Professional Documents
Culture Documents
20
Essentials of Turbo
Machinery in CFD
Edited by :
Ideen Sadrehaghighi, Ph.D.
Unsteady
Unsteady
Flow in
Flow in Axial
Radial
Turbomachines
Turbomachines
(ANSYS)
(ANSYS)
Blade
Interaction
(NUMECA)
Goldman
Annular
Turbine
Cascade
(NASA-TGRID)
ANNAPOLIS, MD
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
2
Table of Contents
1 Introduction ................................................................................................................................ 20
1.1 Preliminaries......................................................................................................................................................... 20
1.2 Full Engine Simulation Methodology .......................................................................................................... 21
10.8.3.3 Blade Twist Effect at Low Mass Flow Rate Conditions ........................................... 295
10.8.4 Conclusion.............................................................................................................................................. 296
10.8.5 References.............................................................................................................................................. 297
313
12.5.2.1 Formulation of Problems .................................................................................................... 313
12.5.2.2 Conclusions ............................................................................................................................... 314
12.5.3 Boundary Mesh Resolution ............................................................................................................. 314
12.5.4 Periodic Meshing ................................................................................................................................. 315
12.6 Boundary Conditions ...................................................................................................................................... 315
12.7 Turbulence Modeling...................................................................................................................................... 317
12.8 Aero-Mechanics ................................................................................................................................................ 317
12.8.1 Nodal Diameter .................................................................................................................................... 319
12.9 Near Wall Treatment ...................................................................................................................................... 319
12.10 Transition Prediction ..................................................................................................................................... 319
12.11 Numerical Consideration .............................................................................................................................. 320
12.12 Convergence Criteria ...................................................................................................................................... 320
12.13 Single or Double Precision ........................................................................................................................... 320
12.14 Heat Transfer Prediction .............................................................................................................................. 320
12.14.1 Keeping it Cool in Gas Turbine ...................................................................................................... 321
12.15 Literature Review and Parallel Processing Tools ............................................................................... 321
12.16 Concluding Remarks ....................................................................................................................................... 322
List of Tables
List of Figures
Figure 5.30 Datum (Left) And Optimized (Right) Rotor 37 Separated Flow Contours (Orange).
Flow Direction Right To Left.................................................................................................................................... 120
Figure 5.31 Lift Plots For The Datum and Optimized ............................................................................... 121
Figure 5.32 R37 Optimized Characteristic Vs Datum – Courtesy of [John et al.] ........................... 121
Figure 6.1 Complex Flow Phenomena Compressors ................................................................................. 123
Figure 6.2 Fan Tip Section Geometry .............................................................................................................. 124
Figure 6.3 Flow Structures with 5 to 6 Orders of Magnitudes Variations in Length and Time
Scales (LaGraff et al., 2006)...................................................................................................................................... 125
Figure 6.4 Shock Structure in Transonic Fan ............................................................................................... 126
Figure 6.5 Pressure Contour of Wake Flow .................................................................................................. 127
Figure 6.6 Instantaneous Absolute Velocity Contour at Nozzle Exit [Matsunuma, 2006]......... 128
Figure 6.7 Unsteady Wakes Convecting in Blade Passage ...................................................................... 129
Figure 6.8 Flow over an unshrouded tip gap ............................................................................................... 130
Figure 6.9 Classical Secondary Flow Model .................................................................................................. 131
Figure 6.10 Modern Secondary Flow Model ................................................................................................. 132
Figure 6.11 Vortex pattern of Latest Secondary Flows ............................................................................ 133
Figure 6.12 Turbine Secondary Flow Model (Takeishi et al.) ............................................................... 135
Figure 6.13 Illustration of formation of hub corner stall together with limiting streamlines and
separation lines ............................................................................................................................................................. 137
Figure 6.14 Three-dimensional separations: traditional view and scope of current
investigation ................................................................................................................................................................... 137
Figure 6.15 Illustration of the near wall flows as taken through oil and dye surface flow
visualization (reproduced with permission of the publisher from ASME) ......................................... 138
Figure 6.16 (Top) - Measurements of the Horseshoe Vortex just upstream of the Vane at the
Vane-End-Wall Juncture (Bottom) - Actual Hardware Showing Effects of the Horseshoe Vortex
on a First Vane (Courtesy of ASME) .................................................................................................................... 139
Figure 6.17 Illustration of Different Vortical ................................................................................................ 140
Figure 6.18 Contours of Non-Dimensional Heat Transfer Coefficients (Reproduced with
Permission ASME) ...................................................................................................................................................... 141
Figure 6.19 Fillet and Bulb Designs as Shown by (Becz et al.) ............................................................... 142
Figure 6.20 CFD Prediction of Streamlines Across a ................................................................................. 143
Figure 6.21 Blade Passage and Slice of The Computational Domain.................................................. 144
Figure 6.22 Computed Mid-Span Mach Number Distribution............................................................... 146
Figure 6.23 End-Wall Stanton Number (103) Distributions Computed with Grid B in
Comparison with the Measurement Data: (1) k-ω Turbulence Model, (2) M-SST, (3) v2-f, (4)
Experiment ..................................................................................................................................................................... 147
Figure 6.24 Effect of Grid Refinement on the End-Wall Stanton Number (x103) Prediction with
the M-SST Turbulence Model: (1) Grid С, (2) Grid D, (3) Grid E, (4) Experiment. ............................ 148
Figure 6.25 End-Wall Streak Line Visualization.......................................................................................... 149
Figure 6.26 Grid on the Solid Surfaces of the Geometry .......................................................................... 152
Figure 6.27 (a) Casing surface mesh showing multiblock structure and (b) hub surface mesh
showing multiblock structure ................................................................................................................................. 153
Figure 6.28 Total temperature (T0) and total pressure (P0) at the blade inlet ............................... 154
Figure 6.29 Instantaneous and time-averaged „dashed… hub surface.............................................. 156
Figure 6.30 Instantaneous and time-averaged (dashed) hub heat transfer distribution for a
wake passing .................................................................................................................................................................. 155
Figure 6.31 Time-Averaged Casing Pressure ............................................................................................... 157
Figure 6.32 Time-Averaged Hub Pressure .................................................................................................... 157
Figure 6.33 Difference in hub pressure distribution between the time-averaged and steady
results................................................................................................................................................................................ 158
14
Figure 6.34 Difference in casing pressure distribution between the time-averaged and steady
results................................................................................................................................................................................ 158
Figure 6.35 Time-averaged casing adiabatic wall temperature ........................................................... 159
Figure 6.36 Time-averaged hub adiabatic wall temperature ................................................................ 159
Figure 6.37 Difference in casing adiabatic wall temperature distribution between the time-
averaged and steady results .................................................................................................................................... 160
Figure 6.38 Difference in hub adiabatic wall temperature distribution between the time-
averaged and steady results .................................................................................................................................... 160
Figure 6.39 Time-averaged casing heat transfer rate ............................................................................... 161
Figure 6.40 Time-averaged hub heat transfer rate .................................................................................... 161
Figure 6.41 Difference in casing heat transfer rate between the time-averaged and steady
results................................................................................................................................................................................ 162
Figure 6.42 Difference in hub heat transfer rate between the time averaged and steady results
162
Figure 6.43 Difference in hub Nusselt number between the time averaged and steady results
163
Figure 7.1 The Schematic of a Modern Gas Turbine Blade with Common Cooling Techniques
(Courtesy of Je-Chin Han) .......................................................................................................................................... 166
Figure 7.2 Typical high-pressure turbine stage showing rim seal and wheel-space ................... 167
Figure 7.3 Schematic of Flm Cooling Concept .............................................................................................. 168
Figure 7.4 Measured Adiabatic Wall Temperatures for Coolant Exiting a Combustor/Vane
Leakage Slot (reproduced with permission from ASME) ............................................................................ 169
Figure 7.5 Contours of Adiabatic Effectiveness for Two Film-Cooling Hole Patterns (left and
center) With a Mid-Passage Gutter for the Cooling Hole Pattern in the Center (Right)
(Reproduced With Permission From the Publisher of ASME) .................................................................. 170
Figure 7.6 Vane Section with Ten Cooling Channels and Temperature Distribution Computed
171
Figure 7.7 Surface Temperature Distribution on the Suction Side (Left) and The Pressure Side
(Right) of the Vane ....................................................................................................................................................... 172
Figure 7.8 Flow Streamlines Colored Cooling Air Temperature into Passages ............................. 173
Figure 7.9 2D Hybrid Mesh around the T106 Blade ................................................................................. 177
Figure 7.10 Blade Profile vs. Pressure Coefficient (Courtesy of De La Calzada et al.) ................. 177
Figure 7.11 Flow Field at the Front and Middle Parts of the Separation Bubble (Courtesy of De
La Calzada et al.) ........................................................................................................................................................... 178
Figure 7.12 Heat Transfer Coefficient for Different Negative Incidences (Courtesy of Calzada et
al.) 180
Figure 7.13 Stanton Number for Different Negative Incidences (Courtesy of De La Calzada et
al.) 181
Figure 7.14 Stanton Number vs. Reynolds Numbers (Courtesy of De La Calzada et al.) ........... 182
Figure 7.15 Heat Transfer Coefficient vs. Reynolds Number (Courtesy of Calzada et al.)......... 182
Figure 8.1 Pressure Ratio by Normalized Mass Flow (Courtesy of Simoes) ................................... 186
Figure 9.1 Schematics of 3D Concept at IGV/Rotor/Stator Interface................................................. 193
Figure 9.2 Interface Between Rotor/Stator .................................................................................................. 194
Figure 9.3 Difference between Passage and Stages ................................................................................... 194
Figure 9.4 Block Computational Domain for a Rotor with guiding Vanes........................................ 196
Figure 9.5 Axial Rotor/Stator Interaction (Schematics Illustrating the Mixing Plane concepts)
197
Figure 9.6 A Compressor Pressure Distribution on a Surface using a Mixing Plane.................... 197
Figure 9.7 Schematic of an Artificial Interface Between a Rotor and a Stator (left) and the
Virtual Control Volume Formed by Displacing Two Adjacent Domains (right)................................. 198
15
Figure 10.28 Spanwise Slice Of The Datum And Optimized R37 Geometries At 60% Span –
Courtesy of [John et al.].............................................................................................................................................. 272
Figure 10.29 Optimized R37 Bump (Blue) Added To The Datum Blade Geometry (Grey) ....... 272
Figure 10.30 Datum (Left) And Optimized (Right) Rotor 37 Static Pressure Contours. Flow
Direction Right To Left – Courtesy of [John et al.] .......................................................................................... 273
Figure 10.31 Datum (Left) And R37 Optimized (Right)........................................................................... 274
Figure 10.32 Datum (Left) And Optimized (Right) Rotor 37 Separated Flow Contours (Orange).
Flow Direction Right To Left.................................................................................................................................... 274
Figure 10.33 Lift Plots For The Datum and Optimized............................................................................. 275
Figure 10.34 R37 Optimized Characteristic Vs Datum – Courtesy of [John et al.] ........................ 275
Figure 10.35 Meridional geometry of NASA rotor 37, from AGARD................................................... 279
Figure 10.36 Measurement Stations within NASA Rotor 37, from AGARD ..................................... 280
Figure 10.37 Parameterization of a Compressor Airfoil .......................................................................... 281
Figure 10.38 Multiblock Grid Used in the Simulations ............................................................................. 282
Figure 10.39 Comparison Between Calculated and Experimental Mach Number Contours at
90% Span and m˙ /m˙ choke = 0.98 :Maximum Mach = 1.576, Minimum Mach = 0, and Contour
Interval = 0.031. ............................................................................................................................................................ 282
Figure 10.40 Blade Geometries of the Baseline and Optimized Configurations ............................ 283
Figure 10.41 Mach Number Contours at 95 and 50 % Span of Baseline and Optimized
Geometries ...................................................................................................................................................................... 285
Figure 10.42 Mach Number Contours on The Suction Surface of the Baseline and Optimized
Blades ................................................................................................................................................................................ 286
Figure 10.43 Different Cases of Blade Twist................................................................................................. 288
Figure 10.44 Mesh Generated by TURBOGRID for the Baseline Case ................................................. 290
Figure 10.45 Performance Curves for the Studied Cases ........................................................................ 292
Figure 10.46 Mach Number Distribution for High Mass Flow Point .................................................. 293
Figure 10.47 Total Pressure Over Span for High Mass Flow points at domain outlet ................ 294
Figure 10.48 Entropy distribution for high mass flow points ............................................................... 294
Figure 10.49 Mach number distribution for low mass flow points..................................................... 295
Figure 10.50 Entropy Distribution for Low Mass Flow Points ............................................................. 296
Figure 11.1 Centrifugal impeller with a highly polished surface likely to improve performance
299
Figure 11.2 Cut-Away View of a Turbocharger showing the Centrifugal Compressor ............... 300
Figure 11.3 Jet Engine Cutaway Showing the Centrifugal Compressor among others ............... 301
Figure 11.4 Ninety Degree Inward-Flow Radial Turbine Stage ............................................................ 303
Figure 11.5 Outward Flow Radial Turbine .................................................................................................... 304
Figure 12.1 Different Flow (2D, Q3D, and full 3D) ..................................................................................... 307
Figure 12.2 Full Blade Simulation using Harmonic Balanced Method (Courtesy of CD-adapco)
309
Figure 12.3 Transient Blade Row Extensions Enable Efficient Multi-Stage CFD Simulation
(Courtesy of ANSYS.com) .......................................................................................................................................... 311
Figure 12.4 Typical Meshing of a Turbomachinery Stage ....................................................................... 312
Figure 12.5 Multi-Block Grid for the Space Shuttle Main Engine Fuel Turbine ............................. 312
Figure 12.6 Pressure contour plot, 2nd order spatial discretization scheme................................. 316
Figure 12.7 Analysis provided vibration required for flutter analysis .............................................. 318
Figure 12.8 Examples of Nodal Diameter ...................................................................................................... 319
18
19
Preface
This note is intended for all undergraduate, graduate, and scholars of Turbomachinery. It is not
completed and never claims to be as such. Therefore, all the comments are greatly appreciated. In
assembling that, I was influenced with sources from my textbooks, papers, and materials that I
deemed to be important. At best, it could be used as a reference. I also would like to express my
appreciation to several people who have given thoughts and time to the development of this article.
Special thanks should be forwarded to the authors whose papers seemed relevant to topics, and
consequently, it appears here©. Finally I would like to thank my wife, Sudabeh for her understanding
and the hours she relinquished to me. Their continuous support and encouragement are greatly
appreciated.
Ideen Sadrehaghighi
June 2018
20
1 Introduction
1.1 Preliminaries
Fluid mechanics and thermodynamics are the fundamental sciences used for turbine aerodynamic
design and analysis. Several types of fluid dynamic analysis are useful for this purpose. The concept
through-flow analysis is widely used in axial-flow turbine performance analysis. This involves
solving the governing equations for inviscid flow in the hub-to-shroud plane at stations located
between blade rows. The flow is normally considered to be axisymmetric at these locations, but still
three-dimensional because of the existence of a tangential velocity component. Empirical models are
employed to account for the fluid turning and losses that occur when the flow passes through the
blade rows. By contrast, hub-to-shroud through-flow analysis is not very useful for the performance
analysis of radial-flow turbomachines such as radial-inflow turbines and centrifugal compressors.
The inviscid flow governing equations do not adequately model the flow in the curved passages of
radial turbomachines to be used as a basis for performance analysis. Instead, a simplified “pitch-line”
or “mean-line” one-dimensional flow model is used, which ignores the hub-to-shroud variations.
These also continue to be used for axial-flow turbine performance analysis. Computers are
sufficiently powerful today that there is really no longer a need to simplify the problem that much for
axial-flow turbomachinery. More fundamental internal flow analyses are often useful for the
aerodynamic design of specific components, particularly blade rows. These include 2D flow analyses
in either the blade-to-blade or hub to shroud (Through Flow) direction, and Quasi-3D flow analyses
developed by combining those 2D analyses. Wall boundary layer analysis is often used to supplement
these analyses with an evaluation of viscous effects1.
Viscous CFD solutions are also in use for turbines. These are typically 3D flow analyses, which
consider the effects of viscosity, thermal conductivity and turbulence. In most cases, commercial
viscous CFD codes are used although some in-house codes are in use within the larger companies.
Most design organizations cannot commit the dedicated effort required to develop these highly
sophisticated codes, particularly since viscous CFD technology is changing so rapidly that any code
developed will soon be obsolete unless its development continues as an ongoing activity.
Consequently, viscous CFD is not covered here beyond recognizing it as an essential technology and
pointing out some applications for which it can be effectively used to supplement conventional
aerodynamic analysis techniques.
Prediction of the flow through cascades of blades is fundamental to all aspects of turbomachinery
aerodynamic design and analysis. The flow through the annular cascades of blades in any
turbomachine is really a 3D flow problem. But the simpler two-dimensional blade-to-blade flow
problem offers many advantages. It provides a natural view of cascade fluid dynamics to help
designers develop an understanding of the basic flow processes involved. Indeed, very simple two-
dimensional cascade flow models were used in this educational role long before computational
methods and computers had evolved enough to produce useful design results. Today, blade-to-blade
(B2B) flow analysis is a practical design and analysis tool that provides useful approximations to
many problems of interest. Inviscid blade-to-blade flow analysis addresses the general problem of
two-dimensional flow on a stream surface in an annular. Two-dimensional boundary layer analysis
can be included to provide an approximate evaluation of viscous effects. That approach ignores the
effect of secondary flows that develop due to the migration of low momentum boundary layer fluid
across the stream surfaces. Its accuracy becomes highly questionable when significant flow
separation is present. These limitations require particular care when analyzing the diffusing flow in
compressor cascades. They are less significant for analysis of the accelerating flow in turbine
cascades, but designers still must recognize the approximations and limitations involved. Previously,
it have been emphasized the influence of the blade surface velocity distributions on nozzle row and
rotor performance. A graph of the blade surface velocity distributions as a function of distance along
the blade surface is often referred to as the blade-loading diagram. The fundamental role of blade
loading diagrams for the evaluation of blade detailed aerodynamic designs was discussed. Blade-to-
blade flow analysis provides a practical method to calculate these blade-loading diagrams. Indeed,
blade-to-blade flow analysis is an essential part of a modern aerodynamic design system.
A Quasi-3D flow analysis employs 2D flow analyses in the hub-to-shroud and blade-to-blade surfaces
to approximate the 3D flow in a blade passage. The fundamental concept is generally credited to Wu2.
The present analysis achieves exceptional computational speed and reliability largely due to its use
of the linearized blade-to-blade flow analysis. But that also imposes some limitations on the method
that are particularly significant for turbines. Its limitation to subsonic or low transonic Mach number
levels excludes a number of turbine applications. As noted, its accuracy is compromised when it is
applied to the rather thick airfoils often used for turbines. It certainly could be extended for more
general use on turbines by substituting a more general blade-to-blade flow analysis such as the time-
marching method. But that would substantially increase the computation time required and
significantly reduce its reliability. It is very doubtful that this Quasi-3D flow analysis would remain
an attractive design tool if that were done. Indeed, it would lose most of its advantages over
commercially available viscous CFD codes while offering a less general solution. We start with some
explanation of Rotating flow, as well as, derivation of Conservation of Angular Momentum concept
which is fundamental in rotating flows, as well as blade to blade passage.
2Wu, C. H., "A General Theory of Three Dimensional Flow in Subsonic and Supersonic Turbomachines of Axial,
Radial, and Mixed Flow Types," National Advisory Committee on Aeronautics, NACA TN 2604, 1952.
22
(often with different modelling degrees of accuracy), leading to a process which is potentially error-
prone from the simulation setup point of view and often resulting in significant mismatches between
numerical and experimental data. (See Figure 1.1).
A one-way coupling approach represents one step further in increasing the accuracy of such
simulations. It can be achieved, for example, by extracting outlet profiles of the flow variables from
individual converged component simulations and applying them as inlet boundary condition profiles
to downstream component runs. Nevertheless, the inter-component interaction is still one-way and
the simulation process and results can suffer from similar drawbacks as in the totally uncoupled
workflow. (See Figure 1.2). In the two-way coupling methodology all the components are coupled
and solved simultaneously in one single simulation. This approach greatly simplifies and accelerates
the simulation workflow. Since all the components are considered simultaneously, there is no need
to prescribe boundary conditions between the various elements of the aero-engine. This avoids
running simulations where the states at the interface between the different components have to be
guessed.
𝛚= ∇×𝐮=0
Eq. 2.1
3 Ting , L., & Klein, R. (1991). Viscous Vortical Flows (Lecture Notes in Physics). Springer.
4 From Wikipedia, the free encyclopedia.
24
Ω = (0, 0, Ω) , r = (x, y, 0)
Eq. 2.2
u = Ω r = (-Ω y , Ωx , 0) → ω = u = (0, 0, 2Ω) = 2Ω
2.1.1.2.2 Irrotational Vortex
If the particle speed u is inversely proportional to the distance r from the axis, then the imaginary
test ball would not rotate over itself; it would maintain the same orientation while moving in a circle
around the vortex axis. In this case the vorticity is zero at any point not on that axis, and the flow is
said to be irrotational.
Ω = (0, 0, r -2 ) , r = (x, y, 0)
Eq. 2.3
u = Ω r = (- yr -2 , xr -2 , 0) → ω = u = 0
x = t sin(t) , y = t cos(t) , z = t
Eq. 2.4
2.1.3 Pressure in Vortex
The fluid motion in a vortex creates a dynamic
pressure (in addition to any hydrostatic
pressure) that is lowest in the core region,
closest to the axis, and increases as one moves
away from it, in accordance with Bernoulli's
Principle. One can say that it is the gradient of Figure 2.3 3D Visualization of a Vortex Curve
this pressure that forces the fluid to follow a
curved path around the axis. In a rigid-body vortex flow of a fluid with constant density, the dynamic
pressure is proportional to the square of the distance r from the axis. In a constant gravity field, the
free surface of the liquid, if present, is a concave
paraboloid.
In an irrotational vortex flow with constant fluid
density and cylindrical symmetry, the dynamic
pressure varies as P∞ − K/r2, where P∞ is the limiting
pressure infinitely far from the axis. This formula
provides another constraint for the extent of the core,
since the pressure cannot be negative. The free surface
(if present) dips sharply near the axis line, with depth
inversely proportional to r2. The shape formed by the
free surface is called a hyperboloid. The core of a vortex
in air is sometimes visible because of a plume of water
vapor caused by condensation in the low pressure and
low temperature of the core; the spout of a tornado is
an example. When a vortex line ends at a boundary
surface, the reduced pressure may also draw matter
from that surface into the core. For example, a dust
devil is a column of dust picked up by the core of an air
vortex attached to the ground. A vortex that ends at the
free surface of a body of water (like the whirlpool that
often forms over a bathtub drain) may draw a column
of air down the core (see Figure 2.4). The forward Figure 2.4 A Plughole Vortex
vortex extending from a jet engine of a parked airplane
can suck water and small stones into the core and then into the engine.
2.2 Impeller
An impeller (also written as impellor ) is a rotor used to increase (or decrease in case of turbines) the
pressure and flow of a fluid. It has been used in variety of everyday equipment such as pumps,
26
compressors, medical devices, mixing tanks, water jets and washing machines. More specifically, an
impeller is a rotating component equipped with vanes or blades used in turbomachinery (e. g.
centrifugal pumps). Flow deflection at the impeller vanes allows mechanical power (energy at the
vanes) to be converted into pump power output. Depending on the fluid flow pattern in multistage
pumps and the impellers' arrangement on the pump shaft, impeller design and arrangements are
categorised as: single-stage, multistage, single-entry, double-entry, multiple-entry, in-line (tandem)
or back-to-back arrangement. Axial and radial flow impellers are rotating industrial mixer
components designed for various types of mixing. Both types of impellers are primarily constructed
from stainless steel. Impellers impart flow. They serve the purpose of transferring the energy from
the motor to the substance of a tank as efficiently as possible. Impellers are organized by their flow
patterns.
2.2.1 Types of Impeller
The impeller of a Centrifugal Pump can be of three types as shown in Figure 2.5 as discussed by
• Open Impeller : where the vanes are cast free on both sides.
• Semi-Open Impeller : when the vanes are free on one side and enclosed on the other.
• Enclosed Impeller : The vanes are located between the two discs, all in a single casting 5.
5 Presented by: Matt Prosoli, “Centrifugal Pump Overview”, Pumps Plus Inc.
6 See Previous.
27
Figure 2.6 Flow Direction of Three Different Pumps/Impellers (Courtesy of Global Spec)
radial flow impellers are the mixing of very viscous fluids. Axial flow impellers impose essentially
bulk motion, and are used on homogenization processes, in which increased fluid volumetric flow
rate is important. Impellers can be further classified principally into three sub-types7
• Propellers
• Paddles
• Turbines
2.2.4 Axial Impellers
Are best for mixing applications that require stratification or solid suspension. Axial impellers are set
up to create effective top to bottom motion in the tank. This motion is highly effective when placed
over the center of a baffled tank. Some common types of axial flow impellers include: marine
impellers, pitched blade impellers, and hydrofoils. Hydrofoil impellers are also known as high
efficiency impellers. They are a popular choice for applications that require a range from general
blending to storage tanks. This is largely due to the greatest pumping per horsepower, cost
effectiveness, and are ideal for shear sensitive applications.
2.2.5 Radial Impellers
are designed in 4-6 blades. In radial flow impellers, the fluid moves perpendicularly to the impeller.
They produce a radial flow pattern which moves the contents of the mixing tank to the sides of the
vessel. The radial flow impacts the side which causes in either an up or down direction which fills the
top and the bottom of the impeller to be ejected once more. It is also important to note that setting
up baffles helps to minimize vortex and swirling motions in the tank, therefore, enhancing agitation
efficiency. Radial impellers are a great fit for low-level applications inside longer tanks based upon
the production of higher shear due to the angle of attack.
2.2.6 Power Number for Impeller
Power number is a value specific to mixing impellers which describes the impeller's power
consumption. The formula for calculating an impeller's power number is
p
Np = Eq. 2.5
n D 5ρ
3
where Np = power number, P = impeller power in watts, ρ = density of tank liquid in kg/m3, n = shaft
speed in revolutions/second and D=impeller diameter in meters. Because of the difficulty in
obtaining many of these values, power numbers can be considered the summary of various
correlated test results (when dealing with standard-sized mixing tank) rather than a precise
specification. Therefore, manufacturers often specify an impeller's power number as a function of its
power and size8.
2.3 Pumps
A pump is a device that moves fluids (liquids or gases), or sometimes slurries, by mechanical action.
Pumps can be classified into three major groups according to the method they use to move the fluid:
direct lift, displacement, and gravity pumps. Pumps operate by some mechanism (typically
reciprocating or rotary), and consume energy to perform mechanical work by moving the fluid.
Pumps operate via many energy sources, including manual operation, electricity, engines, or wind
power, come in many sizes, from microscopic for use in medical applications to large industrial
pumps. Mechanical pumps serve in a wide range of applications such as pumping water from wells,
aquarium filtering, pond filtering and aeration, in the car industry for water-cooling and fuel
injection, in the energy industry for pumping oil and natural gas or for operating cooling towers. In
the medical industry, pumps are used for biochemical processes in developing and manufacturing
medicine, and as artificial replacements for body parts, in particular the artificial heart and penile
prosthesis. (see Figure 2.9),
• Single stage pump : When in a casing only one impeller is revolving then it is called single
stage pump.
• Multi stage pump: When in a casing two or more than two impellers are revolving then it is
called double/multi stage pump.
Pumps are used throughout society for a variety of purposes. Early applications includes the use of
the windmill or watermill to pump water. Today, the pump is used for irrigation, water supply,
gasoline supply, air conditioning systems, refrigeration (usually called a compressor), chemical
movement, sewage movement, flood control, marine services, etc. Because of the wide variety of
applications, pumps have a plethora of shapes and sizes: from very large to very small, from handling
gas to handling liquid, from high pressure to low pressure, and from high volume to low volume.
9 Miha Meznar, “Fluid Flows In Rotating Frames”, University of Ljubljana, March 2005.
10 Here the focus is not on the recirculation flow but rather on the instability patterns in rotating fluids.
31
side. The bottom disk is made of rectified brass, with a black coating to improve visualization
contrast. To allow the differential rotation the radius of the bottom disk is slightly smaller (a tenth of
millimeter) than the radius of the shrouding cylinder. The thickness h of the cell can be varied
between few mm up to several The cell is filled with a mixture of water, glycerol and small anisotropic
flakes. The latter enable us to visualize the fluid flow. The flakes' orientation with the fluid leads to
variations of the reflected light. For example, if the flakes are mainly horizontal, they reflect light, if
they are vertical they do not reflect it so well. The kinematic viscosity ν = μ/ρ lies between 1x10−6 <
ν < 8 x10−6 m2/s due to different concentration of glycerol11. Each of two disks rotate with its own
angular velocity Ωi, where index i = b, where t stands for bottom and top disk respectively. Angular
velocities of the disks range from 0 to 10 rad/s but the upper disk rotates anticlockwise only, whereas
the bottom one can rotate clock- or anticlockwise. Anticlockwise rotation is taken positive. We call
co-rotation the situation where both disks rotate in the same direction (b and t are of the same sign)
and counter-rotation when the disks rotate in the opposite directions (they have opposite signs). If
one of the disks is left fixed, the other rotating, the regime is called rotor-stator regime. We will define
some dimensionless numbers that describe our cell. The first is radius-to-height ratio defined as Γ=
R/h , where R is radius and h height of the cell. The second number is Reynolds number Rei = Ωih2/ν
, where index i = b, t denotes the bottom and top disk respectively, i is the angular velocity of the disks
and ν the kinematic viscosity. The last number is rotation ratio defined as s = Ωb/Ωt = Reb/Ret .
Rotation ratio is positive (s > 0) in the co-rotation regime and negative (s < 0) in the counter-rotation
regime.
2.4.1.1 Recirculating Flow
Each rotation is associated with a meridian recirculating flow, which can be inward or outward
depending on the rotation ratio. For arbitrary positive and small negative rotation ratio s, the radial
recirculating flow is roughly the same as in the rotor-stator case (s = 0): it consists of an outward
boundary layer close to the faster disk and an inward boundary layer close to the slower disk. At
small negative rotation ratio the centrifugal effect of the slower disk is not strong enough to
counteract the inward flow from the faster disk. But as the rotation ratio s is decreased below −0.2,
the slower disk induces a centrifugal flow too, and the radial recirculating flow appears to come
organized into two-cell recirculating structure. At the interface of these two cells a strong shear layer
takes place. The centrifugal flow induced by the faster disk recirculates towards the center of the
slower disk due to the lateral end wall. This inward recirculation flow meets the outward radial flow
induced by the slower disk, leading to a stagnation circle where the radial component of the velocity
vanishes.
2.4.1.2 Instability Flow Patterns
We now turn to the instability patterns of the flow between two rotating disks close to each other (Γ
= 20.9), in both co- and counter-rotating flows. For s ≥ 0 (rotor-stator or co-rotation) and Reb fixed,
on increasing Ret, propagating circular structures are first observed. These axisymmetric vortices
appear close to the landrail wall, propagate towards the center and disappear before reaching the
center of the cell. Above a secondary threshold of Ret, spiral structures appear at the periphery of the
disks, and circles remain confined between two critical radii (Figure 2.11 (a)). These spirals are
called positive spirals (denoted S+) since they roll up to the center in the direction of the faster disk
(here the top one). Increasing Ret further, positive spirals progressively invade the whole cell. Still
increasing Ret, the flow becomes more and more disordered (denoted D, Figure 2.11 (c)). It can be
shown that co-rotation shifts upwards the instability thresholds for circles and positive spirals12.
However, threshold line for circles is parallel to the solid body rotation (b = t) indicating that the
11 G. Gauthier, P. Gondret, F. Moisy and M. Rabaud, “Instabilities in the flow between co- and counter-rotating
disks”, J. Fluid Mech, volume 473, pp. 1-21, 2002.
12 Miha Meznar, “Fluid Flows In Rotating Frames”, University of Ljubljana, March 2005.
32
angular velocity difference Ω = Ωt − Ωb is the only control parameter of this instability and no influence
of the global rotation occurs. By contrast, the borderline for the positive spirals has a larger slope
than the solid body rotation line; in this case the relative angular velocity Ω is not the only control
parameter and an extra velocity of the upper disk is needed for the spirals to arise. The global rotation
in this case has a stabilizing effect.
For s < 0 (counter-rotating case) the onset of the instability patterns depends on the Reynolds
numbers of both disks. For low bottom Reynolds number, −11 < Reb < 0, on increasing the Reynolds
number of the upper disk, the appearance of the instability patterns is the same as in the rotor-stator
or co-rotation case: axisymmetric propagating vortices, positive spirals and disorder. But, for −18 <
Reb < −11, spirals of a new kind appear on increasing Ret. These spirals are said to be negative (and
denoted S−) since they now roll up to the center in the direction of the slower counter-rotating disk
(Figure 2.12 (a)). Unlike circles and positive spirals, negative spirals extend from the periphery to
the center, they invade the whole cell. Also, the onset time for negative spirals is much longer than
for positive ones or circles; when the onset is carefully approached from below, the growth time of
negative spirals can exceed 15 minutes which strongly contrasts circles and positive spirals which
appear almost instantaneously. Increasing Ret further, positive spirals appear as well at the periphery
of the disk, as can be seen in Figure 2.12 (b). Here negative and positive spirals seem to coexist
without strong interaction, which indicates the difference in their origin. The circles and positive
spirals have their origin in the boundary layer instability whereas negative ones, on the other hand,
originate from shear layer instability.
Still increasing Ret, negative spirals disappear and positive spirals alone remain (Figure 2.12(c)).
Increasing Ret yet further, circles appear as in the co-rotation case. Still increasing Ret, the structures
become disorganized and the flow becomes turbulent. For Reb < −18 the negative spirals described
above become wavy, the flow is more and more disorganized and continuously becomes turbulent
without a well-defined threshold. Depending on the Reynolds number, the disorder can be generated
first at the periphery or in the center and then invades the entire cell. Up to now our instability
patterns were limited to radius-to-height ratio Γ = 20.9. Does anything changes if one changes it?
Researchers enlarged the gap h between the disks (Γ diminishes) and observed a new pattern that
consisted of a sharp-cornered polygon of m sides, surrounded by a set of 2m outer spiral arms. These
polygons arise only for small Γs (less than approx. Γ = 10). For higher values the vertical confinement
leads to a saturated pattern where inner arms, connecting the corners of the polygon to the center of
polygons, turn into negative spirals. Another interesting property of the patterns is that they are not
fixed but rather rotate as a whole. Therefore we define the azimuthal phase velocity ωφ in the
laboratory frame. It corresponds to the angular velocity of the global rotation of the spiral pattern.
For the S+ spirals ωφ is always positive (anticlockwise), i.e. the positive spirals rotate in the direction
of the faster (top) disk, regardless of motion of the bottom one. S− spirals, on the other hand change
sign of ωφ. It means that for small Ret the pattern rotates in the direction of the slower (bottom) disk
while at higher Ret it moves with the top (faster) disk. Here only compare the directions of the disks
and phase velocity. The size of phase velocity is only a fraction of the disk velocities. We see that the
co-rotation flow (Reb > 0, right-hand part of the diagram) is qualitatively the same as the rotor-stator
flow (vertical line Reb = 0); the thresholds of instabilities (circles C and positive spirals S+) are found
to increase just with the bottom Reynolds number. By contrast, the counter-rotating case (Reb < 0,
left-hand part) is much more rich.
Figure 2.14 Experimental Pressure Iso-Surfaces ; Left - Without Sweep ;Right - With Forward Sweep
(Courtesy of RÁBAI and VAD [93])
secondary flow created by turning the incoming end-wall boundary layer within the blade passage,
and therefore reduce the cross-passage flow of end-wall fluid toward the suction surface. Aft sweep
will tend to add to the classical secondary flow and there is then more loss and blockage buildup on
the suction surface.
Forward sweep leads to enlargement of the meridional stream-tube near the end-wall ahead of the
leading edge followed by a contraction and then again an opening further downstream. The sequence
leads to a corresponding acceleration,
deceleration, and acceleration of the end-wall
flow relative to that for un-swept blades; the
consequence is usually lower end-wall loss (see
Figure 2.14)13. Aft sweep has the opposite effects
and leads to an increase in end wall loss. Dihedral,
on the other hand, causes the bound vortex and its
image to induce velocities more or less parallel to
the primary flow. If the dihedral is positive, so that
the suction surface makes an copious angle with
the end-wall, the induced velocities tend to reduce
the velocity peak on the suction surface, which
reduces the diffusion near the suction surface
end-wall corner.
13Gergely Rábai And János Vad, “Validation Of A Computational Fluid Dynamics Method To Be Applied To Linear
Cascades Of Twisted-Swept Blades”, Periodica Polytechnica Ser. Mech. Eng. Vol. 49, No. 2, Pp. 163–180 (2005).
35
profiles obtained. In principle, an infinite number of blades are required in order to keep the
periodicity of the blades in the original machine. The two-dimensional flow is representative for the
real three-dimensional flow. The expressions for axial momentum are the same in both
configurations. Moment of momentum with the three-dimensional flow corresponds to tangential
momentum with the cascade (tangential = circumferential)14.
Constant tangential momentum along the blade height with the linear cascade is then equivalent to
constant angular momentum (vur = constant) along the radius in the real machine. This means then
that the three-dimensional flow is a so-called free vortex flow. The term means swirling flow with
constant angular momentum.
14Erik Dick, “Fundamentals of Turbomachines”, Springer Dordrecht Heidelberg New York London. ISBN 978-
94-017-9626-2.
36
𝐮𝐫 = 𝐮 − 𝛚
⏟×𝐫 where 𝛚 = 𝛚 𝐚̂
whirl velocity
Eq. 3.1
In the above, ur is the relative velocity (the velocity viewed from the rotating frame), u is the
absolute velocity (the velocity viewed from the stationary frame), ω x r is the whirl (or moving)
velocity (the velocity due to the moving frame), and â is unit directional vector depending or rotation
direction. When the equations of motion are solved in the rotating reference frame, the acceleration
of the fluid is augmented by additional terms that appear in the momentum equations (Batchelor,
2000)16. Moreover, the equations can be formulated in two different ways:
• Expressing the momentum equations using the relative velocities as dependent variables
(known as the relative velocity formulation).
• Expressing the momentum equations using the absolute velocities as dependent variables in
the momentum equations (known as the absolute velocity formulation).
The exact forms of the governing equations for these two formulations will be provided in the
sections below. It can be noted here that pressure-based solvers provide the option to use either of
these two formulations, whereas the density-based solvers always use the absolute velocity
formulation.
ωω
∂ρ
Mass + ∇. (ρ𝐮r ) = 0
∂t
∂
(ρ𝐮r ) + ∇. (ρ𝐮r 𝐮r ) + ρ (2𝛚
⏟ × 𝐮r + ⏟ 𝛚 × 𝛚 × 𝐫) = −∇p + ∇𝛕rij + 𝐅
∂t Centrifugal
Coriolis
∂
Energy (ρEr ) + ∇. (ρ𝐮r Hr ) = ∇. (k∇T + 𝛕rij ur ) + 𝐒h
∂t
p 1 2 p
Er = h − + [𝐮r − (𝛚 × 𝐫)2 ] , Hr = Er +
ρ 2 ρ
Eq. 3.2
Here, the subscript r is for relative quantities, the momentum equation contains two additional
acceleration terms, the Coriolis acceleration (2ω x ur), and the Centrifugal acceleration (ω x ω x r).
In addition, the viscous stress τijr is defined as before except that relative velocity derivatives are
used. The energy equation is written in terms of the relative internal energy (Er) and the relative total
∂ρ
Mass + ∇. (ρur ) = 0
∂t
∂
Momentum (ρu) + ∇. (ρur u) + ρ (ω
⏟ × u) = −∇p + ∇τij + F
∂t I
∂
Energy (ρE) + ∇. (ρur H + p(ω × r) = ∇. (k∇T + τij . u) + Sh
∂t
Eq. 3.3
In this formulation, the Coriolis and Centripetal accelerations can be collapsed into a single term (I).
Be advised that from now on we will be dealing with linear momentum if noted otherwise19.
