Esplora E-book
Categorie
Esplora Audiolibri
Categorie
Esplora Riviste
Categorie
Esplora Documenti
Categorie
FACOLTÀ DI INGEGNERIA
Tesi di laurea in
INGEGNERIA MECCANICA
RELATORE: CANDIDATO:
RINGRAZIAMENTI………………………………………………………………………………5
INTRODUZIONE.............................................................................................................. 8
CAPITOLO 1 ....................................................................................................... 10
CAPITOLO 2 ....................................................................................................... 22
1
CAPITOLO 3………….....……………………………………………………………….39
BIBLIOGRAFIA……………………….……………………………………………….………..51
INDICE DELLE FIGURE…...…………………………………………………………..……...52
ELENCO DEI SIMBOLI………………………………………………………………………...85
2
CONTENTS
ACKNOWLEDGEMENTS……………………………….……………………………………..5
INTRODUCTION ............................................................................................................. 8
CHAPTER 1 ........................................................................................................ 10
CHAPTER 2 ........................................................................................................ 22
3
CHAPTER 3………………...……………………...…………………………………….39
CFD CALCULATION........................................................................................... 44
BIBLIOGRAPHY…………………………………………………...…………………………..51
TABLE OF FIGURES………………………………………………………………………….52
NOMENCLATURE………………………………………………………………………...…..85
4
RINGRAZIAMENTI
Vorrei ringraziare il Dott. Volker Guemmer il quale mi ha dato la possibilità di scrivere la mia
tesi di laurea specialistica presso la Rolls-Royce Deutschland. Sono grato a lui per la pazienza
e per indicazioni che mi ha dato.
Nel periodio trascorso all’interno del dipartimento “ Compressor and Fan Aerodynamics” ho
conosciuto persone gentilissime. Devo ringraziare il mio amico Domenico Berterame per
avermi aiutato.
Voglio esprimere la mia sincera gratitudine al Prof. Vinicio Magi, ordinario presso
l’ Università degli Studi della Basilicata per la fiducia riposta in me. Nel corso dei miei studi
universitari mi è stato possibile imparare molto da lui ed apprezzare il suo “humour”.
ACKNOWLEDGEMENTS
I should like to thank Dr. Volker Guemmer who gave me the opportunity to write my master
thesis in Rolls-Royce Deutschland. I am grateful for his patience and guidance.
During the time I have spent in Compressor and Fan Aerodynamics department I met very
kind people. Gratitude is due my friend Domenico Berterame for his help.
I express my right and dutiful thanks to Prof. Vinicio Magi , Full Professor at University of
Basilicata, for the confidence he placed in me. During my studies at university I had the
possibility to learn so much from him and to appreciate his sense of humour.
5
INTRODUZIONE
6
• capitolo 2: si affronta la questione riguardante il metodo con descrizione della
procedura sviluppata nel corso dei mesi trascorsi in Rolls-Royce e di quella
convenzionale adoperata;
7
INTRODUCTION
The present work has been developed during an internship at Rolls-Royce Deutschland.
The host company is a world-leading provider of power systems and services for use on
land, at sea and in the air, operates in four global markets: civil aerospace, defence
aerospace, marine and energy. Rolls-Royce Deutschland is a subsidiary of aircraft
engine maker Rolls-Royce plc with facilities at Dahlewitz outside Berlin and at Oberursel
near Frankfurt am Main.
Nowadays the most aircrafts engine utilizes axial flow compressors having the big
advantage of the high flow rate per unit area. This kind of turbo-machine is a multi-stage
unit as the amount of pressure increase by each stage is small. A stage consists of a
row of rotating blades followed by a row of stator vanes able to impart force and
moment to the flow.
The latest trend in axial compressors design has been going towards a reduction of
both the number of stages and blades. A consequence of this is an increase of the
loading of each aerofoil, which now looks smaller and smaller, with a reduction of the
span wise dimension towards a higher axial development, and this imposes greater
attention to the 3D phenomena and to the secondary flow effects.
To handle these issues, there are conventional design techniques for the whole
machine, but also new technologies that allow subsequently improving the performance.
The present research work tries indeed to have a preliminary investigation about one of
these innovative solutions. The idea is to avoid the boundary layer detachment in highly
loaded aerofoil, energizing the surface flow by a high velocity jet, which is created in a
passive way, taking a portion of the air from the pressure side of the blade and
accelerating it in a convergent duct through the blade thickness.
Since the conventional Rolls-Royce aero-design process is not containing proper
capabilities for such an investigation, almost half of the internship timescale has been
spent for developing and validating a new procedure, which in the end has been proved
to deliver reliable results.
Afterward a CFD campaign has been carried out and the outcome of this analysis
shows that the considered technology could be quite beneficial: in the case of stators it
was possible to keep the efficiency level unless a blade number decrease by 35% and
also for rotors interesting improvements were achieved in terms of tip vortex reduction.
Obviously further studies are needed before a concrete use of this kind of devices, but
so far the idea itself seems to be promising, some knowledge about the physical
phenomena has been achieved and important milestones about the design rules have
been fixed.
The present work is divided in three chapters:
8
• Chapter 3: results about a selection of the investigated cases and conclusions.
9
CHAPTER 1
T = m(r2Vϑ 2 − r1Vϑ1 ) ;
W = ω (r2Vϑ 2 − r1Vϑ1 ) .
This, or its derivatives below, is known as the Euler equation for turbo-machinery.
Sometimes it is written in terms of the local blade speed:
W = U 2Vϑ 2 − U 1Vϑ1 ;
And for the special case of axial machines where the flow enters and leaves at the
same radius:
10
In special cases, for example when there is casing treatment to delay the stall, there
may be large effective stresses at the walls and an expression such as the equation
above may give a misleading estimate for the blade work.
In an axial compressor it is often possible to consider flows in rotor passages by merely
adopting a moving frame of reference.
The objective of a compressor is usually to raise the static pressure and this means that
there must be a rise in static enthalpy.
if U1=U2, as it might in an axial machine, then it follows that:
h2 − h1 =
2
(
1 2
W1 − W22 ; )
as is usual for any stationary passage or diffuse. The corresponding expression for the
static pressure rise only valid for loss-free incompressible flow is:
p 2 − p1 =
1
2
(
ρ W12 − W22 . )
If there are losses in incompressible flow the static pressure rise can be written as:
p 2 − p1 =
1
2
(
ρ W12 − W22 − Δploss . )
for compressible flow more care is necessary and it is convenient to begin with the
thermodynamic relation:
dp
dh = + Tds ;
ρ
so that on integrating:
2 dp 2
h 2 − h1 = ∫1 ρ
+ ∫
1
Tds .
The losses for adiabatic flow are contained in the entropy rise.
The process can be shown graphically; fig.1.c is for an axial blade row in which (h0)rel is
constant. With no losses the compression is isentropic and the outlet static pressure is
p2s; with losses the static pressure at outlet will be lower at p2. The losses, by which is
meant the rise in entropy, are produced by processes associated with flow, mainly
shear work (sometimes called viscous dissipation) and mixing of the flow.
In the case U2>U1 some of the static enthalpy and pressure rise comes only from the
( )
term U 22 − U 12 / 2 and since this is unconnected with the flow process it does not have
loss-making processes associated with it and gives rise to no losses and no entropy
increase. The losses tend to increase as the amount of deceleration of the relative flow
is increased and also in proportion to the cube of the relative velocity; in other words the
11
( )
loss may be expected to be related fairly directly to W22 − W12 / 2 and not to the overall
change in static enthalpy h2-h1. The significance of this is that is that part of enthalpy
( )
rise attributable to the change in blade speed, U 22 − U 12 / 2 , is essentially loss free.
Furthermore this can be increased without aerodynamic limit, unlike the enthalpy rise
produced by decelerating the relative flow where excessive reductions in velocity lead to
flow separation. It is these two factors which have favoured the use of radial
compressors: if most of the static enthalpy rise is attributable to the change in blade
speed between inlet and outlet the expected pressure rise will be obtained and the
efficiency will be reasonably high even if the aerodynamic behaviour is poor with large
regions of separated flow. In fact the principal limit on the maximum pressure rise from
radial compressors is the strength of the material from which the impeller is made.
