Physics
5. Kim, I., Chokani, N., 1992, “Navier-Stokes Study of Supersonic Cavity Flowfield
6. Jeng, Y. N., Wu, T, J., 1992, “Numerical Study of Supersonic Open Cavity Flow
Cavity Flows Driven by Thick Shear Layers”, Aeronautical Journal, November, pp.
365-374
8. Zhang, X., Edward, J. A., 1990, “An Investigation of Supersonic Oscillatory Cavity
Flows Driven by Thick Shear Layers”, Aeronautical Journal, December, pp. 355-364.
9. Hunter, Scott, "Source Term Modeling of Endwall Cavity Flow Effects on Gaspath
94
10. Hunter, S.D. and Orkwis, P.D., Endwall cavity flow effects on Gaspath
Aerodynamics in an Axial Flow Turbine: Part II: Source term model development,
13. Zhang, X., Edward, J. A., 1990, “An Investigation of Supersonic Oscillatory Cavity
Flows Driven by Thick Shear Layers”, Aeronautical Journal, December, pp. 355-364.
14. Zhang, X., Edward, J. A., 1995, “Analysis of Unsteady Supersonic Cavity Flow
pp.373-393.
15. Rossiter, J.E., 1964, “Wind Tunnel Experiments on the Flow Over Rectangular
No. 3438.
16. Shapiro, A. H., 1954, “Free Surface Water Table”, Physical Measurements in gas
Dynamics and Combustion” Edited by Ladenburg, R. W., Lewis B., Pease, R. N.,
Taylor, H. S.
17. Orkwis, P. D., Tam, C. J., Disimile, P. J., “Observation on Using Experimental Data
as Boundary Conditions for Computations”, AIAA Journal, Vol. 33, No 1, pp. 176-
178.
18. Adamczyk, J. J., 1984, “Model Equation for Simulating Flows in Multistage
95
19. Hall, E. J., 1997, “Aerodynamic Modeling of Multistage Compressor Flowfields –
21. Orkwis, P.D. and Turner M.G., "Linear Deterministic Source Terms for Hot Streak
22. Orkwis, P.D., Turner, M.G., Barter, J. W., “Deterministic Source Terms for Turbine
Hot Streak Applications Derived from Linear Unsteady Solutions”, AIAA A99-
34144.
23. Busby, J., Sondak, D., Staubach, B. and Davis, R., “Deterministic Stress Modeling of
24. Sondak, D.L., Dorney, D.J. and Davis, R.L., “Modeling Turbo-machinery
Unsteadiness with Lumped Deterministic Stresses,” AIAA Paper 96-2570, July 1996.
steady cavity simulations: Part 2 ", 39 th Aerospace Sciences Meeting and Exhibit,
Jan 2001.
26. Jeny, Y, N, Payne, U. J., 1995, “Numerical Study of Supersonic Open Cavity Flow
and Pressure Oscillation Control”, Journal of Aircraft, Vol. 32, No. 2, pp.336-369.
27. Zhang, X., Edward, J. A., 1988, “Computational Analysis of Unsteady Supersonic
Cavity Flows Driven by Thick Shear Layers”, Aeronautical Journal, November, pp.
365-374..
28. Giles, M., “An Approach for Multistage Calculations Incorporating Unsteadiness”,
96
97