3.1.4 Early Formulation and Consideration
There are more simplified version of equations depending to the type of analysis. One particular
version is axisymmetric which usually governs the aerospace cruse condition. For example a time-
marching finite volume numerical procedure is presented for 3D Euler analysis of turbomachinery
flows by [Soulis]20. Another pioneering and early studies in the subject prediction of turbomachinery
performance, as indicated by the flow in the through-flow, or axisymmetric plane, is done by [Siebert
and Yocum]21 when only steady-state and incompressible flows are considered.
reference frame. In such cases, you must break up the problem into multiple cells zones, with well-
defined interfaces between the zones (MRF). The manner in which the interfaces are treated leads
to three approximate, steady-state modeling methods for this class of problem:
• Single Reference Frame (SRF)
• Multiple Reference Frame (MRF)
• Mixing Plane Method (MPM)
If unsteady interaction between the stationary and moving parts is important, you can employ the
Sliding Mesh approach, or more simplified Non-Uniform Harmonic Balanced Method, to capture
the transient behavior of the flow.
3.2.1 Single Rotating Reference Frame (SRF) Modeling
Many problems permit the entire computational domain to be referred to as a single rotating
reference frame (SRF modeling). In such cases, the equations for a Rotating Reference Frame are
solved in all fluid cell zones. Steady-state solutions are possible in SRF models provided suitable
boundary conditions are prescribed. In particular, wall boundaries must adhere to the following
requirements:
• Any walls which are moving with the reference frame can assume any shape. An example
would be the blade surfaces associated with a pump impeller. The no slip condition is defined
in the relative frame such that the relative velocity is zero on the moving walls.
• Walls can be defined which are non-moving with respect to the stationary coordinate system,
but these walls must be surfaces of revolution about the axis of rotation. Here the so slip
condition is defined such that the absolute velocity is zero on the walls. An example of this
type of boundary would be a cylindrical wind tunnel wall which surrounds a rotating
propeller.
Rotationally periodic boundaries may also be used, but the surface must be periodic about the axis
of rotation. As an example, it is very common to model through a blade row on a turbomachine by
assuming the flow to be
rotationally periodic and using a
periodic domain about a single
blade. This permits good
resolution of the flow around the
blade without the expense of
model all blades in the blade row
(see Figure 3.3). Flow boundary
conditions (inlets and outlets) can
be in most cases prescribed in
either the stationary or rotating
frames. For example, for a velocity
inlet, one can specify either the
relative velocity or absolute
velocity, depending on which is
more convenient. In some cases Figure 3.3 Single Blade Model with Rotationally Periodic
(e.g. pressure inlets) there are Boundaries
restrictions based upon the
velocity formulation which has been chosen. For additional information, reader should refer to
FLUENT 6.3® user manual.
40
3.2.1.1 Case Study - Aerodynamics and Structural Analysis of (HAWT) Wind Turbine Blade
Wind blade design is determined using blade element momentum (Thresher & Dodge, 1998)23. The
blade plays a pivotal role, because it is the most important part of the energy absorption system.
Practical horizontal axis wind turbine (HAWT) designs use airfoils to transform the kinetic energy in
the wind into useful energy and it has to be designed carefully to enable to absorb energy with its
greatest efficiency. There are many factors for selecting a profile. One significant factor is the chord
length and twist angle which depend on various values throughout the blade. In this work, the airfoil
sections used in horizontal axis wind turbine (HAWT) are S818; S825 and S826 airfoils used in NREL
phase 2 and phase 3 wind turbines. They have several advantages in meeting the intrinsic
requirements for wind turbines in terms of design point, off-design capabilities and structural
properties. The lift and drag goes beyond the traditional aim of capacity maximization, contributing
also for organization’s profitability and value. Indeed, lean management and continuous
improvement approaches suggest capacity optimization instead of maximization. The study of
capacity optimization and costing models is an important research topic that deserves contributions
from both the practical and theoretical perspectives. This paper presents and discusses a
mathematical model for capacity management based on different costing models (ABC and TDABC).
A generic model has been developed and it was used to analyze idle capacity and to design strategies
towards the maximization of organization’s value. The trade-off capacity maximization vs
operational efficiency is highlighted and it is shown that capacity optimization might hide operational
inefficiency.
3.2.1.1.1 Introduction
The power efficiency of wind energy systems has a high impact in the economic analysis of this kind
of renewable energies. The efficiency in these systems depends on many subsystems: blades,
gearbox, electric generator and control. Some factors involved in blade efficiency are the wind
features, e.g. Its probabilistic distribution, the mechanical interaction of blade with the electric
generator, and the strategies dealing with pitch and rotational speed control 24. It is a complex
problem involving many factors, relations and constraints. The design of optimal blades involves
aerodynamic, structural and control problems. However, the design cycle can be practically
approached as an iterative and stepped method. For aerodynamic optimization the blade can be
modelled as a series of sections along the pitch axis. Each section has an airfoil shape, chord length
and attach angle which is the result of a collective pitch angle and a local twist one. This last is a
property of the blade while the pitch angle depends on the control strategy of the whole energy
system.
The computation of the wind flow around rotating blades is a very complex problem. For a precise
knowledge of the wind flow and the induced forces in the turbine surfaces it is necessary to solve the
three-dimensional Navier-Stokes equations in a rotating frame, but the computational cost to obtain
such precise solution prohibits their use in the design and analysis environments (Thresher & Dodge,
1998). The blade element momentum theory (BEM) is basically a one-dimensional simplified theory
that is used routinely by wind power industry because it provides reasonably accurate prediction of
performance25. The BEM theory has shown to give good accuracy with respect to time cost, and at
moderate wind speeds, it has sufficed for blade geometry optimization 26-27-28. The BEM theory is the
23 Thresher, R. W., & Dodge, D. M. (1998). Trends in the evolution of wind turbine generator configurations and
systems. WIND ENERGY.
24 K.Y. Maalawi, M.A. Badr, A practical approach for selecting optimum wind rotors, Renewable Energy, (2003).
25 J.L. Tangler, The Nebulous art of using wind-tunnel airfoil data for predicting rotor Performance, 5 (2002).
26 M.M. Duquette, K.D. Visser, Numerical implications of solidity and blade number on rotor performance of
horizontal-axis wind turbines, J. Sol. Energy Eng.-Trans, ASME. 125 (2003) 425–432, 2003.
27 P. Fuglsang, H.A. Madsen, Optimization method for wind turbine rotors, J. Wind E, Ind. Aerodynamics. (1999).
28 M. Jureczko, M. Pawlak, A. Mezyk, Optimisation of wind turbine blades, J. Mater, Proc. Technol. 167 (2005).
41
composition of two different approaches to study the forces in a wind turbine. The first is the
momentum theory that studies the global changes in wind momentary, axial and tangential, in an
ideal turbine. Changes in axial and rotational moment between upwind and downwind induce thrust
and torque respectively in the rotor. The wind flow is split in many differential non-interacting
annular stream tubes. The second theory, the blade element, studies the aerodynamic forces acting
in a local airfoil. As in aeronautics wing theory, the forces are lift, which is perpendicular to the wind
direction, and drag that is in the same direction. Drag is mainly generated by friction between the
viscous fluid and the airfoil surface. It is a dissipative force that generates power loss and lack in
momentum changes. The applications of thick airfoils are extended to the assessment of wind turbine
performance. It is well established that the power generated by a Horizontal-Axis Wind Turbine
(HAWT) is a function of the number of blades B, the tip speed ratio λ (blade tip speed/wind free
stream velocity) and the lift to drag ratio (CL/CD) of the airfoil sections of the blade. The airfoil
sections used in HAWT are generally thick airfoils such as the S818, S825 and S826 of airfoils. These
airfoils vary in (CL/CD) for a given B and λ, and therefore the power generated by HAWT for different
blade airfoil sections will vary. Another goal of this study is to evaluate the effect of different airfoil
sections on HAWT performance using the Blade Element Momentum (BEM) theory.
3.2.1.1.2 Airfoils and General Concepts of Aerodynamics
A number of terms are used to characterize an airfoil. The mean camber line is the locus of points
halfway between the upper and lower surfaces of the airfoil. The most forward and rearward points
of the mean camber line are
on the leading edge and
trailing edges, respectively.
The straight line connecting
the leading and trailing
edges is the chord line of the
airfoil, and the distance from
the leading to the trailing
edge measured along the
chord line is designated as
the chord of the airfoil. The Figure 3.4 Airfoil Cross-Sections used in the Design of the Wind
thickness is the distance Turbine Blades
between the upper and
lower surfaces, also
measured perpendicular to the chord line. Finally, the angle of attack α is defined as the angle
between the relative wind and the chord line. (See Figure 3.4)
3.2.1.1.3 Lift, Drag and Non-Dimensional Parameters
Theory and research have shown that many flow problems can be characterized by non-dimensional
parameters. The most important non-dimensional parameter for defining the characteristics of fluid
flow conditions is the Reynolds number. Force and moment coefficients, which are a function of
Reynolds number, can be defined for two or three-dimensional objects. Force and moment
coefficients for flow around two-dimensional objects are usually designated with a lower case
subscript lift and drag coefficients that are measured for flow around two- or three-dimensional
object are usually designated with an upper case subscript. Rotor design usually uses two
dimensional coefficients, determined for a range of angles of attack and Reynolds numbers, in wind
tunnel tests. The two-dimensional lift and drag coefficients is defined as:
42
L/l D/l
CL = , CD =
(1/2)ρU 2 c (1/2)ρU2 c
Eq. 3.4
where ρ is the density of air, U is the velocity of undisturbed airflow, c is the airfoil chord length and
l is the airfoil span. Other dimensionless coefficients that are important for the analysis and design of
wind turbines include the power and thrust coefficients and the tip speed ratio, the pressure
coefficient, which is used to analyze airfoil flow:
p − p∞
Cp =
(1/2)ρU2
Eq. 3.5
Under similar ideal conditions, symmetric airfoils of finite thickness have similar theoretical lift
coefficients. This would mean that lift coefficients would increase with increasing angles of attack
and continue to increase until the angle of attack reaches 90 degrees. The behavior of real symmetric
airfoils does indeed approximate this theoretical behavior at low angles of attack. For example,
typical lift and drag coefficients for a S818, S825, and S826airfoils,the profiles of which are shown in
Figure 3.5 (a-d) as a function of angle of attack and Reynolds number.
43
Note that, in spite of the very good correlation at low angles of attack, there are significant differences
between actual airfoil operation and the theoretical performance at higher angles of attack. The
differences are due primarily to the assumption, in the theoretical estimate of the lift coefficient, that
air has no viscosity. Surface friction due to viscosity slows the airflow next to the airfoil surface,
resulting in a separation of the flow from the surface at higher angles of attack and a rapid decrease
in lift. This condition is referred to as stall. Airfoils for horizontal axis wind turbines (HAWTs) often
are designed to be used at low angles of attack, where lift coefficients are fairly high and drag
coefficients are fairly low. The lift coefficient of this symmetric airfoil is about zero at an angle of
attack of zero and increases to over 1.0 before decreasing at higher angles of attack. The drag
coefficient is usually much lower than the lift coefficient at low angles of attack. It increases at higher
angles of attack. The lift coefficient at low angles of attack can be increased and drag can often be
decreased by using a cambered airfoil.
3.2.1.1.4 Blade Element Momentum (BEM) Theory
The blade element momentum (BEM) theory is a compilation of both momentum theory and blade
element theory (M. Jureczko, M. Pawlak, A. Mężyk, 2005)29 and (Ernesto Benini, 2003)30. Momentum
theory, which is useful in predicted ideal efficiency and flow velocity, is the determination of forces
acting on the rotor to produce the motion of the fluid. Theory determines the forces on the blade as
a result of the motion of the fluid in terms of the blade geometry. By combining the two theories,
BEM theory, also known as strip theory, relates rotor performance to rotor geometry.
3.2.1.1.5 Aerodynamic Load
Aerodynamic load is generated by lift and drag of the blades airfoil section, which is dependent on
wind velocity, blade velocity, angle of attack and yaw31. The angle of attack is dependent on blade
twist and pitch. The aerodynamic lift and drag produced are resolved into useful thrust in the
direction of rotation absorbed by the generator and reaction forces. It can be seen that the reaction
forces are substantial acting in the flat wise bending plane, and must be tolerated by the blade with
limited deformation. For calculation of the blade aerodynamic forces the widely publicized blade
element momentum (BEM) theory is applied. Working along the blade radius taking small elements
δr, the sum of the aerodynamic forces can be calculated to give the overall blade reaction and thrust
loads.
3.2.1.1.6 Blade Geometry
The main objective in the design of wind turbines is to find a rotor that meets the basic conditions
requested. The most important condition is to get a rotor to deliver output power required at a
particular speed. For this, the first assumption of the aerodynamic rotor is its diameter, which can be
roughly estimated power. In addition, it is necessary to take into account the importance of the
geometry of the rotor, taking into consideration the most important, the aerodynamic performance,
strength and stiffness conditions, and costs. However, power generation through wind turbines also
play a decisive role in the design of the aerodynamics of the rotor, which is influenced by other
parameters such as power generator and control system. The results of the optimal distribution of
the cord and twist for a blade of 43.2 m in diameter and having various profiles.
29 M. Jureczko, M. Pawlak, A. Mężyk, Optimisation of wind turbine blades, Journal of Materials Processing
Technology, Volume 167, Issues 2–3, 2005, Pages 463-471, ISSN 0924-0136,
https://doi.org/10.1016/j.jmatprotec.2005.06.055
30 Ernesto Benini, Significance of blade element theory in performance prediction of marine propellers, Ocean
Engineering, Volume 31, Issues 8–9, 2004, Pages 957-974, ISSN 0029-8018,
https://doi.org/10.1016/j.oceaneng.2003.12.001
31 M.O.L. Hansen, Aerodynamics of Wind Turbines: Rotors, Loads and Structure, James & James Science
Figure 3.6 (Top) Airfoils Superposed on the Wind Turbine Blade and (Bottom) Top View of a Subset of
the Airfoil Cross-Sections illustrating Blade Twisting
32 Saaa E. Mouhsine, Karim Oukassou, Mohammed Marouan Ichenial, Bousselham Kharbouch, Abderrahamane
Hajraoui, “Aerodynamics and Structural Analysis of Wind Turbine Blade”, 11th International Conference
Interdisciplinarity in Engineering, INTER-ENG 2017, 5-6 October 2017, Romania.
33 D. C. Wilcox, Formulation of the k–ω Turbulence Model Revisited, 11, AIAA Journal, 823–2838, 2008.
34 F. R. Menter, “Zonal Two Equation k-ω Turbulence Models for Aerodynamic Flows”, AIAA Paper, Vol. 93, 1993.
45
regions and regions with strong acceleration. Details of the SST 𝑘−𝜔 turbulence model is available in
[Mouhsine et al.]35.
3.2.1.1.9 Mesh Generation
In order to create the computational domain and generate mesh, the commercially available software
“ANSYS Meshing tool” is used to build a wind tunnel model and generate an unstructured mesh
around the blade in the computational domain. As shown in Figure 3.6, a 3D straight untampered
blade is placed inside a computational domain (mimicking a wind tunnel) with inflow and outflow
boundaries. The wall boundary condition is applied to the right and lift surface of the computational
domain. The back and front surfaces of the computational box are set as symmetry boundary
condition due to the free motion of air on these surfaces. As shown, one important part of the mesh
shape is that it must be smooth and dense enough to be suitable for any arbitrary airfoil shape and
3D Horizontal-Axis Wind Turbine. The wind tunnel geometry is always the same, but the airfoil/blade
in the center of the tunnel changes from one generation to the next. This poses a challenge for mesh
generation in 3D. Faces are meshed using quadrilateral cells, and we require that the number of
nodes on opposite faces be identical. To ensure this distribution, we define a set number of nodes
(and not relative node spacing) along each edge. Otherwise, thicker or more cambered airfoil edges
would have more nodes than thinner ones if a relative distribution was used. A refined boundary
layer is carefully constructed around the airfoil to capture the boundary layer behavior.
35Saaa E. Mouhsine, Karim Oukassou, Mohammed Marouan Ichenial, Bousselham Kharbouch, Abderrahamane
Hajraoui, “Aerodynamics and Structural Analysis of Wind Turbine Blade”, 11th International Conference
Interdisciplinarity in Engineering, INTER-ENG 2017, 5-6 October 2017, Romania.
46
Figure 3.8 Static Pressure, Velocity Magnitudes, Deformation and Stress Distribution
47
at the inlet of 12 m/s with turbulent intensity of 5% and turbulent viscosity ratio of 10 and the
Pressure of 1 atm in order to validate the present simulation. As mentioned in the beginning of this
work, the aerodynamic performance of wind turbines are primarily a function of the steady state
aerodynamics that is discussed. The analysis presented provides a method for determining average
loads on a wind turbine. However, a number of important steady state and dynamic effects that cause
increased loads or decreased power production from those expected with the BEM theory presented
here, especially increased transient loads. Figure 3.7 gives the obtained result of the force analysis
on the airfoil sections. Figure 3.8 (a-d) show static pressure, velocity magnitudes, deformation and
stress distribution of the Horizontal-Axis Wind Turbine (HAWT).
3.2.1.1.11 Conclusion
In this study, we applied the finite element model of aerodynamics and static structural analyses of
Horizontal Axis Wind Turbine (HAWT). The first part of the paper focused on the wind turbine
geometry modeling, mesh generation, and numerical simulation of Horizontal-Axis Wind Turbine
(HAWT). The fluid and structural meshes are compatible at the interface and may be employed for
the coupled FSI analysis. These aerodynamic models have been coupled with a nonlinear formulation
describing the structural dynamics to moderate deformations. A comprehensive look at blade design
has shown that an efficient blade shape is defined by aerodynamic calculations based on chosen
parameters and the performance of the selected airfoils. Aesthetics plays only a minor role. The
optimum efficient shape is complex consisting of airfoils sections of increasing width, thickness and
twist angle towards the hub. This general shape is constrained by physical laws and is unlikely to
change. However, airfoils lift and drag performance will determine exact angles of twist and chord
lengths for optimum aerodynamic performance. Due to the large and flexible structure of the wind
turbine blades, there will probably be aeroelastic instability. The displacement of the tip of the blade
at the nominal wind speed (12 m/s) is obtained (0.045 m) a reduction in the power performance of
the turbine, which implies a reduction in the rated power. In order to do the intensive study of the
structural models and aeroelastic behavior of the blade, the aerodynamic is constructed correctly.
3.2.2 Flow in Multiple Rotating Reference Frames (MRF)
Many problems involve multiple moving parts or contain stationary surfaces which are not surfaces
of revolution (and therefore cannot be used with the Single Reference Frame modeling approach).
For these problems, you must break up the model into multiple fluid/solid cell zones, with interface
boundaries separating the zones. Zones which contain the moving components can then be solved
using the moving reference frame equations, whereas stationary zones can be solved with the
stationary frame equations. The manner in which the equations are treated at the interface lead to
two approaches:
➢ Multiple Reference Frame Model (MRF)
➢ Mixing Plane Model (MPM)
➢ Sliding Mesh Model (SMM)
Both the MRF and Mixing plane approaches are steady-state approximations, and differ primarily
in the manner in which conditions at the interfaces are treated. These approaches will be discussed
in following sections. The sliding mesh model approach is, on the other hand, inherently unsteady
due to the motion of the mesh with time. The MRF model is, perhaps, the simplest of the two
approaches for multiple zones. It is a steady-state approximation in which individual cell zones move
at different rotational and/or translational speeds. The flow in each moving cell zone is solved using
the moving reference frame equations. If the zone is stationary (ω = 0), the stationary equations are
used. At the interfaces between cell zones, a local reference frame transformation is performed to
enable flow variables in one zone to be used to calculate fluxes at the boundary of the adjacent zone.
It should be noted that the MRF approach does not account for the relative motion of a moving zone
48
with respect to adjacent zones (which may be moving or stationary); the grid remains fixed for the
computation. This is analogous to freezing the motion of the moving part in a specific position and
observing the instantaneous flow field with the rotor in that position. Hence, the MRF is often
referred to as the frozen rotor approach. While the
MRF approach is clearly an approximation, it can
provide a reasonable model of the flow for many
applications. For example, the MRF model can be
used for turbomachinery applications in which
rotor-stator interaction is relatively weak, and the
flow is relatively uncomplicated at the interface
between the moving and stationary zones. In
mixing tanks, for example, since the impeller-baffle
interactions are relatively weak, large-scale
transient effects are not present and the MRF
model can be used. Another potential use of the
MRF model is to compute a flow field that can be
used as an initial condition for a transient sliding
mesh calculation. This eliminates the need for a
startup calculation. The multiple reference frame Figure 3.9 Mixing Tank Geometry with One
model should not be used, however, if it is Rotating Impeller
necessary to actually simulate the transients that
may occur in strong rotor-stator interactions, the
sliding mesh model alone should be used.
3.2.2.1 Case Study – Mixing Tank
In a mixing tank with a single impeller, you can define a rotating reference frame that encompasses
the impeller and the flow surrounding it, and use a stationary frame for the flow outside the impeller
region. An example of this configuration is illustrated in Figure 3.9. (The dashes denote the interface
between the two reference frames). Steady-state flow conditions are assumed at the interface
between the two
reference frames. That is,
the velocity at the
interface must be the
same (in absolute terms)
for each reference frame.
The grid does not move.
You can also model a
problem that includes
more than one rotating
reference frame. Figure
3.10 shows a geometry
that contains two
rotating impellers side by
side. This problem would
be modeled using three Figure 3.10 Mixing Tank with Two Rotating Impellers
reference frames: the
stationary frame outside
both impeller regions and two separate rotating reference frames for the two impellers. (As noted
above, the dashes denote the interfaces between reference frames.)
49
𝐮 = 𝐮f + 𝛚
⏟×𝐫 + 𝐮t , ∇𝐮 = ∇𝐮f + ∇(𝛚 × 𝐫)
Swirl Velocity
Eq. 3.6
Note that scalar quantities such as density, static pressure, static temperature, species mass fractions,
etc., are simply obtained locally from adjacent cells.
3.2.3.2 Interface Treatment: Absolute Velocity Formulation
When the absolute velocity formulation is used, the governing equations in each subdomain are
written with respect to that subdomain's reference frame, but the velocities are stored in the absolute
frame. Therefore, no special transformation is required at the interface between two subdomains.
Again, scalar quantities are determined locally from adjacent cells.
50
3.2.3.3 Case Study 1 - Experiments and CFD Calculations on the Performance of a Non-Reversible
Axial Fan36
36 BACAK Aykut, ÜNVERDİ, Salih Özen, “Experiments and CFD Calculations on the Performance of a Non-
Reversible Axial Fan”, Proceedings of the 7th International Conference on Heating, Ventilation and Air-
Conditioning, May 30-1st June, 2016, Tehran, Iran ICHVAC7-1813.
37 D. Dwivedi, D.S. Dantodiya, “CFD Analysis of Axial Flow Fans with Skewed Blades”, IJETAE, Issue 10, 2013).
38 T. Köktürk, “Design and Performance Analysis of a Reversible Axial Flow Fan”, Master Thesis, METU, 2005.
51
implementing Latin Hypercube sampling in the design space and multi objective evolutionary
algorithm coupled with response surface approximation surrogate model39-40.
3.2.3.3.2 Experimental Method
Sufficient ventilation of a parking garage depends on the correct fan selection. Fans are selected by
deciding if the air flow rate determined by intersecting the fans’ H-Q characteristic curves with the
system resistance curve is sufficient for the ventilation of the space. For that purpose, database of
various fans’ performance curves and static pressure loss vs. airflow rate curve of the ventilated
system are needed. Fan databases are setup by testing the performance of fans in test systems
that are designed according to the national and/or international standards. One of the commonly
used measurement systems is AMCA 210 test rig shown in Figure 3.13. The flow rate through the
fan is changed by adjusting the flow resistance, i.e. nozzle diameter, which is located in the variable
exhaust system of the AMCA chamber.
39 J.H. Kim, J.W. Kim, K.Y. Kim, “Axial-Flow Ventilation Fan Design Through Multi-Objective Optimization to
Enhance Aerodynamic Performance”, Journal of Fluids Engineering, vol. 133 / 101101-1, 2011.
40 J. Kim, B. Ovgor, K. Cha, J. Kim. S. Lee, K. Kim, “Optimization of the Aerodynamic and Aero acoustic Performance
length for both the upstream and downstream ducts of the axial fan is assumed to be 5D, where D is
the diameter of the fan casing. Both stationary and rotating zones of the MRF model are generated
for one sixth of the axial fan by using periodic boundary conditions. Velocity inlet and pressure outlet
boundary conditions are used at the upstream and downstream ends of the computational domain,
respectively.
[Pascu] built a CFD model of an axial fan which is
3D and 2D long in the upstream and downstream
sides of the fan, respectively42. The complete
surface mesh of the rotating zone obtained by
revolving one slice of the blade and the hub
around the fan axis is shown in Figure 3.14.
Surface mesh is created on both pressure and
suction sides of the fan blades, therefore the
blade thickness is maintained. The cylindrical
outer wall of the AMCA chamber is the boundary
that surrounds the computational model.
Boundary layer mesh is applied throughout
stationary and rotating zone walls and around
the fan blade walls.
There is not a stator in the model. Ten boundary
mesh layers are generated next to the duct wall
and five boundary mesh layers are used around Figure 3.14 Surface mesh of the rotating zone
the blades. Thin boundary mesh layers required
by the k-ε turbulence model are used to resolve the small blade tip clearance and flow around the
blades, which yield accurate prediction of the fan performance and especially the fan noise
generation43-44. Convergence of the results are
tested under grid refinement by generating three Mesh Density Number of Cells
meshes of different resolution, whose number of Coarse Mesh 169477
cells are given in Table 3.1.
Medium Mesh 178410
Reynolds averaged continuity and Navier-Stokes
Fine Mesh 496616
Equations coupled with turbulent kinetic energy
and dissipation rate transport equations of the
Table 3.1 Mesh Resolution and Number of Cells
standard k-ε turbulence model are used in the fan
simulations. RNG k-ε turbulence model chosen in
a similar study of axial flow fans 45. Continuity, momentum and k-ε turbulence model equations are
given below:
∂ρ ∂(ρui )
+ =0
∂t ∂xi
∂ ∂
ρui +
̅̅̅̅ ρu
̅̅̅̅̅
u =
∂t ∂xi i j
42 M. T. Pascu, “Modern Layout and Design Strategy for Axial Fans”, Ph.D. Thesis at Erlangen University, 2009.
43 Srinivas G., Srinivasa Rao Potti: “Numerical Simulation of Axial Flow Fan Using Gambit and Fluent”. IJRET,
Vol.3, No 3, pp. 586-590, 2014.
44 A. Raj. S, P. Pandian P.: “Effect of Tip Injection on an Axial Flow Fan under Distorted Inflow”. IJASER, Vol.3, No
∂κ ∂κ ∂κ ∂κ
ρ
+ ρU + ρV + ρW =
∂t ∂x ∂y ∂z
∂ μt ∂κ ∂ μt ∂κ ∂ μt ∂κ
( )+ ( )+ ( ) − ρε +
∂x σκ ∂x ∂y σκ ∂y ∂z σκ ∂z
∂U 2 ∂V 2 ∂W 2 ∂U ∂V 2 ∂U ∂W 2 ∂U ∂W 2
μt {2 [( ) + ( ) + ( ) ] + ( + ) + ( + ) +( + ) }
⏟ ∂x ∂y ∂z ∂y ∂x ∂z ∂x ∂z ∂x
𝐺
Eq. 3.8
∂ε ∂ε ∂ε ∂ε
ρ
+ ρU + ρV + ρW =
∂t ∂x ∂y ∂z
∂ μt ∂ε ∂ μt ∂ε ∂ μt ∂ε ε2 ε
( )+ ( )+ ( ) − C2 ρ + C1 μt G
∂x σκ ∂x ∂y σκ ∂y ∂z σκ ∂z κ κ
Eq. 3.9
After generating
tetrahedral cells with
appropriate resolution
above the inflation layers,
boundary conditions are
set. In Figure 3.15 and
Table 3.2 rotating and
stationary cell zones and
boundary conditions are
indicated. For more
accurate solutions, second
order discretization is
used for continuity,
momentum and Figure 3.15 Cell zones and boundary conditions
turbulence model
equations.
Turbulence Modeling κ-ε
3.2.3.3.4 Results and Inlet B.C. Velocity Inlet
Discussion Outlet B.C. Pressure Outlet
Reynolds averaged
Blade Pressure Side Periodic
continuity, Navier-Stokes
Blade Suction Side Periodic
and k-ε turbulence model
equations are solved Fluid Air
simultaneously using Fan Speed 1500
Multiple Reference Residual For Iteration Convergence 1E-5
Frames, MRF, method. Near Wall Treatment Standard Wall Function
After simulations, validity
of the results are checked Table 3.2 Solution Methods and Boundary Conditions
54
by verifying that y+ values on the pressure and suction surfaces of the fan blades remain between 30
and 100.
In the following figures comparison of CFD simulation and test results for fan pressure rise, torque
and efficiency vs. volumetric flow rate curves are given for the three grid resolutions. Difference
between the calculated and test torque data is caused by neglecting mechanical friction in the CFD
model.
30 ∗ Q ∗ ∆p
η=
π∗n∗T
Eq. 3.10
where Q is the volumetric airflow rate
in m3/s, ΔP is the fan pressure rise in
Pa, n is the fan speed in rpm, and T is
the sum of the fan aerodynamic and
mechanical friction torques in Nm. In
Figure 3.17, numerical torque values
are almost half of the experimental
ones because the losses caused by the
bearing and transmission friction
torques are ignored. Even though
there are significant variations in the
previous charts, simulation results of
fan static pressure rise vs. volume flow
rate curves are quite similar to the test
result, because mechanical friction of
bearings and the transmission do not
affect these parameters. Following
formula is used to calculate the electric
motor torque in terms of the motor Figure 3.17 Torque vs. Volumetric Flow Rate Curves
power and fan rpm.
60 ∗ p
Tmotor =
2∗π∗n
Eq. 3.11
The torque curve which is obtained by
adding the aerodynamic torque
calculated by postprocessing
CFD results and the motor torque
calculated above and estimated
mechanical friction torque is compared
with the torque value read from electric
control panel. Figure 3.16 shows the
streamline inside the duck at the rotor.
3.2.3.3.5 Conclusion
CFD simulation results of an axial fan are
obtained under grid refinement and are
compared with the fan test data. Use of
the CFD model of a slice of the fan with Figure 3.16 The streamlines inside the duct at a rotor
speed of 1500 rpm
periodic boundary conditions decreases
55
the memory requirements of the simulations because of the substantial decrease of the number of
cells. Acceptable agreement of the axial fan static pressure rise vs. airflow rate curve with the test
data is promising for using CFD as a means of predicting the performance of axial fans. The pressure
rise prediction of the current CFD model is quite well in the neighborhood of the design point of the
fan. Indeed, the calculated pressure rise corresponding to the volume flow rate at the design point is
almost the same as the test data. As the fan operates away from the design point and stall inception
occurs, numerical results begin to move away from the test data. This shows that the standard k-ε
turbulence model with standard wall functions is insufficient in calculating the location of the
separation point and the pressure change on the blade surfaces for separated flows. If electric motor
and mechanical friction torques are not taken into account, aerodynamic torque calculated by CFD is
almost half of the value measured in experiments. As the motor torque and estimated mechanical
friction torque values are considered, numerical and experimental torque vs. air flow-rate curves get
closer. Furthermore, because the calculated efficiency depends on the friction torque, taking into
account of the total torque results in having closer agreement between the predicted and measured
efficiency values. Reversible axial fan performance and noise predictions by CFD simulations and
comparison of the results with the test data is planned as the future research. Furthermore the degree
of improvement of the predicted fan performance at stall condition, provided by applying shear
stress transport, SST, k-ω turbulence model on fine meshes is to be studied in the future.
3.2.3.4 Case Study 2 - Potsdam Propeller CFD Benchmark
Authors : Daniel LaCroix1 and Radek Máca2
Affiliations : 1Pointwise, 2CFD Support
Originally Appeared : LinkedIn – CFD Support
Source : Same as Above
This report presents the benchmark validation of CFD simulation results of the Potsdam Propeller
Test Case (PPTC), using TCFD® with POINTWISE® mesh. PPTC is a marine propulsor that was
extensively measured by SVA Potsdam and related data were published [1] [2] [3]. The aim of this
benchmark was to evaluate the TCFD® , computational fluid dynamics (CFD) software, on the very
advanced mesh, created in POINTWISE® , mesh generation software, and to compare the results with
the measurement data available. The particular goal of this benchmark is to compare the propeller
Efficiency, Torque Coefficient, and Thrust Coefficient vs. Advance Coefficient with the real
experimental measurement of SVA Potsdam Laboratory.
3.2.3.4.1 Potsdam Propeller CFD Benchmark Description
The high demand for improving the accuracy, quality, and credibility of the CFD simulation results,
should be assessed by providing a high qualitative and intensive comparison with experimental
measurement data. The purpose of this benchmark is the validation of CFD simulation software
TCFD® with the mesh created in high-end meshing software
POINTWISE® and to compare the results with the
measurement data available. Potsdam Propeller Test Case
(PPTC) is a marine propulsor that was extensively measured
by SVA Potsdam and related data were published in [1], [2],
and [3].
The particular goal of this benchmark is to compare the
propeller Efficiency, Torque Coefficient, and Thrust
Coefficient vs. Advance Coefficient with the real
experimental measurement of SVA Potsdam Laboratory. The
experimental investigation includes open water test and
Figure 3.18 SVA Potsdam
velocity field measurements at different operation
Laboratory
conditions. A detailed description of the open water tests
56
conducted at the towing tank of the SVA is presented in the SVA report [1], which can be found on
the SVA website (sva-potsdam.de). See Figure 3.18. At the propeller analysis, there are a few
important dimensionless numbers. Those are Advance Coefficient, Thrust Coefficient, Torque
Coefficient, and Efficiency. They are defined as (respectively):
Va T Q JK T
J= , KT = , KQ = , η0 =
n. D ρ. n2 . D4 ρ. n2 . D5 2πK Q
Eq. 3.12
Where Va is the advance speed [m/s], n is the speed of rotation [1/s], D is propeller diameter [m], T
is thrust [N], ρ is water density [kg/m3], Q is torque [Nm]. The measurement results are available
for Advance coefficients from J = 0.5 to 1.6 and the simulation points are chosen accordingly.
Altogether, 11 Advance Coefficient modes were simulated according to the measurements.
3.2.3.4.2 Meshing
An unstructured viscous computational mesh was constructed with Pointwise on the Potsdam
propeller geometry as part of this TCFD validation benchmark. Pointwise, Inc. has previously worked
with variants of the geometry for other studies. For an in-depth discussion of Pointwise technology
and how it can be used for this particular geometry, please consult [4].
A combination of anisotropic and isotropic triangles were used in the surface mesh discretization.
Areas of high curvature - such as the leading edge, trailing edge, and the tip were resolved by utilizing
Pointwise’s T-Rex algorithm. This tool grows anisotropically stretched, right-angled triangles layered
in the normal direction to a boundary [5], as shown below. Using this, areas of high curvature are
able to be resolved without the need to isotopically refine the area. The result is an accurate
adherence to the surface and a reduction in the point count. The interior of the surface mesh was
resolved with isotropic triangles created using a modified Delaunay algorithm. (see Figure 3.19).
After meshing the geometry, the outer boundary of the moving reference frame (MRF) as well as the
outer boundary of the computational domain were meshed utilizing isotropic triangles and the
Delaunay algorithm. The MRF is cylindrical domain approximately 4.8 D long (4.8 x the propeller
diameter) and 1.5 D in diameter (see Figure 3.20). It starts just upstream of the propeller and
extends downstream into the wake. A far field block was generated outside of the MRF
corresponding to 10 D and 2.6 D; these were the limits taken from the file provided.
The volume mesh is a combination of a prismatic core surrounded by isotropic tetrahedral cells. The
prismatic portions of the grid were initialized using T-Rex. Starting from the surface mesh,
anisotropic tetrahedral cells were grown until reaching a desired stop criteria, colliding with another
front, or violating quality criteria. If an element stops advancing this did not prevent adjacent
elements from continuing. After the tetrahedral layers are grown, the cells are combined to form
prisms (or hexagons if the surface mesh is made up of quadrilateral cells). This reduces the total cell
count of the mesh without sacrificing quality. Once the near-wall viscous mesh was generated, the
remainder of the volume was populated with isotropic tetrahedral cells. The total cell count was just
below 4.1 M46 cells. The average maximum included angle was 101, and the maximum was 170. The
average volume ratio was 1.8 with a maximum of 28.
3.2.3.4.3 TCFD® Simulation Setup
The simulation run in TCFD is quite straightforward. The external mesh, created in POINTWISE, is
simply loaded and the simulation parameters are set. The simulation type is the propeller. Time
management is steady-state. The fluid flow model is incompressible. The mesh has two components
(the water tunnel and the cylinder with propeller inside) of total 4.1M cells. The inlet flow velocity
defines the advance coefficient and the velocity varies from 1.7 to 5.9 m/s in 11 points. The outlet
boundary condition is static pressure. For turbulence modeling, the RANS modeling approach has
been used with the k-ω SST turbulence model with the wall functions. The fluid properties of water
are selected. The density of 997.71 kg/m3. Dynamic viscosity of 9.559e-4 Pa.s.
3.2.3.4.4 TCFD® Simulation Post-processing
TCFD® includes a built-in post-processing module that automatically evaluates the required total
quantities, such as propeller efficiency, thrust coefficient, torque coefficient, forces, force coefficients,
flow rates, and much more. All these quantities are evaluated throughout the simulation run, and all
the important data is summarized in tabled .csv files as well as in the HTML report, which can be
updated anytime during the simulation for every run. Furthermore, visual postprocessing of the
volume fields can be done with ParaView. (see Figure 3.22).
46 M= million
59
3.2.3.4.6 Conclusion
The CFD analysis of the PPTC was performed successfully. It has been shown that the TCFD® in
connection with POINTWISE® provides very accurate results that are in perfect agreement with the
measurement data. All the simulation and measurement data are freely available. Potential questions
about TCFD® are to be sent to info@cfdsupport.com . Questions about POINTWISE® are to be sent to
pointwise@pointwise.com .
3.2.3.4.7 References
[1] Barkmann, U., Potsdam Propeller Test Case (PPTC) - Open Water Tests with the Model Propeller
VP1304, Report 3752, SchiffbauVersuchsanstalt Potsdam, April 2011
[2] Barkmann, U., Heinke, H.-J., Potsdam Propeller Test Case (PPTC) Test Case Description, Second
International Symposium on Marine Propulsors smp’11, Hamburg, Germany, June 2011, Workshop:
Propeller performance
[3] Heinke, H.-J., Potsdam Propeller Test Case (PPTC), Cavitation Tests with the Model Propeller
VP1304, Report 3753, SchiffbauVersuchsanstalt Potsdam, April 2011
[4] Carrigan, T., Bagheri, B., “A Study of the Influence of Meshing Strategies on CFD Simulation
Efficiency,” NAFEMS World Congress 2017, NWC17-466, 2017.
[5] Steinbrenner, J. P. and Abelanet, J.P., "Anisotropic Tetrahedral Meshing Based on Surface
Deformation Techniques," AIAA-20060554, AIAA 45th Aerospace Sciences Meeting, Reno, NV
60
Note that it may be desirable to under-relax the changes in boundary condition values in order to
prevent divergence of the solution (especially early in the computation).
3.3.3.1 Mass Conservation Across the Mixing Plane
Note that the algorithm described above will not rigorously conserve mass flow across the mixing
plane if it is represented by a pressure inlet and pressure outlet mixing plane pair. If you use a mass
62
flow inlet and pressure outlet pair instead, we will force mass conservation across the mixing plane.
The basic technique consists of computing the mass flow rate across the upstream zone (pressure
outlet) and adjusting the mass flux profile applied at the mass flow inlet such that the downstream
mass flow matches the upstream mass ow. This adjustment occurs at every iteration, thus ensuring
rigorous conservation of mass ow throughout the course of the calculation. Also note that, since mass
flow is being fixed in this case, there will be a jump in total pressure across the mixing plane. The
magnitude of this jump is usually small compared with total pressure variations elsewhere in the
flow field. Other quantities which will be conserved across Mixing Plane are Swirl and Total
Enthalpy.