On axial machines radius changes may have important effects as well. For a typical
axial compressor stage the static enthalpy Δh might be approximately 0.4U2, where U is
the local blade speed. Suppose that between inlet and outlet to the rotor the distance of
the streamline from the compressor axis increases by 10%. Than it follows that the
( )
quantity U 22 − U 12 / 2 increases by about 10% of U12 too. In other words a small change
in radius can produce “free” changes in static enthalpy of the same order of magnitude
as those produced by the deflection and deceleration of the flow in the blades.
The changes are free because they are without losses and do not contribute to the
tendency of the boundary layer fluid to separate. This often has led to a pronounced
effect at the hub of axial compressors.
cos (γ 1 + φ1 )
tan (χ 1 )s = tan (χ 1 )c − sin φ1 tan ε 1
cos γ 1
Clearly an exactly equivalent expression can be created for the trailing edge or for flow
angles. This expression is solely a geometric relation and gives no consideration to the
fluid mechanism.
• stagger ξ
• solidity σ
• camber angle θ
• camber line shape (usually a circular arc)
• thickness-chord ratio t/c
• thickness-distribution
The succinctness of the term solidity, more common in American usage, makes it
preferable to the equivalent pitch-chord ratio, s/c=σ -1.
The most important variables are the first two, followed by the camber angle. The last
three only become very important for the overall performance as the inlet Mach number
rises and the flow starts to have supersonic patches. An additional geometric variable
that is sometimes important is the surface roughness, characterized by the roughness
height ks. The LE radius may be thought of as a part of the definition of profile shape but
13
sometimes it is considered as a separate variable. It has a large effect at high Mach
numbers.
Although at the TE the direction of the blade is denoted by χ2; the flow does not leave in
this direction. The deflection ε =α1-α2 is then smaller than that which would be implied
by the blade exit angle. The loss in deflection is usually referred to as deviation that is
defined by:
δ = α2 − χ2 ;
this quantity has the advantage for correlation purposes of being fairly small. Defined
in the same way but with very different significance, is the incidence.
i = α1 − χ1 ,
being the angle between the mean flow direction into the blade and the projection of the
camber line at the LE. This is not the same as the definition of incidence used in
aeronautics, referred to here as the angle of attack A, which is the angle between the
inlet flow direction and the chord line. The chord line is inclined to the axial direction by
the stagger angle ξ so it follows that:
A = α1 − ξ
If the camber angle of a blade θ is known or can be estimated and if the camber line can
be approximated by a circular arc, then it is easy to see that the incidence can be found
from i=A-θ/2.
AVDR = ρ 2V x 2 / ρ1V x1 ,
which matters. This abbreviation is used even when the density change is negligible.
Given these aerodynamic inputs or constraints it is possible to decide what is
acceptable or realistic value of outlet flow angle, α2, and the deflection, ε = α1-α2. From
this, it is possible to make decisions about solidity, stagger and camber, as well as
thickness and the distribution of thickness and camber.
The object of the blades is to produce a rise in static pressure or a deflection of the flow
and normally one effect is necessary for the other. The rise in static pressure may be
convenient non-dimensionalized by the dynamic pressure to the blade row:
14
c p = ( p2 − p1 ) / ( p01 − p1 ) ;
⎛1 ⎞
c p = ( p 2 − p1 ) / ⎜ ρV12 ⎟ .
⎝2 ⎠
Neither the deflection nor the pressure rise coefficient is itself a sufficient description of
the blade loading: blades of high stagger can produce little deflection but large pressure
rise whereas for blades of very low stagger a high deflection can be accompanied by
almost no pressure rise.
For all blades there is a loss in stagnation pressure and it is again convenient to non-
dimensionalize the loss by the inlet dynamic pressure so that:
ω1 = ( p 01 − p 02 ) / ( p 01 − p1 ) ;
where p02 denotes the average stagnation pressure measured a short distance
downstream from the blades (usually the mass average value of p02 is used). For most
purpose the blade losses are not of first importance. What matters far more is that the
blades should produce the deflection specified so that the flow leaves the blade inclined
at the desired outlet angle α2.
The universal use of lift and drag coefficients in aeronautics led to their employment in
compressor cascades and fig.1.e shows an isolated blade row whit the axial and
tangential forces on the blade shown as X and Y respectively.
It is assumed that the fluid can be treated as incompressible and the axial velocity in
and out taken as equal. Conservation of momentum then shows that the axial force
applied by the blade to the gas is:
X = ( p 2 − p1 )s ;
X /s =
1
2
( )
ρ V12 − V22 − Δp0 ,
Y = ρVx s(V y1 − V y 2 )
With constant axial velocity it makes sense to define mean direction and velocity by:
15
Vm = V x sec α m ,
1 ⎛ V y1 + V y 2 ⎞ 1
tan α m = ⎜ ⎟⎟ = (tan α 1 + tan α 2 ) .
2 ⎜⎝ V x ⎠ 2
By analogy with isolated aerofoil theory the lift force L can then be defined as the
resultant force perpendicular to the mean velocity and the drag force D as that parallel
to the mean velocity. Thus:
L = X sin α m + Y cos α m .
Δp0 sin α m
CL =
L
=
2
[tan α1 − tan α 2 ]casα − ,
1
ρVm2c σ m
1
ρVm
2 σ
2 2
D = Y sin α m − X cos α m ,
that reduces to
D = sΔp0 cosα m .
D Δp0 cosα m
CD = = .
1 2 1 2 σ
Vm c Vm
2 2
V1 = Vm cos α m / cosα1.
Hence:
16
Δp0 1 cos3 α m ω cos3 α m
CD = = .
2 σ cos α1 σ α
2 2
1 cos
ρV1 1
2
CL =
2
[tan α1 − tan α 2 ]casα − CD tan α m .
σ m
Since CL/CD is typically greater than 40, and tanαm is not very different to unity, it often
suffices to neglect the last term in the equation above. With the loss small, i.e. CD«CL,it
is also easy to show that :
CL = (2 / σ )(ψ / φ )cos α m ,
where ψ = Δh0/U2 and Φ = Vx/U are the stage loading and flow coefficients. By
considering a closed contour around a blade it is easy to see that the circulation around
an individual blade in the cascade is found to be:
With the restriction of constant axial velocity and the neglect of the total pressure loss
term it is possible to write for the lift:
L = ρVm Γ ,
17
difference between the outlet flow angle α2 and the blade outlet angle χ2 referred to as
the deviation defined by δ= α2 – χ2 (Fig.1.d).
The deviation is predominantly inviscid effects to which the boundary layer fluid makes
only a small additional contribution. Out across the passage the flow is inclined to this
direction, the sense of the inclination being that reduces the force on the blades. For
axial blades the deviation is given approximately by δ= 0.3θ√(s/c) where θ is the
camber, s is the blade pitch and c is the chord. The expression for deviation allows
some simple generalization for axial blading. The outlet flow direction is given by:
α2 = ξ– θ/2+ δ;
where ξ is the blade stagger, the inclination to the axial direction of the chord, the line
joining the LE and the TE. Introducing the estimate for the deviation for a solidity c/s ≈
1.0 gives:
α2 = ξ– θ/2+ 0.3θ;
so that:
α2 = ξ – 0.2θ.
In other words the outlet flow direction depends to only a fairly small extent on the
camber whereas it is the stagger angle ξ, which really has a big effect. At low inlet mach
numbers most axial blades area able to tolerate quite a large incidence range, so again
the camber is of secondary importance. At high inlet Mach numbers the blade
performance is strongly affected by incidence and the overriding dependence on
stagger to the relative excursion of camber is no longer so true. Stagger remains a very
important variable at high Mach numbers.
For most purposes it is possible to ignore the unsteadiness by working in a frame of
reference fixed to the component under consideration: for stator blades a coordinate
system is used which is stationary (absolute frame) and for rotor blades the frame of
reference moves at the local blade speed (relative frame).