(a) Rotor-Stator
Interaction (b) Blower
boundary type.) The resultant interior zone corresponds to where the two interface zones overlap;
the resultant periodic zone corresponds to where they do not. The number of faces in these
intersection zones will vary as the interface zones move relative to one another. Principally, fluxes
across the grid interface are computed using the faces resulting from the intersection of the two
interface zones, rather than from the interface zone faces themselves.
Figure 3.28 Initial position and some translation with Sliding Interface
Figure 3.28
In the example shown in Figure 3.28 (b), the interface zones are composed of faces A-B and B-C,
and faces D-E and E-F. The intersection of these zones produces the faces a-d, d-b, b-e, etc. Faces
produced in the region where the two cell zones overlap (d-b, b-e, and e-c) are grouped to form an
interior zone, while the remaining faces (a-d and c-f) are paired up to form a periodic zone. To
compute the flux across the interface into cell IV, for example, face D-E is ignored and faces d-b and
b-e are used instead, bringing information into cell IV from cells I and III, respectively.
65
4 Elements of Turbomachinery
4.1 Background
Turbomachinery is widely used equipment in industry such as compressors and turbines in a jet
engine; steam turbine in power plants, propeller for ships, hydraulic turbines for irrigation, wind
turbines for green energy, small fans for cooling, and so on (Wang, 2010)47. A common feature of
these devices is that they all work with fluid and have rotating component. Gorla48 gives a general
definition of turbomachinery which says “Turbomachinery is a device in which the energy transfer
occurs between a flowing fluid and a rotating element due to dynamic action, and results in a
change in pressure and momentum of the fluid”. The usage of turbomachinery has a long history.
It is recorded that the waterwheel, a kind of primitive turbomachinery, was invented and used for
power generation more than hundred years ago. Although the configuration is simple, it does follow
the same basic principle with other complicated modern turbomachinery’s, for instance the
compressor and the gas turbine in a jet engine. Figure 4.1 represents classification of
turbomachines. Here we concern mainly with axial devices. As the air is compressed in compressor
before entering the combustion chamber where it is mixed with fuel and combustion occurs (a.k.a.,
aggravated stage). Then the gas with high pressure and high temperature flows through gas turbines
and leaves the engine through a nozzle. While expanding through the turbine blades, power is
released from the gas and drives the turbine rotating. This constitutes the modern gas turbine engine
Mixed Flow
Devices Wind Turbine
Turbomachines
Comprssible Gas Turbine
Steam Turbine
Fluid Physics
Pumps
Incompressible
Hydroulic Turbine
47 Xiaodong Wang, “CFD Simulation of Complex Flows in Turbomachinery and Robust Optimization of Blade
Design”, Submitted to the Department of Mechanical Engineering Doctor of Philosophy, July 2010.
48 R. S. R. Gorla. Turbomachinery: Design and Theory. CRC Press, 2003.
66
49 See previous.
50 M. T. Schobeiri. Turbomachinery: Flow Physics and Dynamic Performance. Springer, Berlin, 2005. 2, 37, 38
67
thermal models of solid components, is current area of development. A wide variety of flow
phenomena, which are coupled in nature, occur in Turbomachinery CFD ranging from shock surfaces,
boundary layer, secondary flow, and vortex generating from blade tip and hob. These, makes the flow
analysis of turbo-machinery extremely complex and CFD limited. The number of turbine stages varies
in different types of engines, with high bypass ratio engines tending to have the most turbine stages.
The number of turbine stages can have a great effect on how the turbine blades are designed for each
stage. Many gas turbine engines are twin spool designs, meaning that there is a high pressure spool
and a low pressure spool. The high pressure turbine is exposed to the hottest, highest pressure air,
and the low pressure turbine is subjected to cooler, lower pressure air. The difference in conditions
leads to the design of high pressure and low pressure turbine blades that are significantly different
in material and cooling choices even though the aerodynamic and thermodynamic principles are the
same. Under these severe operating conditions inside the gas and steam turbines, the blades face high
temperature, high stresses, and potentially high vibrations. Steam turbine blades are critical
components in power plants which convert the linear motion of high temperature and high pressure
steam flowing down a pressure gradient into a rotary motion of the turbine shaft.
Figure 4.7 illustrates of a twin spool jet engine. The high pressure turbine (HPT) is connected by a
single spool to the high pressure compressor (purple) and the low pressure turbine is connected to
the low pressure compressor by a second spool (green)51.
Figure 4.2 Component of Turbomachines and their Thermodynamic (Brayton cycle) properties
by the ignition turns the turbine. The turbine is then able to drive both the compressor at the front
of the engine and also some useful load. In airplanes, it produces thrust.
4.4.1 Turbojet
The first jet engine was a turbojet. This is a simple turbine engine that produces all of its thrust from
the exhaust from the turbine section. However, because all of the air is passing through the whole
turbine, all of it must burn fuel. This means it is inefficient, and the solution is the turbofan (see
Figure 4.3).
4.4.2 Turbofan
In a turbofan, the turbine primarily drives a fan at the front of the engine. Most engines drive the fan
directly from the turbine. There are usually at least two separate shafts to allow the fan to spin slower
than the inner core of the engine. The fan is surrounded by a cowl which guides the air to and from
the fan. Part of the air enters the turbine section of the engine, and the rest is bypassed around the
engine. In high-bypass engines, most of the air only goes through the fan and bypasses the rest of the
engine and providing most of the thrust (see Figure 4.4).
4.4.3 Turboprop
In a turboprop, the turbine primarily drives a propeller at the front of the engine. There is no cover
around the prop. Some air enters the turbine, the rest does not. The propeller is geared to allow it to
spin slower than the turbine (see Figure 4.5). Although this diagram shows only a single shaft, many
turboprops have two, with a high pressure shaft driving the compressor and a low pressure shaft
driving the propeller. Turboprops are more efficient at lower speeds since the prop can move much
more air with a smaller turbine than the fan on a turbofan engine. The cover around the turbofan's
large fan allows it to perform better than an open propeller at high speeds, but limits the practical
size of the fan.
At supersonic speeds, turbojets have more of a performance benefit. They develop all of their thrust
from the high velocity turbine exhaust, while turbofans supplement that with the lower velocity air
from the fan. Since the air from the fan is also not compressed nearly as much as the core turbine
flow, it is also harder to prevent the flow from going supersonic and causing losses. The Concorde
used turbojets because it was designed to cruise for long periods at supersonic speeds. Modern
fighter jet engines are turbofans, which
provide a compromise between
efficiency and speed.
Elsewhere in aviation, turbine engines
are used in helicopters, as a turboshaft
engine driving the rotors instead of a
propeller, and with a freewheeling
clutch to enable autorotation’s (see
Figure 4.6). Turbocharged piston
engines use a turbine much differently
from the examples above. Instead of
being the primary power source, the
turbine only assists the piston engine.
A turbocharger uses a turbine to
compress air sent to the engine intake.
The increased compression helps the Figure 4.6 Turboshaft Engine
engine generate more power. The
70
turbine of a turbocharger is driven by engine exhaust gasses, and a supercharger is similar but is
directly powered by the engine52.
4.4.4 How does it work?
Gas turbines are comprised of three primary sections mounted on the same shaft: the compressor,
the combustion chamber (or combustor) and the turbine, as described above. The compressor can
be either axial flow or centrifugal flow. Axial flow compressors are more common in power
generation because they have higher flow rates and efficiencies. Axial flow compressors are
comprised of multiple stages of rotating and stationary blades (or stators) through which air is drawn
in parallel to the axis of rotation and incrementally compressed as it passes through each stage. The
acceleration of the air through the rotating blades and diffusion by the stators increases the pressure
and reduces the volume of the air. Although no heat is added, the compression of the air also causes
the temperature to increase. The compressed air then mixed with fuel injected through nozzles. The
fuel and compressed air can be pre-mixed or the compressed air can be introduced directly into the
combustor. The fuel-air mixture ignites under constant pressure conditions and the hot combustion
products (what we like to call: aggravated gases) are directed through the turbine where it expands
rapidly and imparts rotation to the shaft. The turbine is also comprised of stages, each with a row of
stationary blades (or nozzles) to direct the expanding gases followed by a row of moving blades. The
rotation of the shaft drives the compressor to draw in and compress more air to sustain continuous
combustion. The remaining shaft power is used to drive a generator which produces electricity.
Approximately 55-65 % of the power produced by the turbine is used to drive the compressor. To
optimize the transfer of kinetic energy from the combustion gases to shaft rotation, gas turbines can
have multiple compressor and turbine stages. Because the compressor must reach a certain speed
before the combustion process is continuous – or self-sustaining – initial momentum is imparted to
the turbine rotor from an external motor, static frequency converter, or the generator itself. The
compressor must be smoothly accelerated and reach firing speed before fuel can be introduced and
ignition can occur. Turbine speeds vary widely by manufacturer and design, ranging from 2,000
revolutions per minute (rpm) to 10,000 rpm. Initial ignition occurs from one or more spark plugs
(depending on combustor design). Once the turbine reaches self-sustaining speed above 50% of full
52 Aviation weekly.
71
speed; the power output is enough to drive the compressor, combustion is continuous, and the starter
system can be disengaged.
Simply put, in a compressor, to raise the pressure, the fluid must be slowed down as it passes
through a blade row. In a turbine, to drop the pressure, the fluid must be accelerated as it passes
through a blade row. By having alternate stationary and moving blade rows and making use of
the change of frame of reference, it is possible to always slow down (relative to the blade row)
or always speed up the fluid. For example: In a turbine the flow is accelerated in the stator
(stationary blade row). However, because the rotor row is moving, the flow appears to be moving
more slowly in the relative frame and so can be re-accelerated in the relative frame. This
appears to be a deceleration in the absolute frame53.
4.4.5 What is Thrust?
Thrust is a mechanical force which is generated through the reaction of accelerating a mass of gas, as
explained by Newton's 3rd law of motion. A gas or working fluid is accelerated to the rear and the
engine and aircraft are accelerated in the opposite direction. To accelerate the gas, we need some
kind of propulsion system. For right now, let us just think of the propulsion system as some machine
which accelerates a gas. From Newton's second law of motion, we can define a force T to be the
change in momentum of an object with a change in time. The thrust force, using a simple control
volume around propulsion systems obtained (see Figure 4.8) as
Figure 4.8 A 1D Control Volume around a propulsion system (Courtesy’s of NASA Glen Research
Center)
if Pe ≠ P0 → T = ṁ e Ve − ṁ 0 V0 + (Pe − P0 )Ae
if Pe = P0 → T = ṁ e Ve − ṁ 0 V0
Where T is thrust and ṁ = mass flow rate
Eq. 4.1
We see that there are two possible ways to produce high thrust. One way is to make the engine flow
rate (m dot) as high as possible. As long as the exit velocity is greater than the free stream, entrance
velocity, a high engine flow will produce high thrust. This is the design theory behind propeller
aircraft and high-bypass turbofan engines. A large amount of air is processed each second, but the
velocity is not changed very much. The other way to produce high thrust is to make the exit velocity
very much greater than the incoming velocity. This is the design theory behind pure turbojets,
turbojets with afterburners, and rockets. A moderate amount of flow is accelerated to a high velocity
in these engines. If the exit velocity becomes very high, there are other physical processes which
become important and affect the efficiency of the engine. There is a simplified version of the general
thrust equation that can be used for gas turbine engines. The nozzle of a turbine engine is usually
designed to make the exit pressure equal to free stream. In that case, the pressure-area term in the
general equation is equal to zero. The thrust is then equal to the exit mass flow rate times the exit
velocity minus the free stream mass flow rate times the free stream velocity54.
Axial
Dynamic
Centrifugal Single Acting
Scroll
Screw
Lobe
operational limitations associated with operating gas turbines in combined cycle mode, including
longer startup time, purge requirements to prevent fires or explosions, and ramp rate to full load.
55David Gordon Wilson; "The Design of High-Efficiency Turbomachinery and Gas Turbines", pp 487-492,
published by the MIT Press, Cambridge, Massachusetts, 1984, 5th printing 1991.
75
chord line Two-dimensional blade section were laid convex side up on a flat surface, the chord
line is the line between the points where the front and the rear of the blade section
would touch the surface.
compressor rotary machine that produces a relatively high pressure rise (pressure ratios greater
than 1.1) in a compressible fluid
deflection Total turning angle of the fluid. It is equal to the difference between the flow inlet angle
and the flow exit angle
deviation angle the flow exit angle minus the blade exit angle
diffuser A duct or passage shaped so that a fluid flowing through it will undergo an efficient
reduction in relative velocity and will therefore increase in (static) pressure.
EGV at the exit of the compressor consisting of another set of vanes further diffuses the
fluid and controls its velocity entering the combustors and is often known as the Exit
Guide Vanes (EGV)
effectiveness term applied here to define the heat-transfer efficiency of heat exchangers
efficiency Performance relative to ideal performance. There are many types of efficiency
requiring very precise definitions
entropy A property of a substance defined in terms of other properties. Its change during a
process is of more interest than its absolute value. In an adiabatic process, the increase
of entropy indicates the magnitude of losses occurring
expander A rotary machine that produces shaft power from a flow of compressible fluid at high
pressure discharged at low pressure. Here the only types of expander treated are
turbines
flow exit angle angle between the fluid flow direction at the blade exit and the machine axial direction
flow inlet angle angle between the fluid flow direction at the blade inlet and the machine axial direction
head the height to which a fluid would rise under the action of an incremental pressure in a
gravitational field
hub the portion of a turbomachine bounded by the inner surface of the flow annulus
hub-tip ratio same as hub-to-tip-radius ratio
IGV An additional row of stationary blades that frequently used at the compressor inlet
and are known as Inlet Guide Vanes (IGV) to ensure that air enters the first-stage
rotors at the desired flow angle, these vanes are also pitch variable thus can be
adjusted to the varying flow requirements of the engine
hub-to-tip radio ratio of the hub radius to the tip radius
incident angle the flow inlet angle minus the blade inlet angle
intensive Property that does not increase with mass; for instance, the pressure and temperature
property of a body of material do not double if an equal mass at the same temperature and
pressure is joined to it. (The energy, on the other hand, would double.)
intercoolers heat exchangers that cool a gas after initial compression and before subsequent
compression
isentropic occurring at constant entropy
isothermal occurring at constant temperature
leading edge the front, or nose, of the blade
mean section the blade section halfway between the hub and the tip
meridional a plane cutting a turbomachine through a diametric line and the (longitudinal) axis
plane
nozzle blade same as stator blade, for turbines only
76
pitch the distance in the direction of rotation between corresponding points on adjacent
blades
pressure The concave surface of the blade. Along this surface, pressures are highest
surface
pump A machine that increases the pressure or head of a fluid. In connection with
turbomachinery it usually refers to a rotary machine operating on a liquid.
radius ratio same as hub-to-tip-radius ratio
recuperator a heat exchanger, defined in this book as one with nonmoving surfaces, transferring
heat from a hot fluid to a cold fluid
regenerated See "CBEX."
cycle
regenerator a heat exchanger, defined in this book as one having moving surfaces or valves
switching the hot and cold flows
reheat The effect of losses in increasing the outlet enthalpy, or in decreasing the steam
wetness, in a steam-turbine expansion. Also see "reheat combustor."
reheat a combustor fitted between two turbines to bring the gas temperature at inlet to the
combustor second turbine to approach the temperature at inlet to the first
root The compressor or turbine-blade section attaching it to its mounting platform. Rotor
blade root sections are normally at the hub, and stator- blade roots at the shroud
rotor the rotating part of a machine, usually the disk or drum plus the rotor blades
rotor blade a rotating blade
separation when a fluid flowing along a surface ceases to go parallel to the surface but flows over
a near-stagnant bubble, or an eddy, or over another stream of fluid
shroud the surface defining the outer diameter of a turbomachine flow annulus
solidity the ratio of the chord to the spacing
spacing same as pitch
stagger angle the angle between the chord line and the turbine axial direction (also known as the
setting angle)
stall the condition of operation (usually defined by the incidence) of an airfoil, or row of
airfoils, at which the fluid deflection begins to fall rapidly and/or the fluid losses
increase rapidly
static conditions or properties of fluids as they would be measured by instruments moving
(conditions) with the flow
stator the stationary part of a machine, normally that part defining the flow path
stator blade A stationary blade.
suction surface The convex surface of the blade. Along this surface, pressures are lowest
surge the unstable operation of a high-pressure-ratio compressor whose stalls propagate
upstream from the high-pressure stages or components allowing reverse flow and the
discharge of the reservoir of high-pressure fluid, followed by re-establishment of
forward flow and a repetition of the sequence.
tip The outermost section of the blade or "vane."
total conditions or properties of fluids as they would be measured by stationary
(conditions) instruments that bring the fluid isentropically to rest
trailing edge the rear, or tail, of the blade
transverse the plane normal to the axis of a turbomachine
plane
77
turbine A rotary machine that produces shaft power by extracting energy from a stream of
fluid passing through it, using only fluid-dynamic forces (as distinct from "positive
displacement" or piston-and-cylinder-like machines).
turbomachines As for "turbine," except that the shaft power may be produced or absorbed, and the
energy may be extracted from or added to a stream of fluid.
working fluid Fluid that undergoes compression, expansion, heating, cooling, and other processes in
a heat-engine cycle. In an open-cycle gas turbine the working fluid is air
engine.
4.8.2 Axial Compressor
The compressor is responsible for providing the turbine with all the air it needs in an efficient
manner. In addition, it must supply this air at high static pressures. The example of a large turboprop
axial flow compressor will be used. The compressor is assumed to contain fourteen stages of rotor
blades and stator vanes. The overall pressure ratio (pressure at the back of the compressor compared
to pressure at the front of the compressor) is approximately 9.5:1. At 100% (>13,000) RPM, the
engine compresses approximately 433 cubic feet of air per second. At standard day air conditions,
this equals approximately 33 pounds of air per second. The compressor also raises the temperature
of the air by about 550F as the
air is compressed and moved
rearward. The power
required to drive a
compressor of this size at
maximum rated power is
approximately 7000
horsepower. In an axial flow
compressor, each stage
incrementally boosts the
pressure from the previous
stage.
A single stage of compression
consists of a set of rotor
blades attached to a rotating
disk, followed by stator
vanes attached to a
stationary ring. The flow area
between the compressor
blades is slightly divergent.
Flow area between
compressor vanes is also
divergent, but more so than
for the blades. In general
terms, the compressor rotor
blades convert mechanical
energy into gaseous energy.
This energy conversion
greatly increases total
pressure (PT). Most of the
increase is in the form of
velocity (V), with a small
increase in static pressure
(PS) due to the divergence of
Figure 4.13 Schematic Diagram of fluid properties through an axial
the blade flow paths. The
compressor stage – Courtesy of [T. B. Ferguson, Gravdahl, and
stator vanes slow the air by Egeland]
means of their divergent duct
shape, converting 'the
accelerated velocity (V) to higher static pressure (PS). The vanes are positioned at an angle such that
the exiting air is directed into the rotor blades of the next stage at the most efficient angle. This
79
process is repeated fourteen times as the air flows from the first stage through the fourteenth stage.
Figure 4.13 shows one stage of the compressor and a graph of the pressure characteristics through
the stage. (see [Niazi]57).
The stator removes swirl from the flow, but it is not a moving blade row and thus cannot add any net
energy to the flow. Rather, the stator rather converts the kinetic energy associated with swirl to
internal energy (raising the static pressure of the flow). Thus typical velocity and pressure profiles
through a multistage axial compressor look like those shown in Figure 4.13. Alternatively, assuming
incompressible, constant density, and with no body force, we can use Bernoulli’s equations (PT = PS +
(1/2)ρV2) where PT is the stagnation pressure, a measure of the total energy carried in the flow, p is
the static pressure a measure of the internal energy, and the velocity terms are a measure of the
kinetic energy associated with each component of velocity58. The rotor adds swirl to the flow, thus
increasing the total energy carried in the flow by increasing the angular momentum (adding to the
kinetic energy associated with the tangential or swirl velocity, 1/2rv2). The stator removes swirl
from the flow, but it is not a moving blade row and thus cannot add any net energy to the flow. Rather,
the stator rather converts the kinetic energy associated with swirl to internal energy (raising the
static pressure of the flow).
Thus a typical velocity and pressure profiles through a multistage axial compressor look like those
shown in Figure 4.14. In addition to the fourteen stages of blades and vanes, the compressor also
incorporates the inlet guide vanes and the outlet guide vanes. These vanes, located at the inlet and
the outlet of the compressor, are neither divergent nor convergent. The inlet guide vanes direct air
to the first stage compressor blades at the "best" angle. The outlet guide vanes "straighten" the air to
provide the combustor with the proper airflow direction. The efficiency of a compressor is primarily
Figure 4.14 Pressure and Velocity profile through a Multi-Stage Axial Compressor
57 Saeid Niazi, “Numerical Simulation of Rotating Stall and Surge Alleviation in Axial Compressors”, A Thesis
Presented to the Academic Faculty, Georgia Institute of Technology, 2000.
58 MIT OpenCourseWare
80
determined by the smoothness of the airflow. During design, every effort is made to keep the air
flowing smoothly through the compressor to minimize airflow losses due to friction and turbulence.
This task is a difficult one, since the air is forced to flow into ever-higher pressure zones. Air has the
natural tendency to flow toward low-pressure zones. If air were allowed to flow "backward" into the
lower pressure zones, the efficiency of the compressor would decrease tremendously as the energy
used to increase the pressure of the air was wasted. To prevent this from occurring, seals are
incorporated at the base of each row of vanes to prevent air leakage. In addition, the tip clearances
of the rotating blades are also kept at a minimum by the use of coating on the inner surface of the
compressor case. All components used in the flow path of the compressor are shaped in the form of
airfoils to maintain the smoothest airflow possible. Just as is the case for the wings of an airplane, the
angle at which the air flows across the airfoils is critical to performance. The blades and vanes of the
compressor are positioned at the optimum angles to achieve the most efficient airflow at the
compressor’s maximum rated speed. Any deviation from the maximum rated speed changes the
characteristics of the airflow within the compressor. The blades and vanes are no longer positioned
at their optimum angles. Many engines use bleed valves to unload the force of excess air in the
compressor when it operates at less than optimum speed.59 The example engine incorporates four
bleed valves at each of the fifth and tenth compressor stages. They are open until 13,000 RPM (~94%
maximum) is reached, and allow some of the compressed air to flow out to the atmosphere. This
results in higher air velocities over the blade and vane airfoils, improving the airfoil angles. The
potential for airfoil stalling is reduced, and compressor acceleration can be accomplished without
surge.
4.8.3 Diffuser
All turbomachines and many other flow systems incorporate a diffuser (e.g. closed circuit wind
tunnels, the duct between the compressor and burner of a gas turbine engine, the duct at exit from a
gas turbine connected to the jet pipe, the duct following the impeller of a centrifugal compressor,
etc.)60. Air leaves the compressor through exit guide vanes, which convert the radial component of
the air flow out of the compressor to straight-line flow. The air then enters the diffuser section of the
engine, which is a very divergent duct. The primary function of the diffuser structure is aerodynamic.
The divergent duct shape converts most of the air’s velocity (Pi) into static pressure (PS) with the aid
of Bernoulli equation. As a result, the highest static pressure and lowest velocity in the entire engine
is at the point of diffuser discharge and combustor inlet. Other aerodynamic design considerations
that are important in the diffuser section arise from the need for a short flow path, uniform flow
distribution, and low drag loss. In addition to critical aerodynamic functions, the diffuser also
provides:
59MIT, OpenCourseWare.
60 S. L.
Dixon, “Fluid Mechanics and Thermodynamics of Turbomachinery”, 5th edition, Senior Fellow at University
of Liverpool, 1978-1998.
81
The primary fluid mechanical problem of the diffusion process is caused by the tendency of the
boundary layers to separate from the diffuser walls if the rate of diffusion is too rapid61. The result of
too rapid diffusion is always large losses in stagnation pressure. On the other hand, if the rate of
diffusion is too low, the fluid is exposed to an excessive length of wall and fluid friction losses become
Pre-dominant. Clearly, there must be an optimum rate of diffusion between these two extremes for
which the losses are minimized.
4.8.4 Nozzle
In a large number of turbomachinery components the flow process can be regarded as a purely nozzle
flow in which the fluid receives an acceleration as a result of a drop in pressure (see
Figure 4.7). Such a nozzle flow occurs at entry to all turbomachines and in the stationary blade rows
in turbines. In axial machines the expansion at entry is assisted by a row of stationary blades (called
guide vanes in compressors and nozzles in turbines) which direct the fluid on to the rotor with a
large swirl angle. Centrifugal compressors and pumps, on the other hand, often have no such
provision for flow guidance but there is still a velocity increase obtained from a contraction in entry
flow area. In reality, Nozzle and Diffuser work against each other. A nozzle increases the velocity of a
fluid, while a diffuser decreases the velocity of a fluid. Nozzles can be used by jets and rockets to provide
extra thrust. Conversely, many jet engines use diffusers to slow air coming into the engine for a more
uniform flow.
4.8.5 Combustor
Once the air flows through the diffuser, it enters the combustion section, also called the combustor.
The combustion section has the difficult task of controlling the burning of large amounts of fuel and
air. It must release the heat in a manner that the air is expanded and accelerated to give a smooth and
stable stream of uniformly heated gas at all starting and operating conditions. This task combustion
liners must position and control the fire to prevent flame contact with any metal parts. The engine
under consideration here uses a can-annular combustion section with six combustion liners (cans).
They are positioned within an annulus created by inner and outer combustion cases. Combustion
takes place in the forward end or primary zone of the cans. Primary air (amounting to about one
fourth of the total engine’s total airflow) is used to support the combustion process. The remaining
air, referred to as secondary or dilution air, is admitted into the liners in a controlled manner (Figure
4.15). The secondary air controls the flame pattern, cools the liner walls, dilutes the temperature of
61 See 13.
82
the core gasses, and provides mass. This cooling air is critical, as the flame temperature is above
1930C (3500F), which is higher than the metals in the engine can endure. It is important that the fuel
nozzles and combustion liners control the burning and mixing of fuel and air under all conditions to
avoid excess temperatures reaching the turbine or combustion cases. Maximum combustion section
outlet temperature (turbine inlet temperature) in this engine is about 1070C (>1950F). The rear third
of the combustion liners is the transition section. The transition section has a very convergent duct
shape, which begins accelerating the gas stream and reducing the static pressure in preparation for
entrance to the turbine section.
4.8.6 Axial Gas Turbine
This example engine has a four-stage turbine. The turbine converts the gaseous energy of the
air/burned fuel mixture out of the combustor into mechanical energy to drive the compressor, driven
accessories, and, through a reduction gear, the propeller. The turbine converts gaseous energy into
mechanical energy by expanding the hot, high-pressure gases to a lower temperature and pressure.
Each stage of the turbine consists of a row of stationary vanes followed by a row of rotating blades.
This is the reverse of the order in the compressor. In the compressor, energy is added to the gas by
the rotor blades, then converted to
static pressure by the stator vanes. In
the turbine, the stator vanes increase
gas velocity, and then the rotor blades
extract energy. The vanes and blades
are airfoils that provide for a smooth
flow of the gases. As the airstream
enters the turbine section from the
combustion section, it is accelerated
through the first stage stator vanes.
The stator vanes (also called nozzles)
form convergent ducts that convert
the gaseous heat and pressure energy Figure 4.17 Turbine Flow Characteristics
into higher velocity gas flow (V). In
addition to accelerating the gas, the vanes "turn" the flow to direct it into the rotor blades at the
optimum angle. As the mass of the high velocity gas flows across the turbine blades, the gaseous
energy is converted to mechanical energy. Velocity, temperature, and pressure of the gas are
sacrificed in order to rotate the turbine to generate shaft power. Figure 4.17 represents one stage
of the turbine and the characteristics of the gases as it flows through the stage. A multi-stage turbine
is illustrates in Figure 4.15. The efficiency of the turbine is determined by how well it extracts
mechanical energy from the hot, high-velocity gasses. Since air flows from a high-pressure zone to a
low pressure zone, this task is accomplished fairly easily. The use of properly positioned airfoils
allows a smooth flow and expansion of gases through the blades and vanes of the turbine. All the air
must flow across the airfoils to achieve maximum efficiency in the turbine. In order to ensure this,
seals are used at the base of the vanes to minimize gas flow around the vanes instead of through the
intended gas path. In addition, the first three stages of the turbine blades have tip shrouds to
minimize gas flow around the blade tips. We can apply the same analysis techniques to a
turbine. Again, the stator does no work. It adds swirl to the flow, converting internal energy into
kinetic energy. The turbine rotor then extracts work from the flow by removing the kinetic
associated with the swirl velocity.
4.9 Difference in
Blading Between Compressor
Compressor and
Turbine • Area increase: pressure rise
There is quite a difference • Flow deceleration: thick boundary layers
between Compressor and • Little flow turning: many stages
Turbine blading. Aside
from shape of it, they are
number of stages and
arrangement of it. While Turbine
Compressor blades are • Area decrease: pressure drop
generally thin and straight,
and resemble a tiny • Flow acceleration: thin boundary layers
rectangular wing with low • Large flow turning: few stages
camber thickness. Turbine
84
blades are more curved. In particularly large and recent engines, where efficiency is critical, turbine
blades will often be full of tiny holes for cooling effects. The difference best described below and
examples of blade shown in Figure 4.18. To distinguish between high pressure and low pressure
stages (compressor or turbine does not matter), the length of the blade and its torsion (i.e. how much
the aerodynamic profile turns around the axis of the blade going from the root to the tip) are key:
shorter and more twisted blades will be high pressure ones, longer and straighter blades will be low
pressure. Note that two blades of the same length could come one from a high pressure stage and the
other from a low pressure one of a different engine: "short" and "long" are relative to the engine size.
rotor to a fluid, or the other way around. The working principle of the compressor and the turbine is
therefore quite similar.
4.11.1 Euler’s Equation for Turbomachinery 62
Let’s examine a rotor, rotating at a constant angular velocity ω. The initial radius of the rotor is r 1,
while the final radius is r2. A gas passes through the rotor with a constant velocity c. The rotor causes
a moment M on the gas. The power needed by the rotor is thus P = Mω. It would be nice if we can find
an expression for this moment M. For that, we first look at the force F acting on the gas. It is given by
d(mc)
dFu = = ṁc
dt
Eq. 4.2
Where we have used the assumption that c stays constant. Only the tangential component Fu
contributes to the moment. Every bit of gas contributes to this tangential force. It does this according
dF𝑢 = ṁdc𝑢
Eq. 4.3
Where cu is the tangential velocity of the air. Let’s integrate over the entire rotor. We then find that
2 2 2
Furthermore, another form of Euler’s Turbomachinery equation, with aid of the steady flow energy
equation:
It relates the temperature ratio (and hence the pressure ratio) across a turbine or compressor to the
rotational speed and the change in momentum per unit mass. Note that the velocities used in this
equation are what we call absolute frame velocities (as opposed to relative frame velocities).63 It is
given fact that:
• If angular momentum increases across a blade row, then T2 > T1 and work was done on the
fluid (a compressor).
• If angular momentum decreases across a blade row, then T2 < T1 and work was done by the
fluid (a turbine)
63MIT, OpenCourseWare.
64 Breuer, B., Bissinger, N., C., “Encyclopedia of Aerospace Engineering – Volume 8 - Chapter EAE 573-Basic
Principles – Gas Turbine Compatibility – Gas Turbine Aspects”.
87
component (cu, 1 = 0), and the exit flow angle of the rotor blade remains unchanged (in the rotor frame
of reference). With these assumptions, an increase in work input according to equation 1 can only be
achieved by an increase of cu,2. According to the dependencies shown in Figure 4.20, this requires
reducing the axial velocity component of the flow behind the rotor. Because of conservation of mass
flow through the rotor, also the axial velocity at rotor inlet will be reduced, leading to an increased
incidence of the flow to the rotor blade. Translating the state of flow behind the rotor from the
rotating frame of reference into the stationary one, Figure 4.20 also shows that an increase of work
delivered to the flow by the rotor increases the incidence to the subsequent stator row as well.
Therefore, an increase of work input to the flow means increasing incidences to both rotor and stator
airfoils. Therefore, an increase of work input to the flow means increasing incidences to both rotor
and stator airfoils.
Very much like aircraft wings, these airfoils have certain operating limits in terms of airfoil angle of
attack or incidence. With increasing incidence, rotor airfoils provide for a larger work input and
hence pressure rise, but at the same time the aerodynamic loading increases, up to a point where the
flow separates. On a larger scale, the pressure rise capability of a compressor is typically depicted
using a compressor map where pressure rise is depicted as a function of compressor mass flow for
different rotational speeds.
An example map is provided
with Figure 4.21, and for the
sake of illustration, it also
relates different regimes of
compressor operating range
to an aircraft operating at
different angles of attack. At
low pressure ratios, the
airfoils operate with negative
to small incidence, and usually
elevated losses. When the
pressure ratio is increased,
airfoil incidences now
approach a condition with
minimum losses. Further
increasing the pressure ratio
is equivalent to further rise of
airfoil aerodynamic loading
and losses increase due to Figure 4.21 Compressor Operating Map
formation of regions of
separated flow. At the upper end of a speed line, there is a point where regions of separated flow have
enlarged to an extent where no further pressure rise is achievable, in analogy to aircraft wings
reaching the stall limit where no further increase of lift can be provided65. The upper operational
limit of a compressor map is called the surge line, representing a condition where large flow
separation prevents further pressure rise. The surge line represents an operational limit for engine
operation, since the occurrence of compressor surge (sometimes also referred to as compressor stall)
leads to a highly unsteady flow field within the engine, quite often also entailing periods of reversed
flow, that is air flowing in the “wrong” direction through the compressor. Surge is associated with
large fluctuations of power output. Furthermore, it is accompanied by increased structural loads
caused by the rapid changes of flow field state. Compressor maps are usually established (either
numerically or by means of testing) for a standard set of inlet conditions. These inlet conditions are
65 See 71.
88
typically derived from simplified installation assumptions and assume a simplified inlet profile with
radial variations only, but uniform in circumferential direction. Intake distortion considerations deal
with conditions that deviate from these design assumptions and aim to identify the consequences of
these deviations with regard to engine operation.
since it decreases the net power output of the gas turbine68-69. With this in mind, during the design
phase of gas turbine it is very important to optimize the cooling flow if you are considering both the
performance and reliability. Cooled Gas turbine design is quite complicated and requires not only the
right methodology, but also the most appropriate design tools, powerful enough to predict the results
accurately from thermodynamics cycle to aerothermal design, ultimately generating the 3D blade.
Different cooling methods that are employed depend on the extent of the cooling required. The
cooling flow passes through several loops internally and is then ejected over the blade surface to mix
with the main flow. The mixing of the cooling flow with the main flow alters the aerodynamics of the
flow within the turbine cascade. The cooling flow that is injected into the main flow needs to be
optimized, not only in terms of thermodynamic parameters, but also in terms of the locations to
ensure the turbine vanes and blade surfaces are maintained well below the melting surface. The
spacing between the holes, both in horizontal and vertical direction, affects not only the surface
temperature of the blade, but also the strength of the blade and its overall life.
Performing a 3D analysis for optimizing the flow, spacing, and location of cooling flow is
computationally expensive. One has to resort to reduced order 1D flow and heat network simplifies
the task of not only arriving at the optimal configuration of cooling holes and location, but also in
aerothermal design of the gas turbine flow path and generation of the optimized 3D blades with
reduced overall design cycle time. Designers are faced with the challenge of simplifying the complex
3D cooled blade and accurately modelling it.
68 Amjed Ahmed Jasim AL-Luhaibi, Mohammad Tariq, “Thermal Analysis of Cooling Effect on Gas Turbine Blade”,
eISSN: 2319-1163 | pISSN: 2321-7308.
69 Posted by: Abdul Nassar, “Optimizing the Cooling Holes in Gas Turbine Blades”, SoftInWay® Incorporated,
2016.
90
ṁ√RT02 p03 ωD
, , and η
p02 D2 p02 √RT02
Eq. 4.8
These parameter groups are known as the mass flow parameter group, the pressure ratio, the
shaft speed parameter group and the efficiency. The efficiency can be either polytrophic or
isentropic. (These two efficiencies depend on each other anyway). The relation between the four
dimensionless parameters can be captured in a graph, known as a characteristic. An example of a
characteristic is shown in Figure 4.24. When applying dimensional analysis to a turbine, the same
results will be found. However, this time the initial and final pressures are p04 and p05. The initial and
final temperatures are T04 and T05.
4.14.1 Stall
Let’s examine the air entering the
rotor71. Previously, we have
assumed that this air has exactly
the right angle of incidence “i” to
follow the curvature of the rotor
blade. In reality, this is of course
not the case. In fact, if the angle of
incidence is too far off, then the
flow can’t follow the curvature of
the rotor blades. The other
phenomena associated with Stall is
if there are pockets of low axial
velocity covering one or two blade
passages (see Figure 4.26). This
is called stall and usually starts at
one rotor blade. However, this stall
alters the flow properties of the air
around it. Because of this, stall Figure 4.24 Characteristics Graph of a Compressor
spreads around the rotor. And it
does this opposite to the direction of rotation of the rotor. This phenomenon is called rotating stall.
Often, only the tips of the rotor blades are subject to stall. This is because the velocity is highest there.
This is called part span stall. If, however, the stall spreads to the root of the blade, then we have full
span stall. For high compressor speeds ω, stall usually occurs at the last stages. On the other hand,
for low compressor speeds, stall occurs at the first stages. Generally, the possibility of stalling
increases if we get further to the left of the characteristic. (See also Figure 4.24).
70 MIT OpenCourseWare.
71 See previous.
91
72 MIT OpenCourseWare.
73 Saeid Niazi, “Numerical Simulation of Rotating Stall and Surge Alleviation in Axial Compressors”, A Thesis
Presented to the Academic Faculty of Georgia Institute of Technology, 2000.
74 See previous.
92
reaches supersonic velocities. Choked flow can also occur at the compressor. If we look at the right
side of Figure 4.24, we see vertical lines. So, when we change the pressure ratio p03/p02 at constant
compressor speed ω, then the mass flow remains constant.
75Magdi Ragheb and Adam M. Ragheb (2011). “Wind Turbines Theory - The Betz Equation and Optimal Rotor
Tip Speed Ratio, Fundamental and Advanced Topics in Wind Power”, Dr. Rupp Carriveau (Ed.), ISBN: 978-953-
307-508-2, InTech, Available from: http://www.intechopen.com/books/fundamental-and-advanced-topicsin-
wind-power/wind-turbines-theory-the-betz-equation-and-optimal-rotor-tip-speed-ratio.
93
76 Xiaodong Wang, “CFD Simulation of Complex Flows in Turbomachinery and Robust Optimization of Blade
Design”, Submitted to the Department of Mechanical Engineering Doctor of Philosophy at the Vrije Universiteit
Brussel July 2010
77 X. D. Wang, Sh. Kang, “Solving stochastic burgers equation using polynomial chaos decomposition”, J. Eng.
78 Boyer, K., M., “An Improved Streamline Curvature Approach for Off-Design Analysis of Transonic Compression
Systems “, PhD. Dissertation, Virginia Polytechnic Institute and State University, 2001.
79 J. F. Escuret, D. Nicoud, and Ph. Veysseyre,”Recent advances in compressors aerodynamic design and analysis”,
Design”, Submitted to the Department of Mechanical Engineering Doctor of Philosophy at the Vrije universiteit
Brussel July 2010.
81 Wu, C. H. “A general through flow theory of fluid flow”, NACA paper TN2302, 1951.
95
suited to calculating transonic flow but are not yet highly developed further use in through flow
calculations. Problems with calculating transonic flow are currently much more severe in steam
turbines than in gas turbines. The traditional use of streamline curvature method (SCM)
approaches, as most often discussed in the literature during the preliminary design phase, are
discussed in detail in82. The stream surface represented by
s(r, ψ, z) = 0
Eq. 5.1
As depicted in Figure 5.3. The through-flow solver provides a preliminary blade shape, continually
refined through solutions from higher-order and secondary flow models. One way to calculate a 3D
flow field is to solve two sets of equations, one dealing with axis-symmetric flow in the meridional
plane, commonly referred to as the “S2” surface, and the other with blade-to-blade flow on a stream
surface of revolution, the “S1” plane (see Figure 5.2). The traditional formulation for the governing
momentum equation(s) is a first-order velocity gradient representation, one in the radial and one in
the tangential direction approach for off-design analysis along an axis-symmetric S2 surface. It is
generally accepted that any streamline curvature solution technique will yield satisfactory flow
solutions as long as the deviation, losses, and blockages are accurately predicted83.