18
1.2.6 STAGE AND BLADE LOADING
The enthalpy rise of a stage is related to the square of the rotational speed Δh/U2 where
U can either be taken at the blade tip speed or the speed at mid-radius, the latter being
quite common for axial machines. The enthalpy change can be the static or stagnation
enthalpy, depending on the context, though stagnation is more common. For
compressors in which the pressure rise is small compared to the absolute pressure,
such as low-speed machines, the density may be reasonably approximated as constant
and it is convenient to define the pressure rise coefficient:
Δp/ρU2,
ψ=Δh/U2;
although sometimes ψ is used for Δp/ρU2, it follows that for incompressible operation:
Δp/ρU2 = ηΔh/U2,
where η is the efficiency, and both non-dimensional groups are very similar in
magnitude and often in their trends. The ratios Δh/U2 or Δp/ρU2 provide a measure of
the actual input to the potential work available, i.e. to U2. Clearly the demand on the
stage is more taxing if a large enthalpy input is required from a low blade speed
machine than a high-speed one and the magnitude of ψ gives a measure of this.
V2 ΔVϑ
DF = 1 − +
V1 2σV1
where V1,V2 are the average velocities into and out of a blade row in a frame of
reference fixed to the blade, ΔVθ is the change in whirl velocity in the row and s is the
solidity, equal to blade chord/blade pitch. Values of DF in excess of 0.6 are thought to
indicate blade stall and a value of 0.45 might be taken as a typical design choice. Over
the last few years attention has been focussed more on the endwall region as the limit
19
for loading and the weight given to the diffusion factor has decreased. The criterion to
be
adopted for endwalls loading or pressure rise is less clear, mainly because the fluid
mechanics is still not understood.
Methods analogous to that produced by de Haller (1953) are still current; de Haller
deduced that the velocity out of a blade should not be less than about 0.75 times the
inlet velocity in the performance is to be satisfactory. This is equivalent to requiring that
the static pressure rise at the wall should not exceed about 0.44 times the dynamic
pressure into the a blade row. The de Haller criterion has not been found to be entirely
satisfactory. A more recent method relates the pressure rise capability to the mean
height (i.e. mid-span) solidity averaged over the stage; it is based on a large number of
measurements in multistage compressors.
The most common method of assessing what is acceptable loading at the wall is
probably by reference back to previous designs by the same manufacturer, it is now
very rare for an organization to be designing an axial compressors for the first time. The
general view seems to be that a stage pressure rise not exceeding about 0.4rU2 is
reliable.
The limit on the maximum Mach number is flexible and depends to a large extent on the
balance between high efficiency and high pressure ratio per stage being sought. The
loss in the efficiency with Mach number is nowhere near as a serious as was once
thought. As the Mach number is increased the operating range reduces, i.e. the
difference between the mass flow for choke and surge is reduced. An important reason
for keeping the speeds of industrial compressors down is to maintain the widest
possible operating range.
Decisions have to be taken regarding the blade chord and the number of blades.
Increasing the chord reduces the aspect ratio (height/chord) and increases solidity
(chord/pitch) for the same annulus and number of blades. These trends are evident in
Fig. 1.f. The rise in solidity and fall in aspect ratio can both be attributed in the main to a
rise in chord length. With these trends for aspect ratio and solidity there is the striking
rise in pressure rise per stage and the increase in the overall pressure ratio possible
and utilizable for a single compressor.
It should be emphasized that the single most important decision in the design process is
the choice of a realistic stage loading. An over-ambitious choice may lead to untold
problems later with possibility of actually achieve the combination of efficiency, pressure
ratio, mass flow and range originally intended.
Back in the 1950s it was believed that the trend would be towards high aspect ratio
blades to give a short compressor, mainly, it seems, because the blade behaviour well
away from the endwalls was comparatively well understood and this was the direction of
development which consideration of the blades seems to indicate. The trend was
reversed mainly because large chord blades are more effective in the endwall regions
and it is these regions which are crucial in determining both the efficiency and the stall
point.
High aspect ratio blades were long and thin and had atrocious vibration problems.
The change towards low aspect ratios was not the result of an understanding of the
processes involved but consideration of the trends for performance of different designs.
There are several performance goals to be compared, in particular pressure rise,
efficiency and operating range (operating range might be defined as the ratio of the
difference between maximum and minimum mass flow to the design value).
20
The evidence suggests that for a good compressor near the design point the efficiency
tends to be slightly lower if the solidity is on the high side (and the aspect ratio low) but
the pressure rise and operating range are greater.
The major trend over the last 30 years has shown a rise in efficiency but a more marked
rise in overall pressure rise.
There are special problems that arise from combining stages to form multistage
compressor, usually referred to as matching. The ability to handle the matching of
compressor and the operation of several rows of variable stagger stator blades has
made possible the very large increase over the years in the pressure ratio for a single
compressor spool.
Increasing Mach number by increasing rotational speed can lead to mechanical
problems. The limiting condition for a compressor with large pressure ratio is normally
reached at the rear hub; this is largely a materials problem connected with the high
temperatures. High solidity blading exacerbates the problem because of its greater
mass of blade metal.
Increased rotational speed makes it possible to increase the flow per unit area. With the
emphasis on blade design for an axial compressor it is easily overlooked that the overall
meridional flowpath (that is the flowpath in a longitudinal cross-section showing axial
and radial components) has a crucial effect on the design and the performance of a
compressor. The aerodynamic problems are, for example, greatly relieved if the hub
radius can increase from front to back, whilst they are made worse if the annulus area is
too large towards exit.
Fundamental to all the aerodynamic design are the basic decisions of an aerodynamic
nature. At the blade mid-height (sometimes known as the pitch-line radius) a choice
must be made for the local flow coefficient Φ=Vx/U and the stage loading ψ=Δh0/U2 (or
alternatively Δp0/ρU2). Sometimes the degree of reaction R=Δhrotor/Δhstage (or the
equivalent in terms of static pressure rise) is treated as important too. Such decisions
are separate from choice of solidity, blade section etc., although solidity does have a
marked effect on the choice of loading.
21
CHAPTER 2
This approach was designed and developed for conventional compressor geometries,
therefore is not showing enough capabilities for the actual study, which intends to
investigate the effect of slots on the blade surfaces.
In order to introduce this kind of devices, create a new 3D shape of the aerofoil and grid
the resulting geometry, has been necessary to substitute the first two steps of the
previously described method by the following operations:
Later on, after carrying out this last point, it has been possible to go back to the
traditional path staring from the third step.
The features of each program involved in both procedures will be analyzed in the details
thought the following paragraphs.
Obviously running computations based on grids obtained with different mesh
generators, jumping from a structured approach to an unstructured one, could have an
effective influence on the final results and drive to significant discrepancies. As reported
in the following chapters, sensitivity studies about this issue have been also produced
and calculations were made on the same geometry meshed by the two considered
programs. The outcome of such an investigation shown minor differences in the CFD
predictions, therefore the proposed modification to the traditional method was
considered absolutely reliable and validated.
22
2.2 PADRAM
Padram stands for “Parametric Design and Rapid Meshing”. It is able to change
parametrically the geometry and rapidly mesh it. There are two methods used to
produce suitable computational grids for the geometry that is changing during an
optimization process.
In the first one the mesh is perturbed to accommodate the new geometry whilst the
second approach relies on a completely new mesh to be generated. Although, the
former is quite popular, it is the author’s experience that producing good quality viscous
meshes in three-dimensions for large assemblies is very difficult. This is particularly true
when the geometry is changing significantly from its datum shape for example as part of
a heuristic optimization run. When generating new meshes not only the topology but
also the grid density needs to be kept as close as possible to the original geometry in
order to minimize the errors in computing the flow sensitivities. This is particularly
difficult to control when using unstructured grid generators.
Furthermore, both the aforementioned techniques require a relatively long
computational time to generate large meshes for complex three dimensional geometries
making them unsuitable for the inclusion in automatic optimization loop. Padram allows
obtaining a good quality viscous mesh by an automatic methodology. It requires few
minutes. Padram makes uses of both transfinite interpolation and elliptic grid generators
to create hybrid “C-O-H meshes” (Figure 2.cc, Figure 2.dd), which means structured O-
meshes around the blade surface and unstructured meshes in the other grid blocks. An
orthogonal body-fitted “O-mesh” (Figure 2.ee) is used to capture the viscous region of
an aerofoil whilst an “H-mesh” is used near the periodic boundaries and where
stretched cells are required for example in the wake. The mesh is independently
generated for every stream section, hence no mapping is required to transfer the
meshes radially. This ensures the good quality meshes are created at every height even
if the geometry is considerably changing from hub to tip.
The “O-mesh” is primarily used for the blade and in particular should be extended to
capture the viscous regions of the aerofoil. The “C-mesh” is used for splitter, and semi-
infinite bodies, such as the Pylon and RDF. The “H-mesh” is used in the passage to link
the “O-mesh” to the periodic boundaries and is used for the upstream and the
downstream blocks.
Padram generates a mesh on the unwrapped-plane (θ-m plane) of each stream-section
that is read in and then stack-up in the radial direction to produce the so-called “master
mesh”. The CFD mesh is then interpolated from the “master mesh” to achieve a
particular number of radial planes.
23
2.3 THE HYDRA SOLVER
Is useful give an overview of the HYDRA user suite. The main elements of this suite
are:
• JM52 for pre-processing mesh files and setting up HYDRA input files;
• JM54 for generating multi-grid levels;
• JM56 for multi-passage and multi-stage applications including casing treatment;
• JL09 for post-processing and extracting quantitative data from unstructured mesh
solutions.
There is also a Visual 3 based post-processor directly linked to HYDRA called “spy”.
JL09 is the preferred post-processor, but spy is able to access a small number of data
arrays (e.g. y+) that are not currently available within JL09.
The following sections give a brief summary of each of the main codes in the HYDRA
suite.
2.3.1 M52
JM52 is the primary preprocessing tool for the HYDRA suite. The program may be run
interactively using the “Graphical User Interface” (GUI) or in script mode, which enables
commands to be defined in a simple text file, which controls the execution of the
program written using a simple scripting command structure. The scripting option is
particularly powerful for driving the HYDRA calculations through an optimizer or a job
scheduling script.
The main function of JM52 is to translate meshes from a number of in-house and
commercial mesh formats into HYDRA format for subsequent use in JM56. The data
formats that can be read by JM52 are: SLIQ, CFDS, JH01, AU3D, SC03 (.ffm files),
CINDY, ITP g2d, DPLOT, JA63, Fluent (v5.6), HYDRA, Centaur, Plot3D, CGNS,
Triangle, UNSFLO and NEWT (.mcv files). In addition to the HYDRA grid format, mesh
data may be output in a number of alternate formats: CGNS, Plot3D, JH01 and
Fieldview Unstructured. Plot3D and JH01 files may only be created for structured grid
data.
JM52 may also be used to initialize the flow data. A number of different options are
available; a simple 1D axial profile defined by the user, a uniform flow condition defined
by the user, linear interpolation from an existing SWIPE Boundary Conditions file or by
reading an existing flow solution from one of the data formats defined above.
Flow solver control parameters such as CFL number, number of multigrid levels,
number of multigrid cycles, etc. may be defined. JM52 recognizes the multi-zone
structure used by HYDRA for multi-stage and multi-component calculations and allows
the flow to be initialized either on a zone by zone basis or for the entire mesh. The mesh
can be checked using the JM52 diagnostic checks. These diagnostics check for
negative cell volumes, cell degeneracies (i.e. cells with multiple entries of the same
node index in their connectivity table), mesh periodicity (for periodic boundary cases
only), cell connectivity (i.e. all cells are a valid part of the mesh having only one
neighbouring cell or boundary face for each individual cell face) and for point
24
redundancy. In GUI mode, a mesh viewer is available. Meshes may be rescaled in any
direction by a user defined factor and the rotation axis of the system may be changed.
HYDRA requires that axis of rotation is the x-axis and it may be necessary to change
the system axis to conform to this convention.
JM52 is used to define the boundary conditions for the solver run. This includes the
definition of the boundary surface types as inviscid or viscous walls, inflow or outflow,
periodic surfaces, etc. Boundary conditions such as fixed wall temperatures and relative
rotational speeds, inflow and outflow boundary profiles can also be defined or may be
taken from previously generated boundary condition files or directly from SWIPE
Boundary Condition data. For structured grids j-periodicity may be enforced to ensure a
perfectly periodic mesh.
Circumferential lines for radial equilibrium and/or mixing plane calculations in HYDRA
can be generated within JM52. The distinction between JM52 and JM56 is that JM52 is
primarily a single passage/component tool for setting up and initializing HYDRA using a
variety of mesh formats. At the end of JM52, a complete set of input for running the
steady HYDRA solver is generated and the user is able to launch the calculation.
JM56 is the multistage/multicomponent design system. It works entirely within the
HYDRA data structures using constituent meshes and flow files that have been pre-
processed by JM52. JM56 reads all the HYDRA input files directly and so accesses all
the HYDRA multi-grid mesh levels and flow field with a single command. At the end of
JM56, a complete set of input for running steady multi-stage calculations with mixing
planes or unsteady stage calculations with sliding planes is generated.
JM52 also provides the facility for setting up flutter, forced response and acoustic
calculations, defining wake shapes from existing profiles and mode shapes. JM52 also
sets up the input needed to run the adjoint version of HYDRA.
The meshes that are written by JM52 contain only the primitive mesh connectivity data.
The derived data, such as the edge connectivity and edge weights, are computed in
JM56. Hence, even for ingle component meshes, JM56 must be executed after JM52.
The version 6.05 for JM52 and 6.07 for JM56 must be used since the following versions
of are not able to write a proper HYDRA mesh matching PADRAM and CENTAUR
output meshes.
2.3.2 JM56
For all meshes, JM56 is employed to compute:
• Derived connectivity data, such as edge connectivity and edge weights, needed
by HYDRA.
• Sequence of coarser meshes for the multigrid solver.
25
HYDRA mesh and flow solution files are read into JM56. The graphics window for JM56
displays the surface mesh for the grid which has been read and any other meshes
already imported into the system.
After reading the mesh the user may import it into JM56, this two step process allows
the user an opportunity to check that the mesh that has been read is the correct data,
until it is imported the mesh is not part of the multistage geometry and may be rejected.
Once meshes have been imported, the user is then able to build multistage models by
importing further meshes into JM56 and is also able to construct multipassage models.
File output is controlled by the user. The user must actively select which mesh zones
are to be written.
JM56 will strip-out a user specified number of K-planes from the original structured grid
lying between the outer casing wall and the blade tip.
The interface region between the two mesh regions is designated as a Sliding Plane
boundary condition.
26
2.3.4 NON-LINEAR BOUNDARY CONDITIONS
There are 3 ways of specifying boundary conditions for non-linear HYDRA calculations:
3. Prescribed values at every boundary node using the array boundary conditions
stored on the initial solution file (this corresponds to the internal HYDRA array
qb).
Global values are used in 2D cases or for freestream boundaries where the boundary
conditions are uniform over the boundary. Profiles are the standard boundary conditions
for 3D calculations and allow the boundary conditions to vary as a function of one of the
coordinate directions. Prescribing boundary conditions at every node on one or more of
the boundaries is available through JM52 and allows conditions that are not simple 1-D
profiles (e.g. 2-D profiles) to be defined.
27
2.3.6 PROFILE BOUNDARY CONDITIONS
The standard 3D boundary conditions are defined as functions of one of the coordinate
directions and are specified using a separate file. The data contained on the file
depends on the type of boundary condition to be applied. However, the principle behind
the file format and the way it is handled within hydra is common.
Inviscid walls are identified by setting the group type value in input.dat to 1. Inviscid
walls do not require any boundary conditions and so they are specified by a single line
in input.dat. Inviscid walls do require the definition of the direction normal to the wall in
order to apply the flow tangency condition. The array of normal vectors is pre-computed
in JM56 and stored on the mesh file.
Viscous walls are identified by setting the group type value in input.dat to 2.
When checking the values of y+ it should be remembered that this is a function of near-
wall velocity and so checking the initial y+ field is highly dependent on the initial guess
(in fact if the initial guess has zero velocity, then the initial y+ values are also zero).
However, a check of the initial y+ field can often reveal disparities in the near-wall mesh
spacing on different surfaces (e.g. blade and endwalls). A similar near-wall spacing on
all viscous surfaces leads to mesh cells at the junction of two viscous surfaces which
have an aspect ratio close to unity. Aspect ratios much bigger than unity at corners (e.g
28
blade and endwall or blade surface and blade tip) have been observed to cause
convergence problems in hydra.