82 Chung-Hua Wu, “A General Theory of Three Dimensional Flow in Subsonic and Supersonic Turbomachines of
Axial-Radial- and Mixed Flow Types”, National Advisory Committee for Aeronautics, Technical Note, 1952.
83 Boyer, K., M., “An Improved Streamline Curvature Approach for Off-Design Analysis of Transonic Compression
Systems “, PhD. Dissertation, Virginia Polytechnic Institute and State University, 2001.
97
to calculate transonic flows with weak shock with transonic flow and they have now largely been
abandoned. Velocity potential methods can waves but they have seen limited use in turbomachinery.
The numerical methods described above are inviscid and need to be coupled to a boundary layer
calculation if they are to be used to predict blade loss. For com pressor blades the boundary layer
blockage must be included in the inviscid calculation as it significantly affects the blade surface
pressure distribution84. For most turbine blades the boundary layer is so thin that it may be
calculated separately after obtaining the surface pressure distribution from an inviscid calculation. A
recent alternative (N–S) equations which predict the boundary native to coupled inviscid/boundary
layer calculations is the direct solution of the Navier– layer growth as part of the main calculation.
These demand a much finer grid near to the blade surfaces than do inviscid calculations and so are
considerably more ‘expensive’. Nevertheless the N–S equations for blade to-blade flow are now
routinely solved as part of the design process, requiring only a few minutes CPU time on a modern
workstation. There remains controversy about the best turbulence and transition models to use and
about how many mesh points are necessary within the boundary layer.
A variety of blade–to-blade solvers are currently available in the design system. They range from
potential and streamline curvature method up to fully viscous, time marching solvers. The main use
of the blade-to-blade codes is to ensure that the vector diagrams set by thorough Flow are achievable
and within the bounds of blade thickness, loading and efficiencies. For examples, in turbine design
the suction surface diffusion is taken as a primary indicator as to the condition of the boundary layer.
The blade-to-blade code solves for the suction surface velocity ratio, or diffusion factor, and the
geometry is adjusted accordingly. Most of these codes are very similar to those available in other
design systems and have also been described elsewhere. However, three codes (TAYLOR, AEGIS and
NOVAKED2D) are different and worth mentioning85.
5.3.3 Case Study – Turbine Airfoil Optimization Using Inviscid Quasi 3D (Q3D) Analysis Codes
Citation : Ng, E., & Yi, M. (1998). Computation of Q3D Viscous Flows in Various Annular Turbine Stages
with Heat Transfer. International Journal of Rotating Machinery, 4, 25-33.
Turbine airfoil design has long been a domain of expert designers who use their knowledge and
experience along with analysis codes to make design decisions86. The turbine aerodynamic design is
a three-step process that is pitch line analysis, through-flow analysis, and blade-to-blade analysis, as
depicted in Figure 5.4. In the pitch line analysis, flow equations are solved at the blade pitch, and a
free vortex assumption is used to get flow parameters at the hub and the tip. Using this analysis the
flow path of the turbine is optimized, and number of stages, work distribution across stages, stage
reaction, and number of airfoils in each blade row are determined. In the through-flow analysis, the
calculation is carried out on a series of meridional planes where the flow is assumed to be
axisymmetric and the boundary conditions of each stage are determined. The axisymmetric through-
flow method allows for variation in flow parameters in the radial direction without using the free
vortex assumption and accounts for interactions between multiple stages. In the blade-to-blade
analysis, airfoil profiles are designed on quasi-3D surfaces using a computational fluid dynamics
code.
The primary sources of losses in an airfoil are profile loss, shock loss, secondary flow loss, tip
clearance loss, and end-wall loss. Profile loss is associated with boundary layer growth over the blade
profile causing viscous and turbulent dissipation. This also includes loss due to boundary layer
84 Calvert, W. J. and Ginder, R. B., “Quasi-3D calculation system for the flow within transonic compressor blade
rows”, ASME paper 85-GT-22, 1985.
85 Ian K. Jennions, “Elements of a Modern Turbomachinery Design System”, GE Aircraft Engines, One Neumann
Transfer”, International Journal Of Rotating Machinery, 1998, Vol. 4, No. 1, pp. 25-33.
98
separation because of conditions such as extreme angles of incidence and high inlet Mach number.
Shock losses arise due to viscous dissipation within the shock wave which results in increase in static
pressure and subsequent thickening of the boundary layer, which may lead to flow separation
downstream of the shock. End-wall loss is associated with boundary layer growth on the inner and
outer walls on the annulus. Secondary flow losses arise from flows, which are present when a wall
boundary layer is turned through an angle by an adjacent curved surface. Tip clearance loss is caused
by leakage flows in the tip clearance region of the rotor blade, where the leaked flow fails to
contribute to the work output and also interacts with the end-wall boundary layer. The objective of
the design is to create the most efficient airfoil by minimizing these losses. This often requires
trading-off one loss versus another such that the overall loss is minimized.
To compute all these losses a 3D viscous analysis is required; however, due to the computational load
of such a code, a quasi-3D analysis code is often used in the design process. Thus the impact of the
blade geometry on 3D losses cannot be determined and only 2D losses can be minimized, that is,
profile and shock losses. A viscous quasi-3D analysis though less computationally intense is still too
expensive for use in design optimization, and an inviscid quasi-3D code is used instead. Consequently,
viscous losses are not computed from the analysis code and airfoil performance is gauged by the
characteristics of the Mach number distribution on the blade surface. The most practical formulation
for low-speed turbine airfoil designs still remains the direct optimization formulation based on 2D
inviscid blade-to-blade solvers.
87 Goel, Sanjay,” Turbine Airfoil Optimization Using Quasi-3D Analysis Codes”, University at Albany, USA.
99
function, and the visual perspective of the designer is the only proven metric available, it must be
captured in a suitable numerical algorithm to provide a measure of quality of an airfoil. The current
work employs curve fitting coupled with design heuristics to compute quality metrics from the Mach
number distribution and the airfoil geometry.
These metrics are weighted for different designs based on individual designer preferences. Primary
evaluation metrics that have been defined are diffusion, deviation, incidence deviation, and leading
edge crossover. A physical interpretation of these metrics is presented below. Diffusion is defined as
the deceleration of the flow along the blade surface. It is measured as the cumulative aggregate of all
flow diffusions at each point along the airfoil surface. As the flow diffuses, the boundary layer
thickens, and the momentum loss in the boundary layer increases. In this case, the increased drag
causes a significant loss of momentum; flow separation may result, causing much larger losses. Thus,
the objective of the design is to minimize the diffusion effect. Since the impact of diffusion on the
pressure and suction sides is different, separate terms are defined for the suction and pressure sides.
In the test case presented here, a low-pressure turbine nozzle is optimized. The flow-path of the low-
pressure turbine used in the investigation is shown in Figure 5.5. The radial distances in the figure
are measured with reference to the centerline of the engine and the axial distances are measured
with reference to a point upstream of the first stage of the turbine. The horizontal lines in the figure
represent the streamlines of the flow. Thirteen streamlines are shown, the top and bottom of which
coincide with the casing and the hub respectively. The vertical lines represent the edges of the blade
rows and the location of the frame. The turbine has six stages, each stage composed of two blade
rows. The first blade row consists of nozzles and the second blade row consists of buckets. The stages
are numbered from 1 to 6 in the Figure 5.5.
locations as shown in Figure 5.7. To ensure slope and curvature smoothness of the geometry,
second- order polynomials were used to represent the radial distribution of geometry parameters.
1 dmcθ2
(p + dp) (r + dr) dθ − p r dθ − (p + dp) dr dθ =
2 r
Eq. 5.2
Writing dm = ρ r dϴ dr and ignoring terms of 2nd order we obtain:
1 dp cθ2
=
ρ dr r
Eq. 5.3
For an incompressible fluid and using
thermodynamic relations the Radial Equilibrium
Equation can be written as:
dh 0 ds
− T = cx x + θ
dc c d
(rcθ ) or
dr dr dr r dr
dc
cx x + θ
c d
(rcθ ) = 0
dr r dr
1 dp 0 1 dp dc dc
= + c x x + cθ θ or Figure 5.8 Radial Equilibrium
ρ dr ρ dr dr dr
1 dp 0 dc
= cx x + θ
c d
(rcθ )
ρ dr dr r dr
Eq. 5.4
101
This equation clearly states that equal work is delivered at all radii and the total pressure losses
across a row are uniform with radius. It may be applied to two sorts of problem: the design (or
indirect) problem, in which the tangential velocity distribution is specified and the axial velocity
variation is found, or the direct problem, in which the swirl angle distribution is specified, the axial
and tangential velocities being determined.
⏟
𝐮 = 𝐰
⏟ + ⏟
𝐯 =𝐰+𝛚×𝐫
Absolute Relative Coordinates
Eq. 5.5
Introducing this into the mass conversation and after some manipulation we obtain,
∂r ρ
+ ∇. (ρ𝐰) = 0
∂t
Eq. 5.6
Comparing with non-inertia frame of reference, it seems to keep the same expression where
subscript r refers to the rotating frame of reference. Without causing confusion, the subscript r can
be omitted in general. The total derivative (acceleration) is also can be redefined as
Du ∂w ∂v
= + + w. (∇w) + 2 w
⏟× ω + ω ⏟× v
Dt ∂t ∂t Coriolis Centrifugal
Eq. 5.7
The first item on right-hand side expresses the local acceleration of the velocity field within the
rotating frame of reference. The second term and third item denote the angular velocity acceleration
Figure 5.9 Coriolis and Centripetal forces created by the Rotating Frame of Reference
102
and the convective term within the rotating frame of reference, respectively. While, the fourth item
and last item are the Coriolis acceleration and the Centrifugal acceleration, respectively, which are
fictitious forces produced as a result of transformation from stationary frame to rotating frame of
reference. Figure 5.9 shows the directions of the velocity and the acceleration, and relationship
between the absolute velocity, relative velocity and rotation (Schobeiri, 2005). Substituting the
acceleration in Error! Reference source not found. distinctly, for an incompressible flow equations o
f motion and energy, in rotating frame of reference can be obtained:
( w ) ( v)
Momentum : + + w.(w ) + ω v + 2ω w = μw − p + F
t t
w
2
v
2
D ρ h + +
2 2 p
=
Energy : + (kT ) + (τ w ) + w F + q H
Dt t
Eq. 5.8
Which can written in scalar form of (r, ϴ, and z) with the aid of cylindrical coordinates. It should be
noted that WF is the work of body forces in rotating frame of reference, F is the body force, while the
subscript r is omitted here. The detailed derivation process of governing equations in rotating frame
of reference can be found in [Schobeiri]88. Alternatively, we can choose more compact form of
integral representation with arbitrary control volume V and differential surface area dA in a relative
frame of reference rotating steadily with angular velocity ω:
dV + F − G dA = S dV where W = ρ , ρu , ρE and
dW
v = u − rω
T
V dt V
88 M. T. Schobeiri, “Turbomachinery: Flow Physics and Dynamic Performance”, Springer, Berlin, 2005.
89“Simulation of unsteady turbomachinery flows using an implicitly coupled onlinear harmonic balance method”,
Proceedings of ASME Turbo Expo 2011, GT2011.
90 Damian Vogt,” Turbomachinery Lecture Notes”, 2007.
103
Eq. 5.10
5.6.1 Isentropic Efficiency
Depending on which process is taken as ideal process efficiencies are referred to as isentropic or
polytrophic efficiencies. In case of an isentropic efficiency the ideal process is represented by an
isentropic change of state from start to end pressure, i.e. the same pressures as for the real process.
This is illustrated in Figure 5.11 for an expansion process by means of an enthalpy-entropy diagram
(h-s diagram). In the above depicted process the changes in total energy are referred to, which is
expressed by indexing the efficiency by “tt”, i.e. “total-to-total”. With the aid of h0 = h + (1/2) c2 where
c is the flow velocity, the total-to-total isentropic efficiency (expansion and compression) is thus given
by
actual change in energy Δh 0 h − h 02
For Expansion : η tt = = = 01
ideal change in energy Δh os h 01 − h 02s
Eq. 5.11
ideal change in energy Δh 0s h 02s − h 01
For Compression : η tt = = =
actual change in energy Δh 0 h 02 − h 01
Note: For adiabatic real processes the entropy must always increase during the change of state. Due
to this increase in entropy the real change in energy is smaller than the ideal during expansion. In
other words, you get out less energy from the real process than you could have from an ideal one For
the compression process the increase in entropy signifies that you need to put in more energy to
compress a fluid than you would have in an ideal process Therefore the efficiency is always smaller
or equal to unity The only way to reduce entropy would be to cool a process. However in such case
we do no longer look into adiabatic processes. In certain cases the kinetic energy that is contained in
the fluid (i.e. the amount of energy that is due to the motion) cannot be used at the end of a process.
An example for such a process is the last stage of an energy producing turbine where the kinetic
energy in the exhaust gases is not contributing to the total energy produced. In such case a so-called
total-to-static isentropic efficiency is used, identified by indexing the efficiency by “ts”, i.e. “total-to-
static”. Note that it is necessary to include total and static states in this case. The total-to-total
isentropic efficiency (expansion) is thus given by:
104
−1
actual change in energy h 01 − h 02 Δh 0 1 c 22
η ts = = = = + Eq. 5.12
ideal change in energy h 01−h 2s c 22 η tt 2h 0
Δh 0s +
2
This relation shows that for values of c2 > 0 the total-to-static efficiency is always smaller than the
total-to-total efficiency. For further detailed aspects of efficiency in turbomachines the readers
should consult with 91-92.
91 S.L. Dixon, B.Eng., PH.D., “Fluid Mechanics, Thermodynamics of Turbomachinery”, Senior Fellow at the
University of Liverpool, UK.
92 Damian Vogt, “Efficiencies”, Turbomachinery Lecture Notes, 2007.
93 Kalitzin, G. & Iaccarino G., “End wall heat transfer computations in a transonic turbine cascade”, XVII Congresso
design of the turbine cooling system. The objective of the present work is to use this database to
evaluate the influence of turbulence models on the accuracy of heat transfer predictions in
complex three-dimensional flows in turbine geometries. The sensitivity of the heat transfer
coefficient prediction to the turbulence model used is analyzed using two different models: the
Spalart-Allmaras one equation model and Durbin's four equation v2-f model. The use of two different
flow solvers, the NASA research code CFL3D and the commercial package FLUENT©, increases
confidence in the results and allows the elimination of effects related to the numerical discretization
of the equations.
5.7.1 Numerical Methods
The present results have been computed using two different RANS flow solvers: the NASA code
CFL3D and the commercial software Fluent® is a compressible, finite-volume code for multi-block
structured grids. The mean flow fluxes are computed with the Roe flux difference splitting scheme.
Turbulence models are solved segregated from the mean flow in an elimination of effects related to
the numerical discretization of the equations. The CFL3D is a compressible, finite-volume code for
multi-block structured grids. Turbulence models are solved segregated from the mean flow in an
implicit manner using three-factored Approximate Factorization. The v2-f model has been
implemented in this code in an implicit manner. The resulting linear algebraic system is solved with
a three or two-factored Approximate Factorization scheme. Fluent® solves the time-dependent RANS
equations on structured and unstructured meshes using a control-volume-based technique; the
diffusion terms are discretized using a second-order central-difference scheme while a second-order
upwind scheme is employed for the convective terms. An Euler implicit discretization in time is used
in combination with a Newton-type linearization of the fluxes. The resulting linear system is solved
using a point Gauss-Seidel scheme in conjunction with an algebraic multi-grid method. The additional
equations for the turbulent quantities are solved in a segregated fashion using a 1st or 2nd order
upwind discretization scheme with explicit boundary conditions.
5.7.2 Mesh Generation
The large scale linear cascade investigated in the experiments consists of twelve blades with an axial
chord of 10.7 cm. A part of the cascade is shown in Figure 5.13 A. The high blade count of the
cascade ensures good periodicity. This allows us to consider only one blade and only the region
between end wall and symmetry plane in the computations. The actual computational domain is
Shock
Reflection
shown in Figure 5.13 B. The block boundaries of the structured 3-block mesh and the boundary
conditions used are highlighted in the Figure 5.12. An O-mesh topology around the blade has been
chosen to ensure a high quality mesh near the blade surface. The two-dimensional mesh consisting
of 48x192 cells has been generated through simple geometric interpolation. After generating the
outer boundary as an arbitrary line between two blades and distributing lines connecting the outer
boundary with the blade, O-lines have been interpolated using a stretching function. The three-
dimensional mesh has been generated by copying the described 2D grid in the span-wise direction
and clustering the grid points at the end-wall. Two meshes, mesh A and mesh B, have been generated
with 40 and 52 cells span-wise, respectively. All block dimensions have been chosen to contain
factors of the power 2 to exploit multi-grid. The mesh has been transformed into an unstructured
mesh for the flow computations with Fluent©. The multi-block decomposition disappears for an
unstructured solver. The height of the first cell above the wall has an average y+ value of about 1. The
height has been adjusted after initial computation.
5.7.3 Heat Transfer Results for 2D & 3D
In the simulation of three-dimensional flow, the computational grid is often a compromise between
a desired resolution and computational accord ability. In two dimensions, however, it is easier to
carry out a complete grid sensitivity study. With this objective in mind, the flow in the symmetry
plane has been computed in a two-dimensional plane. The structured grid or default mesh, for this
report is shown in Figure 5.13(A). It is the same used at each span wise location in the three-
dimensional calculations. It contains 11008 cells. The unstructured grid, shown in Figure 5.13(B)
is obtained through successive refinement in regions with high pressure gradients and large strain
rates like shock waves, boundary layers, and wakes. This mesh contains 71326 cells. The Mach
contours plotted for both grids show a very complex shock wave pattern in the wake of the blades.
The accelerating flow within the passage generates an oblique shock wave on the pressure side of a
blade (see red circles in Figure 5.13). This shock is reflected on the suction side of the successive
blade. It then interacts with the viscous wake of the blade from which it originated. Partly due to
reflection in pressure BC. Somehow the new development by ANSYS© claims that Average Pressure
Specification at Pressure
Boundary which allows the f=0 f = 0.5 (Old)
exit pressure to vary across
the boundary, but
maintains an average
equivalent to the specified
exit pressure value94. It also
claims that it is less
reflective than previous
version with improved
results. The Pressure
blending factor ‘f’ (default
value 0.0) may need to
change f > 0.0 in cases
where stability is degraded.
For f = 0 recovers the fully Figure 5.14 Average Pressure Specification at pressure boundary
averaged pressure, and f = 1
recovers the specified pressure. The results of this improvement displayed in Figure 5.14.
A second shock wave is generated on the suction side near the trailing edge. The default mesh does
not resolve the shock wave the wake. Only the two shocks at the trailing edge are clearly visible. The
heat transfer at the wall depends significantly on the thermal conductivity of the fluid. The effect of
using a constant thermal conductivity at reference temperature is demonstrated with the FLUENT©
results reported in the same figure. The overall Stanton number is under-predicted. This explains the
difference observed between the FLUENT and CFL3D Stanton number distributions at the end wall
reported. It has to be noted that the constant thermal conductivity is the default option in. The
pressure distributions on blade and end-wall are not very sensitive to the grid resolution and inflow
profile for the case considered. Both flow solvers predicted a reasonable agreement with the
experiment as reported in. We note, however, that the pressure distribution on the blade and the
shock structure is sensitive to the treatment of the periodic boundary since it is located relatively
close to the blade surface. In this paper we will focus primarily on the analysis of the heat transfer
distribution, on the dependence of the Stanton number distribution on inflow profile and grid
resolution.
5.7.4 Experimental Data
The experimental data for the end-wall show some interesting features that will help to differentiate
the predictive capabilities of the models tested (Figure 5.15). The horseshoe vortex generated by
the rolling up of the incoming boundary layer enhances the wall heat transfer, and its structure is
clearly visible in the higher Stanton number (Region A). A second distinct heat transfer peak is
measured near the stagnation point (Region A). Within the passage, four additional interesting
features are present: the first is a localized peak in the Stanton number related to the impingement
of the suction-side leg of the horseshoe vortex on the blade surface (Region B). The second feature is
the presence of a shock wave on the pressure side near the trailing edge that increases the heat
transfer on the end wall (Region C). Third, there is a gradual increase of heat transfer at the end wall
which is related to the acceleration of the fluid in the passage (Region D). And finally, the presence of
a corner vortex on the suction side of the blade (Region E) is indicated in experiments by a low heat
transfer region. In the wake, a very sharp peak in the Stanton number is measured just downstream
of the trailing edge (Region C). The numerical predictions of the Stanton number show most of the
features observed in the experiments but, in general, fail to predict the quantitative heat transfer on
the end wall correctly.
5.7.5 Effects of Turbulence
The increased heat transfer beneath the horseshoe vortex is captured by both turbulence models.
The S-A model seems to spread this high Stanton number region and shift it towards the suction side.
Spreading of the horseshoe vortex is related to the turbulence generation in the vortex shear layer.
The v2-f model tends to produce a thinner vortex. The secondary peak on the suction side (Region
B), which is related to the stagnation of high temperature fluid convected by the horseshoe vortex, is
predicted by both models. The v2-f model predicts a higher value for the Stanton number. The SA
model predicts slightly larger values for the gradual increase in heat transfer within the passage
(Region D). The trailing edge peak (Region C) and the low heat transfer region on the suction side of
the blade (Region E) are reproduced by both models. A quantitative comparison of the heat transfer
on the blade surface is shown for three stations in Figure 5.16 for the v2-f and SA model,
respectively.
The heat transfer in the stagnation region, the location where span is 0, is accurately predicted by
both models at 25% and 50% span (solid line). Both stations are located outside of the incoming
boundary layer specified at the inlet. The station at 10% span, however, is located well inside of this
boundary layer, and both models over-predict the heat transfer here by 25%. The higher heat
transfer indicates that the turbulence intensity is too high at this location. This observation is
supported by a computation in which the turbulence levels inside the end wall boundary layer have
been reduced by setting the turbulence quantities at the inlet to a uniform value corresponding to
25% turbulent intensity (dotted line). This lowers the Stanton number in the stagnation region to
the value measured in the experiments. In addition, it delays the transition on the upper surface of
the blade. The SA model shows a large sensitivity to the reduced boundary layer turbulence across
the entire span on the pressure side. The heat transfer on the pressure side of the blade is consistently
under-predicted at each station by both models. The same has been observed for the 2D computation
shown in Figure 5.16. At this stage it is not clear whether this is due to the specification of the inlet
109
5.8 Case Study 2 - Using Shock Control Bumps To Improve Transonic Compressor
Blade Performance96
Citation : John, A., Qin, N., and Shahpar, S. (March 2, 2019). "Using Shock Control Bumps to Improve
Transonic Fan/Compressor Blade Performance." ASME. J. Turbomach. August 2019; 141(8):
081003. https://doi.org/10.1115/1.4042891
Shock control bumps can help to delay and weaken shocks, reducing loss generation and shock-
induced separation and delaying stall inception for transonic turbomachinery components, as
described by [John et al.]97. The use of shock control bumps on turbomachinery blades is investigated
here for the first time using 3D analysis. The aerodynamic optimization of a modern research fan
blade and a highly loaded compressor blade are carried out using shock control bumps to improve
their performance. Both the efficiency and stall margin of transonic fan and compressor blades may
be increased through the addition of shock control bumps to the geometry. It is shown how shock
induced separation can be delayed and reduced for both cases. A significant efficiency improvement
is shown for the compressor blade across its characteristic, and the stall margin of the fan blade is
increased by designing bumps that reduce shock-induced separation near to stall. Adjoint surface
Figure 5.17 Contours Of Casing Static Pressure Beneath A High-Speed Rotor (550 M/S
Tip Speed) With Pronounced Negative Camber. From Prince – Courtesy of [Prince]
95 Kalitzin, G. & Iaccarino G., “Computation of heat transfer in a linear turbine cascade”, Center for turbulence
Research Annual Research Briefs, 1999.
96 Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
sensitivities are used to highlight the critical regions of the blade geometries, and it is shown how
adding bumps in these regions improves blade performance. Finally, the performance of the
optimized geometries at conditions away from where they are designed is analyzed in detail.
5.8.1 Introduction and Motivation
Shocks are a major source of loss for transonic fans and compressors. They cause entropy generation,
boundary layer thickening and shock induced separation. The impingement of the shock on the blade
suction surface (and the resulting, strong, adverse pressure gradient) can cause the boundary layer
to detach, leading to larger blade wakes, reduced efficiency, lower blade stability and reduced stall
margin. Any method that can be used to alleviate shock strength (and the associated negative effects)
therefore has the potential to significantly improve transonic fan/compressor performance.
5.8.2 Shock Control for Turbomachinery & Literature Survey
Relatively little work on designing geometries directly to weaken the shock waves in transonic
turbomachinery components can be found in the literature, though it has been known for some time
that reducing the pre-shock Mach number of transonic compressors can improve their efficiency98.
It was clear to transonic compressor designers in the 70s and 80s that shock strength was increased
by the amount of convex curvature on the suction side between the leading edge and the shock99.
Nearly flat suction surfaces that minimized the expansion were therefore favored, with the next step
to try designs with concave curvature (often referred to as negative camber). Geometries with
negative camber result in gradual compression along the suction surface which weakens the shock.
The concave curvature of the blade surface and the reduction of flow area in the flow direction leads
to a deceleration of the supersonic flow through compression waves, and therefore a weaker shock.
[Prince]100 designed a rotor with pronounced negative camber (see Figure 5.17). This lead to a rise
in static pressure along the suction surface prior to the shock as intended, but the resulting efficiency
was disappointing due to the strong shock on the pressure surface. [Ginder and Calvert]101 had more
success in designing a rotor with negative camber. With negative camber, the Mach number ahead of
the shock was reduced to 1.4 (compared to 1.5 for the traditionally designed blade) which drastically
reduced the amount of boundary layer separation and loss.
Recently, it was demonstrated by [John et al.]102 how the freeform shaping of a compressor blade can
improve blade efficiency by delaying and weakening the shock and reducing separation. The flexible
parameterization method used allowed an s-shaped design to be generated that included a pre-
compression geometry around mid-span. This s-shaped, pre-compression geometry is similar to the
negative camber designs described above. The effect of the pre-compression geometry on the shock
and separation is described in Figure 5.18. The current work proposes the use of shock control
bumps as an alternative method to reduce shock related loss to those described above. Shock control
bumps have the benefit that relatively small modifications to the original geometry are required to
achieve the desired effect.
98 Ginder, R., and Calvert, W., 1987. “The design of an advanced civil fan rotor”. Journal of turbomachinery,
109(3), pp. 340–345.
99 Cumpsty, N. A., 1989. Compressor aerodynamics. Longman Scientific & Technical.
100 Prince, D. C., 1980. “Three-dimensional shock structures for transonic/supersonic compressor rotors”. Journal
Figure 5.18 Schematic of Shock Structures (A) Datum, (B) S-Shaped Design. Courtesy of [John et al.]
103 Ashill, P., and Fulker, J., 1992. “92-01-022 a novel technique for controlling shock strength of laminar-flow
aerofoil sections”. DGLR BERICHT, pp. 175–175.
104 Drela, M., and Giles, M. B., 1987. “Viscous-inviscid analysis of transonic and low Reynolds number airfoils”.
reduction”. Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering,
222(5), pp. 619–629.
112
Figure 5.19 Datum Geometry and Optimized Shock Control Bumps on The Mid- Section of Nasa Rotor
67- From Mazaheri et al..
The work by [Mazaheri & Khatibirad]109 demonstrated the benefit that bumps may provide at both
on and off-design conditions, and their potential to improve stall margin. The simplified 2D analysis
lacks accuracy however as the complex behavior of radial and separated flow cannot be predicted.
For a thorough understanding of the potential for the use of shock control bumps, 3D analysis and
the design of 3D bumps is needed to truly assess their effect.
5.8.4 Test Case - NASA Rotor 37
The case studied here is NASA Rotor 37. This has a very strong shock wave (with a relative tip Mach
number of nearly 1.5) which causes large separation, decreasing the blade efficiency. It is a well-
documented case, having been extensively tested and simulated as part of a turbomachinery
validation study. It is a transonic rotor with inlet hub-to-tip ratio 0.7, blade aspect ratio 1.19, rotor
tip relative inlet Mach number 1.48 and rotor tip solidity 1.29. It has historically been a challenge for
CFD simulation. The very high pressure ratio, strong shock wave-boundary layer interaction, large
tip leakage vortex and highly separated flow mean that it poses challenges for turbomachinery
108 Mazaheri, K., and Khatibirad, S., 2017. “Using a shock control bump to improve the performance of an axial
compressor blade section”. Shock Waves, 27(2), pp. 299–312.
109 See Previous.
113
solvers. Rotor 37 has been the subject of review articles that highlight the complexity of matching
experimental and computational measurements and the associated uncertainties.
The CFD setup is shown in Figure 5.20. At the inlet, a radial distribution of total pressure and
temperature (based on the original experimental values) is specified. The inlet turbulence intensity
is 1%. At the outlet, a value for circumferentially mixed-out and radially mean-mass capacity (non-
dimensional mass flow) is used. Periodic boundaries are used to represent full annulus flow.
Figure 5.20 The R37 CFD Domain Used – Courtesy of [John et al.]
Stationary walls are treated as adiabatic viscous walls and the rotational speed of the non-stationary
portions of the domain is 1800.01rads1, as specified in the experiment. Rolls-Royce CFD solver
Hydra is used for all of the simulations presented here, using the Spalart-Allmaras turbulence model
(fully turbulent). The 4.27 million cell mesh is generated by PADRAM, has y+ of the order of one on
all surfaces with 30 cells in the tip gap.
5.8.5 Validation
As previously alluded to, many studies have struggled when matching simulations of Rotor 37 to the
experiment. A wide range of work has been undertaken to investigate the discrepancy found between
simulation and experiment, with the primary work being the 1994 ASME/IGTI blind test case study
in which a range of codes were used to simulate the rotor, with no knowledge of the experimental
values. A large variation was seen between the different predictions, prompting analysis by
114
[Denton]110. Recent work has also been carried out by [Chima]111 and [Hah]112. The differences are
usually attributed to uncertainty in the experimental measurements, the lack of real geometry in the
simulations (e.g. the upstream hub cavity is usually missing) and also the difficulty in fully resolving
the complex flows. The pressure ratio agreement is reasonable across the characteristic, but the
efficiency prediction is about 2% below the experimental value at the design point (98% choke). This
matches the trend of previous results, where the better the PR prediction, the worse the efficiency
match. This ’trade-off’ has been seen in a range of previous simulations. Figure 5.21 gives the radial
profiles of total PR and efficiency at 98% of simulated choke compared to the experimental values at
98% experimental choke. The radial trends have been captured fairly well, although there is an offset
from the experiment for both. The choke mass flow found in the simulations was 20.91kg/s, matching
quite closely the experimental of 20.93kg/s.
110 Denton, J., 1997. “Lessons from rotor 37”. Journal of Thermal Science, 6(1), pp. 1–13.
111 Chima, R., 2009. “Swift code assessment for two similar transonic compressors”. In 47th AIAA Aerospace
Sciences Meeting including The New Horizons Forum and Aerospace Exposition, p. 1058.
112 Hah, C., 2009. “Large eddy simulation of transonic flow field in NASA rotor 37”. 47th AIAA Aerospace
Sciences Meeting including The New Horizons Forum and Aerospace Exposition, p. 1061.
115
separate and a large wake forms. It is at this design point that Rotor 37 will be optimized, as a
reduction in this separation could significantly increase blade efficiency.
5.8.7 Validation
Due to experimental data for this geometry not being available, simulation validation was carried out
using a similar fan blade geometry that has experimental data available. The related blade has very
similar performance parameters, and the simulation set up is identical. The results are given here. A
comparison of the simulations of this related blade against experimental data can be seen in 113. Both
the pressure ratio and efficiency curves match the experimental data well, though there is a slight
offset to the overall values and stall margin. The radial curves show good comparison to experimental
data, although the radial variation in efficiency is underpredicted compared to the experiment.
Overall, the simulation compares well, lying within 1% across the range of flow rates.
Figure 5.22 (a) 3d Separation (Orange) On The R37 Geometry (Flow Right To Left), (b) Rel. Mach
No. Contour At 60% Span – Courtesy of [John et al.]
113 Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
Fan/Compressor Blade Performance”, GT2018-77065.
114 Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
area to investigate the benefit of shock control bumps. It is the shock-induced separation that is
responsible for limiting the operating range of the blade, and if this separation can be reduced then
it is expected that this will extend the stable working range of this fan.
Figure 5.23 Shock Region Flow Features For RR-Fan At Points A) A, B) B, C) C, D) D, E) E, F) F. Flow
Direction – Courtesy of [John et al.]
RIGHT TO LEFT.
5.8.9 Adjoint Sensitivity Analysis
Adjoint sensitivity analysis is a useful tool that can be used to provide information on the sensitivity
of an objective function to changes in the geometry. Here, the adjoint sensitivity used is the
sensitivity of efficiency (as a percentage) to surface deformation (in mm) normal to the surface. This
can be used to inform which regions of the blade will have the greatest impact when modified, and
are therefore most important to control during an optimization. Hydra Adjoint115 is used to provide
the blade surface sensitivities: A primal Hydra simulation is first used to provide the flow solution,
followed by Hydra adjoint which calculates the flow-adjoint sensitivity and provides the sensitivity
of the objective function to changes in the flow. Once these two relatively expensive simulations are
completed, the mesh sensitivities are then mapped onto the surface. This finds the relationship
between changes in the flow to changes in the blade surface mesh. Combining these provides the
sensitivity (gradient) of the objective function (efficiency) to perturbations of the blade surface. The
adjoint surface sensitivity analysis for Rotor 37 at design point and RR-FAN at point D. It can be seen
115 Duta, M. C., Shahpar, S., and Giles, M. B., 2007. “Turbomachinery design optimization using automatic
differentiated adjoint code”. ASME Turbo Expo 2007: Power for Land, Sea, and Air, American Society of
Mechanical Engineers, pp. 1435–1444.
117
that the most sensitive regions of both geometries are focused around the shock on the suction
surface. This indicates that geometry changes in this region will have a significant impact on the blade
efficiency, and therefore if shock control bumps are applied here some benefit should be found. For
complete details, please consult the [John et al.]116
5.8.10 Shock Bump Parameterization & Optimization
The CST (Class Shape Transformation) method is used in this work to define the bump geometries.
The CST method uses Bernstein polynomials to create smooth (second derivative continuous)
contour bumps. For this project 3rd order CST bumps are used, constructed from four Bernstein
polynomials.
Controlling the
weighting (amplitude)
of these polynomials
modifies the bump
height and asymmetry.
Figure 5.24 shows
how the bump
geometry (solid black
line) to be added to the
blade surface is a sum of
the four Bernstein
polynomials (colored
dashed lines). The CST
bump parameterization
provides a high degree
of flexibility, enabling
the generation of Figure 5.24 Example 2d CST Bump (Solid Line) And The Four Polynomials
Used To Construct It (Dashed Lines) – Courtesy of [John et al.]
smooth, asymmetric
bumps in 2D and 3D.
The CST bump parameterization
technique was implemented inside of
the PADRAM geometry and meshing
software. The technique modifies each
2D radial section of the blade geometry,
adding a bump. The properties of these
2D bumps are smoothly interpolated in
the radial direction from control
sections. The resulting geometry is
controlled by the bump start and end
positions, the four Bernstein
polynomial amplitudes and the span-
wise distribution. This allows 3D
variation of the bumps in the radial
direction. Both continuous (where Figure 5.25 a) Example Individual Bump Geometry
bump amplitudes are smoothly And B) Example Continuous Bump Geometry – Courtesy of
interpolated radially) and individual [John et al.]
(where the bump amplitude returns to
116Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
Fan/Compressor Blade Performance”, GT2018-77065.
118
zero periodically in the radial direction) CST bumps were tested. Examples of the blade with
individual and continuous bumps added is shown in Figure 5.25.
During this work, a study was carried out (not detailed here for brevity) to compare the benefit of
using individual bumps (where a series of discrete bumps is added to the datum geometry in the
radial direction) with a continuous bump (note continuous bumps are still ’3D’ and their shape,
position and amplitude can vary in the radial direction). It was concluded that, for these cases, the
individual bumps needed to have greater amplitude than the continuous bumps to offer the same
benefit, leading to increased separation downstream of the bump position. The continuous bumps
tested offered greater benefit, and therefore only results using the ’continuous’ bump geometry
approach are presented here.
5.8.11 Optimization Method
In this work the Multi-point
Approximation Method (MAM) is
used for the optimization studies. It is
a gradient based method that uses
localized Design of Experiments (DoE)
and trust regions to efficiently search
through the design space. When using
MAM, an initial generation of
simulations (chosen by DoE) is carried
out around the start point. A response
surface is constructed for this region
and the sub-optimal point found. The
search is then moved to this point,
where a new generation is constructed Figure 5.26 Spanwise Slice of the Datum and Optimized
and the process repeated until the R37 Geometries At 60% Span – Courtesy of [John et al.]
search converges on the optimal
design. The MAM method has been shown to be an efficient and consistent approach for a wide range
of highly constrained optimization problems, working successfully for design spaces made up of
hundreds of parameters.
5.8.12 Rotor 37 Bump Optimization
For the Rotor 37 optimization, the bump geometry was
controlled at 5 radial heights (to allow radial variation of
the parameters) with the geometry smoothly interpolated
between the control stations using a cubic B-spline.
Towards the tip the bump placement and movement range
are increased in chord-wise position as the shock is sat
further downstream at the tip. The initial design used at
the start of the optimization process had bumps positioned
with approximately 60% of the bump downstream of the
datum shock, as is known to be beneficial from previous
work. The objective function for the optimization was
blade efficiency and the simulations were carried out at
98% simulated choke. The optimizations were carried out
on the Rolls-Royce CFMS cluster using the MAM method.
The geometry of the optimized shock bump can be seen in
Figure 5.26. A slice at 60% span is shown. The 3D Figure 5.27 Optimized R37 Bump
geometry compared to the datum is shown in Figure 5.27. (Blue) Added To The Datum Blade
Geometry (Grey)
The bump applied to the datum geometry varies radially,
119
with the maximum bump amplitude and width localized between 40 and 60% span. This makes sense
as the strongest shock location, largest separation and maximum adjoint sensitivity occur around
mid-span for Rotor 37, and therefore greater shock control is needed in this region. The resulting
variation from hub to tip of the geometry demonstrates the benefit provided by optimizing the
geometry. Without optimization it would be difficult to manually specify the bump position, width,
amplitude and asymmetry, which would result in reduced benefit.
5.8.13 Analysis of the R37 Optimized Bump Design
The flow features for the resulting, optimized, continuous bump design is compared to the datum in
Figure 5.28. The datum shock position is shown via a white line on the optimized geometry. It can
be seen how the use of bumps has delayed the shock. The reduction in separation for the optimized
design can be seen in Figure 5.30. The delay of the shock position has reduced the separation
initiation point and the volume of separated flow. The performance of this geometry is compared to
the best
individual PR Delta PR / % Efficiency / % Delta efficiency / %
bumps Datum 2.05 - 85.45 -
geometry
Individual 2.06 0.51 86.21 0.76
(not
Cont. 2.08 1.2 86.93 1.48
described
in detail Table 5.1 Rotor 37 Optimized Bump Performance Comparison – Courtesy of [John et al.]
here) and
the datum
in Table 5.1. It can be seen that the efficiency benefit is greatest for the continuous bump design.
The efficiency is increased by 1.48%, while the pressure ratio is also increased. A summary of
Figure 5.28 Datum (Left) And Optimized (Right) Rotor 37 Static Pressure Contours. Flow Direction
Right To Left – Courtesy of [John et al.]
previous optimization results for Rotor 37 by various researchers is given by [John et al.]117. The
maximum efficiency benefit achieved by those studies was around 1.7-1.9% (without decreasing PR).