Must be checked that on viscous walls the distance is zero and close to a viscous wall
the contours follow the shape of the viscous wall.
For 3D annular calculations, the inflow total temperature and pressure are specified in
the frame of reference defined by omega_bc on the input.dat file. This is relative to the
rotation of the zone to which the boundary belongs. For 2D and 3D non-annular
calculations the inflow total temperature and pressure are specified in the absolute
frame. For calculations using SI units, the total temperature must be specified in
degrees K and the total pressure in Pascal; otherwise, they are non-dimensional.
For inviscid or laminar calculations the only inflow conditions needed are total
temperature and pressure and the two flow angles. For calculations using the Spalart-
Allmaras turbulence model, the user must prescribe inflow values of the Spalart variable
at each point on the profile. The Spalart variable is the last variable on each line of
the inflow file. The value typically chosen is 0.17616E-03; this corresponds to an inflow
turbulent viscosity which is 10 times the laminar viscosity.
For calculations using the k–ε turbulence model, the user must prescribe inflow values
of k and ε at each point on the profile. k and ε then are the last variables on each line of
the inflow file.
For Cartesian cases, the inflow angles are specified using the same convention as for
freestream boundaries. The annular test cases (or Cartesian cases with the inlet profile
29
specified as a function of radius) the inlet flow angle is specified in terms of a tangential
(or whirl) angle, α and a radial (or pitch) angle, β (Figure 2.ff).These angles are in the
frame rotating at -zone + -bc. The whirl angle is the angle in the x-y plane (or x-µ plane
for annular cases) between the velocity vector and the x-axis. The pitch angle is the
angle in the x-z plane (or x-r plane for annular cases) between the velocity vector and
the x-axis. Within hydra, α corresponds to the variable atinl and β corresponds to the
variable arinl. Important: the sign convention for α is opposite to the convention for θ.
This is to be consistent with the standard conventions used in the ROLLS-ROYCE
system. Hence, if qx is the axial component of velocity,
q y , qϑ = q x tan α
q z , qr = q x tan β
(
q = q x 1 + tan 2 α + tan 2 β )
1
cx =
1 + tan α + tan 2 β
2
tan α
cϑ =
1 + tan 2 α + tan 2 β
tan α
cr =
1 + tan 2 α + tan 2 β
cos α cos β
cx =
cos 2 β + cos 2 α sin 2 β
sin α cos β
cϑ =
cos β + cos 2 α sin 2 β
2
cos α sin β
cr =
cos 2 β + cos 2 α sin 2 β
30
hydra solves internally using Cartesian coordinates the direction cosines in the r and µ
directions, are converted into y and z direction cosines at each boundary node, using
the following expressions:
ycr + zcϑ
cy =
r
− ycϑ + zc r
cz =
r
where y, z and r are the coordinates at each boundary node. This is simply the
expression for the transformation of velocity components, but with the sign of the µ
component reversed to match the convention for the sign of the whirl angle.
There are four options for fixing the mass flow through either an inflow or outflow
boundary:
31
When specifying a fixed flow rate for multistage turbomachinery calculations, the value
should correspond to the flow per passage (or the number of passages used in the
calculation if there are more than one).
Fixed mass outflow with prescribed exit pressure: in input.dat, the boundary type must
be set to 19 for boundary group(s) with fixed mass outflow and a prescribed pressure.
Fixed mass outflow with radial equilibrium pressure: In input.dat the boundary type must
be set to 22 for boundary group(s) with fixed mass outflow and a radial equilibrium
pressure profile.
Fixed mass flow guidelines: it is not well-posed to use solely fixed mass inflows and
outflows, even if the specified mass flows match exactly, in a calculation. There must be
at least one standard inflow or outflow in order to establish a pressure level in the
calculation. Specifying a fixed mass flow boundary type is best done using JM52.
However, the fixed mass flow boundaries do not require any additional topological
information and so existing inflow or outflow boundaries can be converted to fixed mass
boundaries by modifying the input.dat and boundary condition files.
The hydra formulation of the mixing plane boundary condition works in terms of primitive
variables. Circumferentially flux averaged values of density, axial velocity,
32
circumferential velocity, radial velocity and static pressure are transferred from the
inflow boundary of the mixing plane to the nodes on the outflow boundary and vice
versa. At the nodes on both boundaries the perturbations of the transferred flux
averaged primitive variables from their local values are transformed into perturbations in
the 1D characteristic variables using the following equation.
Where un and ut are the normal and tangential velocities to the boundary. On the inflow
boundary the fourth characteristic is an outgoing wave so its perturbation is set to zero
and on the outflow boundary the first three characteristics are outgoing waves so their
perturbations are set to zero. The modified perturbations in characteristic variables are
then transformed back into perturbations in primitive variables using the following
equation.
These perturbations are then added to their local values to set primitive variables at the
boundary nodes which are then used to calculate boundary fluxes.
As given, the mixing plane boundary types are 11 and 12 for the standard mixing plane.
The language convention used by hydra is that boundary type 11 corresponds to the
mixing plane at the inflow boundary to a blade row. Similarly, boundary type 12 is the
mixing plane at the outflow to a blade row. The boundary indices in input.dat identify the
type of a mixing plane, but do not pair the inflow and outflow planes. The pairing is
defined via the separate boundary condition files for each group.
It is, therefore, important that the pairing indices in the input.dat file are modified
accordingly. In general, it is better to let JM56 create a new set of input files than to
make all the changes manually.
33
2.4 MATLAB CODE
The Matlab code, completely programmed at Rolls-Royce Deutschland, is capable to
read in the so called “Blade_Definition” file, traditionally employed in the company’s
methods for storing the geometry information, split each section profile in a number of
arbitrary segmentations chosen by the user, drive him/her through the slots definition
and save the outcome in the ”Knowledge fusion” format.
In the following paragraphs is proposed a briefing description on how those functions
are implemented by the code.
34
The algorithm behind does not actually interpolate sections but in fact the segment
parameters. This is achieved by an explicitly defined polynomial for each of the 24
parameters of every segment. Depending on the number of sections utilized by the
interpolation process, the order of the polynomial is increased. If are used only the first
and last section as nodes, results in a polynomial of the first order (straight line). Using
n nodes, results in a polynomial of the order n-1.
2.5 UNIGRAPHICS
Unigraphics, well known commercial CAD software, is employed in the actual method
for generating the final 3D slotted geometry to be meshed starting from the
segmentations created by the Matlab code and saved in the ‘knowledge fusion’ format.
The following paragraphs give a detailed description of the necessary operations.
35
deleted.
Sometimes the operation does not only split a surface but also deletes one of the
resulting parts. A solution to this is to use a different sequence of segments for the split
operation (figure 2.d). Another bug that occurs occasionally is the creation of a set of
surfaces instead of several single surfaces. The problem with this is the absence of the
possibility to delete single surfaces. Also the boundaries of the surrounding control
volume need to be changed.
As in every aerodynamic study, the solid we wish to create is not the blade itself but the
a volume with the blade inside as a cavity. To achieve this we need to create a hole in
the casing and/or hub surfaces (figure 2.e). Using the available surfaces for this will not
succeed (no continuous intersection). It is more practical to use splines that were
created to form the surface at the hub and casing and extrude them in radial direction
(figure 2.f).
At this point a number of additional operations can result beneficial for preparing the
geometry for the next step, the grid generation by Centaur . This software is unable to
mesh a body that contains surfaces with very high curvature and cope with surfaces
that describe a loop.
As a consequence of this, it is necessary to split in (at least) two parts the original blade
and all the segmentations. The easiest way for that is to employ a surface provided by
default which roughly follows the mean camber lines of every section (figure 2.g) and is
extended beyond the profiles in such a way to allow an intersection with the blade itself
and also an optional fillet.
The first thing that has to be checked is the existence of every surface that is needed to
perfectly describe a solid body. Large gaps or surfaces poking out are not allowed. The
unity of all surfaces has to create an absolutely watertight domain.
Sometimes is necessary increase slightly the tolerance in the sewing dialog, but this
can cause many problems in the mesh generation process. There is a global option that
might prevent solid bodies from being created. Increasing the tolerance is a quick
solution but not the best one. This step is important to avoid curves and surfaces
interfering with the mesh generation.
• use four panels at LE, TE, PS, SS: in this case it is possible to split the blade by
a Unigraphics tool that allows an isoparametric trim along the splines which
defines the radial sections and obtaining a proper panel overlapping in Centaur.
The main problem of this procedure is that not all Unigraphics tools can be
utilized to modify the geometry, because many of them produce the disappearing
of some panels when the IGES file is loaded in Centaur. In theory the mesh
generator should be able to create new panels using the contour lines but usually
they create problems in the griding process. In the end this restrictions impose to
36
create many intersections and new surfaces by extrusion useful for the trim tool
(figure 2.j, figure 2.k).
• use three panels at LE, PS, SS: this way is much easier. Using the Isoparametric
Trimming tool, the PS and SS surfaces are created by a plane generated in the
Matlab code. This plane follows the mean camber line of each radial section.
Sometimes it creates problems in the TE smoothness as a consequence of a not
proper panels overlapping. The increase in cells number of this area has been
obtained by a Centaur tool.
37
2.6 CENTAUR
Why Hybrid Grids? Traditional approaches to mesh generation have been via block-
structured (usually composed of hexahedra) or unstructured (tetrahedra) techniques.
There are pros and cons to both approaches and thus the strategy chosen has been
dependent on the particular application. The Centaur grid generator combines the both
techniques via a hybrid (prismatic/hexahedral/pyramidal/tetrahedral) meshing strategy.
The prismatic and hexahedral elements are utilized in regions of high solution gradients,
and tetrahedra are utilized elsewhere with pyramids utilized in some locations to allow
for a transition between the prisms/hexahedra and the tetrahedra. The main advantages
of the hybrid system are summarized below:
38
possible to merge every pair of curves with one another.
Setting the tolerance to 1e-02 seems to be the right choice. For each case must be
defined the inlet, outlet, viscous walls, wake and periodic surfaces.
Once the output files are written coming back to Setupgrid only if errors occur or some
regions require adjustment, is the iterative process that allows of obtaining a proper
grid.
Several global parameters for the mesh are stored within the files casename.sin,
casename.pin and casename.tin. The mentioned values may change during the
refinement process. As the present ones still do not produce a reliable mesh, must be
modified to create a reasonable one.
Starting Makegrid should be possible to create the mesh an then it is possible to convert
the casename.hyb file to a Fluent case using the hybconvert tool.
To make sure that the Fluent case file is free from errors and unnecessary information,
it is possible run Fluent.
• Three panels splitting: this solution has been useful when Centaur has been not
able to join properly between each other the SS, PS and TE panels because the
curvature in the connecting areas was lost. To obtain the same smoothness of
39
the previous case, a Centaur curve source has been utilized. Not using this
option the transition from high clustering area (TE) to low clustering area (PS or
SS) results more gradual.
Sometimes the TE or LE radius of the blade segmentations has really low values. In this
case obviously the cells size in these areas must be smaller than the other ones on the
TE and LE of the original blade. It means also that the curvature gradients are more
pronounced with a high probability of a bad quality mesh in the connecting zones of the
panels. In order to avoid this issue just one panel has been utilized to overlap e.g. SS
and TE of the segmentation one (figure 2.t).
When it is intended to open slots which cover a considerable percentage of the blade
height, it is likely to get an increase of the grid number of cells. Whether going back to
reasonable mesh dimensions is required, it is necessary to stretch in span direction the
cells inside the duct too.
As previously mentioned one of the most sensible areas in the flow field is the blade
wake, because in this zone most of the losses and vortexes are generated, therefore a
proper description of what is happening is strictly necessary. Small changes in the grid
could drive to effective changes in CFD results. Centaur is able to generate a wake
mesh with prismatic elements grown on both sides of the wake surface allowing for the
transition from the boundary layer to the wake region to be modelled accurately. The
wake surface boundary condition is a special setting that allows for prismatic elements
to be grown from both sides of the panel. The geometry is automatically modified to be
topologically valid (each curve is attached to 2 panels, etc.) in the area of the wake
panel. This boundary condition is specified just like any other boundary condition when
the group is created.
The dip in prism thickness at the trailing edge is due to aspect ratio issues that can be
corrected: the asp section of the expert grid generator file controls automatic prismatic
pullback due to high aspect ratio (tall) prisms created during the process. By default the
prismatic generator will locally reduce prism thickness to ensure the prisms are not
overly thick. This ensures that the tetrahedra will easily generate a good quality mesh
near the outer prismatic surface.
While the default values will typically generate a good quality hybrid grid, for some
cases, it is necessary to limit the prisms to only grow to a different aspect ratio. This will
allow for an even higher quality tetrahedral grid for these cases.
In other cases, the aspect ratio constraint can cause the prisms to be overly constrained
and appear to not be thick enough to resolve the desired flow features. In this case, the
aspect ratio constraints may need to be raised to allow for taller prismatic elements on
the final prismatic layer. Care must be taken to not raise these parameters to too high a
value or the tetrahedra will have difficulty generating in that region.
40
Although for the actual grid the result has been not fully satisfying, it seems reasonable
to exclude any possibility to obtain a further enhancement. Without going in the details,
the main reason of this is a TE surface described in 2D by a circular arch. All features in
Centaur user-guide refer to a sharpened TE. Many parameters have been forced to
obtain a mesh similar to Padram O-mesh and Padram wake mesh (figure 2.u).
As a consequence of that, the grid containing the so defined wake surface has shown a
bad convergence in CFD calculations made on blade configuration with high slots: in
these cases a fitting volume for locally changing the size of tetrahedric elements at the
blade outlet has been employed (figure 2.v).
This solution is more similar to the Padram one because in the wake zone there are no
cells with a size comparable to the o-mesh cells utilized for investigating the boundary
layer. The main problem, which occurs when a fitting volume is utilized (figure 2.x), is
the low control in cells amount. Is not possible obtain a gradual cells size growth from
the middle of the wake to the transition zone (figure 2.w)
Casename.sin
41
4. ! Factor for analytic curvature clustering interior to panels
2. ! Factor for proximity clustering
2. ! Factor for CAD clustering
Casename.pin
Casename.tin
Casename.egg
adp 2
T
T
asp 3
2
2
15
smo 2
20
2
wak 1
20
42
By the first file is possible control the surface mesh generation process but usually also
surfaces sources are useful to improve the grid in critical areas. The second one
controls the prismatic mesh generation avoiding an excessive pull-back between
surfaces with angles less of 90°, but the main target is a properly grid generation in the
boundary layer as by Padram. The casename.tin file controls the tetrahedral mesh
generation. The last one is able to modify particular features of the mesh generator.
43
CHAPTER 3
CFD CALCULATION
44
In the computational domain, mixing plane, inlet and outlet surfaces have been placed
approximately at mid-distance between two consecutive blades.
As inlet conditions, radial profiles taken from the throughflow were used, while for the
outlet a radial equilibrium was fixed and calculations with different back-pressures were
run in order to get a full stage-characteristic for each different configuration.
45
3.4.1 ORIGINAL AEROFOIL
Compressor stator 5 row is featuring 114 blades. Unless the general performance could
be considered quite satisfactory, since a previous design and optimization process on it
has been already carried out, the CFD investigation shown significant losses at the
casing zone, generated by some secondary flow. Based on that it was thought to open
a slot right at the blade tip in order to clean it out the separation.
As first change to the original stator, a slot was positioned at the casing zone in the front
portion of the blade (from 25% to 50% of the chord), extending from 90% to 100% of
the span (Figure 3.f), with the aim to locally reduce the loss level. As mentioned, the
purpose is to take a portion of the flow from the PS and accelerate it in a convergent
duct in order to obtain a jet energizing the low Mach number flow area on the SS.
The results related to such a geometry show a small decrease of the stage efficiency
(figure 3.g), which is also confirmed by the losses radial distribution plot (Figure 3.h).
The reasons for this bad outcome is that the slot is not properly working, since no jet at
its exit is generated, and the only effect of that device is having a thicker blade wake
due to the presence of two trailing edges in the geometry.