These optimizations were able to modify parameters such as blade camber, thickness, lean and
sweep though, so had greater design flexibility than the current shaping approach. This shows that
Figure 5.30 Datum (Left) And Optimized (Right) Rotor 37 Separated Flow Contours (Orange). Flow
Direction Right To Left.
boundary layer, and secondary flows (like the tip-leakage vortex). Secondary flows and their
interactions with other phenomena are another major source of flow complexity. Indeed, Denton and
Dawes, 1999, suggest the prediction of blade surface and end-wall corner separations to be one of
the most challenging tasks of 3D, viscous solvers, largely due to the obvious dependence on
turbulence model. Additionally, the use of blade twist, sweep (viewed from the meridional plane)
and lean (observed looking axially through the machine) contributes to the 3D flow effects.
A significant consideration in the design of transonic fan blades is the control of shock location and
strength to minimize aerodynamic losses without limiting flow. Custom-tailored airfoil shapes are
required to “minimize shock losses and to provide desired radial flow components. Figure 6.2 shows
features of the tip section geometry typical of a transonic fan. The shape of the suction surface is key
as it:
• Influences the Mach number just ahead of the leading edge passage shock, and
• Sets the maximum flow rate.
124
As noted by Wisler, 1987, the cascade passage area distribution is chosen to provide larger-than-
critical area ratios; thus, maximum flow is determined by the first captured Mach wave, location
determined by the forward suction surface (induction surface). This maximum flow condition is often
referred to as leading edge choke, or in cascade parlance, “unique incidence” (note that “unique”
incidence is really a misnomer; here, “choking” incidence will be used). The flow induction surface
and fan operating condition (incoming relative Mach number at the airfoil leading edge) set the
average Mach number just ahead of the leading edge passage shock. A “traditional” convex suction
surface results in a series of Prandtl-Meyer expansion waves as the flow accelerates around the
leading edge. Increasing the average suction surface angle (relative to the incoming flow) ahead of
the shock reduces the average Mach number, and presumably reduces the shock losses. Common for
modern transonic fan tip sections is a concave induction surface, the so-called “pre-compression”
airfoil. As indicated in previous chapter, there are four major area of research going on in
turbomachinery, namely: Unsteady Flow, Film cooling, Turbulence and 3D Flow. We start with the
unsteadiness first.
119Lakshminarayana, B. “An assessment of computational fluid dynamic techniques in the analysis and design of
turbomachinery”, the 1990 freeman scholar lecture, J. Fluids Engineering Vol. 113(No. 3): 315-352, 1991.
125
1. Various forms of secondary flow caused by viscosity and complex geometry, which is
dominated by vortex flows: passage, leakage, corner, trailing, horseshoe and scraping
vortices, etc. These form three- dimensional and rotational nature of the flow.
2. Inherent unsteadiness (see below) due to the relative motion of rotor and stator blade
rows in a multi stage environment.
3. The flow pattern in the near-wall region includes: laminar, transitional and turbulent
flows; besides separated flows are often exist.
4. The flow may be incompressible, subsonic, transonic or supersonic; some
turbomachinery flows include all these flow regimes.
5. Due to the limitation of flow space, there are strong interactions of the solid wall surfaces
with above complicated phenomena. Besides, in gas turbines, the use of cooling gas makes
the flow more complex.
A good understanding of the unsteady flow in turbomachinery is necessary for advanced design as it
shown in Figure 6.3 with broad spectrum. According to Greitzer120, the unsteady flow in
turbomachinery can be classified into two groups: inherent unsteadiness and conditional
unsteadiness. The conditional unsteadiness is mainly caused by the sudden changes of the working
condition. For example when turbomachinery is working on the start stage, acceleration stage or off-
design condition, the fluctuation of working condition might lead to the unsteady rotating stall,
surge, flutter and flow distortion of turbomachines. Sometimes, the distortion of inlet flow or the
asymmetric outlet condition of vector nozzle also might lead to the unsteadiness. The inherent
unsteadiness is mainly due to the relative motion and interaction between rotor and stator and,
generally speaking, it could be divided as:
Figure 6.3 Flow Structures with 5 to 6 Orders of Magnitudes Variations in Length and Time
Scales (LaGraff et al., 2006)
120 E. M. Greitzer, “Thermoaldynamics and fluid mechanics of turbomachinery”, AS1/E 9713, NATO, 1985.
126
A - Mach
number
contours
and operating range over which the fan must operate121. Figure 6.4 (A-B-C) illustrates some typical
features – leading edge oblique shock, aft passage normal shock below peak efficiency, and a near-
normal, detached bow shock near peak efficiency (and higher) loading conditions. Note that
throughout this report, loading refers to flow turning. For high tip-speed fans (inlet relative Mach
numbers greater than 1.4), the trend seems to be to design for an oblique leading edge shock through
higher loading conditions (near and at peak efficiency). This trend seems reasonable given the
continued need to reduce losses. Other flow field considerations in transonic fans include the
interrelationship between the rotor tip-clearance vortex structure and passage shock, high Mach
number stator flow, most notably in the hub region, and strong shock – boundary layer interaction.
121Boyer, K., M., “An Improved Streamline Curvature Approach for Off-Design Analysis of Transonic Compression
Systems “, PhD. Dissertation, Virginia Polytechnic Institute and State University, 2001.
T.Matsunuma, “Unsteady flow field of an axial-flow turbine rotor at a low Reynolds number”, ASME-GT06,
122
instantaneous absolute velocity contour pattern at the nozzle exit shown in Figure 6.6. The
experimental results suggest that the secondary vortices are periodically and three-dimensionally
distorted at the rotor inlet. A curious tangential high turbulence intensity region spread at the tip
side is observed at the front of the rotor, which is because of the axial stretch of the nozzle wake due
to the effects of the nozzle passage vortex and rotor potential flow field.
Figure 6.6 Instantaneous Absolute Velocity Contour at Nozzle Exit [Matsunuma, 2006]
• When the wake passes, the wake-induced turbulent spots form within attached flows in front of
the separation point, the turbulent spots continue to grow and enter into the separation zone,
and consequently inhibit the formation of separation bubble. The calmed region trails behind
the turbulent spots. It is a laminar-like region, but it has a very full velocity profile. The flow of
the calmed region is unreceptive to disturbances. Consequently, it remains laminar for much
longer than the surrounding fluid and can resist transition and separation. It is the combination
of the calming effect and the more robust velocity profile within the calmed region that makes
this aspect of the flow so important. After the interaction of the wake, boundary layer separation
occurs in the interval between the two wakes.
about the leakage flow unsteady interactions in compressor. For example, [Sirakov & Tan (2003)]
investigated the effect of upstream unsteady wakes on compressor rotor tip leakage flow. It was
found that strong interaction between upstream wake and rotor tip leakage vortex could lead to a
performance benefit in the rotor tip region during the whole operability range of interest. The
experimental result of [Mailach et al. (2008)] revealed a strong periodical interaction of the incoming
stator wakes and the compressor rotor blade tip clearance vortices. As a result of the wake influence,
the tip clearance vortices are separated into different segments with higher and lower velocities and
flow turning or subsequent counter-rotating vortex pairs. The rotor performance in the tip region
periodically varies in time. Compared with in compressor, very little published literature is available
on the unsteady interactions between leakage flows and adjacent blade rows in turbine. [Behr et al.
(2006)] indicated that the pressure field of the second stator has an influence on the development of
the tip leakage vortex of the rotor. The vortex shows variation in size and relative position when it
stretches around the stator leading edge.
• Profile losses due to blade boundary layers and their separations and wake mixing; in high
speed condition, shock/boundary layer interaction may exist.
• End-wall boundary layer losses, including secondary flow losses and tip clearance losses.
• Mixing losses due to the mixing of various secondary flows, such as the passage vortex and
tip leakage vortex.
Among all these losses, the most complex one is the secondary flow loss. That is why considerable
research on the secondary flow phenomena has been done in last decades. Secondary flow is
defined as the difference between the real flow and a primary flow, which is related to the
development of boundary layer on end-wall and blade surface, the evolution of vortices in
passage, and detached flows or simply, the secondary flow in a blade row can be defined as any
123 Sh. Kang, “Investigation on the Three Dimensional within a Compressor Cascade with and without Tip
Clearance”, PhD thesis, Vrije Universiteit Brussel, September 1993.
131
flow, which is not in the direction of the primary or stream wise flow 124. Based on topology
analysis and experiments, as well as the numerical simulations in recent decades, a couple of
secondary flow models are proposed which are presented below.
6.8.1 Classical View
The so-called classical secondary flow model, as illuminated in Figure 6.9 (a-b), is proposed by
124 Lei Qi and Zhengping Zou, “Unsteady Flows in Turbines”, Beihang University China.
132
Hawthorne125 for the first time according to the theory of inviscid flow in 1955. This model presents
the components of vorticity in the flow direction when a flow with inlet vorticity is deflected through
a cascade. The main vortex, so-called passage vortex, represents the distribution of secondary
circulation, which occurs due to the distortion of the vortex filaments of the inlet boundary layer
passing with the flow through a curved surface. The vortex sheet at the trailing edge is composed of
the trailing filament vortices and the trailing shed vorticity whose sense of rotation is opposite to
that of the passage vortex. The classical vortex model attributes the secondary flow losses to the
generation and evolution of vortex passage. However, this model is relatively simple, in which the
interaction between the inlet boundary layer and blade force was not considered. Moreover, the
vortex system within passage is only single passage vortex in half of the passage height range with
other vortices absences. The secondary flow losses can be visualized by absence/presence of
secondary vortex on
Figure 6.9 (b).
6.8.2 Modern View
When a shear flow along
the solid wall approaches a
blade standing on the wall,
the shear flow will be
separated from the wall
and roll up into a vortex in
front of the blade leading
edge. This vortex is called
horseshoe vortex due to its
particular shape. This well-
known phenomenon is
firstly observed in the flow
around cylinders. The oil
(a) Kline 1966
flow visualizations by
[Fritsche]126 show the
evidence of the horseshoe
vortex in accelerating
cascades. In 1966, Klein
presents a finer cascade
vortex model with both the
passage and horseshoe
vortices as depicted in
Figure 6.10 (a). While, the
pioneering work for
detailed analysis of
secondary flow patterns in
turbine cascades in general (b) Langston, 1977
is done in 1977 by
[Langston et al.]127 who Figure 6.10 Modern Secondary Flow Model
proposed the well-known
125 W. R. Hawthorne,” Rotational flow through cascades part 1: the components of vorticity.” Journal of Mechanics
and Applied Mathematics, 8(3):266–279, 1955.
126 A. Fritsche. Str¨omungsvorg¨ange in schaufelgittern. Technische Rundschau Sulzer, 37(3), 1955.
127 L. S. Langston, “Three-dimensional flow within a turbine blade passage”, Journal of Engineering for Power,
99(1):21–28, 1977.
133
The first point from Langston is supported by the light sheet experiment by Marchal and
128 C. H. Sieverding, “Recent progress in the understanding of basic aspects of secondary flows in turbine blade
passages”, Journal of Engineering for Gas Turbines and Power, 107(2):248–257, 1985.
129 L. S. Langston, “Three-dimensional flow within a turbine blade passage”, Journal of Engineering for Power,
99(1):21–28, 1977.
130 C. H. Sieverding, “Recent progress in the understanding of basic aspects of secondary flows in turbine blade
passages”, Journal of Engineering for Gas Turbines and Power, 107(2):248–257, 1985.
134
Sieverding131 in 1977. While, the results of this experiment also show the counter-rotating vortex,
called counter vortex by Langston, in the trailing edge plane on the mid span side of the passage
vortex rather than in the corner, which is not consistent with the second point from Langston.
6.8.3 Latest View
In 1987, [Sharma and Butler]132 proposed a secondary flow pattern which is slightly different to that
from Langston. This pattern, shown in Figure 6.11 (a), demonstrates that the suction side leg of the
horseshoe vortex wraps itself around the passage vortex instead of adhering to the suction side. This
result is similar to the results of [Moore]133 and [Sieverding]134. However, in 1988, another pattern is
given by [Goldstein and Spores]135, shown in Figure 6.11 (b), which is different to Sharma’s again.
Based on mass transfer results, they suggested that the suction side leg of the horseshoe vortex stays
above the passage vortex and travels with it. This flow pattern is similar to that suggested by [Jilek]136
in 1986. The major difference among these three models is the location of the suction side leg of the
horseshoe vortex. Since it is difficult to be detected due to the small size, most literatures cannot
demonstrate develop of this vortex clearly. In 1997, a very detailed secondary flow visualization
study was performed by Wang137. They proposed a more comprehensive but more complicated
secondary flow the passage vortex and travels with it. This flow pattern is similar to that suggested
by Jilek138 in 1986. The major difference among these three models is the location of the suction side
leg of the horseshoe vortex. Since it is difficult to be detected due to the small size, most literatures
cannot demonstrate develop of this vortex clearly. In 1997, a very detailed secondary flow
visualization study was performed by [Wang]139. They proposed a more comprehensive but more
complicated secondary flow pattern, as illustrated in which includes the passage vortex, the
horseshoe vortex, the wall vortex and the corner vortex. The development of the horseshoe vortex
nearby the end-wall is effected by the boundary layer on end wall and the blade surface. In modern
advanced blade, the leading edge radius of blade is so small that can be compared with the thick of
boundary layer. Hence, the separation of boundary layer on end wall generates the multi-vortex
structures at the leading edge of blade. Due to a strong pressure gradient the pressure side leg of the
horseshoe vortex moves toward the suction side after it enters the passage. Meanwhile it entrains
the main flow and the inlet boundary layer forming a multi-vortex leg. In 2001, [Langston]140
reviewed these new models after the [Sieverding’s] review. Laster in the same year, [Zhou and
Han]141 gave a more comprehensive review of all these models. They concluded that the good
131 P. Marchal and C. H. Sieerding, “Secondary flows within turbomachinery blading’s”, CP 214, AGARD, 1977.
132 O. P. Sharma and T. L. Butler, “Prediction of the end wall losses and secondary flows in axial flow turbine
cascade. Journal of Turbomachinery”, 109:229–236, 1987.
133 J. Moore and A. Ransmayr, “Flow in a turbine cascade part 1: losses and leading edge effects”, ASME, 1983.
134 C. H. Sieverding and P. Van den Bosch,” The use of colored smoke to visualize secondary flows in a turbine-
understanding of the secondary flow in turbomachinery can help greatly to control the vortices
within passage and decrease the losses, help greatly to control the vortices within passage and
decrease the losses.
6.8.4 Comparing and Contrasting Secondry Flow in Turbine and Compressors
Another view begins by comparing and contrasting turbine and compressor secondary flows,
together with conclusions on the way forward to design in compressors142. A large amount of
material has been published on secondary flow effects in axial flow turbomachinery, both turbines
and compressors. Only a brief summary of these is given here. As will be seen in the next section,
non-axisymmetric end wall profiling has been pursued in recent years principally in the field of axial
flow turbines. Consequently, it is useful to compare and contrast turbine and compressor secondary
flows.
Comprehensive reviews of
turbine secondary flows are
given in [Sieverding and
Langston], and of secondary
loss generation in [Denton].
Whilst secondary flows are
induced by any total
pressure profile that enters a
blade row and is
subsequently deflected by it,
the clearest understanding
has been obtained for the
case when the total pressure
profile is just due to the
incoming end wall boundary
layers. Figure 6.12 shows a
diagrammatic
representation of turbine
end wall secondary flows
taken from [Takeishi et al.]
(note that the rotation of the
vortices is generally
exaggerated) which has
been describe this more Figure 6.12 Turbine Secondary Flow Model (Takeishi et al.)
fully, but the basic elements
are:
• Rolling up of the inlet boundary layer into the horseshoe vortex at the airfoil leading edge.
The pressure surface side leg of this becomes the core of the passage vortex. The passage
vortex is the dominant part of the secondary flow and beneath it on the end wall a new
boundary layer is formed, referred to as cross-flow "B" in Figure 6.12, which starts in the
pressure side end wall corner.
• Upstream of this the inlet boundary layer is deflected across the passage (over turned),
referred to as cross-flow "A". The end wall separation line marks the furthest penetration of
the bottom of the inlet boundary layer into the passage and divides it from the new boundary
layer forming downstream of it. The dividing streamline between the suction and pressure
142N W Harvey, “Some Effects of Non-Axisymmetric End Wall Profiling on Axial Flow Compressor Aerodynamics.
Part I: Linear Cascade Investigation”, Proceedings of GT2008.
136
side flows is shown as the attachment line in Figure 6.12. It intersects with the separation
line at the saddle point.
• The new end wall boundary layer, cross-flow "B", carries up onto the airfoil suction surface
until it separates (along the airfoil "separation line") and feeds into the passage vortex. The
suction side leg of the horseshoe vortex, referred to as the counter vortex in Figure 6.12,
remains above the passage vortex and moves away from the end wall as the passage vortex
grows.
• A small corner vortex may occur in the suction surface/ end wall corner rotating in the
opposite sense to the passage vortex. This has the effect of opposing the overturning at the
end wall, although at the cost of additional loss.
One additional source of “classical” secondary flow that must be mentioned is the trailing edge
vorticity that originates as a vortex sheet downstream of the blade trailing edge due to the variation
in circulation along the span of the airfoil (and not shown). The scope for reducing this by modifying
the end wall flows does not appear to be great and has not been part of this study. The basic features
of compressor secondary flows are the same as those in a turbine blade row. However, there are a
number of important differences in the details between the two, [Cumpsty]143:
• The turning in a compressor blade row is much lower; typically 30 – 40 degree , compared to
100 degree in a turbine.
• From classical secondary flow theory, this would be expected to result in lower secondary
flows in a typical compressor row, for a comparable inlet total pressure profile.
• An additional feature, often overlooked, for turbine secondary flows is that once they have
rolled up into vortices any further acceleration of the flow will stretch them feeding in more
kinetic energy (of rotation), [Patterson]. This may have the effect of amplifying the benefit of
anything that delays the initial development of secondary flows on the end walls.
• Since the flow through a compressor blade row diffuses such vortex stretching will not occur.
Rather the diffusion will encourage more rapid mixing out of the vortices. It is suggested that
this is the reason why the smaller vortices (counter and corner) seen in turbine rows are not
often identified for compressor ones. In addition end wall over-turning in a compressor row
will be much more likely to result in flow separation, especially when the static pressure rise
across the row increases if the compressor moves up its characteristic.
6.9 3D Separation
A number of different flow regimes come under the heading of “three-dimensional separation”:
• If the aerodynamic loading is low enough, then the low momentum fluid in the airfoil suction
side/ end wall corner will separate off the blade surfaces (as in turbine secondary flows) but
will still have forward momentum.
• Where the loading is such that reverse flow does occur, then this may initially only be on one
of either the end wall or the airfoil suction surface refers to the former as “wall stall” and the
latter as “blade stall”.
• The combination of these two is known as “corner stall”. The resulting flow patterns are
illustrated in Figure 6.13. where the illustration of formation of hub corner stall together
with limiting streamlines and separation lines, (Lei et al.).
144 Martin N. Goodhand and Robert J. Miller, “The Impact of Real Geometries on Three-Dimensional Separations
in Compressors”, Journal of Turbomachinery, 2012.
138
on global flow parameters, such as blade loading, inlet boundary layer skew, etc., and thus ignoring
small deviations such as those highlighted above. The results show that any deviation which causes
suction surface transition to move to the leading edge over the first 30% of span will cause a large
growth in the size of the hub separation, doubling its impact on loss. The geometry deviations that
caused this, and are thus of greatest concern to a designer, are changes in leading edge quality and
roughness around the leading edge, which are characteristic of an eroded blade. 3D separations
always occur on compressor blades in the corner between the suction surface and the end wall145
(see Figure 6.14). Blades can be designed such that these are relatively small and benign; however,
as loading or incidence is increased, their size and thus detrimental effect can increase significantly.
In practice, it is these separations which limit the total blade loading by their impact on loss, blockage,
and deviation.
145 Gbadebo, S. A., Cumpsty, N. A., and Hynes, T. P., 2005, “Three-Dimensional Separations in Axial
Compressors,” ASME J. Turbomachinery , 127, pp. 331–339.
146 Karen Thole, “Mechanical Engineering Department Penn State University”, Park, PA 16802-1412.
147 Langston, L. S. “Cross flows in a Turbine Cascade Passage,” ASME J of Engineering for Power 102 -1980.
139
depicted in Figure 6.16-(Top). The flow model shows that the inlet boundary layer separates from
the approaching end-wall to form what is known as a horseshoe vortex. One leg of the horseshoe
vortex, present on the pressure side of the airfoil (concave side), is convected into the passage and is
promoted by the inherent pressure gradient between the two airfoil surfaces. This pressure side leg
of the horseshoe vortex develops into what is known as the passage vortex. The other leg of the
horseshoe vortex, present on the suction side of the airfoil (convex side), has an opposite sense of
rotation to the larger passage vortex and develops into what is known as a counter vortex. The
counter vortex can be thought of as a planet rotating about the axis of the passage vortex (sun). While
this picture represents a time averaged representation, measured data indicates that the vortex is
not steady. While the development of the vortical structures originates in the end-wall regions, the
growth can be such that the passage vortex occupies a large portion of the airfoil exit. This vortical
structure growth extends up to 30-40% of the total span for older vane designs and has been reduced
to approximately 10-15% of the total span
in the last 15 years. The flow patterns
previously described make it difficult to
cool the end-wall, particularly when
considering the near end-wall flow, as
illustrated in Figure 6.16148. The surface
flow visualization in Figure 6.16, achieved
through an oil and surface dye technique,
illustrates the strong cross flows that occur.
The cross flows are driven by the inherent
pressure gradients from the pressure to the
suction side of adjacent airfoils. For
example, these cross flows influence how
the film-cooling jets exit from the holes as
well as influence the end-wall heat transfer
coefficients149.
6.10.1 Theoretical Development of End-
Wall Flows
As the end-wall boundary layer approaches
a turbine airfoil, the flow stagnates,
whereby the total pressure becomes the
static pressure along the span of the vane.
Given that the fluid nearer to the end-wall
has a lower velocity, a stronger deceleration
in the boundary layer occurs for the higher
speed fluid than for the lower speed fluid. As
a result of these differences in the
deceleration, a transverse static pressure
gradient occurs along the vane span causing
Figure 6.16 (Top) - Measurements of the Horseshoe
the higher speed fluid to turn toward the Vortex just upstream of the Vane at the Vane-End-Wall
end-wall plate. Subsequently, the formation Juncture (Bottom) - Actual Hardware Showing Effects
of a horseshoe vortex occurs just upstream of the Horseshoe Vortex on a First Vane (Courtesy of
ASME)
148 Friedrichs, S., Hodson, H. P. and Dawes, W. N., “Distribution of Film-Cooling Effectiveness on a Turbine End-
wall Measured Using the Ammonia and Diazo Technique,” J of Turbomachinery 118 -1996.
149 Karen Thole, “Mechanical Engineering Department Penn State University”, Park, PA 16802-1412.
140
of the turbine vane. One of the legs of the horseshoe vortex wraps around the pressure side of the
vane and the other leg wraps around the suction side of the vane. Figure 6.16 (top) shows
measurements of the horseshoe vortex upstream
of the turbine vane and Figure 6.16
(Bottom)150 shows extreme damage from the
effects of this vortex on actual hardware151.
These measurements were made for a simple
case with an isothermal flow and an approaching
boundary layer that was 9 % of the vane span
(Z/S = 0.09). From the combined contour/vector
plot in Figure 6.16 (top), one can see where the
flow separates from the upstream end-wall and
how the flow is then rolled into a vortex. It is also
clear to see that, if a film-cooling jet were injected
into this region, it would be difficult to maintain
the coolant along the end-wall.
The flow fields discussed thus far have described
the condition for a uniform, isothermal flow field
with an approaching two-dimensional boundary
layer along the end-wall. In practice, this
idealized flow situation rarely happens since an
upstream combustor is present whereby there
can be large variations in the exiting flow. Non-
uniformity of inlet profiles in addition to the
viscous boundary layers along the end-wall are
caused by temperature gradients at the
combustor exit. The development of crossflow
and vortical motions in a curved passage, such as
an airfoil passage, can be understood by
considering flow along two streamlines, as
shown in Figure 6.17. Two idealized cases are
considered: Top) a gradient of velocity due to a
turbulent inlet boundary layer with an Figure 6.17 Illustration of Different Vortical
isothermal flow, Bottom) a linear temperature Patterns that are possible for two idealized flow
gradient typical at the exit of a combustor with a conditions: Top) Isothermal with an Inlet
uniform velocity field152. Assuming steady, Boundary Layer ; and , Bottom) Inviscid Flow
incompressible, inviscid flow with negligible with a Temperature Profile
variation of velocity in the n-direction, the
centripetal acceleration for the streamlines A and B must be balanced by the pressure gradient across
the pitch:
2 2
∂p VSA ∂p VSB
| =ρ , | =ρ
∂n A RA ∂n B RB
150 Friedrichs, S., Hodson, H. P. and Dawes, W. N., “Distribution of Film-Cooling Effectiveness on a Turbine End-
wall Measured Using the Ammonia and Diazo Technique,” J of Turbomachinery,1996.
151 M. Kang, A. Kohli, and K.A. Thole, “Heat Transfer and Flow-field Measurements in the Leading Edge Region of
1996).
141
Eq. 6.1
From boundary layer theory ∂p/∂n, for both streamlines from the developing boundary layer on the
sidewalls should be equal. Therefore
∂p ∂p
| = |
∂n A ∂n B
Eq. 6.2
Due to the viscous turbulent boundary layer at the end-wall, it is evident that VsB < VsA. So for the
boundary layer assumption to hold, the radius of curvature of the streamline at B must be reduced.
This creates a crossflow in the boundary layer from the pressure surface towards the suction surface
of the blade, and thus generates secondary flow (flow that is not aligned with the stream-wise
direction) as depicted in Figure 6.17(top). Now consider Figure 6.17(Bottom), a constant
velocity profile with linear temperature profile. The same physics hold for this case; however, the
resulting vortex is reversed in direction. In this instance the temperature at A is greater than at B,
therefore ρA < ρB. Now RB must be greater than RA for the normal pressure gradient to balance and
the cross flow is generated towards the pressure side of the adjacent vane row. The change in
streamline curvature would be less severe in this case compared to Figure 6.17(top) since the
velocity term is squared in the relationship of Eq. 6.1.
As one can see from these simple, idealized flow situations, there can be large variation in the
expected secondary flow pattern that can be derived in a turbine vane passage. The important driver
for how the flow develops in a turbine vane passage is the total pressure profile entering the passage.
As this total pressure profile becomes the static pressure along the vane stagnation, the flow will be
driven from a high pressure region to a low
pressure region. In most turbine designs there is
a flow leakage slot between the combustor and
the turbine whereby cooler fluid is injected into
the main hot gas path. This leakage can also have
an effect on the secondary flow patterns that
develop. The bottom line when considering end-
wall flows is that the profile exiting the
combustor, which is often referred to as the
combustor pattern factor that ultimately enters
the turbine, should be known to fully predict the
secondary flows that will develop in the turbine
passage. In practice, the combustor pattern
factor is one of the parameters used in designing
cooling schemes for the airfoils and their
associated platforms.
6.10.2 End-Wall Heat Transfer
The heat transfer coefficients given in Figure
6.18 are represented in terms of a non-
dimensional Stanton number based on exit
velocity. In the region upstream of the vanes,
there is a high heat transfer region that occurs Figure 6.18 Contours of Non-Dimensional Heat
Transfer Coefficients (Reproduced with
between the stagnation point and the
Permission ASME)
reattachment of the flow on the suction side of
the airfoil. This is the area which experiences
very high acceleration. As the flow moves through the passage, it is apparent that the location of the
peak Stanton numbers (peak heat transfer) is being swept from the outer pressure surface towards
142
the suction side of the central vane153. Note that the heat transfer data shown here were taken with
a two-dimensional inlet boundary layer under low speed conditions, to allow for highly-resolved
data, with matched Reynolds number conditions. There is evidence in the literature that the
secondary flow patterns remain the same at both low and high Mach number conditions. Rather,
these secondary flows are a stronger function of the airfoil geometry and inlet profile conditions that
of Mach number. There is no data in the literature that discusses how the heat transfer on the end-
wall is altered depending upon an inlet flow condition that is relevant to that exiting a combustor. As
was stated previously, it is important to consider that the profile exiting the combustor can vary
greatly from that of a two-dimensional boundary layer assumption. For further info and cooling
effects, please see chapter 7.
6.10.3 Leading Edge Modifications
Because industry is concerned with problems at the vane leading edge-end-wall juncture, a number
of more recent studies have begun to evaluate geometric modifications to airfoils in this region. At
this point there have been three different geometric concepts tested for an asymmetric airfoil
geometry that include the following classifications: fences, fillets, and bulbs. Methods using flow
control such as suction combined with injection have also been reported but are generally not
feasible for gas turbines where gas temperatures exceed airfoil melting temperatures. [Chung and
Simon] first presented their concept for secondary flow control in 1993 that encompassed using a
fence placed in mid-passage between two turbine airfoils154. While their tests indicated a reduction
in strength of the passage vortex, industry’s concern was in cooling the fence and that it acted as a fin
conducting heat to the platform as it was exposed to the hotter main gas path fluid.
Three-dimensional end-wall contouring, which includes a more comprehensive geometric
modification than simply a modification to the end-wall-airfoil leading edge juncture, has also been
investigated computationally by Harvey, et al. and experimentally verified by [Hartland, et al.]155. To
design the end-wall contour, they used a linear
sensitivity matrix in conjunction with superposition
methods prior to applying an inverse design
algorithm. The results of the experimental
verification confirmed a predicted reduction in exit
flow angle deviations. Moreover, the experiments
indicated a 30% reduction in loss, which was higher
than predicted. In a later study, [Brennan, et al. and
Rose, et al.] applied similar computational and
experimental (respectively) methodologies as
[Harvey et al. and Hartland, et al.]. They applied
these methods to a high pressure turbine for a single
stage in both the vane and blade passages. They Figure 6.19 Fillet and Bulb Designs as Shown
reported stage efficiency improvements of 0.59%, by (Becz et al.)
which exceeded their predicted improvement of
0.4%. Using end-wall contouring and leading edge modifications show promise in reducing
secondary flows; however, there are numerous effects that need to be considered. Because this
153 M. Kang, A. Kohli, and K.A. Thole, “Heat Transfer and Flow field Measurements in the Leading Edge Region of
a Stator Vane End-wall,” J of Turbomachinery, 1999.
154 J. T. Chung and T. W. Simon, “Effectiveness of the Gas Turbine End-wall Fences in Secondary Flow Control at
Part II – Experimental Validation,” J of Turbomachinery 122 (2000), ;Neil W. Harvey, Martin G. Rose, Mark D.
Taylor, Jonathan Shahrokh and David G. Groegory-Smith, “Non-axisymmetric Turbine End Wall Design: Part I –
Three-Dimensional Linear Design System,” J of Turbomachinery, 2000.
143
modification must be practically feasible, required manufacturing, space limitations, and cooling are
all practical issues that must be addressed. Moreover, some of these designs may be sensitive to the
inlet flow conditions which need to be considered156. See Figure 6.19.
6.10.4 Blade Tip Heat Transfer
The performance of a turbine engine is a strong function of the maximum gas temperature at the
rotor inlet. Turbine blade designers concentrate on finding adequate cooling schemes for high
pressure turbine blades, particularly the tip region where heat transfer is quite high. The clearance
between a blade tip and its associated outer casing, also known as the blade outer air seal, provides
a flow path across the tip that leads to aerodynamic losses and high heat transfer rates along the
blade tip. The flow within this clearance gap is driven by a pressure differential between the pressure
and suction side of the blade, and is also affected by the viscous forces as the fluid comes into contact
with the walls of the gap. A complete description of blade tips and the associated problems is given
by Bunker and is well described by Figure 6.20. Gap size, rotational effects, blade geometries and
Reynolds numbers were all highly influence the heat transfer coefficients.
One method for improving the thermal
environment along the blade tip is to inject
coolant into the tip region. In a review
paper on tip heat transfer, Bunker states
that for blade tips there have been very
few film-cooling studies reported in the
literature even though film-cooling is
widely used18. The discussion given in this
section is relevant to tip film-cooling (see
chapter 7) since that is what is typically
used in industry. Many tip heat transfer
and film-cooling studies have been
completed without rotational effects. In
general, there is evidence in the literature
that supports a widely variable effect of
rotation, which warrants further studies.
One of the first pioneers in this region is
[D. Metzger] where by [Kim et al.] presents
a summary of his work157. In addition to Figure 6.20 CFD Prediction of Streamlines Across a
concluding that there is only a weak effect Blade Tip (Reproduced With Permission From ASME).
of the relative motion between a
simulated blade and shroud on tip heat transfer coefficient, they stated that there is a strong
dependency of cooling effectiveness for a tip on the shape of the hole and injection locations. Four
hole configurations were discussed by [Kim et al.] that included the following: discrete slots located
along the blade tip, round holes located along the blade tip, angled slots positioned along the pressure
side and round holes located within the cavity of a squealer tip. The studies reported by [Kim et al.]
were performed in a channel that simulated a tip gap, whereby no blade with its associated flow field
was simulated. In comparing the discrete slots to the holes, their data indicated a substantial increase
in cooling effectiveness using the discrete slots for all blowing ratios tested. Injection from the
pressure side holes provided cooling levels of similar magnitude to the holes placed on the tip with
156Karen Thole, “Mechanical Engineering Department Penn State University”, Park, PA 16802-1412.
157Y. W. Kim and D. E. Metzger, “Heat Transfer and Effectiveness on Film Cooled Turbine Blade Tip Models,” J of
Turbomachinery 117 (1995);Y. W. Kim, J. P. Downs, F. O. Soechting, W. Abdel-Messeh, G. Steuber, and S.
Tanrikut, “A Summary of the Cooled Turbine Blade Tip Heat Transfer and Film Effectiveness Investigations
Performed by Dr. D. E. Metzger,” J of Turbomachinery 117-1995.
144
better spreading occurring in the case of the pressure side injection. Kim et al. also reported that an
increase in coolant mass flow generally yielded improved cooling with tip surface holes, but for
pressure side holes, increased coolant flow yielded decreased cooling effectiveness158.
6.10.5 Case Study 1 - Effects of Grid Refinement and Turbulence in 3D Flow Structure and End-
Wall Heat Transfer in Transonic Turbine Blade Cascade
Results of numerical simulation of 3D turbulent flow and end-wall heat transfer in a transonic turbine
cascade are presented by [Levchenya & Smirnov]159. Employing several turbulence models (k-ω
model by Wilcox, Menter SST model, v2-f model by Durbin), an analysis CFD predictability was done in
comparison with measurements in a linear cascade at the NASA Glenn Research Center transonic
turbine blade cascade facility. It has been concluded in particular that rather fine computational grids
are needed to get grid-
independent data on the end-
wall local heat transfer
controlled by complex 3D
structure of secondary flows.
With CFD codes of second-
order accuracy, one should
use grids comprised of about
or more than 2 M cells (for
each full blade passage) to get
a definite conclusion on
preference of one or another
turbulence model for
predictions of phenomena
under consideration. Figure 6.21 Blade Passage and Slice of The Computational Domain
158 Karen Thole, “Mechanical Engineering Department Penn State University”, Park, PA 16802-1412.
159 Alexander M. Levchenya and Evgueni M. Smirnov, “CFD-Analysis of 3d Flow Structure and End-wall Heat
Transfer in a Transonic Turbine Blade Cascade: Effects of Grid Refinement”, West-East High Speed Flow Field
Conference 19-22, November 2007.
160 Giel, P.W., and Gaugler, R.E., “NASA Blade 1. End-wall heat transfer data. Version 1,” NASA-Glenn Research
specified on the end-wall, starting from the position of 0.3xCx (Cx referred to axial chord of blade)
upstream of the blade leading edge. Remaining walls were treated as adiabatic. Periodic boundary
conditions were used in the pitch wise direction. For computational purposes, only half of the real
span was considered, with the symmetry boundary condition at mid-span.
6.10.5.2 Computational Aspects
The 3D in house, incompressible/compressible Navier-Stokes code is based on the second order
finite-volume spatial discretization using the cell-centered variable arrangement and body-fitted
block-structured grids. For transonic flow analysis, a high-order version of the Jameson’s H-CUSP
scheme161 is implemented. For the present analysis, a set of 3D non-uniform grids have been
generated assuming the flow symmetry with respect to the passage middle plane. All the grids are of
3-block H-O-H structure (see Figure 6.21). Each mesh covers one half of the blade channel height,
and was obtained by translation of a 2D grid along the span wise direction. Grids of the best
resolution have 45 nodes along this direction, clustered to the end-wall. The distance from the first
cell center to the end-wall was equal to 0.2×10-4 x Сx that produced the area averaged yp + value of 0.8.
As a result of special computations, it has been established that a further grid refinement with respect
to the span wise direction is not necessary.
Below the main attention is paid to the effects of grid refinement in the planes parallel to the end-
wall (in fact, a starting 2D grid refinement), especially in the leading edge region where 3D vortex
structures arise. In order to characterize the grid quality in the LE region, we have introduced an
averaged cell size, Δ*, evaluated for the cells that
are placed in the middle between the saddle Mesh Cell # Δ*/Cx
(separation) point and the blade LE, except the A 360000 0.027
boundary layer region adjacent to the blade. Note
that this cell size is measured in the line of the LE, B 730000 0.022
and the cell aspect ratio in this region did not
exceed 2.0 for all the grids. Table 6.1 covers data C 750000 0.022
for five of the grids, results for which are presented D 760000 0.017
below. For the finest grid, the averaged cell size
introduced is of 1% of the blade axial chord (or E 1200000 0.010
about 5% of the blade LE radius). Note also that the
grid slice given in Figure 6.22 corresponds to grid Table 6.1 Parameters of the Grids Used
B (every second grid line of the grid is omitted
there for clarity).
To ensure a direct comparison of computational heat transfer results with the measurement data,
local Stanton numbers were calculated using the same procedure as developed and justified by [Giel
et. al]162. Remarkably that this procedure results in Stanton numbers that are practically
independent of the surface heat flux rate varied in the experiments. Under the operation conditions
under consideration, the choice of temperature difference used as the driving potential and the
choice of a reference temperature for gas thermos physical properties significantly affects the heat
transfer coefficient. Having performed a comparison of various definitions, [Giel et. al]163 suggested
to define the Stanton number as follows
161 Jameson, A., “Positive schemes and shock modelling for compressible flows”, Int. J. Num. Meth. Fluids, Vol. 20,
743-776 (1995).
162 Giel, P.W., Thurman, D.R., Van Fossen, G.J., Hippensteele, S.A, and Boyle, R.J., “End-wall heat transfer
qw
St =
ρref Uin Cp (Tw − Taw )
Eq. 6.3
where the local adiabatic wall temperature, Taw, is defined as
Taw 1−r
=r+ 2
Tin 1 + 0.5(Υ − 1)Mis
Eq. 6.4
In expression Eq. 6.4 the local isentropic Mach number, Mis, is determined from the wall static
pressure, and the recovery factor, r, is evaluated as r = Pr1/3. The density, ρref, included in Eq. 6.3,
unlike the value used for the Reynolds number is not the actual physical density. It is defined as
pin
ρref = , Tref = Tis + 0.5(Tw − Tis ) + 0.22(Taw − Tis )
RTref
Eq. 6.5
The isentropic temperature, Tis, is evaluated using the local isentropic Mach number and the free-
stream total temperature.
6.10.5.3 Results and Discussion
Flow field computed is illustrated in Figure 6.22. Here, as an example, results obtained with grid B
and the SST version of the Menter model (M-SST) are given. It should be emphasized, however, that
for the Mach number field and the static pressure distribution over the blade all other combinations
of grids A to E and the turbulence models (introduced above) produced practically identical results.
Figure 6.22 shows mid-span Mach
number contours. The distributions
computed are in an excellent agreement
with the NASA GRC experimental data. The
strong affection of the span wise location on
the static pressure distribution over the
suction surface is well reproduced in the
CFD-analysis. Near the trailing edge the
computations predicts an increase in
pressure on the suction side that is due to
the flow overexpansion and viscous-
inviscid interaction phenomena in the
trailing edge region. In contrast to the blade
surface pressure distribution,
computational results for local end-wall
heat transfer are very sensitive both to the
turbulence model and the grid quality.
Figure 6.23 presents Stanton number
distributions over the end-wall computed Figure 6.22 Computed Mid-Span Mach Number
Distribution
with three turbulence models in
comparison with the measurement data.
These distributions were obtained using grid B. Generally one can conclude that all the models
capture the main trends in formation of the local heat transfer pattern under action of 3D vortex
structure developing in the blade passage. However, there are pronounced distinctions in the blade
LE region where spots of enhanced heat transfer are observed. In particular, the k - ω model predicts
147
Figure 6.23 End-Wall Stanton Number (103) Distributions Computed with Grid B in Comparison
with the Measurement Data: (1) k-ω Turbulence Model, (2) M-SST, (3) v2-f, (4) Experiment
a zone of the highest Stanton numbers that adjoins directly the leading edge, but the latter is in
contradiction with the measurements. Both the M-SST model and the v2-f model predict crescent
zones of extreme heat transfer placed slightly upstream of the blade LE. Such a crescent zone is seen
in the experimental Stanton number distribution as well. At that, the v2-f model gives a Stanton
number distribution that is much more non-uniform as compared with the measurement data. All
the models did not predict a spot of enhanced heat transfer observed in the experiments at the blade
suction side not far from the leading edge. As a whole, the MSST model has prediction superiority
among the turbulence models examined.
6.10.5.4 End-Wall Heat Transfer Sensitivity w.r.t Grid and Turbulence Models
Now we concentrate at analysis of grid-sensitivity of local end-wall heat transfer. This analysis is
performed for the k-ω and the M-SST turbulence models taking into account that currently they are
rather popular in predictions of wall-bounded flows. Our computations have shown that the k-ω
model prediction results are considerably less sensitive to grid refinement as compared with the
MSST model. With the k-ω model, grids B to E produced practically identical results. Small
distinctions in the end-wall Stanton number distributions are observed only when comparing results
148
obtained using the initial grid (grid A) with those of one of the finer grids as we discuss the reasons
of relatively week grid-sensitivity of the k-ω model results for the problem under consideration.
Figure 6.24 Effect of Grid Refinement on the End-Wall Stanton Number (x103) Prediction with the
M-SST Turbulence Model: (1) Grid С, (2) Grid D, (3) Grid E, (4) Experiment.
Figure 6.24 illustrates the effect of grid refinement on the end-wall Stanton number distributions
computed with the M-SST turbulence model (in order to sharpen the effect, in this figure the upper
limit of the color legend is decreased as compared with Figure 6.23). The simulation results are very
sensitive to the grid quality in the region placed upstream of the blade LE, where horseshoe vortex
structures arise. Grid refinement results in formation of two crescent zones of extreme heat transfer
(see the St maps for grid D and E), whereas only one such a zone was observed when using coarser
grids. Remarkably that even for grids C and D, comprised of about same numbers of cells, there is a
considerable difference between the results for zones of high Stanton numbers. Grid E produces the
most detailed pattern of the Stanton number distribution. It should be recognized however that the
grid refinement has not resulted in a considerably better agreement between the computational and
measurement data. In particular, as in the coarser grid case, the finest grid computations do not
149
predict the high-St spot observed in the experiments at the blade suction side not far from the leading
edge. As well, the St values are underestimated at the pressure side near the blade trailing edge.
Let’s discuss now the reasons of considerable distinctions between the end-wall heat transfer
prediction results obtained for the blade LE region with the k-ω turbulence model and the M-SST
model. A detailed analysis of the flow structure in the end-wall boundary layer just upstream of the
blade LE has shown that the k-ω model produces a much simpler flow topology than the M-SST
model, provided fine grids are used in both the cases. For the M-SST model case, Figure 6.25 (A)
presents a pattern of limiting streamlines on the end-wall computed with grid E. One can see trails of
a number of well-resolved vortex structure determining the end-wall heat transfer peculiarities in
the blade cascade under consideration. The section A-A marked in the blade LE region was used for
visualization (see Figure 6.25 (B)) of the near-end-wall flow topology in the normal plane. With a
fine grid, the MSST model predicts a complicated vortex structure: with the main horseshoe vortex,
a counter rotating secondary vortex located closer to the end-wall and a tertiary vortex. Application
of a similar visualization technique to the flow field computed with the k-ω model has shown that a
structure with one horseshoe vortex is predicted, and intensity of this vortex is reduced as compared
with the main vortex predicted by the M-SST model.
Previously, such a kind of distinctions was reported by [Levchenya et al]164 when analyzing numerical
simulation results for the 3D turbulent flow and end-wall heat transfer in a cascade of thick vanes. In
that contribution it was reported also that, at least in the region of the horseshoe vortex formation,
the k-ω model produces a higher level of the eddy viscosity than the M-SST model, and it is a main
reason of distinctions in the flow topology upstream of the blade leading edge.
6.10.5.5 Summary
With an finite-volume Navier-Stokes code of second-order accuracy, effects of computational grid
refinement have been investigated. The problem of 3D turbulent flow and end-wall heat transfer in
164Levchenya, А.М., Ris, V.V., and Smirnov, E.M., “Testing of turbulence models as applied to calculations of 3D
flow and end-wall heat transfer in cascades of thick vane blades”, Proc. 4th Russian National Heat Transfer Conf.,
MPEI Publishers, Moscow, Russia, Vol.2, 167-170 (in Russian, 2006)
150
a linear transonic turbine cascade with a large turning angle. Three turbulence models were used at
the computations (k-ω model by Wilcox, Menter SST model, v2-f model). The main attention for the
grid-sensitivity aspects was paid to the cases of the k-ω and the M-SST since currently they are rather
popular in predictions of wall bounded flows. It has been established that the M-SST model prediction
results are considerably more sensitive to grid refinement as compared with the k-ω model,
especially for the flow and heat transfer region placed upstream of the blade leading edge, where
horseshoe vortex structures arise. A less grid-sensitivity of the k-ω model is due to the fact that
generally it produces a higher level of the eddy viscosity, and it results in prediction of a simplified
flow topology as compared with the M-SST model165. Other similar studies were performed by
[Ivanov et al.]166 and [V.D. Goriatchev, et al.]167.
6.10.6 Case Study 2 - Comparison of Steady and Unsteady RANS Heat Transfer Simulations of Hub
and End all of a Turbine Blade Passage
The necessity of performing an unsteady simulation for the purpose of predicting the heat transfer
on the end wall surfaces of a turbine passage is addressed by [El-Gabry and Ameri]168. This is
measured by the difference between the two solutions obtained from a steady simulation and the
time average of an unsteady simulation. The heat transfer coefficient (Nusselt number) based on the
adiabatic wall temperature is used as the basis of the comparison. As there is no film cooling in the
proposed case, a computed heat transfer coefficient should be a better measure of such difference
than, say, a wall heat flux. Results show that the effect of unsteadiness due to wake passage on the
pressures and recovery temperatures on both hub and casing is negligible. Heat transfer on the end
walls, however, is affected by the unsteady wake; the time-averaged results yield higher heat
transfer; in some regions, up to 15% higher. The results for the end wall heat transfer were compared
with results in open literature and were found to be comparable.
6.10.6.1 Introduction
Gas turbine heat transfer remains an important topic of concern as turbine inlet temperatures
continue to rise and combustor exit profiles continue to flatten with the goal of maximizing power
output and efficiency. This in turn means higher gas-side heat transfer to the hot gas path
components, including the vane and blade and, in particular, the end walls, which are now
experiencing temperatures nearly as high as the peak temperatures near the midspan.
Of importance to understanding the end wall heat transfer is describing and characterizing end wall
flows. As early as 1976, [Langston et al.]169 wrote that “the literature is certainly not lacking in
experimental studies of end wall flows in turbine cascades” but is lacking a “complete analytical
solution” of the end wall flows. In this 1976 paper, [Langston et al. ] cited the potential of numerical
models to fill in that gap, which, one can argue, has yet to be entirely filled, which, that is, where the
present research and other papers on computational fluid dynamics (CFD) modeling of end wall flow
165 Alexander M. Levchenya* and Evgueni M. Smirnov, “CFD-Analysis of 3d Flow Structure and End-wall Heat
Transfer in a Transonic Turbine Blade Cascade: Effects of Grid Refinement”, West-East High Speed Flow Field
Conference 19-22, November 2007.
166 Nikolay Ivanov, Vladimir Ris, Evgueni Smirnov, and Denis Telnov, “Numerical Simulation Of End-wall Heat
Transfer In A Transonic Turbine Cascade”, Conference On Modelling Fluid Flow (CMFF’03) The 12th
International Conference On Fluid Flow Technologies Budapest, Hungary, September 3-6, 2003.
167 V.D. Goriatchev1, N.G. Ivanov2, E.M. Smirnov2, V.V. Ris2, “CFD Analysis of Secondary Flows and Pressure Losses
in a NASA Transonic Turbine Cascade”, 1- Department of Mathematics, Tver State Technical University, 170026
Russia; and 2- Department of Aerodynamics, St.-Petersburg State Polytechnic University, 195251, Russia.
168 Lamyaa A. El-Gabry, Ali A. Ameri, “Comparison of Steady and Unsteady RANS Heat Transfer Simulations of
Hub and End wall of a Turbine Blade Passage”, Journal of Turbomachinery, JULY 2011.
169 Langston, L. S., Nice, L. M., and Hooper, R. M., 1976, “Three Dimensional Flow Within a Turbine Cascade
and heat transfer continue to play an important role in our understanding of this critical region of
the turbine passage.
[Langston et al. ] used ink to visualize flow near the end walls of a cascade and measured pressures
and velocities at axial locations within the passage. [Gregory-Smith et al.]170 presented the flow
visualization of the end wall and made measurements in a cascade with the goal of calculating the
vorticity in the passage.
There have also been several review papers that survey literature on secondary flows in turbine
passages and that highlight results of turbine end wall aerodynamics and heat transfer studies. The
dominant secondary flow structures highlighted include the cross-passage flow that crosses from the
pressure side to the suction side and what is commonly referred to as the horseshoe
vortex that forms at the stagnation point on the leading edge, where the flow separates into a
pressure side and a suction side leg. There are additional forms of secondary flow, including corner
vortices. There are three “corners”: the first is between the pressure side surface and the hub, the
second is between the suction side surface and the hub, and the third is between the leading edge
surface and the hub.
Near the blade tip, the dominant secondary flow feature is the tip leakage vortex that manifests itself
along the suction side of the airfoil. The losses due to tip leakage flow can account for up to 1/3 of the
total stage losses, as suggested by [Boyle et al.]171, making them important from an aerodynamics
perspective as well as their impact on heat transfer. Research in the topic of near-tip flows and heat
transfer started with basic research using very basic geometry to represent flat and grooved
rectangular tip models. These studies, among other things, established that the effect of relative
motion between the blade and casing on heat transfer is negligible, which was consequently followed
by a series of linear cascade tests using three-blade, four-blade, and five-blade cascades to measure
static pressure and local heat transfer distribution on the airfoil, tip, and casing surfaces for a variety
of tip geometries.
Experimental data obtained in stationary cascades offer detailed measurements; however, results
may not scale to the actual engine conditions or be truly representative of a true rotating blade in an
engine. Therefore, experimental data have also been obtained at engine conditions in rotating rigs.
[Haldeman and Dunn]172 at the Ohio State University Gas Turbine Laboratory measured the heat
transfer for the vane and blade of a rotating high pressure turbine stage operating at design corrected
conditions using a large shock-tunnel facility.
For the blade, Stanton number is reported at 20% and 96% spans at the blade tip and on the shroud.
[Polanka et al.]173 also made pressure and heat flux measurements on the tip and shroud under
rotating conditions at the U.S. Air Force Turbine Research Facility, which is a full scale rotating rig.
These test setups are far more representative of engine conditions, however, the test data are sparse
and a limited number of discrete measurements are available as compared with the cascade tests.
In addition to the experimental research on end wall aerodynamics and heat transfer, there have
been a number of computational studies on the subject, several of which complement the
experimental work. The 9H tip heat transfer experiments of [Bunker et al.]174 were modeled using in-
170 Gregory-Smith, D. G., Graves, C. P., and Walsh, J. A., 1988, “Growth of Secondary Losses and Vorticity in an Axial
and Loading—Part A: Parameter Effects Including Reynolds Number, Pressure Ratio, and Gas-to-Metal
Temperature Ratio,” ASME J. Turb., 125, pp. 97.
174 Bunker, R. S., Bailey, J. C., and Ameri, A., 2000, “Heat Transfer and Flow on the First-Stage Blade Tip of a Power
Generation Gas Turbine: Part 1—Experimental Results,” ASME J. Turb., 122, pp. 263–271.
152
house Reynolds-averaged Navier–Stokes _RANS_ codes and the OSU tests were modeled using the
commercial CFD code STARCD _17_ and an in-house GE-developed code Tacoma . [Polanka et al.]175
presented a comparison between numerical predictions and experimental results of [Polanka et al.].
The conclusions were that the agreement between the test and the CFD is “very good” on the airfoil
and “moderately good” on the end walls and tip; the tip heat transfer was overpredicted by the CFD
and although the shroud pressure distribution was accurately predicted with the 3D RANS solver,
the heat transfer there was overpredicted by a factor of 2.
Wake effects and unsteadiness were investigated by [Pullman]176, who used test measurements to
show the vortex formation at the stator exit and their development within the rotor passage to the
rotor exit. Using the experimental measurements at the stator exit as boundary conditions to a rotor
CFD model, the predictions of the aerodynamic losses _using steady and unsteady analyses are
calculated and the flow field at the rotor exit is predicted and compared with test data. The steady
simulation predicted 10% less loss in the rotor row than the unsteady simulation.
The present work focuses on the effect of unsteadiness due to wake passage on the end wall heat
transfer as well as pressures and
temperatures as predicted using an
unsteady 3D RANS solver. The vehicle
for this computational study is the E3
high pressure turbine blade, which has
been reported extensively in literature.
Time-resolved test measurements are
not available for the E3 HPT blade;
therefore, qualitative comparisons of the
CFD results will be made with HPT data
in open literature, to serve as a check on
the CFD results. The primary question
this research seeks to answer is whether
an unsteady analysis gives the same
answer as a steady analysis. For this size
model ∿2x106 nodes, it takes 500 clock
hours to reach a converged steady state
solution. Also, it takes the same model
25,000 clock hours to converge the
unsteady simulation. The question is: Is
the added time and complexity of Figure 6.26 Grid on the Solid Surfaces of the Geometry
running unsteady worth it?
6.10.6.2 Computational Method
A 3D RANS code developed at the NASA Glenn Research Center has been used to predict pressures,
temperatures, and heat transfer in a single blade passage of the GE-E3 gas turbine. The code was
Glenn-HT and was described in detail by [Ameri et al.]177. It uses a finite volume discretization
scheme and is second order accurate in time and space. The turbulence model used in the calculations
was the low Reynolds number k-ω model by [Wilcox], which is integrated to the wall; a
175 Polanka, M. D., Clark, J. P., White, A. L., Meininger, M., and Praisner, T. J., 2003, “Turbine Tip and Shroud Heat
Transfer and Loading Part B: Comparison Between Prediction and Experiment Including Unsteady Effects,” ASME
Paper No. GT2003-38916.
176 Pullman, G., 2004, “Secondary Flows and Loss Caused by Blade Row Interaction in a Turbine Stage,” ASME
nondimensional grid spacing y+ of near 1 is maintained on all wall surfaces, including the blade and
end walls. The code was run on 48 processors of a Xeon Linux Cluster using message passing interface
for parallel processing.
The E3 gas turbine has 46 vanes and 76 blades and the blade rotates at 8400 rpm. The blade tip
clearance is 2% of the blade span. The present research focuses on the effect of the upstream vane
wake on the blade. The domain of this computation is restricted to a single blade passage; this
simplification of using a vane/blade ratio of 1 is based on a separate preliminary study, which showed
that for purposes of computing average heat transfer, the wake frequency for a 1:1 ratio produces a
very similar result as compared with a 2:3 ratio. The 2:3 ratio is an approximation to the actual
vane/blade count, which was 46:76.
The inlet of the domain is located at 15% axial chord upstream of the blade leading edge _midway
between the vane trailing edge and the blade leading edge_ and the exit of the domain is at 50% axial
chord downstream of the blade trailing edge. To create the multiblock structured grid needed for the
solver GRIDPRO™, commercially available software for generating structured meshes was used.
shows the surface mesh on the solid airfoil and hub. Figure 6.27a shows the grid on the casing
surface highlighting the multiblock structure of the grid blocks. Figure 6.27b shows the grid
structure on the hub surface. The grid consists of 164 blocks and a total of 1.8 x 106 nodes. There are
65 nodes across the tip clearance gap in the radial direction and 101 nodes from the hub to tip of the
blade. The grid independence of the solution was established.
Figure 6.27 (a) Casing surface mesh showing multiblock structure and (b) hub surface mesh showing
multiblock structure
A dimensionless time step of 0.005 was used based on earlier investigations in which the time step
was varied from 0.001 to 0.01. For the selected time step of 0.005, 320 steps were required to
complete one period; (i.e., the passing of a wake across a single blade passage), which is sufficiently
fine, a resolution for purposes of this study. Based on the selected time step, blade count, and RPM,
there were 320 time steps over the period of one wake passage.
6.10.6.2.1 Boundary Conditions
A separate computation of the vane flow was used to establish the inlet boundary conditions used in
the present blade analysis. The total pressure and temperature at the vane exit were calculated using
the same code and methodology and the wake profile is taken from the midspan of the vane
154
computation and used for the entire span of the inlet section to the blade. The total pressure in the
vane wake was approximated with a trigonometric function
Qw
h=
Taw − Tw
Eq. 6.9
where Qw is the wall heat flux, Tw is the wall
temperature, and Taw is the adiabatic wall
temperature. Therefore, in order to calculate heat
transfer coefficient defined as such, it is necessary to
run two separate analyses; one in which the walls
have zero heat flux in order to determine the
adiabatic wall temperatures and a second with a
prescribed wall temperature in order to calculate the
heat flux. For the runs with a prescribed wall
temperature, a constant temperature of 0.7
(normalized using inlet total temperature) is applied Figure 6.28 Total temperature (T0) and
to all walls. The steady state boundary conditions are total pressure (P0) at the blade inlet
based on the average of the unsteady computations.
6.10.6.3 Results and Discussion
Figure 6.30 shows the instantaneous pressure distribution on the hub surface at equal times over
the period of one wake passage. The image outlined with the dashed line shows the time-averaged
pressure distribution. These time-averaged results are compared with a steady simulation in which
there is no wake at the inlet. The inlet boundary condition for the steady simulation is the area
average of the unsteady boundary condition in Figure 6.28.
155
Figure 6.29 Instantaneous and time-averaged (dashed) hub heat transfer distribution for a wake
passing
The instantaneous pressure distributions in Figure 6.30, particularly the position and extent of the
low pressure region on the suction side, illustrate the effects of the wake passage. This region of
minimum pressure is indicative of a vortex that is traveling along the suction side (perhaps the
suction side (SS) leg of the horseshoe vortex). As time passes from t=0 to t=T/4, this low pressure
region moves along the SS edge and extends in the blade to blade direction. As it approaches the
trailing edge (TE), it appears to move away from the SS edge toward the pressure side (PS) trailing
edge and begins to diffuse. At some time between T/4 and 3/8T, a second low pressure region
emerges near the leading edge before the high point of curvature (high-C) that also moves along the
SS edge extends and diffuses as before. The periodic growth, diffusion, and movement of this low
pressure region indicate an interaction between the end wall secondary flows and the upstream
wake.
Likewise, the instantaneous heat flux images in Figure 6.29 suggest an interaction between the
wake passage and the end wall heat transfer on the hub. In discussing the pressure, we focused on
the low pressure region along the suction side and noted the presence of two low pressure regions
at the same time during part of the period: one region is diffusing into the passage near the trailing
156
edge and one is beginning to grow as it moves along the SS edge near the leading edge. In Figure
6.29, let us focus our attention on the high heat transfer regions shown in red. At the leading edge,
there is high heat transfer due to stagnation that extends along the PS edge and grows. There is a low
heat transfer region between the passages that moves along the SS edge and grows then diffuses
across the passage to PS edge. In doing so, it mixes with the high heat transfer fluid and reduces the
heat transfer, resulting in what appears to be a break in the high heat flux zone along the PS edge.
The latter piece of the high heat transfer region along the PS edge shrinks in extent and decreases in
magnitude as it diffuses and appears to “shed” at the trailing edge.
The pressure distribution on the casing and hub surfaces is plotted in Figure 6.31 and Figure 6.32,
respectively, nondimensionalized by the inlet total pressure. Figure 6.32 is actually a repeat of the
time-averaged contour plot in Figure 6.29 outlined with a dashed line. The airfoil outline shown in
Figure 6.31 and Figure 6.34 and other casing images to follow is the projection of the tip surface
on the casing.
On the casing surface, there is a region of high pressure in the forward part of the passage above the
pressure side surface of the airfoil whose outline is shown in Figure 6.31. This high pressure
corresponds to the stagnation region of the airfoil near the tip. There is a low pressure region near
the trailing edge of the airfoil, which likely corresponds to higher tip flow speeds.
157
̅aw − Taw
T
̅aw − Tw
T
Eq. 6.10
This is a more physical
representation of the difference as Figure 6.36 Time-averaged hub adiabatic wall temperature
the denominator is indicative of the
driving temperature for heat
transfer. Figure 6.37 shows that the time-averaged casing surface adiabatic wall temperatures are
about 4-6% lower than the steady results. For the hub surface, Figure 6.37 shows that the time-
averaged hub surface adiabatic wall temperatures are up to 8% lower than those predicted by a
steady state simulation. The region of largest difference is in the forward region of the hub surface.
160
Figure 6.37 Difference in casing adiabatic wall temperature distribution between the time-averaged and
steady results
The dimensionless wall heat flux, as defined in the nomenclature, is plotted in Figure 6.39 and
Figure 6.40. Figure 6.39 shows the time averaged wall heat flux distribution on the casing surface.
The area of highest heat transfer is the region of the casing above the pressure side edge of the airfoil
tip due to the entry of tip leakage flow into the clearance gap, consistent with Figure 6.35.
The heat transfer rates at the hub surface shown in Figure 6.40 repeat of the dash-outlined image
in Figure 5.30 are lower than those found on
the casing surface Figure 6.39. The area of
highest heat transfer is on the hub surface
near the leading edge in the stagnation region
and penetrates at a lower magnitude along
the pressure side surface aft toward the
trailing edge. The heat transfer rate on the
end walls obtained through time averaging is
compared with the steady state-predicted
heat fluxes and the difference between the
time-averaged and steady results for the
casing and hub are shown in Figure 6.41 and
Figure 6.42, respectively.
Figure 6.38 Difference in hub adiabatic wall
temperature distribution between the time-averaged
and steady results
161
Lamyaa A. El-Gabry, Ali A. Ameri, “Comparison of Steady and Unsteady RANS Heat Transfer Simulations of
178
Hub and End wall of a Turbine Blade Passage”, Journal of Turbomachinery, JULY 2011.
162
Lamyaa A. El-Gabry, Ali A. Ameri, “Comparison of Steady and Unsteady RANS Heat Transfer Simulations of
179
Hub and End wall of a Turbine Blade Passage”, Journal of Turbomachinery, JULY 2011.
163
180 Epstein, A. H., Guenette, G. R., Norton, R. J. G., and Yuzhang, C., 1985, “Time Resolved Measurements of a
Turbine Rotor Stationary Tip Casing Pressure and Heat Transfer Field,” AIAA Paper No. 85-1220.
181 Ameri, A. A., Rigby, D. L., Steinthorsson, E., Heidmann, J., and Fabian, J. C., 2008, “Unsteady Analysis of Blade
and Tip Heat Transfer as Influenced by Upstream Momentum and Thermal Wakes,” ASME Paper No. GT2008-
51242.
182 See 165.
183 Tallman, J. A., Haldeman, C. W., Dunn, M. G., Tolpadi, A. K., and Bergholz, R. F., 2006, “Heat Transfer
Measurements and Predictions for a Modern High Pressure, Transonic Turbine, Including End walls,” ASME Paper
No. GT2006-
90927.
184 Lamyaa A. El-Gabry, Ali A. Ameri, “Comparison of Steady and Unsteady RANS Heat Transfer Simulations of
Hub and End wall of a Turbine Blade Passage”, Journal of Turbomachinery, JULY 2011.
185 R. F., 2006, “Heat Transfer Measurements and Predictions for a Modern High Pressure, Transonic Turbine,
[Figure 24 (Not shown here)]186 presents the time-averaged Nusselt number distribution for the hub
surface of the E3 blade; the data is identical to [Fig. 19 (same source)] presented earlier; however,
the range has been modified to better highlight features and draw comparisons to the Stanton
number distribution for the hub in [Fig. 23]. The contours in [Figs. 23 and 24]187 are similar: The
heat transfer peaks at the leading edge due to stagnation and decreases with a relatively high heat
transfer region along the PS edge. There is a much narrower low heat transfer region along the
suction side, similar to the blade analyzed.
6.10.6.5 Conclusions
The end wall pressures, temperatures, and heat transfer have been computed using an unsteady
RANS calculation, time averaged, and compared with steady state predictions to evaluate the effect
of unsteadiness due to wake passage.
Results obtained using the k-omega turbulence model show that the effect of unsteadiness on the
average end wall pressures on both hub and casing is negligible; the unsteadiness increases the
pressure prediction in some areas by up to 2%. The area, where the pressure is most affected by
unsteadiness, is on the suction side part of the passage for both hub and casing and the effect of
unsteadiness seems to extend over a slightly larger area on the hub than the casing.
The contribution of unsteadiness due to wake passage on the recovery temperature on the casing is
negligible over most of the surface. On the hub surface, the time-averaged results yield lower
recovery temperature predictions in the forward region of the passage.
Heat transfer on the end walls is affected by the unsteady wake more than the pressures and
temperatures. The time-averaged results yield higher heat transfer; in some regions, the increase is
up to 15%. On the casing surface, the increase in heat transfer is more localized and appears on the
suction side region of the passage, where the peak pressure difference between the time averaged
and steady results occurred. The effect on the hub surface is less localized and the increase in heat
transfer predictions can be seen all along the pressure side edge and penetrate well into the passage.
As there are no experimental data available to validate the CFD results for the E3, the results for the
end wall heat transfer were compared with results in open literature. For the casing, the results were
comparable to experimental measurements of [Epstein et al.] quantitatively in terms of the influence
of the above-rotor tip area to the overall shroud heat flux and qualitatively in terms of the pattern of
heat transfer on the surface. For the hub surface, the results were compared with results from
[Tallman et al.] and were also found to be similar, thereby verifying the unsteady Reynolds-averaged
Navier-Stokes (URANS) calculations of the present work.
7 Blade Cooling
As the turbine inlet temperature increases, the heat transferred to the turbine blade also increases.
The level and variation in the temperature within the blade material, which cause thermal stresses,
must be limited to achieve reasonable durability goals188. The operating temperatures are far above
the permissible metal temperatures. Therefore, there is a critical need to cool the blades for safe
operation. The blades are cooled with extracted air from the compressor of the engine. Since this
extraction incurs a penalty on the thermal efficiency and power output of the engine, it is important
to understand and optimize the
cooling technology for a given turbine
blade geometry under engine
operating conditions. Gas turbine
cooling technology is complex and
varies between engine manufacturers.
Figure 7.1 shows the common cooling
technology with three major internal
cooling zones in a turbine blade as film
cooling in the leading edge, pressure
and suction surfaces with Vanes, and
blade tip region. The leading edge is
cooled by jet impingement with film
cooling, the middle portion is cooled
by serpentine rib-roughened passages
with local film cooling, and the trailing
edge is cooled by pin fins with trailing
edge injection. Interested readers are
referred to several recent publications
that address state-of-the-art reviews
of turbine blade cooling and heat
transfer.
The first high pressure stage of a
modern gas turbine operates at very Figure 7.1 The Schematic of a Modern Gas Turbine Blade
high temperatures that require with Common Cooling Techniques (Courtesy of Je-Chin Han)
complex blade-cooling systems to
guarantee high performance and efficiency of the gas turbine while maintaining a very low level of
energy losses, though using compressed air for cooling189. An accurate and efficient Conjugate Heat
Transfer (CHT) solver is thus necessary to compute the flow and temperature fields of the air within
the cooling channels and of the gas around the blades by means of the Navier Stokes and energy
equations as well as the blade temperature field, by means of the heat conduction equation. Due to
the very high geometrical complexity of the cooling channels within the blades, generating a body
fitted mesh for the three domains air, gas and blade is extremely difficult and time consuming. And
Nevertheless, many turbine blade cooling simulations have been performed with success, though at
large computational cost190.
188 Je-Chin Han, “Recent Studies in Turbine Blade Cooling”, International Journal of Rotating Machinery, 2004.
189 D. De Marinisa, M. D. de Tullioa, M. Napolitanoa,* and G. Pascazioa, “A conjugate-heat-transfer immersed-
boundary method for turbine cooling”, Energy Procedia, 2015.
190 Luo G. and Razinsky E. H., “Conjugate heat transfer analysis of a cooled turbine vane using the V2F turbulence
[Mahmood and Acharya]191 considered adding fillets at the junction of end wall and blade/vane
leading edge in the cascades. The blade cascade operates at low speed atmospheric conditions while
the vane cascade operates at high speed conditions with the exit Mach number near to 1.0.
191Gazi I. Mahmood, Sumanta Acharya, “Blade and Vane Leading Edge Fillet on End wall Cooling in Linear
Turbine Cascades”, Proceedings of the 15th International Heat Transfer Conference, IHTC-15 August 10-15,
2014, Kyoto, Japan.
168
effectiveness since these ratios essentially gives the ratio of the coolant to hot mainstream thermal
capacitance.
In general, higher the pressure ratio, better the film cooling protection (i.e., reduced heat transfer to
the airfoil) at a given temperature ratio, while lower the temperature ratio, better the film cooling
protection at a given pressure ratio. However, too high of pressure ratio (i.e., blowing too much) may
reduce the film cooling protection, because of jet penetration into the mainstream (jet lift-off from
the surface). Therefore, it is important to optimize the amount of coolant for airfoil film cooling under
engine operating conditions (Reynolds number 106, Mach number0.9 at exit conditions). For
designing a better film cooling pattern of an airfoil, turbine cooling system designers also need to
know where heat is transferred from hot mainstream to the airfoil.
As mentioned earlier, these film-hole pattern (i.e., film hole location, distribution, angle and shape)
would affect film cooling performance. There are many film cooling papers available in the open
literature. This paper is limited to review a few selected papers to reflect recent development in
turbine blade film cooling. Specifically, this review focuses on the following topics: rotor blade film
cooling, nozzle guide vane film cooling, airfoil end wall film cooling, as well as airfoil leading edge and
blade tip region film cooling.
Several papers for rotator blade, stator
vane, and end wall film cooling, under
representative engine flow conditions,
are reviewed and discussed. Some
parametric studies for large-scale
cascade blade, cascade vane, cascade
end wall, simulated blade tips and
leading edge region film cooling under
low speed flow conditions are also
reviewed and discussed.
Figure 7.3 shows the schematic of
film cooling concept. Typically, the
heat load to the surface without film Figure 7.3 Schematic of Flm Cooling Concept
cooling is represented as heat flux q’=
h0(Tg- Tw), where h0 is the heat transfer coefficient on the surface with wall temperature Tw and
oncoming gas temperature (Tg). When coolant is injected on the surface, the driving temperature is
Tf, film temperature, which is a mixture of gas (Tg) and coolant temperature (T), q"= h(Tf- Tw), where
h is the heat transfer coefficient on the surface with film injection. Also, a new term film effectiveness
0/) is introduced, where
Tg − Tf
η=
Tg − Tc
Eq. 7.1
The η values vary between 0 and with as the best film cooling effectiveness. Therefore, the heat flux
ratio can be written as:
q" h Tf − Tw h Tg − Tc
=( )[ ] = ( ) × {1 − η [ ]}
q"0 h0 Tg − Tw h0 Tg − Tw
Eq. 7.2
To obtain any benefit from film cooling, the heat load ratio, q"/q’, should be below 1.0. The heat
transfer coefficient ratio (h/ho) is enhanced due to turbulent mixing of the jets with the mainstream
and is normally greater than 1.0. The temperature ratio (Tf - Tw)/(Tg - Tw), which is related to the film
effectiveness should be much lower than 1.0 such that the heat load ratio is lower than 1.0. The best
169
film cooling design is to reduce the heat load ratio (i.e., smaller h/ho enhancement with greater η) for
a minimum amount of coolant available for a film cooled airfoil 192.
7.1.1.1 End-Wall Film-Cooling
One method of combating the high heat transfer coefficients along the end-wall is through the use of
film-cooling holes whereby cooler air is injected through discrete holes in the end-wall. Film-cooling
hole placement, particularly in the end-wall region, has traditionally been based upon designer
experience whereby a number of design variables are considered. For example, one should take into
account leakage from a slot at the combustor turbine interface whereby cooler gases leak into the
main gas path (combustor film-cooling carryover). Most turbines are designed such that pressures
outside the main gas path are higher than those found
in the main gas path to insure that the hot
combustion gases are not ingested below the
platform. If designed properly, this leakage flow
could be relied upon as a source of coolant. It is also
important to remember that the secondary flows that
develop are affected by the leakage flows at the vane
inlet, as indicated by the previous discussion in this
section.
Other design variables include roughness effects,
film-cooling migration (as will be discussed), film-
cooling limitations resulting from internal cooling
schemes, and cooling hole manufacturing.
Manufacturing of cooling is generally done through
the use of electro-discharge machining (EDM) or
laser drilling with both being subjected to access
limitations. Generally, EDM is more expensive than
laser drilling and conversely laser drilling can tend to Figure 7.4 Measured Adiabatic Wall
be more limited in terms of hole geometries that can Temperatures for Coolant Exiting a
be manufactured. Consider a leakage slot flow Combustor/Vane Leakage Slot (reproduced
between the combustor and turbine whereby the with permission from ASME)
coolant is 0.75% of the core flow. The cooling to the
end-wall that can be provided is shown in Figure 7.4. Coolant exits across much of the width of the
slot albeit in a very non-uniform manner. This non-uniform slot flow arises from the static pressure
distribution and secondary flow development along the end-wall. Although there is a large uncooled
region, often referred to as a bow wake, around much of the vane at the leading edge and along the
pressure side of the airfoil, there is also a well-cooled region in the center of the passage. As the slot
flow increases, the cooling potential also increases to a point after which the benefit is small193.
As was previously mentioned, end-wall film-cooling has largely been based on designer experience.
One difficulty in designing the film-cooling hole pattern is knowing before-hand the local static
pressure along the end-wall, which varies greatly along the end-wall as the flow accelerates through
the passage. If one considers that a single supply feeds all the film-cooling holes and inviscid flow
through the holes, it can be shown that a global, ideal (loss-free) blowing ratio for all the film-cooling
holes is given by the following equation,
192 Je-ChinHan and Srinath Ekkad, “Recent Development in Turbine Blade Film Cooling”, International Journal of
Rotating Machinery, 2001, Vol. 7, No. 1, pp. 21-40.
193 Karen Thole, “Mechanical Engineering Department Penn State University”, Park, PA 16802-1412.
170
ρc Uc ρc Po,c - ps,in
Mglobal = √ .
ρin Uin ρin Po,in - ps,in
Eq. 7.3
where Po,in and ps,in are defined at the inlet to the turbine. If one then wants to compute the local, ideal
blowing ratio for each cooling
ρc Uc ρc Po,c - ps,∞
M= =√ .
ρin Uin ρin Po,in - ps,∞
Eq. 7.4
hole (now taking into consideration that the local velocity varies), the following equation can be used
where ps,∞ is the local static pressure defined at the exit of the cooling hole. Eq. 7.4 indicates that the
same blowing ratio will occur for each hole placed along a constant static pressure line. Designing an
end-wall cooling pattern to achieve a uniform blowing ratio is one methodology that some companies
have used. Figure 7.5 (a) shows a cooling hole pattern whereby the holes in the passage were placed
along a constant static pressure line. Figure 7.5 (b) shows an end-wall cooling hole pattern with
holes that were placed on lines parallel to the incoming flow direction. The end-wall in Figure 7.5
(b) also provides the space for including a mid-passage gap that occurs between vanes as the vanes
are mated on the turbine disk. Vanes are generally cast in either doublets or singlets and then placed
in a turbine disk with some type of seal between the vane platforms. Generally, relatively low levels
of coolant leak from this mid-passage gap (less than 0.3%). Figure 7.5 (c) shows the same film-
cooling hole pattern as that in Figure 7.5 (b) with the exception being the mid-passage gap is
present in Figure 7.5 (c). The coolant flow conditions for all three vanes shown in Figure 7.5 are
the same with 0.75% of the core flow exiting in the form of coolant from the upstream slot, 0.5%
Figure 7.5 Contours of Adiabatic Effectiveness for Two Film-Cooling Hole Patterns (left and center)
With a Mid-Passage Gutter for the Cooling Hole Pattern in the Center (Right) (Reproduced With
Permission From the Publisher of ASME)
171
coolant exiting the film-cooling holes and no net coolant flow through the mid-passage gap for Figure
7.5(c). The contours plotted in Figure 7.5 represent the non-dimensional adiabatic wall
temperatures. As discussed in the previous section, adiabatic surface temperatures represent the
local fluid temperature. From all three contour plots in Figure 7.5, it is clear that the upstream slot
can definitely be an important part of cooling mid portion of the end-wall. In comparing Figure 7.5
(c) to the rest, however, it is also seen that the mid-passage gap limits the area of the cooling present.
The mid-passage gap provides a convective coolant trough through which coolant from the upstream
slot enters and passes through the gap along the mid-passage before exiting the gap near the trailing
edge of the airfoil.
The streamlines superimposed upon the experimental measurements of the adiabatic wall
temperatures presented were computed using CFD. These streamlines were extracted from the CFD
simulations at a location very near to the end-wall (less than 2% of the span). The streamlines, for
the most part, can be used to predict the trajectory of the coolant flow exiting the cooling holes. There
are some regions, however, where the streamlines differ from the jet trajectories, such as at the
entrance to the passage closer to the pressure side. In comparing all of the cooling hole patterns in
Figure 7.5, it is clear that more uniform cooling can be achieved by placing cooling holes along
constant static pressure lines as in Figure 7.5 (a). All three patterns illustrate the difficulty of cooling
the end-wall along the pressure side of the vane and along the leading edge-end-wall juncture. In
general these are very difficult areas to cool because of the secondary flows that were described
previously. The horseshoe vortex in the stagnation region can lead to coolant being swept upstream
of the holes at low blowing ratios, whereas at high blowing ratios the coolant separates from the end-
wall and impacts the vane surface rather than the end-wall surface. Despite the cooling holes injecting
coolant towards pressure side of the vane, the sweeping motion of the passage vortex prevents
cooling at the juncture between the pressure side of the vane and the end-wall194.
As an animation example on the right, courtesy of siemens blog,
consider the cooling flow which is meant to blanket and protect the
blade surface. The turbulence between the primary flow and the
blanket of cooling flow streams has a massive impact on the GTFilmCooling.mp4
effectiveness in shielding the metal parts from high heat loads.
194 Karen Thole, “Mechanical Engineering Department Penn State University”, Park, PA 16802-1412.
172
al.]195. The locations of vanes and their characteristics are most important. More recently there was
an inquiry by [ Jasim et al.]196 where it was mainly focused on gas turbine blade heat transfer analysis
and effect of increasing the number of external film cooling holes rows.
7.2.1 Vane NASA C3X
The first case is the NASA C3X Vane
Experiment. The experimental
object was represented as a linear
array. Each cascade employed three
vanes, characteristic of a first-stage
turbine (see Figure 7.6). Geometry
parameters of vane and cooling
channels were taken from [Hylton et
al. (1983)]. Each of the vanes was
cooled by an array of ten radial
cooling holes. The vanes were
fabricated of ASTM type 310
stainless steel, which has a relatively
low thermal conductivity197. The Figure 7.7 Surface Temperature Distribution on the Suction
average heat transfer coefficient for Side (Left) and The Pressure Side (Right) of the Vane
each cooling hole was calculated
from empirical relation of Nusselt
Number with Reynolds and Prandtl Numbers as:
D
NuD = F(0.022. Pr 0.5 . Re0.8
D ) = α.