An explanation for this could be that in this preliminary case the slot was designed in a
wrong way, first of all because it was positioned where there was not a proper pressure
delta between the slot inlet and exit, able to generate a jet on the stator suction side
(figure 3.i). In other words it was realized that the slot position and the axial Mach
number distribution (Figure 3.f) were not fitting with each other, since not all the velocity
peak difference between the two stator surfaces were used.
Furthermore the purpose to generate a jet on the front portion of the suction side
seemed to be not suitable, since normally the detachment of the boundary layer to be
cleaned happens in the rear part of an aerofoil, close to the trailing edge.
Based on this concern, it was clear that the original aerofoil shape had to be changed,
designing a rear loaded tip section and shifting the slot towards the second half of the
blade chord.
46
In order to better catch this effect, it was considered to have a variable position for the
slot in span direction (fig. 3.j), since as a consequence of the variable camber style of
the transition zone of the blade, the pressure delta peak was corresponding to different
values of the chord for each radial section (fig. 3.l). Also the dimension of the slot itself
was increased and the current one goes from 75% to 100% of the span (fig. 3.j).
The overall performance of such a design are quite promising (figure 3.m), since a
minor improvement at high flows making a comparison with the original stator is
achieved. It is interesting to see the loss radial profile at blade exit (fig 3.n), which
surprisingly shows a worsening at the casing and improvements from 10% to 30% and
form 60% to 75% of the span. Analyzing the flow flied (fig. 3.o) the following reasons for
such a behavior were estimated:
• The slot is not properly generating a jet at the blade tip, where the flow field is
featuring low velocity due to viscous effects due to the endwall, which are here
prevailing;
• The presence of a double trailing edge is widening the blade wake increasing the
loss level at the tip;
• The slot is somehow decreasing the secondary flow and this allows a drop of the
losses from 60% to 75% of the span;
• As a consequence of the higher blockage at the casing the flow is pushed to go
though the lower part of the stator and this brings a decrease of the losses at the
hub.
• The slot has to be positioned in the rear part of the blade, where the boundary
layer detachment occurs;
• The slot has to be axially positioned where there is the maximum difference
between the pressure on suction and pressure side, in order to achieve a strong
jet at its exit;
• As a consequence of the first two points, a proper rear loaded blade has to be
designed;
• The slot loses its efficiency at the endwalls where the viscous effects are
prevailing.
47
According to these constrains, an iterative process were implemented in order to
optimize both the blade design, paying main of the attention to the camber distribution,
and the slot shape and positioning. In the meanwhile also a progressive reduction of the
number of blade was carried out, since as mentioned this represents the main target of
this approach.
The outcome of this procedure has been a stator extremely different from the starting
one and featuring the following solutions:
Obviously this kind of stator could never be working without a proper slot, since the
separation due to the high loading and to the unconventional caber style would
generate huge losses (fig. 3.s).
Therefore a slot was positioned in the rear part of the aerofoil and opened from 5% to
95% of the span (fig 3.r). The results of such a design showed a proper working of the
slot itself, but nevertheless a decrease of the performance making a comparison with
the original compressor, since the efficiency has dropped by 0,30%. Certainly a reason
for that was the reduced number of blades, but actually the main portion of the losses
has been a consequence of the vortexes generated at the endwalls for viscous reasons.
What is happening there? The air coming out from the slot is much faster than the flow
at casing and hub, therefore curls are provoked by the viscosity drag (figure 3.t). This
unexpected behavior suggested a step back to the configuration featuring an open slot
also at the endwalls, because it was estimated that in a trade off between the two
possible solutions, the old one would have a minor impact and would have been less
painful for the efficiency.
A fully slotted stator was indeed designed (fig. 3.r), which was now featuring two
completely separated and axially overlapped segmentations. The performance of such
a stator were better than the previous one and still worst than the base case, but the
gap has been now reduced to 0,18% (fig. 3.u). Once again the reason for that is the
high losses level at the enwalls, where as expected the slot is working at its best.
Instead it looks absolutely interesting the behavior at mid-height, where the latest
version is better than the original one, unless the lower number of blade.
3.5 CONCLUSIONS
The study carried out has got as main achievement the definitions of some fundamental
milestones for the design of stators featuring slots taking a portion of the flow from the
PS and accelerating it in a convergent duct in order to obtain a jet energizing the
48
boundary layer on the SS. The most important concepts of the achieved knowledge can
be summarized as following:
• Slots must be positioned in the rear part of the aerofoil, where normally
separations occur, in order to energize the boundary layer and reduce losses.
For doing this a proper rear loaded blade has to be designed, since for
generating a suitable jet at slot exit, the maximum pressure difference between
the suction and pressure side has to be in the second half of the chord;
• Slots are properly working at mid-span and this allow a reduction of the number
of blades;
• Slots lose efficiency at the endwalls where the viscous effects are prevailing;
• In the connection zone between a slotted section and a conventional one, the
radial velocity gradients could generate vortexes.
The final design of the study could be considered definitively positive and promising,
since the performance of a conventional optimized stator were barely matched with a
blade reduction by 35%.
Obviously further studies have to be carried out, because it seems likely that further
margins for improvement are existing. Especially at the endwalls the phenomena are
not fully understood and high losses are still generated, therefore a better design has to
be found.
49
the pressure ratio characteristic and a worsening of the stall margin too. The estimated
reasons for this behavior could be resumed as following:
• The rotor casing is generating more losses (fig. 3.z) and this drops pressure
ratio and stability. Such a result is not fully understood, since the contour plots
of that area (fig. 3.aa) show a visible improvement in term of reduction of the
tip vortex;
• It is interesting to see that the stator casing is now working better and this
drives a general improvement of the stage efficiency.
3.7 CONCLUSIONS
The preliminary investigation on the slotted rotor has showed once again that using
such a device could be beneficial for the performance. Obviously further analysis,
similar to what carried out for in the case of the stator, are necessary, but the contour
plots taken at the clearance zone show a promising reduction of the tip vortex and it is
easy to estimate that an optimization process of the slot shape and position could likely
bring bigger benefits.
50
BIBLIOGRAPHY
[1] “Compressor Aerodynamics” N.A. Cumpsty.
[2] “The jet Engine” © Rolls-Royce plc 1986 Fifth edition Reprinted 1996 with revisions.
[4] Padram Introduction and User Guide; Author: S.Shshpar (February 2004).