λ
Eq. 7.5
Where D is the hole diameter, λ is thermal conductivity. The Re for each cooling channel was
determined from the measured flow rate, the cooling hole diameter, and viscosity based on the
average coolant temperature. F is a function of Pr, ReD, and x/D, which corrects the Nu expression for
a fully developed thermal boundary layer to account for thermal entrance region effects. Each of the
vanes was cooled by an array of ten radial cooling holes.
Another test cases with more vanes is the Convective Heat Transfer in The Steam Turbine Vane as
depicted in Figure 7.7. The investigations on the thermal efficiency potential of steam cooling, have
been carried out experimental work [Bohn et al (2002)]. Geometrical configuration consists of a
rectangular duct containing a three-vane cascade. The central vane can be convectively cooled by
supplying steam to 22 straight radial cooling passages198.
195 A.V. Rubekina, A.V. Ivanov, G.E. Dumnov, A.A. Sobachkin (Mentor Graphics Corp., Russia); K.V. Otryahina
(PAO NPO Saturn, Russia), “Keeping it Cool in Gas Turbines”, Mentor Graphics white paper.
196 Hadeel Raheem Jasim, Narsimhulu Sanke and Khaled Al-Farhany, “Heat Transfer Simulation of Gas Turbine
Blade with Film Cooling”, International Journal of Modern Engineering and Research Technology, 2018.
197 A.V. Rubekina, A.V. Ivanov, G.E. Dumnov, A.A. Sobachkin, “ Using Modern CAD-Embedded CFD Code For
Numerical Simulation Of Heat Transfer In Vanes And Blades”, Mentor Graphics White Paper.
198 A.V. Rubekina, A.V. Ivanov, G.E. Dumnov, A.A. Sobachkin, “ Using Modern CAD-Embedded CFD Code For
Numerical Simulation Of Heat Transfer In Vanes And Blades”, Mentor Graphics White Paper.
173
Figure 7.8 Flow Streamlines Colored Cooling Air Temperature into Passages
199 Explanation of Figure 7.6 - (top left) and section view of stagnation row of holes (top right), as well as Gas
temperature distribution (bottom left), cooling air streamlines with the Mach number distribution (bottom
right)
200 A.Hasanpour, M. Farhadi and H.R. Ashorynejad, “Hole Configuration Effect on Turbine Blade Cooling”, World
simulated. The average temperature magnitude in mid-plan section of each configuration is obtained
and finally the lower temperature value is selected such as best arrangement.
where the subscripts s and f refer to solid and fluid respectively, k indicates the thermal conductivity,
and nw is the unit vector normal to the wall. In order to impose the boundary conditions above, at the
solid interface cells the effect of the boundary conditions imposed by the flow upon the temperature
and heat-flux at the body surface has to be accounted for, exactly as the same conditions are to be
enforced at the fluid-interface cells. A direct coupled solution of the URANS and heat conduction
equations is unfeasible. Thus, an iterative procedure is to be employed. A first-order-accurate
approximation is obtained as:
Eq. 7.7
where m is the current iteration and βf is the inverse of the distance between the fluid interface cell
and the wall Eq. 7.7 provides the heat fluxes (apart from the thermal conductivity) at all fluid cells.
Such values are then used to provide those pertaining to all Solid Cell Projection Point (SCPP) by
means of a distance-weighted interpolation and then, the Neumann condition at all SCPPs201.
7.3.1 Case Study - Heat Transfer in Separated Flows on the P. S. of Turbine Blades
Heat transfer in separated flows on the pressure side of a typical high lift turbine profile is
numerically investigated by [De La Calzada, et, al.] . The numerical code was first validated on
attached flows in turbine blades. To obtain flow separation cases, the profile is subject to large
negative incidences so that a separation bubble is obtained at the pressure side. The numerical
results are compared to available experimental data for code validation. The aim of the present
investigation is to perform a detailed numerical study of the heat transfer phenomena in separated
flows at flow conditions representative of LPT airfoils. A comparison with experimental data is
performed, hence allowing the validation of the code and the confirmation of the main flow features.
The relationship between the dynamic and thermal boundary layers and their importance of the
velocity component perpendicular to the wall in creating injection of flow towards the wall or
ejection of flow from the wall and their effect in the heat transfer is analyzed.
7.3.1.1 Literature Survey
Much attention has been paid to the investigation of large flow separation in simple cases, including
both velocity related measurements and heat transfer measurements. These include experimental
investigations of backward-facing steps as [Vogel and Eaton]202 or [Sparrow et, al.]203, where the
Backward-Facing Step “, Heat and Mass Transfer, vol. 107, pp. 922-929, 1985.
203 E. M. Sparrow, S. S. Kang, and W. Chuck, Relation Between the Points of Flow Reattachment and Maximum
Heat Transfer for Regions of Flow Separation, Int. J. Heat Mass Transfer, vol. 30, no. 7, pp. 1237-1246, 1987.
175
relationship between the separation region and the heat transfer features was studied.
Corresponding numerical investigations have been performed on similar configurations like the one
by [Kaminejad et al.]204 where only laminar conditions and very low Reynolds numbers are
considered. The effect of turbulence was taken into account for example by [Rhee and Sung]205, where
good agreement with experimental data was also found for very low Reynolds numbers. More
recently, Rhee and Sung also investigated the effect of local forcing on the separation and reattaching
flow. However, very little attention has been paid to the heat transfer in large separated flow regions
in turbine representative conditions. [Bassi et al.]206 present CFD results on the separated flow region
of a HPT airfoil with cutter trailing edge with no cooling ejection, but only a short discussion about
the separated flow physics is included. Regarding experimental investigations, [Rivir et al.]207 have
measured the flat plate heat transfer in a region of turbulent separation, and [Bellows and Mayle]208
have measured the heat transfer on a blunt body leading edge separation bubble both for cases of
high Reynolds number209. More recently, [Calzada and Alonso]210 performed a numerical
investigation of large flow separation region at the pressure side of a turbine profile but not
comparison with experimental results was included. [Lutum and Cottier] presented a similar
investigation, but results indicated that simulations were not able to reproduce experimental heat
transfer results at the pressure side separation region especially for low turbulence levels. From
experimental investigations, it is known qualitatively that separated flow regions are characterized
by large and rapid variations of the heat transfer (e.g., Rhee and Sung). Furthermore, the heat transfer
presents a local minimum and a local maximum in the vicinity of separation and reattachment points
respectively, with regions where the heat transfer coefficient (HTC) is much larger than that of
attached flows211. Taking into account that separated flow regions are usually characterized by high
turbulence levels and large scale unsteadiness, there is a tendency in the heat transfer community to
explain the heat transfer phenomena in separated flows in terms of the generation of turbulence
rather than in terms of the dynamic and thermal boundary layers relationship212.
7.3.1.2 CFD Modeling
A CFD solver used based on [Jameson et al]213 been used. Convective terms are discretized using a
cell vertex scheme, and the viscous terms are computed by means of the Hessian matrix. Integration
in time is performed using an explicit five stage Runge-Kutta scheme where the viscous and artificial
dissipation terms are evaluated in the first, third, and fifth stages. The code runs on unstructured
204 H. Kazeminejad, M. Ghamari, and M. A. Yaghoubi, “A Numerical Study of Convective Heat Transfer from a
Blunt Plate at Low Reynolds Number”, Int. J. Heat Mass Transfer, vol. 39, no. 1, pp. 125-133, 1996.
205 G. H. Rhee, and H. J. Sung, “A Low-Reynolds Number, Four Equation, Heat Transfer Model for Turbulent
Separated and Reattaching Flows”, Int. J. Heat Fluid Flow, vol. 18, pp. 38-44, 1997.
206 F. Bassi, S. Rebay, M. Savini, S. Colantuoni, and G. Santoriello, “A Navier-Stokes Solver with Different
Turbulence Models Applied to Film-Cooled Turbine Cascades”, Paper No. 41, AGARD-CP-527, 1993.
207 R. B. Rivir, J. P. Johnston, and J. K. Eaton, “Heat Transfer on a Flat Surface under a Region of Turbulent
Reattached Flows”, AIAA Journal, Vol. 26, no. 2, pp. 144-150, 1988.
210 P. De La Calzada and A. Alonso, “Numerical Investigation of Heat Transfer in Turbine Cascades with Separated
meshes which are built by a quasi-structured layer all along the walls, where viscous effects are
expected to be dominant and by a fully triangular unstructured mesh in the rest of the flow domain
obtained by Steiner triangulation214. For turbulence simulation, the two equations κ-ω model from
Wilcox215 is implemented. More details of the numerical code can be found in Corral and Contreras216.
Numerical results are post processed to obtain heat and mass transfer relevant parameters at the
walls. This is performed by computing velocity and temperature variation in the direction
perpendicular to the wall so that heat transfer and friction coefficient as well as Stanton number can
be computed, as defined below. The local Stanton number is the equivalent for the temperature to
the local skin-friction coefficient for the velocity. Although the local Stanton number variations do
not represent variations in heat flux alone but also take into account the local velocity value, it is the
most adequate parameter to describe the thermal boundary layer behavior and to develop special
correlations for heat transfer estimation (i.e., Reynolds-Colburn analogy).
T
− k
HTC x n w τw
Ch x = = , Cf x = Eq. 7.8
ρ e C p u e (T0 − Tw )ρ e C p u e 1
ρ e u e2
2
Note, that the total temperature is used in the definitions instead of the adiabatic wall temperature,
even though compressible effects and therefore viscous dissipation may be important since the
representative cases for LPT usually imply an exit Mach number of around 0.5, as we have in our
study. However, the difference between the aforementioned coefficients and the corresponding
compressible definitions can be kept sufficiently low if the wall temperature for the computations is
properly chosen. In our particular cases, the total temperature is defined as in the experiments and
the wall temperature is taken around 25 K higher than the fluid temperature, which keeps the
difference between compressible and incompressible heat and mass transfer coefficient values lower
than 2% even at regions with Mach numbers around 0.5. This wall temperature value also develops
a thermal boundary layer whose magnitude is large enough to avoid high sensitivity to any random
numerical errors in the resolution of the temperature field around the wall.
7.3.1.3 Description of the Blade Computational Grids and Results for Attached Flow
The T106-300 blade section has been used as a generic geometry representative of a typical highly
loaded LPT airfoil217 (see Figure 7.9). In this investigation, the blade profile is subject to extremely
large negative incidences in order to have a large separation bubble on the pressure side. Mach and
Reynolds numbers are varied around typical LPT values. The generated grid is hybrid in nature with
higher definition in regions adjacent to the wall, trailing edge and leading edge, as shown in Figure
7.9. Due to the expected flow separation at the pressure side when the profile is subject to high
negative incidence, on this investigation the viscous mesh is extended to a region larger than attached
flows would require for this Reynolds number. The objective of this large region with high definition
is to capture the shear layers and flow features on the pressure side large bubble. However, in order
214 R. Corral and J. Fernandez-Castañeda, “Surface Mesh Generation by Means of Steiner Triangulation”, Proc.
29th AIAA Fluid Dynamics Conference, vol. 39, pp. 176-180, Albuquerque, New Mexico, 1998.
215 D. C. Wilcox, “Reassessment of the Scale Determining Equation for Advanced Turbulence Models”, AIAA J.,
to-Blade Computations”, Proc. 45th ASME Gas Turbine and Aero engine Congress, Exposition and Users
Symposium, ASME Paper 2000-GT-515, Munich, 2000.
217 H. Hoheisel, “Test Case E/CA-6, Subsonic Turbine Cascade T106, Test Cases for Computation of Internal Flows
Figure 7.10 Blade Profile vs. Pressure Coefficient (Courtesy of De La Calzada et al.)
178
along the airfoil is presented in Figure 7.10. The separation region is characterized by low velocities
and a fairly constant pressure distribution. However, at the reattachment region the static pressure
increases reaching a local maximum (i.e., local minimum value of Cp in Figure 7.10), which indeed
indicates the reattachment point. The experimental results indicate, to some extent, a shorter
separation bubble which reattaches earlier, hence starting earlier also the acceleration towards the
trailing edge. This may be related to the already identified higher pressure at the pressure side
predicted by the CFD, which might indicate some slight difference in local incidence angle between
the experiments and the simulations 218.
Some more detail about the flow can be identified by comparing experimental and numerical heat
transfer results shown in Figure 7.10. Both experimental and numerical results show two local
minima and maxima between the extreme values achieved at leading and trailing edges (these
Figure 7.11 Flow Field at the Front and Middle Parts of the Separation Bubble (Courtesy of De La
Calzada et al.)
218P. De La Calzada, M. Valdes, and M. A. Burgos, “Heat Transfer in Separated Flows on the Pressure Side of
Turbine Blades”, Industria de Turbopropulsores S. A., Spain, 2007.
179
extreme values are not shown in the graph). The first minimum occurring at around 0.03 x/L
corresponds to the expected reduction in heat transfer rate. In Figure 7.11 (A-D) the Mach number,
total Pressure, Temperature, Velocity vectors fields are plotted where the large bubble at the
pressure side can be clearly identified. Helped by the streamlines traces, the multiple bubble
configuration can be also identified. In this particular case, two bubbles appear. As confirmed in
Figure 7.11(D) by the velocity vectors, one small bubble is stretched towards the pressure wall,
developing at the center of the full separation region whose vortex is rotating counterclockwise, and
one large bubble, rotating clockwise is extending up to the external shear layer along the major part
of the pressure side, having its vortex core at the rear part of the separation region while extending
its vortex influence also to the front part. Detail of the temperature field and flow velocities in the
regions of flow separation and reattachment are plotted in Figure 7.11 (C-D). At points 1 and 3, the
flow is separating and a large component of the velocity perpendicular and directed away from the
wall exists. This flow configuration takes heat from the side walls and ejects it, creating an ejection
stagnation or fountain-like region where the wall thermal field is penetrating the main flow helped
by the perpendicular component of the velocity, hence increasing the effective thermal boundary
layer and decreasing the heat transfer rate.
This phenomena is particularly clear in front of point 3, where the increase in the thermal boundary
layer thickness can be easily identified by the extension of the high temperature region close to the
wall in Figure 7.11. Point 3 corresponds to the separation of an internal second bubble, which must
also exist in the experiment configuration since the local minimum can be also identified in
measurements in Figure 7.11. Points 2 and 4 correspond to reattachment points where there is an
important component of the velocity perpendicular and towards the wall, hence taking fresh fluid to
the wall and creating an injection stagnation region where the thermal boundary layer is reduced
and heat transfer is increased. To further investigate the thermal boundary layer developing through
the separation region the temperature profiles developing along straight lines perpendicular to the
wall, marked a-d in Figure 7.11 (B). Note, that the y coordinate is non-dimensionalized with the
thermal boundary layer thickness, which is basically coincident with the thickness of the separated
region. Dotted lines represent the temperature distribution of the corresponding adiabatic wall case
where static temperature only varies as a result of the velocity profile (being the stagnation
temperature fundamentally constant), while solid lines show the temperature distribution within the
re-circulation region at different distances from the wall in the case with heat transfer and heated
wall219.
The lowest wall temperature gradient is obtained at the leading edge separation point 1 (line a),
where even with the thermal boundary layer being relatively thin the fluid temperature shows a low
gradient specially close to the wall driven by the ejection of heated flow from the wall through the
ejection stagnation region configuration. On the contrary, the highest heat flux is achieved at the main
bubble reattachment point 4 (line d) where, additionally to the thin thermal boundary layer
thickness, the fluid temperature variation is mainly concentrated at the wall in a region about 10%
of that thickness, hence increasing the temperature gradient at the wall. This reduction of the
effective boundary layer thickness is driven by the injection stagnation region configuration, where
the velocity component perpendicular to the wall is forcing the thermal boundary layer to be
squeezed towards the wall. As an additional proof showing that there is low coupling between the
dynamic and thermal boundary layers and their gradients in separated regions, the relationship
between the velocity parallel to the wall and temperature gradients at the wall is investigated. It is
widely accepted that the Reynolds-Colburn analogy is only reliable in attached flows only for modest,
near-zero, pressure gradients, and with a constant wall temperature. The computed local skin-
219P. De La Calzada, M. Valdes, and M. A. Burgos, “Heat Transfer in Separated Flows on the Pressure Side of
Turbine Blades”, Industria de Turbopropulsores S. A.,.Spain, 2007.
180
Figure 7.12 Heat Transfer Coefficient for Different Negative Incidences (Courtesy of Calzada et al.)
friction coefficient (absolute value), the Stanton number, and the Reynolds-Colburn analogy are
shown in Figure 7.12 to further demonstrate that the analogy between dynamic and thermal
boundary layer is not valid for separated flows even when no pressure gradient exists. Only at the
rear acceleration region where attached flow is ensured, the Reynolds analogy tends to follow the
correlation showing a conventional relationship between dynamic and thermal boundary layers and
their gradients at the wall. Furthermore, unlike the Stanton number, the skin friction approaches
zero not only at the separation and reattachment regions but also along the major part of the
separated flow region, hence confirming that the Reynolds-Colburn analogy is not applicable. This is
clear proof that there is a very weak coupling between velocity parallel to the wall and thermal
boundary layers in separated flows. On the contrary, it is the convective transport of fluid in a
direction normal to the wall and the fluid conduction effects in low velocity regions what drive the
heat transfer phenomenon, hence supporting once again the prime role of the stagnation region
configurations on the heat transfer mechanism.
7.3.1.5 Inlet Flow Angle Effects
Inlet flow angles of 90°, 100°, 110° (i.e., -37.7, -27.7°, and -10.7° incidence angle, all with separated
flows at pressure side), and 127.7° (i.e., 0° incidence angle, with pressure side attached flow) have
been simulated for the nominal isentropic exit Re = 1.5 x 105 and an isentropic exit Mach number of
0.5. The results in terms of HTC and Stanton number are presented in Figure 7.12 and Figure
7.13220. As expected, the size of the bubble is decreasing with the reduction of the negative incidence
angle as can be concluded from the location of the maximum values of Stanton numbers in Figure
7.13. All the I separated flow cases show relatively large bubbles varying the reattachment points
from 0.5 x/L for -17.7° incidence up to 0.6 x/L for -37.7° incidence. Unlike the HTC whose local
maximum value at reattachment point is maintained almost constant along the attached acceleration
region up to the trailing edge region, the Stanton number clearly generates a more pronounced local
maximum value at the reattachment point driven by the combination of maximum heat flux and static
pressure (i.e., minimum external velocity).
P. De La Calzada, M. Valdes, and M. A. Burgos, “Heat Transfer in Separated Flows on the Pressure Side of
220 220
It is noticeable that in all separated flow cases the same multiple bubble configuration is obtained as
indicated by the presence, within the separated region, of one additional local minimum and one
additional local maximum in heat transfer parameters although it is less evident in the slight negative
incidence case (i.e., -17.7° incidence). Generally, it can be concluded that the separated region always
generates a redistribution in the heat flux by decreasing the value at the front separation region and
by increasing it at the rear reattachment region. The higher or lower surface averaged effective value
will depend on the particular geometry and conditions.
Figure 7.13 Stanton Number for Different Negative Incidences (Courtesy of De La Calzada et al.)
F. M. White, “Viscous Fluid Flow”, McGraw-Hill, pp. 162 and 248, 1991.
221
H. W. Kim and D. R. Jeng, “Convective Heat Transfer in Laminar Boundary Layer Near the Separation Point”,
222
ASME Proc. of the 1988 National Heat Transfer Conference, vol. 3, pp. 471-476, 1988.
182
Figure 7.15 Stanton Number vs. Reynolds Numbers (Courtesy of De La Calzada et al.)
Reynolds number, as shown in Figure 7.14. By applying the definition relationship between HTC
and Stanton number Ch, it can be seen that the ratio HTC/Ch must retain a dependence on Reynolds
to the power of 1. Therefore, the Stanton number dependence on the Reynolds number should vary
with and exponent of -0.7 in these cases according to the exponent 0.3 found for the HTC, which is
indeed confirmed by the maximum values at the reattachment point shown in Figure 7.14.
Figure 7.14 Heat Transfer Coefficient vs. Reynolds Number (Courtesy of Calzada et al.)
183
One interesting feature is that, for all Reynolds numbers investigated, the size of the bubble and the
internal structure (i.e., multiple bubble configuration) is the same, as can be concluded from the
location of local maxima and minima in the figures. Although it could be expected that increasing the
Reynolds number would reduce the size of the separation bubble, in this particular case the
reattachment is driven by the acceleration of the flow and the role played by the Reynolds and the
corresponding boundary layer instability and potential transition is expected to be very minor. This
is a completely different behavior compared to cases in which there is no flow acceleration, and the
reattachment is driven by boundary layer transition.
In these latter cases, the increase of turbulence produces an early transition and reattachment and
an increase in heat flux due to the stronger reattachment vortex on a blunt flat plate subject to
pulsating conditions. Although the implemented numerical turbulence model was able to produce
high turbulence and the corresponding boundary layer transition on the suction surface to avoid the
back surface separation, at the pressure side the turbulence generation is concentrated on the
external shear layer and from there it is convected downstream to the trailing edge and the
downstream wake. Therefore, it is thought that in cases like this the turbulence is not a strong enough
mechanism to force sufficient flow entrainment and perturbation to the shear layer to produce an
early reattachment of the boundary layer, and it is then expected that the size of the separation
bubble will depend weakly on the Reynolds number and turbulence. See Figure 7.15.
7.3.1.6 Concluding Remarks
A better understanding of the flow physics and the heat transfer mechanisms in large separated flow
regions have been achieved by means of a numerical investigation on the T106-300 typical LPT airfoil
subject to large negative incidence. Flow separation is characterized by a pronounced reduction in
HTC at the separation region, close to the leading edge where the minimum value is achieved, and by
an increase at the reattachment region where the maximum value is achieved. Those are extreme
values, much lower and higher than the ones obtained for attached flows. It is concluded that the
velocity component perpendicular to the wall is the main contributor to the generation of ejection
and impingement stagnation configurations, where the flow is taken from or towards the wall, hence
affecting the thermal field in those regions and contributing to create a lower or higher temperature
gradient at the wall and the corresponding HTC values.
By analyzing the Reynolds-Colburn analogy all along the pressure side of the profile, the low coupling
between the velocity component parallel to the wall and the thermal field and their gradients within
the separation region is confirmed. Additionally, it is also shown that an important variation in HTC
values can occur within the separation region due to the presence of secondary separation bubbles
which can create additional separation and reattachment points. This is confirmed by both numerical
and experimental results for the high negative incidence (i.e.,-37.7° incidence), which show the
presence of one additional local maximum and one local minimum in HTC values that must indicate
the presence of additional corresponding reattachment and separation points, hence indicating the
presence of and the additional secondary separation bubble. Moreover, it is also shown by the
numerical results that the multiple bubble configuration is found for all separated cases investigated
here (i.e., negative incidence varying from -17.7° to -37.7°). Finally, the variation of the heat transfer
with the Reynolds number is investigated. The numerical results show no variation of the separation
bubble size with Reynolds number varying from 150,000 to 400,000. A dependence of the HTC on
the Reynolds number to the power of 0.3 is obtained in the separation region, in particular at the
maximum value occurring at the main bubble reattachment point on the rear part of the separation
bubble223.
223P. De La Calzada, M. Valdes, and M. A. Burgos, “Heat Transfer in Separated Flows on the Pressure Side of
Turbine Blades”, Industria de Turbopropulsores S. A., Madrid, Spain.
184
185
∂ui ∂uj 2
−ρu′i u′j = μT ( + ) − ρκδij
∂xj ∂xi 3
Eq. 8.1
coupled with algebraic expressions or, at most, differential equations for the turbulent velocity and
length scales to which the turbulent viscosity is related. This framework is accepted as being
adequate for thin shear flows and is able to reproduce transition in simple boundary layers, if
combined with appropriately constructed and calibrated transport equations for the variation of the
scales in low-Reynolds-number conditions. However, it fails to resolve turbulence anisotropy and to
represent correctly the effects of normal straining and curvature on the turbulent stresses.
As an example of 1st order turbulence models, [Simões, et. al.]224 compare different turbulence
models available in commercial codes such as κ-ε, κ-ω and Mentors SST. It was concluded that the
SST model was the most accurate results among the three models evaluated, when comparing them
to the experimental data of the compressor performance as depicted in Figure 8.1. The principal
aerodynamic characteristics of most turbomachine flows are governed mainly by a balance between
pressure gradient and convection, while turbulence here tends to affect mainly secondary flow
features and the losses225. This is at least so in low-load conditions in which the boundary layers are
relatively thin and attached. In high-load and off design conditions, however, turbulence can
contribute substantially to the aerodynamic balance and is thus a process of major practical interest.
In such circumstances, the boundary layers grow rapidly, separation can ensue on both suction and
224 Marcelo R. Simões, Bruno G. Montojos, Newton R. Moura, and Jian Su,” Validation of Turbulence Models For
Simulation of Axial Flow Compressor”, Proceedings of COBEM 2009.
225 W.L. Chen, F.S. Lien, M.A. Leschziner, “Computational prediction of flow around highly loaded compressor
cascade blades with non-linear eddy-viscosity models”, International Journal of Heat and Fluid Flow 19, 1998.
186
pressure sides (depending on the blade geometry and the incidence angle) and stream wise vorticity
is intense. The sensitivity of major mean flow features to turbulence is especially high when the flow
enters the blade passage at an angle which departs materially from the design value, thus causing
leading-edge separation and high flow displacement, followed by transition in the separated shear
layer. More generally, transition tends to be a highly influential process in the majority of off-design
flows in that details of the location and evolution of transition can dictate the sensitive response of
the boundary layers to pressure gradients.
This framework is accepted as being adequate for thin shear flows and is able to reproduce transition
in simple boundary layers, if combined with appropriately constructed and calibrated transport
equations for the variation of the scales in low-Reynolds-number conditions. However, it fails to
resolve turbulence anisotropy and to represent correctly the effects of normal straining and
curvature on the turbulent stresses. The last two deficiencies are especially important in blade flows;
first, because the state of turbulence at the leading-edge impingement region is crucially important
to the transitional behavior further downstream, and second, because the blade curvature causes
significant damping or augmentation of turbulence transport in the boundary layers on the suction
and pressure sides, respectively. It is now generally accepted that the substantial variability in the
strength of the interaction between different strain types and the turbulent stresses can only be
resolved, in a fundamentally rigorous sense, through the use of second-moment closure, in which
separate transport equations are solved for all Reynolds-stress components. In particular, the very
different stress-generation terms contained in these equations give rise to that closure's ability to
resolve anisotropy and hence the influence of curvature, rotation and normal straining on the
stresses. However, this type of closure is complex, poses particular challenges in respect of its stable
integration into general computational schemes and is costly to apply in practice [Lien and
Leschziner]226. A simpler and more economical alternative, though one which rests on a weaker
fundamental foundation, is to use nonlinear stress/strain relations which can be made to return,
upon the introduction of physical constraints and careful calibration, some of the predictive
capabilities of 2nd moment closure.
226Lien, F.S., Leschziner, M.A. “Second-moment closure for three dimensional turbulent flow around and within
complex geometries”. Computers and Fluids 25, 237, 1996.
187
8.1 Case Study 1 - Turbulence Comparisons for a Low Pressure 1.5 Stage Test
Turbine
In a gas turbine engine, secondary flows have a unfavorable effect on efficiency. This phenomena
was investigated by [Dunn et al.]227. The current numerical study is aimed at determining which
turbulence model in a commercially available CFD code is best suited to predicting the secondary
flows. Experimental validation is used to determine the appropriateness of the model. It was found
that the Baldwin-Lomax, Spalart - Allmaras and κ-ε predicted the magnitude of the velocity well, but
did not capture the velocity magnitude profile well. The κ-ω and the SST κ-ω captured the profile
better, but did not predict the average value as well as the other models tested. As evident with
reference to the data, Baldwin-Lomax is still one of the best all-purpose models for turbomachinery.
The SST k-ω shows some promise as it predicted the velocity magnitude reasonably well, but the
Baldwin-Lomax predicted the radial and tangential velocities better. Due to the complex nature of
secondary flow it may still be sometime before computational hardware and the numerical models
are such that the complexities can be appropriately modelled. Until such time it is important that new
models be investigated and validated against experimental data. It is also recommended that the
computational expense be carefully weighed against the required accuracy.
227Dwain Dunn, Glen Snedden, and T.W. von Backström, “Turbulence Model Comparisons for a Low Pressure 1.5
Stage Test Turbine”, ISABE-2009-1258.
188
[2], Birch [3], Chien-Kim [4], Launder-Spalding [5], Myong-Kasagi [6], etc.
Depending on the number of transport-equation used, the turbulent models can be classified into
zero-, one- and two-equation and higher-order models. Theoretically speaking, the more the number
of transport equations involved, the more
accurate the prediction is, as less assumptions
are used. The aim of this paper is to evaluate the
different types of turbulent models including
zero-, one- and two-equation models using a
Q3D Navier-Stokes [7] and full energy equation
unless otherwise stated, in one host code, with
experimental data for axial turbomachinery
application.
8.2.2 Turbulent Models
8.2.2.1 Baldwin-Lomax’s Zero-Equation
Model [1]
In zero-equation model, the concept of mixing
length is used. Dimensional analysis of variables
shows that the turbulent viscosity, μT, divided
by the density p has the same dimensions as a
length multiplied by a velocity. Hence
momentum arguments can be used to show that
μT is a function of the flow density, a length scale
in the flow and the local mean flow velocity. The
relationship is omitted here, while it is given in
[NG, E.Y.K. and TAN, S.T., “Comparison of Various
Turbulence Models in Rotating Machinery Blade-
to-Blade Passages”, 1999], and [Sadrehaghighi,
I., “Turbulence Modeling”, 2021] in full details.
8.2.2.2 Birch’s One-Equation Model [3]
In one-equation model, the turbulent viscosity
is related to the turbulent kinetic energy k.
Again, readers are encouraged to consult the
[NG and TAN, “Comparison of Various
Turbulence Models in Rotating Machinery Blade-
to-Blade Passages”, 1999] for additional
information.
8.2.2.3 Standard Two-Equation Model [5]
A standard k-ε model developed by Launder and
Spalding [5] is also coded. This is the most
commonly used model for CFD calculations. We
will avoid the formulation since it is available in
[NG and TAN, “Comparison of Various
Turbulence Models in Rotating Machinery Blade- Figure 8.2 (a) Grid generated for transonic
to-Blade Passages”, 1999],as well as, in compressor rotor (mesh: 86x 45); (b) Measured
[Sadrehaghighi, I., “Turbulence Modeling”, Mach contour at 45% span [9].
2021]. For I.Cs. and B.Cs. , users are encouraged
to consult the source as well.
189
Figure 8.3 Mach contour at rotor mid-span using (Left) Zero Equation (Middle) One Equation (Right)
Two Equation Turbulence Models
The Mach contours computed in Figure 8.3 show that, on the suction side, strong acceleration just
after the leading edge followed by a weak oblique shock. Different models produce slightly different
results. The shock predicted by zero equation model smears out into a wider number of grid.
However, both the one- and two-equation models predict a sharper shock which are closer to the
experimental result. All the models predict the shock at about 18-20% chord of the suction surface.
For the same axial chord, the one-
equation model produces the
highest Mach number, the zero
equation model gives the lowest
value while the two-equation model
predicts an in between value that is
closest to the experimental results.
In brief, the predicted result by each
model agrees qualitatively well
against the experimental data.
8.2.3.2 UTRS Turbine Blades
The next test cases are the UTRC
turbine blades. Experimental
results [10] have been widely
published. The blade surface
pressure-coefficient distributions
(Cp = (Pi - P)/O.5pi Ui2) based on the Figure 8.4 Blade pressure-coefficient prediction for UTRC
inlet condition for UTRC turbine turbine stator
stator at the nominal operating
190
point are compared in Figure 8.4. It shows similarity between the predictions using each turbulence
model. On the suction surface at same X/Cx chord position, the zero-equation model predicts a lower
value as compared to prediction by one- and two-equation models. However, on the pressure
surface, all the models predicted similar Cp values and are very close to the experimental result. As
the boundary layer at pressure surface is relatively thin and no significant separation occurs
therefore all the turbulent models are
able to predict more accurate results.
Included also is the Cp prediction by
laminar flow having similar shape to
turbulent assumption but with a much
lower Cp value at suction side.
Similarity, Figure 8.5 compares the Cp
distributions for turbine rotor. On both
the blade surfaces at same X/Cx position,
zero-equation model predicts a lower
value as compared to that by one- and
two equation models as well as the
experimental result. The latter models
are able to predict results which are very
close to experimental result on pressure
surface and second half of the suction Figure 8.5 Blade pressure-coefficient prediction for
surface. At the first 50% chord on the UTRC turbine rotor
suction surface, the experimental result
is lower than the prediction by one or two-equation models. Same as in case of stator the Cp
prediction by laminar flow has similar shape with that predicted by turbulent models but with a
much lower Cp value at suction side. In the measurement, the flow has passed through the stator,
which results in wake forming, flow distortion and non-uniformity at the rotor inlet in contrast to the
computation. Large error at entrance of rotor was also found in the calculation by Lee et al. [11].
8.2.3.3 C4 Compressor
The final case is on C4 compressor
blade where experimental data are
available from [12]. Figure 8.6
compares blade surface pressure
coefficient distributions at the
nominal operating point, showing
similarity between the predictions
using each turbulence model. On
suction surface, at same position, it
is shown that zero- and one
equation models predict similar
values which are slightly lower than
two-equation model as well as the
experiment values. However, on
pressure surface, zero and one
equation models predictions agree
well with measured data. Figure
8.6 also shows that the two- Figure 8.6 Blade pressure-coefficient prediction for C4
equation model agrees very well compressor blade
with the measurement at both
191
blade surfaces. The prediction of Cp for laminar flow is similar to that predicted by turbulent models
at most portion of the blade surfaces, except the region near trailing edge of the pressure side where
laminar model predicts a small
boundary separation bubble
and hence lower pressure
recovery as compared to
turbulent models.
Local velocity profiles along Y-
axis are plotted at 64% chords
and compare with experimental
result [12] as shown Figure
8.7. The Y/Ye term is the
nondimensionalized normal
distance from the blade surface
Y with the boundary layer
thickness Ye (defined as the grid
point with a speed less than
98% different compared to
adjacent grid). The speed U is
non-dimensionalized with Figure 8.7 Non-dimensionalized Y-direction distance vs. velocity
speed Ue which is the speed at curve at 64% chord of suction side of C4 compressor blade
Ye. At both chord-location, the
velocity profiles predicted by
two-equation model agree very well with the experiment. At above 20% of boundary layer thickness,
zero and one equation models predictions show good agreement with the experimental result.
However, below 20% of boundary layer thickness, zero equation model predicts lower value while
the one-equation model gives a higher value. In brief, all the models are able to predict the growth of
boundary layer and boundary separation. The laminar flow prediction is found to be quite similar to
the zero-equation model. More details could be found in [13].
8.2.4 Concluding Remarks
The computed results were compared with measurement to validate the code and assess the quality
of the numerical solution. The performance of the turbulence models to predict the flow through a
blade passage depends on the number of transport equations used and on the inlet flow conditions.
Another observation from the models used is their different separation behavior within the boundary
layers. It is shown that in most cases the two-equation model produces results which are closest to
the experimental results followed by one-equation model. As the less simplification is made the
closer to physics it will be. Finally, for accurate simulations of fluid machinery, extension to three-
dimension with transition and higher-order of turbulent model are needed.
8.2.5 References
[1] B. Baldwin and H. Lomax (1978). ’Thin layer approximation and algebraic model for separated
turbulent flows’, AIAA Paper, no. 78-257.
[2] T. Cebeci and A.M.O. Smith (1974). ’Analysis of turbulent boundary layers’, Applied Mathematics
and Mechanics, Vol. 15, Academic Press, New York.
[3] N.T. Birch (1987). ’Navier-Stokes predictions of transition loss and heat transfer in a turbine
cascade’, ASME 87-GT-22.
[4] Y.S. Chien and S.W. Kim (1987). ’Computation of turbulent flows using an extended k-c turbulence
model’, NASACR- 179204.
[5] B.E. Launder and D.B. Spalding (1974). ’The numerical computation of turbulent flow’, Comp.
Math. Appl. Mech. Eng., 3, 269-289.
192
[6] H.K. Myong and N. Kasagi (1990). ’A new approach to the improvement of the k-c turbulence
model for bounded shear flows’, JSME Int. J., Ser. B, 33, 63-72.
[7] E.Y.K. Ng and Y. Miao (1996). ’Viscous flow prediction of the single stage axial rotating
machineries’, CFD Journal, 403-420.
[8] H.K. Versteeg and W. Malalasekera (1995). An Introduction to Computational Fluid Dynamics,
Longman Scientific & Technical, Loughborough.
[9] L. Fottner (1990). ’Test cases for computation of internal flows in aero-engine components’, NASA
AGARD Advisory Report No: 275 (VI.4 Test Case E/CO-4 by Dunker, R.) pp. 245-285.
[10] R.P. Dring, H.D. Joslyn, L.W. Hardin and J.H. Wagner (1982). ’Turbine rotor and stator
interaction’, ASME Journal of Engineering for Power, 104, 729-742.
[11] Y.T. Lee, T.W. Bein, J. Feng and C.L. Merkle (1993). ’Unsteady rotor dynamics in cascade’, ASME
Journal of Turbomachinery, 115, 85-93.
[12] E.Y.K. Ng (1992). ’A high resolution coupled parabolic/elliptic Navier-Stokes solver for
turbomachinery flows’, Ph.D. Thesis, University of Cambridge, UK.
[13] S.T. Tan (1998). Numerical investigation of viscous flow in rotating cavity with different
turbulence models, M. Eng. Thesis, Nanyang Technological University, Singapore.
[14] E.Y.K. Ng, S.T. Tan and H.N. Lim (1997). ’Simulation of instability flow in compressor system’,
Proceedings of 8th Aerospace Technology Seminar, Singapore, pp. PR 4/1-4/7.
193
228 Xiaodong Wang, “CFD Simulation of Complex Flows in Turbomachinery and Robust Optimization of Blade
Design”, Submitted to the Department of Mechanical Engineering Doctor of Philosophy at the Vrije Universiteit
Brussel July 2010.
229 Arathi K. Gopinath, Edwin van der Weidey, Juan J. Alonsoz, Antony Jamesonx, Stanford University, Stanford,
CA 94305-4035, Kivanc Ekici {and Kenneth C. Hallk, Duke University, Durham, NC 27708-0300, “Three-
Dimensional Unsteady Multi-stage Turbomachinery Simulations using the Harmonic Balance Technique”
194
proper simulation requires solution of all blades in each row. However, some vendors such as
ANSYS© has developed a suite of tools that enables more efficient solution for a number of analysis
types. The key attribute of these tools is that the full wheel solution can be obtained by solving only
one or at most a few blades per row230.
Turbomachines”, Siemens Industrial Turbomachinery Ltd., Waterside South, Lincoln LN5 7FD, UK.
234 Ding Xi Wang, “An Improved Mixing-Plane Method for Analyzing Steady Flow Through Multiple-Blade-Row
Turbomachines”, Siemens Industrial Turbomachinery Ltd., Waterside South, Lincoln LN5 7FD, UK.
235 Denton J. D. “The calculation of three dimensional viscous flow through multistage turbomachines”,1990.
236 Adamczyk J J, Mulac R A, Celestina M L. “A model for closing the inviscid form of the average-passage
➢ Keep strong conservation of flow solutions across the interface, such as mass flow rate,
momentum, total enthalpy, etc.;
➢ Obtain the aerodynamic parameters with little deviation from the experiment;
➢ Be robust by employing non-reflective boundary conditions on the interface and special
interface treatments for reversed flow.