51
TABLE OF FIGURES
Figure 1. a: velocity triangles into and out of an axial rotor row ......................... 54
Figure 1. b: Idealized rotor. Flow enters at radius r1 and leaves at r2. ................ 54
Figure 1. c: enthalpy diagram for axial rotor with equal blade speed at inlet and
outlet. ............................................................................................... 54
Figure 1. d: blade-to-blade geometry and notation. ............................................ 55
Figure 1. e: momentum balance about a blade in cascade................................ 55
Figure 1. f: trend in compressor geometry (solidity and aspect ratio) and
performance (stage loading and spool pressure ratio) with time. ..... 55
52
Figure 3. d: structured-unstructured pressure radial distribution for several back
pressures. .................................................................................................... 71
Figure 3. e: structured-unstructured axial-velocity radial distribution for several
back pressures. ........................................................................................... 71
Figure 3. f: modified blade (left); Mach number vs. %axial chord (right); ............ 72
Figure 3. g: comparison of overall performance between the base case and the
modified aerofoil. ......................................................................................... 72
Figure 3. h: losses radial distribution for the base case and the modified aerofoil
at several back pressures. ........................................................................... 73
Figure 3. i: Mach number contour plots at 95% in span and at TE, at 335.000 Pa
back pressure for the original aerofoil (left) and for the modified aerofoil
(right). .......................................................................................................... 73
Figure 3. j: partial rear loaded profile with slot going from 75% to 100% along the
span............................................................................................................. 74
Figure 3. k: sections 21 comparison between original aerofoil e partial rear loaded
aerofoil. ........................................................................................................ 74
Figure 3. l: Mach number vs. %axial chord at three positions along the span in the
rear loaded area. ......................................................................................... 75
Figure 3. m: comparison between aerofoils performance with and without slot. . 75
Figure 3. n: losses radial distribution for the partial rear loaded case and the
modified aerofoil at several back pressures................................................. 76
Figure 3. o: Mach contour plots at 80% (top), 90% in span and at TE for partial
rear loaded blade (left) and slotted blade (right). Back pressure of 350.000
Pa. ............................................................................................................... 77
Figure 3. p: DF/DF* (where DF* is the DF value for the original aerofoil at section
11) comparison between original stator and fully rear loaded stator........ 78
Figure 3. q: sections 21 comparison between original aerofoil and fully rear
loaded aerofoil. ............................................................................................ 78
Figure 3. r: Mach number vs. %axial chord (left), modified aerofoil by partial slot
and full slot (right). ....................................................................................... 79
Figure 3. s: Mach number contour plots for fully rear loaded blade without any
slot; .............................................................................................................. 79
Figure 3. t: Mach number contour plots for partial slot going from 5% to 95% in
span (top), and full slot. Back pressure: 335.000 Pa. Radial cuts (left) at 50%
in span. X-cuts at TE (right). ........................................................................ 80
Figure 3. u: overall performance comparison between original aerofoil case, rear
loaded aerofoil with slot from 5% to 95% and “full slot” case for several back
pressures. .................................................................................................... 81
Figure 3. v: losses comparison at H05SE between original aerofoil case, rear
loaded aerofoil with slot from 5% to 95% and “full slot” case for several back
pressures. .................................................................................................... 81
Figure 3. w: losses comparison at H05SE between original aerofoil, rear loaded
aerofoil with slot from 5% to 95% and “full slot” case for several back
pressures. .................................................................................................... 82
Figure 3. x: modified aerofoil of stage 5. ............................................................. 82
Figure 3. y: overall performance comparison between base case and rotor with
slot case. ..................................................................................................... 83
Figure 3. z: losses comparison between base case and rotor with slot case...... 83
Figure 3. aa: radial cuts at 95% in span (top), at 98% (bottom). Back pressure:
340.000 Pa. ................................................................................................. 84
53
Figure 1. a: velocity triangles into and out of an axial rotor row
Figure 1. b: Idealized rotor. Flow enters at radius r1 Figure 1. c: enthalpy diagram for axial
and leaves at r2. rotor with equal blade speed at inlet and
outlet.
54
Figure 1. e: momentum balance about a
Figure 1. d: blade-to-blade geometry and notation.
blade in cascade.
Figure 1. f: trend in compressor geometry (solidity and aspect ratio) and performance (stage
loading and spool pressure ratio) with time.
55
Figure 2. a: streamline sections of the blade.
56
Figure 2.d: splitting result.
Figure 2.c: trimming planes.
Figure 2.g: planes on the mean camber lines. Figure 2.h: 4 panels splitting.
57
Figure 2.i: 3 panels splitting. Figure 2.j: additional surfaces created.
Figure 2.k: additional surfaces created. Figure 2.l: portions of the original surface.
58
5:PS-TE orig.rotor
4:LE-PS segm.2
2:SS-TE segm.1
7:Duct bottom
1:LE
8:SS-TE orig.rotor
Figure 2.n: wake surfaces ( red). Figure 2.o: splitting planes (red).
59
1 2
60
Figure 2.r: cells distributions in span for Structured and unstructured mesh.
61
Figure 2.s: effect of the anisotropic mesh on the LE surface
62
Figure 2.u: wake mesh obtained by wake surfaces.
63
Figure 2.w: Padram wake mesh.
64
Figure 2.y: In-out surfaces and prismatic layers on viscous walls.
Figure 2.z: effects example of the viscous wall setting on several surfaces.
65
Figure 2.aa: prismatic layers on the blade surface (X-Cut plane).
66
RHS H-mesh
Upper Periodic boundary Tip-Gap
θ Upper–H-mesh mesh
LHS H-mesh
O-mesh
Aerofoil
Lower–H-mesh
OGV Corners
r Points Lower Periodic boundary
Axial Chord
z-direction
67
Figure 2.ff: angles used to define inlet velocity direction
Figure 2.gg: definition of inlet pitch angle in case of zero axial flow component.
68
Figure 3. a: stage 5, side view.
69
TARGET FLOW
SS
SU
Figure 3. c: results comparison between structured (SS) and unstructured (SU) case
70
TARGET FLOW
SS
SU
TARGET FLOW
SS
SU
71
Figure 3. f: modified blade (left); Mach number vs. %axial chord (right);
Figure 3. g: comparison of overall performance between the base case and the modified aerofoil.
72
Figure 3. h: losses radial distribution for the base case and the modified aerofoil at several back
pressures.
Figure 3. i: Mach number contour plots at 95% in span and at TE, at 335.000 Pa back pressure
for the original aerofoil (left) and for the modified aerofoil (right).
73
Figure 3. j: partial rear loaded profile with slot going from 75% to 100% along the span.
Figure 3. k: sections 21 comparison between original aerofoil e partial rear loaded aerofoil.
74
Figure 3. l: Mach number vs. %axial chord at three positions along the span in the rear loaded
area.
TARGET FLOW
NO SLOT
SLOT_75%-100%
75
TARGET FLOW
NO SLOT
SLOT_75%-100%
Figure 3. n: losses radial distribution for the partial rear loaded case and the modified aerofoil at
several back pressures.
76
Figure 3. o: Mach contour plots at 80% (top), 90% in span and at TE for partial rear loaded blade (left) and
slotted blade (right). Back pressure of 350.000 Pa.
77
DF vs. span sections
1,30
1,25
1,20
1,15
Original Newak
1,10 Rear Loaded Newak
DF/DF*
1,05
1,00
0,95
0,90
0,85
0,80
1 3 5 7 9 11 13 15 17 19 21
sections
Figure 3. p: DF/DF* (where DF* is the DF value for the original aerofoil at section 11) comparison
between original stator and fully rear loaded stator.
ORIGINAL AEROFOIL
Figure 3. q: sections 21 comparison between original aerofoil and fully rear loaded aerofoil.
78
Figure 3. r: Mach number vs. %axial chord (left), modified aerofoil by partial slot and full slot (right).
Figure 3. s: Mach number contour plots for fully rear loaded blade without any slot;
79
Figure 3. t: Mach number contour plots for partial slot going from 5% to 95% in span (top), and full slot.
Back pressure: 335.000 Pa. Radial cuts (left) at 50% in span. X-cuts at TE (right).
80
MRTP-MM{inp}[IN]
Figure 3. u: overall performance comparison between original aerofoil case, rear loaded aerofoil
with slot from 5% to 95% and “full slot” case for several back pressures.
Figure 3. v: losses comparison at H05SE between original aerofoil case, rear loaded aerofoil with
slot from 5% to 95% and “full slot” case for several back pressures.
81
Figure 3. w: losses comparison at H05SE between original aerofoil, rear loaded aerofoil with slot
from 5% to 95% and “full slot” case for several back pressures.
82
TARGET FLOW
NO SLOT
SLOT
Figure 3. y: overall performance comparison between base case and rotor with slot case.
TARGET FLOW
NO SLOT
SLOT
Figure 3. z: losses comparison between base case and rotor with slot case.
83
Figure 3. aa: radial cuts at 95% in span (top), at 98% (bottom). Back pressure: 340.000 Pa.
84
85
NOMENCLATURE
LIST OF ABBREVIATIONS
RR Rolls-Royce
CFD Computational Fluid Dynamics
PS Pressure Side
SS Suction Side
LE Leading Edge
TE Trailing Edge
VARIABLES USED
Geometric Variables
c Blade chord
d,D Diameter
g Staggered gap, pitch resolved normal to the flow direction
h Blade height
m Distance in meridional direction
r,R Distance in radial direction
s Blade pitch
s Distance along streamlines
t Blade thickness
x Distance in axial direction
y Distance in the pitchwise direction
z Distance normal to x and y
σ Solidity c/s
Flow Variables
85
α2 flow outlet angle (measured from the axial direction)
V1 inlet flow velocity
V2 outlet flow velocity
Subscripted velocity
Special angles
i incidence (angle between inlet flow direction and blade inlet direction)
A angle of attack (angle between inlet flow direction and the chord line)
δ deviation (angle between outlet flow angle and blade outlet angle)
General Variables
87