A popular class of simple mixing models simulate the flow mixing process by simply circumferentially
averaging the flow variables on the interface. Since there are only five independent flow variables in
three-dimensional compressible flow, different selections of the independent flow variables result in
different simple mixing models. In the past decade, some novel mixing models have also been
proposed. All the results demonstrate that different mixing models have various effects on the
computation robustness, flow solution conservation and thus the flow details. However, no
comparative investigation of the aforementioned novel mixing models has been carried out in the
open literatures. With the development of computer capacity, more emphases are put on URANS. The
unsteady computation methods for multi-stages include the phase-lagged method, blade scaling
technique, time-inclined method and frequency domain methods, such as nonlinear harmonic
method and harmonic balance method. Due to the eases of implementation and extension to multi-
stages, the blade scaling technique has been widely applied in the unsteady turbomachinery flow
computations. The exchange of two-dimensional flow fields on the interface between adjacent blade
rows is the most crucial issue for unsteady flow computation of multi-stages because of the non-
matched grid points between the two sides of the interface and the relative motion between rotor
and stator. In such cases, it is necessary to develop an interpolation method strongly maintaining the
conservation and continuity of flow variables across the interface.
237 J. D. Denton, “The calculation of three-dimensional viscous flow through multistage Turbomachinery”,
Journal of Turbomachinery, 114(1):18–26, 1992.
238 Release 12.0 © ANSYS, Inc. 2009-01-22.
197
the issue of how to transfer solution information across such an artificial interface in a steady flow
field analysis.
In reality, upstream wake mixes out gradually when it is transported downstream. The mixing loss is
also expected to rise gradually. With a mixing-plane treatment, the strong circumferential non-
uniformity of an upstream wake mixes out significantly across an interface, leading to a nearly
circumferential uniform flow field on the downstream side. Consequently, across an interface from
its upstream side to its downstream side there is an abrupt rise of loss. Independent research shows
that the artificial mixing loss across the interface of a turbine stage by a steady mixing-plane analysis
is significant, thus leading to higher overall loss for the turbine stage in comparison with the loss
from an unsteady sliding-plane analysis. However, some research demonstrates that the artificial
mixing loss can be trivial, leading to lower overall loss than that by an unsteady sliding-plane analysis.
Nevertheless, investigations by both [Fritsch and Giles]241 and [Pullan]242 indicate that the loss by a
steady mixing-plane analysis grows at a slower rate through the downstream blade row in
comparison with that by an unsteady sliding-plane analysis. Modern multiple-blade-row
turbomachines usually have a small inter row gap. This situation makes it not only desirable but also
necessary for a mixing-plane method to be non-reflective. Otherwise artificial reflections from a
mixing plane very close to a blade leading edge or a blade trailing edge can deter convergence and
spoil solution. Apart from conservation and non-reflectiveness, as pointed out in243, an ideal mixing-
plane method should also be robust so that it can handle reverse flow, which can exist either in a
solution process or in a converged flow field.
9.4.2 Principles of Flux Conservation
An artificial interface between two adjacent blade rows is normally a revolution surface, which is a
single curve connecting the hub contour and the casing contour in the meridional plane as shown in
Figure 9.7244. The conservation law states that the fluxes of mass, momentum, energy, and other
scalar quantities through an arbitrary segment of the interface from the domain on one side of the
Figure 9.7 Schematic of an Artificial Interface Between a Rotor and a Stator (left) and the Virtual
Control Volume Formed by Displacing Two Adjacent Domains (right)
241 Fritsch, G., and Giles, M. B., “An Asymptotic Analysis of Mixing Loss,” ASME J. Turbomachines, 1995.
242 Pullan, G., 2006, “Secondary Flows and Loss Caused by Blade Row Interaction in a Turbine Stage,” ASME J.
Turbomachine., 128, pp. 484–491, July 2006.
243 Holmes, D. G., “Mixing Planes Revisited: A Steady Mixing Plane Approach Designed to Combine High Levels of
segment should be equal to those into the domain on the other side of the segment. This can be easily
understood through forming a virtual control volume, the hatched part as shown in Figure 9.7, by
displacing the two adjacent domains in the stream wise direction. There is no flux through the upper
and lower sides of the control volume; therefore, the flux entering the control volume from one
domain must equal the corresponding flux exiting the control volume into the other domain in a
steady flow analysis. For an arbitrary segment with the length of ds as shown in Figure 9.7, the
fluxes of mass, momentum, energy, and scalar quantities through the segmental revolution surface
are defined as follows:
2π
2π
S = ∫ dsrdθ
0
Eq. 9.2
Dividing Eq. 9.1 by this area gives the circumferential area averaged fluxes. It is obvious that, for a
fully converged steady solution of a multiple-blade-row turbomachine, circumferential area
averaged fluxes of mass, momentum, energy, and other scalar quantities across an arbitrary segment
of an interface, as defined in Eq. 9.1/Eq. 9.2, should be conserved. It should be noted the
circumferential area averaged fluxes as defined in Eq. 9.1/Eq. 9.2 can be calculated over a pitch
angle that is less than 2π. Before a solution converges, the fluxes calculated using flow variables from
domains on its two sides of an interface are usually not conserved or equal. The task of a mixing-
plane method is to make use of the fluxes to drive the differences to zero. Two existing methods will
be explained with the proposition of a new method later. To get the flow primitive variable
200
differences, 3 methods is been proposed by [Wang]245. The differences of the three methods lie in
how to calculate the incoming flow disturbances according to circumferential area-averaged flux
terms on two sides of an interface. For further information, please consult 246.
9.4.3 Case Study 1 - Comparison of Flux
Balanced Mixing Models on Q-1.5 Steady Mixing Model Nomenclature
Stage Rotor 67 Momentum-Averaged MA
To study the performance of flux balanced Entropy-Averaged EA
mixing models on simulating the complex Flux Balanced MA
flow, such as shock wave and reversed (FBMA)
flow on the interface, the quasi-1.5-stage Flux Balanced
Flux
Rotor 67 with small geometric Balanced EA (FBEA)
modification is then investigated . An Experiment:
artificial step of about 6% span is imposed TA
Time-Averaged
on the casing contour ahead of the second Experiment:
interface to produce reversed flow on the MP
Total Pressure Ratio
interface. The designed and modified
casing contours are shown in Figure 9.8. Table 9.2 Nomenclature for different Mixing Models
The grid with the same topology and cell Used in Study
number used for flow solver validation is
generated. Table 9.2 lists the nomenclature for Mixing models for steady computation. For a listing
of relative deviations of flow variables across the 1st and 2nd interfaces of the modified quasi-1.5
stage Rotor 67 vs. the circumferential velocity, see [YaLu et al.]247. On both interfaces, the unsteady
computation has an absolute
advantage on the flow
conservation over all the
mixing models. Compared
with the simple mixing models,
the relative deviations of most
of flow variables are decreased
by the flux balanced models.
By the flux balanced models,
the total temperature ratio
across all the interfaces is
almost strictly maintained,
demonstrating the superior
performance of flux balanced
mixing models on flow
conservation.
Compared with other flow
variables, the relative Figure 9.8 Sketch of Casing Treatment of Rotor 67 (Courtesy of 157)
deviation of tangential velocity
across the first interface is extremely large, especially for MA and EA models. This is because the
tangential velocity on the interface is quite small. The entropy by MA and EA models decreases across
245 Ding Xi Wang, “An Improved Mixing-Plane Method for Analyzing Steady Flow Through Multiple-Blade-Row
Turbomachines”, Siemens Industrial Turbomachinery Ltd., Waterside South, Lincoln LN5 7FD, UK.
246 See Previous.
247 ZHU YaLu, LUO JiaQi & LIU Feng, “Flow computations of multi-stages by URANS and flux balanced mixing
models”, Science China, Technological Sciences, July 2018 Vol.61 No.7: 1081–1091.
201
Figure 9.10 Contour of Relative Mach Number and Iso-Surface of Axial Velocity of Modified Rotor 67
202
wise distributions of flow variables by different computation methods are not consistent with each
other as shown in Figure 9.9. Although the positions of shock wave on the interface are almost the
same for all the methods, the detailed shock wave patterns are slightly different as indicated by the
zones with low static temperature in Figure 9.11. The shock wave patterns by MA and EA models
are similar, and those by FBMA and FBEA models are also close, which are consistent with the span-
wise distributions of total pressure ratio and total temperature ratio. However, none of shock wave
patterns by the mixing models matches well with that of unsteady computation. The patterns of
rarefaction waves after the shock waves perform the similar variations.
Figure 9.11 Temperature Contours on the 1st Interface of Modified Rotor 67 - (a) FBEA ; (b) FBMA ; (c)
EA ; (d) MA ; (e) TA (Courtesy of YaLu et al.)
9.4.4 Case Study 2 - Modeling of Secondary Flows in Single Blade Rows using Mixing Plane
Approach
A computational modeling of secondary flows in single blade rows and a performance assessment in
3D turbine stages computations using wall functions were made by [Xisto et al.] 248 . The analysis of
the flow in turbine blades has been extended from 2D to 3D, and from pure Euler equations to Navier-
Stokes modeling, including turbulent flow. This later has only been possible due to fast development
of computer power of modern desktop computers. Most of these analyses were carried out for
isolated blade rows. However, this approach is not accurate in many circumstances, due to a strong
coupling and interaction between the several blade rows. To fully account for the rotor-stator
interaction, a 3D unsteady Navier-Stokes analysis is required, but such an analysis is too CPU-
intensive and expensive in terms of computing power, so we will restrict our approach to the mixing
plane model . The mixing plane approach is applied at the blade row interface between the stator-
rotor. It can also be applied to several stages in series. In this approach one assumes that the flow is
totally mixed out and is axis-symmetric between the blade rows. Actually, it can only include the
effects of radial variation in an approximate way and cannot account for any circumferential
variations, such as those created by wakes, leakage or secondary flows. Although this, it is important
to clarify that the pitch wise averaging does not affect the span wise variation in flow. Actually, the
span wise variation of pressure, velocity, flow angle, etc., at all stations between hub and tip is
obtained from the full 3D Navier-Stokes computation.
The computation of the flow, for a single blade row, can nowadays routinely be made using a low-Re
248Carlos M. C. Xisto , José C. Páscoa e Emil Göttlich, “Computational modeling of secondary flows in single blade
rows and performance assessment in 3D turbine stages computations using wall functions”, November 2009.
203
turbulence model and resolving the boundary layer, even with desktop computers. The computation
of a whole stage is more computing demanding and, at least with our current capabilities, can only
be accomplished with the use of wall functions. This introduces the reason for the current work,
which is to analyze the performance assessment limitations when using wall functions, for turbine
stage computations, instead of resolving the boundary layer. A major problem that arises in the
design and performance analysis of axial turbines is the understanding, analysis, forecasting and
control of secondary flows . A pioneer work in the understanding of this phenomenon could be found
in . Here Langton presents the evolution of a tree-dimensional flow in a turbine cascade. Were at the
end wall of the cascade the inlet boundary layer separates at the saddle-point and forms the
horseshoe vortex. The pressure leg of this vortex will become the passage vortex and the suction leg
will become the counter-vortex and as an opposite sense of rotation to the passage vortex, see
Figure 6.11.
9.4.4.1 Transonic Turbine Stage Meshing and Flow Details
A high level of detail for the geometry was considered, including the fillets at hub and tip sections and
the rotor gap. The stator comprises 24 blades and the rotor has 36 blades, which represents a ratio
of 2:3 between the stator and rotor blades. In our case, using the mixing plane model we can perform
the computation using only one blade for the stator and rotor rows. The first phase of the
computations performed for the stage was made using isolated blade rows for the stator and rotor.
By solving each flow in an isolated blade row we were able to detect any flow convergence problems,
typically created by poor mesh quality. After this fine tuning of the mesh we proceed into the full
stage computation. For this test case we solved the Navier-Stokes equations using the Spalart-
Allmaras turbulence model. In this computation an implicit discretization using double precision
was retained. The computations started using a first-order discretization in space and later on were
toggle to second order accuracy. The mesh comprises 15 H blocks, with 8 blocks in the stator and 7
in the rotor. The overall mesh comprises 224136 nodes, these were 131136 for the stator and 93000
for the rotor. The stator blade
comprises 49 points in the
inter blade region and 93
points in the axial flow
direction, with 65 points used
to define the blade surface.
For the radial direction we
have distributed 30 points.
For the rotor blade 29 points
were applied in the inter-
blade zone and 93 on the axial
direction, with 65 points used
to define the blade geometry.
In the radial direction 30
points were used, the mesh
can be seen in Figure 9.12.
The flow field at inlet of the
stator is completely subsonic,
transonic flow is restricted to
minor zones around the
stator trailing edge. Thus, at Figure 9.12 Mesh for Transonic Turbine Stage - Upper Image
stage inlet we have imposed Depicted the Mesh at the Hub Surface while the Lower Image
stagnation pressure and Represented Mesh used for the Blade Span
temperature with the
204
corresponding flow angles and at stage outlet static pressure is imposed. At the mixing plane
interface also characteristic boundary conditions are imposed, namely stagnation pressure,
temperature, flow angles at rotor inlet and static pressure at stator outlet. In order to apply the
Spalart-Allmaras turbulence models we have considered a turbulence intensity of 10% and a length
scale of 1% pitch at stator mid span. These turbulence quantities are usually applied in modeling
turbomachinery flows. The initial computations were performed with an explicit approach and using
pure characteristic boundary conditions to extrapolate the variables at the boundaries.
Unfortunately convergence was
not attained, the residues got
stuck at a minor value.
Convergence was attained only
when using the implicit
formulation and applying
nonreflecting boundary
conditions. Due to computing
power restrictions only results
for the Spalart-Allmaras
turbulence model using wall
functions were obtained. Figure
9.13 presents the numerical
results of velocity obtained for a
section at stator mid span.
Although, numerical result
shows a good agreement with the Figure 9.13 Results of the Velocity Contours for a Radial Section
experimental data, however, at Stator Mid Span using the Mixing Plane Approach
more experimental data is
necessary for a precise validation of the model. Only with a span wise distribution of experimental
and computed variables we can assess, in full, the capability of the mixing plane model in the
prediction of this flow field. For further details, readers are encourage to consult [Xisto et al.]249 .
9.4.5 Case Study 3 - Improvement Methods for Mixing Plane Models
For modern turbomachines, the trend of design is to reach higher aerodynamic loading but with still
further compact size. In such a case, the traditional mixing-plane method has to be revised to give a
more physically meaningful prediction. [Pengcheng & Fangfei]250, presented a novel mixing-plane
method, and three representative test cases including a transonic compressor, a highly-loaded
centrifugal compressor and a high pressure axial turbine were performed for validation purpose.
This novel mixing-plane method can satisfy the flux conservation perfectly. Reverse flow across the
mixing-plane interface can be resolved naturally, thus making this method numerically robust.
Artificial reflection at the mixing-plane interface is almost eliminated, and then its detrimental impact
on the flow field is minimized. Generally, this mixing-plane method is suitable to simulate steady
flows in highly-loaded multistage turbomachines.
From the authors’ point of view, the mixing-plane method should not just be a pure numerical
procedure to transfer the circumferentially averaged flow variables across the interface. A physical
correspondence for this pitch wise mixing can be found, i.e., we can just make the gap between the
two adjacent blade rows long enough so as to mix out all the non-uniformities (as shown from Figure
249 Carlos M. C. Xisto , José C. Páscoa e Emil Göttlich, “Computational modeling of secondary flows in single blade
rows and performance assessment in 3D turbine stages computations using wall functions”, Conference Paper·
November 2009.
250 Du Pengcheng a, Ning Fangfei, “Validation of a novel mixing-plane method for multistage turbomachinery
9.14-(a) and (b)), while the span wise mixing is assumed to be suspended in the ‘‘extended mixing
region”. Therefore, for a fully converged flow field in the case as shown in Figure 9.14-(b), we can
find such an intermediate position where the flows are pitch wise uniform. At this position, if we cut
out an infinitely thin slice as denoted by two lines ‘‘ml” and ‘‘mr” in Figure 9.14-(b), the following
governing equations expressed in cylindrical coordinate system hold:
̅
∂𝐐
= 𝐅mr − 𝐅ml , ̅ = [ρ, ρv, ρv, ρv, ρe]T
𝐐
∂t
𝐅 = [ρU, ρUvx + nx p , ρUvθ + nθ p , ρUvr + nr p , ρUH]T
Eq. 9.3
with t being the pseudo time, ρ the density, (vx; vɵ; vr) the absolute velocity components expressed
in cylindrical coordinate (x , ɵ , r), e the total energy, U = nxvx + nɵvɵ + nrvr the advective velocity
normal to the blade row interface, p the static pressure and H the total enthalpy. The unit vector n =
(nx , nɵ , nr)T denotes the normal direction of blade row interface, and n is actually equal to zero
because the interface is a revolution surface.
is very suitable for demonstrating the effectiveness of the proposed mixing-plane model in dealing
with the typical circumferential non uniform flow field featured by strong shock and thick wakes.
Further details can be obtained from .
(A) Static
(B) Entropy
Pressure
K S PS = K R PR Eq. 9.4
Where, KS and KR are relative prime which stand for the number of passages in the stator domain and
rotor domain, respectively. Ps and PR denote the pitch of stator and rotor separately. An
approximation of the blade number can be made if Ks and Kr are large in order to reduce the
computational cost, which is called Domain Scaling. For instance a turbine with 29 blades of stator
and 31 blades of rotor can be approximated by a turbine with 30 blades of both stator and rotor, then
only one passage is needed to mesh for both stator and rotor. However, the simulation results are
only the approximated result to the real model. The Frozen Rotor method is used firstly by [Brost et
al.]251 in simulations of an axial turbine where the simulated results have a good accordance with the
transient results of the measurement. While, the flow field in a passage usually changes a lot during
V. Brost, A. Ruprecht, and M. Maih, “Rotor-Stator interactions in an axial turbine, a comparison of transient
251
and steady state frozen rotor simulations”, Conference on Case Studies in Hydraulic Systems-CSHS03, 2003.
207
252 NUMECA International, Brussels, “Fine/Turbo User Manual V8 (including Euranus)”, October 2007.
253 J. M. Weiss, K. C. Hall, “simulation of unsteady turbomachinery flows using an implicitly coupled nonlinear
harmonic balance method”, Proceedings of ASME Turbo Expo 2011, GT2011.
254 M. Rai, “Application of domain decomposition methods to turbomachinery flows”, ASME Advances and
Figure 9.17 Half Stencil and Full Stencil Reconstruction with: A) Intersection, B) Halo-Cell
257 S. Khelladi, X. Nogueira, F. Bakir and I. Colominas, “Toward a higher-order unsteady finite volume solver Based
on reproducing kernel particle method”, Computer Methods in Applied Mechanics and Engineering, 2011.
258 Hongsik, Xiangying Chen, Gecheng Zha, “Simulation of 3D Multistage Axial Compressor Using a Fully
Conservative Sliding Boundary Condition”, multistage turbomachinery are developed and implemented;
Proceedings of the ASME, 2011 International Mechanical Engineering Congress & Exposition IMECE2011,
November 11-17, 2011, Denver, Colorado, USA.
259 L. He, “Modelling issues for computation on unsteady turbomachinery flows. In Unsteady Flows in
of terms you wish to consider, starting from the first. More the number of terms considered, closer is
the series to the original signal. In the literature, no consensus exists concerning the number of
harmonics needed to achieve convergence for turbomachinery stage configurations. It is shown that
the convergence of Fourier-based methods is closely related to the impulsive nature of the flow
solution, which in turbomachines is essentially governed by the characteristics of the passing wakes
between adjacent rows. As a result of the proposed analysis, a priori estimates are provided for the
minimum number of harmonics required to accurately compute a given turbomachinery
configuration. Their application to several contra-rotating open-rotor configurations is assessed,
demonstrating the practical interest of the proposed methodology. This method solves the steady
transport equations for the time-averaged flow and the time harmonics. For turbomachinery, the
Blade Passing Frequencies (BPF) are the fundamentals in time domain of the periodic disturbances
from the adjacent blade rows. The solving of the generated perturbation amplitudes in a row is
performed in the
frequency domain by a
steady transport
equation associated
with BPFs and
subharmonics. The
deterministic stresses
are calculated directly
from the in-phase and
out-of-phase
components of the
solved harmonics.
Using this method, only
one passage is needed
that saves the
computational cost
Figure 9.18 Relative Velocities Obtained using HB Techniques
greatly. He et al., Vilmin
et al. validated this
method with simulations on a 3D radial turbine and a multistage axial compressor. Therefore, this
method is adopted in the unsteady simulation of a low speed axial turbine. The physical quantity can
be decomposed into a time-averaged value and a sum of perturbations, which in turn can be
decomposed into N harmonics261. Figure 9.18 displays Harmonic function method in obtaining
relative velocities (courtesy of NUMECA.com). Since the Harmonic method is widely used, it is
warranted a bit more exploring which will be dealt in the coming section.
9.5.3 Profile Transformation (Pitch Scaling)
In typical turbomachinery applications, it is very common that one or both blade rows have a prime
number of blades per wheel. Formerly in such cases, it was necessary to model the whole 360° wheel
in order to attain the required level of accuracy. It is possible to reduce the size of the computational
problem (memory and computational time) by solving the blade row solution for one or two passages
per row, while still obtaining reasonably accurate solutions, therefore providing a solution to the
S. Vilmin, E. Lorrain, and Ch. Hirsch, ” Unsteady flow modeling across the rotor/stator interface using the
261
U R1 (r, θ, z, t ) = U R2 (r, θ + PR , z, t − Δt )
U R1 (r, θ, z, t ) = U R2 (r, θ + PR , z, t)
US1 (r, θ, z, t ) = US2 (r, θ + PS , z, t − Δt ) Eq. 9.5
US1 (r, θ, z, t ) = US2 (r, θ + PS , z, t )
Using the following set of space-time transformations to the problem above as:
Δt
r = r , θ = θ , z = z, t = t − Eq. 9.6
PR − PS
The equations that are solved are in the computational (r’, ϴ', z’, t’) transformed space-time domain
and need to be transformed back to physical (r, ϴ, z, t) domain before post-processing. The periodicity
is maintained at any instant in time in the computational domain and it is evident that the rotor and
stator passages are marching at different time step sizes. We have the time step sizes
in the rotor and stator related by their pitch ratio as:
PR Δt S
= Eq. 9.7
PS Δt R
Where nΔtS = PR/VR and nΔtR = PS/VR. The simulation time step size set for the run is used in the stator
passage(s) ΔtS and program computes the respective rotor passage time step size ΔtP based on the
rotor-stator interface pitch ratio as described above. When the solution is transformed back to
physical time, the elapsed simulation time is considered the stator simulation time. Required that the
pitch ratio fall within a certain range, as described by the inequality:
Mω P Mω
1− S 1+ Eq. 9.8
1 − Mθ PR 1 + Mθ
Where Mω is the Mach number associated with the rotor rotational speed (or signal speed in the case
of an inlet disturbance problem), Mϴ is the Mach number associated with the tangential Mach
number, and the ratio of PS to PR is the pitch ratio between the stationary component and the rotating
component. For most compressible turbomachinery applications (for example, gas compressors and
turbines), Mω is in the range of 0.3-0.6, enabling pitch ratios in the range of 0.6-1.5. Note that
according to ANSYS CFX© these limits are not strict, but approaching them can cause solution
212
instability.
9.5.5 Revisiting Non-Linear Harmonic Balance (NLHB) Methodology
Given the time periodic nature of these flows, one can model the unsteady flow in turbomachines
using nonlinear, harmonic balance techniques. Roughly speaking, the family of nonlinear harmonic
methods expands the unsteady flow field in a Fourier series in time and solves for the Fourier
coefficients. [He]266, and [Ning]267 developed a harmonic method in which the unsteady harmonics
are treated as perturbations. [Hall, Thomas, and Clark]268 developed a full harmonic balance method,
which allows for arbitrarily large disturbances and any number of harmonics. The method is
computationally efficient and stores the unsteady nonlinear solutions as the working variables at
several time levels over one period of unsteadiness, rather than storing the Fourier coefficients them-
selves. [Gopinath and Jameson]269 and others have applied this approach to turbomachinery
applications. For an excellent recent survey of Fourier methods applied to turbomachinery
applications, see the survey paper by [He]270. In all these methods, the harmonic balance equations
are solved by introducing a pseudo-time derivative term and then marching the coupled equations
to a steady state. Using the frequency-domain or time-linearized technique, it is possible to first
compute the time-mean (steady) flow by solving the steady flow equations using conventional CFD
techniques. One then assumes that any unsteadiness in the flow is small and harmonic in time (eiωt).
The governing fluid equations of motion and the associated boundary conditions are then linearized
about the mean flow solution to arrive at a set of linear variable coefficients equations that describe
the small disturbance flow. The time derivatives d/dt are replaced by jω where ω is the frequency of
the unsteady disturbance, so that time does not appear explicitly. The resulting time-linearized
equations can be solved very inexpensively, but unfortunately cannot model dynamic nonlinearities.
9.5.5.1 Temporal & Spatial Periodicity Requirement
Consider unsteady flows that are temporally and spatially periodic. In particular, temporal and
spatial periodicity requires that
U (x , t) = U (x , t + T)
Eq. 9.9
U (x + G , t) = U (x , t + Δt)
Where T is the temporal period of the unsteadiness, G is the blade-to-blade gap and Δt is the time lag
associated with the inter blade phase lag. Similarly, for cascade flow problems arising from vibration
of the airfoils with fixed inter blade phase angles σ, or incident gusts that are spatially periodic. As an
example, consider a cascade of airfoils where the source of aerodynamic excitation is blade vibration
with a prescribed inter blade phase angle σ and frequency ω. Then T = 2π/ω and Δt = σ/ω. Because
the flow is temporally periodic, the flow variables may be represented as a Fourier series in time with
spatially varying coefficients.
9.5.5.2 Boundary Conditions
We first consider the flow field kinematics of two adjacent blade rows where the first row has B1
266 He, L., 1996. “Modelling issues for time-marching calculations of unsteady flows, blade row Interaction and
blade flutter”, VKI Lecture Series “Unsteady Flows in Turbomachines”, von Karman Institute for Fluid Dynamics.
267 Ning, W., and He, L., 1998. “Computation of Unsteady Flows around Oscillating Blades Using Linear and Non-
blades spinning with rotational rate ω1 rad/s and the second has B2 blades spinning with rotational
rate ω2 rad/s. The flow field within the stage can be decomposed into a Fourier series in the
rotational direction characterized by a set of Nm1, m2 nodal diameters as
Where m1 and m2 can take on all integer values. In the frame of reference of the first and second
blade row, the frequency of the unsteady disturbance associated with any nodal diameter is
Note that in either row the unsteady frequency associated with a given nodal diameter is a function
of the blade count and relative rotation rate of the adjacent row. Furthermore, associated with each
unsteady frequency is an inter blade phase angle:
B2 B1
σ1,m2 = m2 2π , σ m1 ,2 = m1 2π Eq. 9.12
B1 B2
In the frame of reference of the second row. Clearly the inter blade phase angles associated with a
given nodal diameter are a function of the pitch ratios between the two rows. Note that the pitch in
each row is given by G1 = 2π/B1 and G2 = 2π/B2 in the first and second rows, respectively. Solution
Method Since the solution U is periodic in time, we can represent it by the Fourier series:
M
U ( x, t) = Uˆ
m=− M
m ( x) e imt
Eq. 9.13
1 N −1 ~
where ˆ
U m ( x) =
N n =0
U n ( x, t n ) e -imt n
can considered complex conjugate of each other. Here, ω is the fundamental frequency of the
disturbance, M is the number of harmonics retained in the solution: Û m are the Fourier coefficients,
and Ũ n are a set of N = 2M + 1 solutions at discrete time levels tn = nT/N distributed throughout one
period of unsteadiness, T. At any U is vector of conserved variables and can be expressed as
ρ (x, t) = R n (x, t) eint , ρu (x, t) = U n (x, t) eint , v( x, t) = Vn (x, t) eint ,.....
n n n
Eq. 9.14
At any location in the flow field domain we can transform the time level solutions into Fourier
coefficients and vice versa using a discrete Fourier transform operator [E] and its corresponding
inverse E-1 as follows
~ ~
Uˆ = E U or ˆ
U = E−1 U
Eq. 9.15
Where E and E−1 are square matrices of dimension N × N, and the Fourier coefficients and time level
solutions have been assembled into the vectors Ũ as
214
~ ~ ~ ~ ~
U = [U 0 , U1 , U 2 , ....... U N−1 ]T Eq. 9.16
The solutions at each discrete time level are obtained by applying the governing equations to all the
Ũ simultaneously
̃
∂𝐔
∫ ⃗⃗]. d𝐀
dV + ∮[𝐅⃗ − 𝐆 ⃗⃗ = ∫ 𝐒̃dV
V ∂t V
Eq. 9.17
Where the flux and source vectors F͂͂,͂͂ G̃ , and S͂͂͂͂ are evaluated using the corresponding time level
solution. The time derivative in Eq. 9.19 is evaluated by differentiating Eq. 9.15 with respect to time
as follows:
~
U E −1 ˆ E −1 ~ ~
= U= U = [ D] U
t t t Eq. 9.18
Where [D] is the pseudo-spectral, N × N matrix operator. Substituting appropriately, yields the
desired harmonic balance equations:
⃗⃗]. 𝐝𝐀
̃ dV + ∮[𝐅⃗ − 𝐆
∫ [𝐃]𝐔 ⃗⃗ = ∫ 𝐒̃dV
V V
Eq. 9.19
The harmonic balance equations are discretized using a cell centered, polyhedral-based, finite-
volume scheme. Second order spatial accuracy is achieved by means of a multi-dimensional, linear
reconstruction of the solution variables. The convective fluxes are evaluated by a standard upwind,
flux-difference splitting and the diffusive fluxes by a second-order central difference. A pseudo-time
derivative of primitive quantities, ∂Q/∂τ, with Q = {p, u, T}, is introduced into Eq. 9.19 to facilitate
solution of the steady harmonic balance equations by means of a time marching procedure. An Euler
implicit discretization in pseudo-time271 produces the following linearized system of equations:
U S U
+ Δτ [ A] − + [D] ΔQ = −Δτ R Eq. 9.20
Q Q Q
where R' is the discrete residual of Eqn. (18.17), and ΔQ' are the resultant primitive variable
corrections across one pseudo-time step, Δτ. Operator [A] is the Jacobian of the discrete inviscid and
viscous flux vectors with respect to primitive variables Q and introduces both center coefficients as
well as off-diagonals arising from the linearization of the spatially discretized fluxes. The coupled
system given by Eqn. (18.17) contains equations from all time levels linked at every point in the
domain by the pseudo-spectral operator [D]. The result is a large system, and solving it all at once
would be rather intractable. However, we can exploit the point coupled nature of the system and
employ approximate factorization to produce the following two step scheme:
271Weiss, J. M., Maruszewski, J. P., and Smith, W. A., 1999, “Implicit Solution of Preconditioned Navier-Stokes
Equations Using Algebraic Multigrid”, AIAA Journal, 37(1), Jan., pp. 29–364
215
U S ~
+ Δτ [ A] − ΔQ = −Δτ R
Q Q
Eq. 9.21
U −1 U ~
[ I ] + Δτ [D] ΔQ = ΔQ
Q Q
Where ΔQ̃ ' represents provisional corrections to the solution. In the first step, the time levels are no
longer coupled and we can solve for the ΔQ̃ ' one time level at a time. With the exception of the physical
time derivative appearing, the evaluation of fluxes, accumulation of the residual, and the process of
assembling and solving at each time level proceeds exactly as for a single, steady-state solution in the
time domain. Here we employ an algebraic multigrid (AMG) method to solve the linear system and
obtain the provisional ΔQ̃ ’. In the second step the complete corrections ΔQ' for the current iteration
are obtained by inverting at each point in the domain given all the ΔQ̃ ' computed in step one.
9.5.5.3 Fourier 'Shape Correction' for Single Passage Time-Marching Solution
The Fourier modelling approach to nonlinear flows was proposed in 1990 for time-marching
solutions of unsteady turbomachinery flows272. This was the first Fourier method for
turbomachinery. The objective at the time was to enable an unsteady flow solution to be carried out
in a single blade passage domain but without requiring a large amount of computer memory, as in
the Erdos's Direct Store method. The main ingredient is to carry out the temporal Fourier transform
at the ‘periodic boundaries of the single blade passage domain. Then the Fourier harmonics
(temporal shape) are used to correct the corresponding boundaries according to the phase shift
periodicity. The method was then called ‘Shape Correction’. The validity of the single passage Shape-
Correction method can be examined by comparing with the direct multi-passage solution. Figure
9.21 shows Stagnation Pressure contours under inlet distortion for NASA Rotor 67 where the Left
shows whole passage annulus solution, and the Right, single passage solution as reconstructed. It was
shown that the Fourier modelling as implemented in the Shape-Correction can capture flow
disturbances and responses with large nonlinearity (e.g. a large scale shock oscillation in fan blade
passage under an inlet distortion of long circumferential wave length. Given only 3-5 harmonics were
Figure 9.21 Stagnation Pressure Contours under inlet distortion for NASA Rotor 67
272L. He, "An Euler Solution for Unsteady Flows around Oscillating Blades", ASME, Journal of Turbomachinery,
Vol.112, No.4, pp.714-722, 1990.
216
required for capturing sufficiently accurately the temporal variation, the computer memory
requirement is very low compared to the Erdos’s Direct Store approach. A key advantage of splitting
flow components represented by Fourier harmonics is the ability in dealing with multiple
disturbances with distinctive frequencies (He 1992). The generalized shape correction has been
applied to unsteady flows in multi-
rows (IGT-rotor-stator) with vibrating
rotor blades for optimization of intra-
row gap effects on both aerothermal
performance and flutter stability.
9.5.5.4 Case Study 1 – 2D
Compressor Stage
In this section we compare results
obtained from the implicitly coupled,
non-linear harmonic balance method
described above with solutions from a
full, unsteady simulation based on the
standard dual time-stepping approach.
The test case consists of a model 2D
compressor stage; specifically, the first
stator and second rotor rows of the five Figure 9.22 Computational Mesh for HB and TRS Methods
row. There are three stator blades to
every four rotor blades. The two blade
rows are separated by an axial gap equal
to 0.25 times the aerodynamic chord of
the rotor. The Mach number at the inlet
to the stator is 0.68 and the relative
Mach number entering the rotor is 0.71.
The static-to-total pressure ratio across
the stage is 1.2. Three separate Euler
calculations are made using the
nonlinear harmonic balance method in
which one, two and three harmonics,
respectively, are retained for the blade
passing frequencies in both the stator
and rotor. Contours of instantaneous
pressure, representative of the flow field
within the compressor stage and
computed using three harmonics in each
blade row, are shown in Figure 9.23
using nonlinear harmonic balance
method. Note that computations are
performed on just the center blade
passage outlined in each row. The
solutions shown in the passages above
Figure 9.23 Instantaneous Pressure Distribution Within
and below are phase-shifted the Compressor Stage Using (NLHB)
reconstructions included for clarity.
9.5.5.5 Case Study 2 - 3D Flow in Turbine Cascade
217
3D flow in turbine cascade in which the Harmonic Balance (HB) method is applied for modeling
rotor/stator interaction and pressure fluctuations near trailing edges. Computational results are
compared with Transient Rotor/Stator (TRS) results which shows importance of unsteady effects273.
The harmonic balance method requires only a single blade passage be meshed. A structured HOH
mesh is generated for each of the two blade rows, as shown in Figure 9.22. The inlet and exit grid
planes for each of the blade rows correspond to the axial planes where test data is available. The
blade passage mesh is made up of 1.3 M cells with a near wall spacing of y+ = 1.0 - 2. The HB solver
models the fluid as an ideal gas with turbulence closure provided by the Spalart-Allmaras turbulence
model. The solver is run with a CFL number of 5.0, and separate trials are conducted retaining one,
three, and five modes. The solver has converted to a periodic, unsteady solution within 4000 - 5000
iterations. The TRS solver uses a time step is equal 510-5s. This value correspond 5 steps per vane
passing (10 inner iterations per time step). Assessment of the effectiveness of the two methods of
calculation is carried out on the basis of the comparison of time required for obtaining of non-
stationary periodic solutions on the interval of time, sufficient for the passage of at least one rotation
of the impeller. The HB-results showed that CPU time is increased in 7 times for 5 modes compared
time when calculating with one mode. Using 3 modes CPU time is increased (for one iteration) in 3.3
times in comparison with one-mode approximation. The acceleration of the calculation, which is
defined as the ratio of the CPU time required to obtain a periodic solution using the TRS method to
the CPU time of the decision on the HB method is 1:2 - five modes, 1:1 - three modes and 3:1 - one
mode. Hence, the substantial savings (three times) is observed only in the case of one – mode
approximation. It is important to note that in all cases the calculations were carried out at the same
calculation grid, including two blades. Figure 9.24 present instantaneous predictions of turbulent
viscosity at mid - span for the HB and TRS solutions. The stator wake enters the rotor passage and
grows both laterally and in the stream wise direction. This process continues as the stator wake is
“chopped” by the leading edge of the rotor blade and convects downstream.
TRS HB
Figure 9.24 Instantaneous Predictions of Turbulent Viscosity at Mid-Span for HB and TRS Solutions
9.5.6 Assessment of 2D Steady and Unsteady Adjoint Sensitivities for Rotor-Starter Interaction
Adjoint-based CFD turbomachinery optimization has gained increasing popularity over the last years
due to the advantages in dealing with problems characterized by a large number of design variables
at affordable computational cost [Rubino et al.]274. Thanks to this, adjoint-based optimization offers
the possibility to fully exploit the ever increasing computational power to accomplish novel and
unconventional turbomachinery design. To date, despite the intrinsically time-varying nature of
turbomachinery flows, adjoint-based methods mostly rely on steady state approaches. However, the
use of unsteady-based design could lead to major steps forward in performance improvement for the
next generation of turbines and compressor, allowing to tackle multidisciplinary problems. Time-
accurate adjoint methods are well-established but their industrial use is very limited, due to the
excessive computational cost of both the direct flow solution and the I/O overhead associated with
the reverse adjoint mode. The harmonic balance (HB) method is a cost-effective alternative to time
accurate adjoint for non-linear time periodic flow problems, thus it is highly attractive for
turbomachinery applications.
The objective here is to perform an assessment between steady and HB-based unsteady design
sensitivities, investigating the impact of unsteady effects on the aerodynamic design of
turbomachinery. Computational cost, memory requirements are considered as comparison terms,
by using both steady and unsteady methods. The steady-state adjoint calculations are performed by
resorting to mixing plane (MP) approach, whereas the unsteady analysis is carried out with a sliding
mesh interface and solved with a harmonic balance (HB) method. A duality preserving approach is
used in order to ensure robust convergence of the adjoint equations without any restrictive
assumption on the turbulence viscosity. The two methods are implemented in the open-source SU2
software (Palacios et al., 2013; Economon et al., 2015), whose adjoint has been extended in this work
to compute multi-row HB-based sensitivities. The investigation is performed on an axial turbine
stage for both subsonic and transonic conditions, thus resembling the typical flow characteristics of
gas turbine stages. Further information regarding the method of solution can be obtained at [Rubino
et al.]275.
9.5.6.1 Case Study
The test case considered for the present study is a 2D axial turbine stage, adapted from the 1.5 stage
experimental setup of the Institute of Jet Propulsion and Turbomachinery at RWTH Aachen,
Germany276. The mid-span geometries of the first two blade rows, from the above mentioned setup,
are selected for the subsequent analysis reported in this paper. In order to compare the design
sensitivities, for both unsteady and steady state adjoint computations, the proposed test case is
simulated under subsonic (Case1)
and transonic (Case2) conditions, Parameter Case 1 Case 2
thus resembling the typical flow Stator inlet blade angle 0 0
characteristics of a gas turbine stage. Total temperature 305.8 305.8
The main simulation parameters are Pressure ratio 1.5 1.9
reported in Table 9.3. The Rotational speed 3210 4258
simulations are performed using the Inlet turbulence intensity 5% 5%
Roe scheme for the discretization of
the convective fluxes; second order
Table 9.3 Axial turbine simulation parameters
accuracy is achieved by MUSCL
reconstruction. For both unsteady
274A. Rubino, S. Vitale, M. Pini and P. Colonna, “Assessment of fully-turbulent steady and unsteady adjoint
sensitivities for stator-rotor interaction in turbomachinery”, GPPS-NA-2018-130.
275 A. Rubino, S. Vitale, M. Pini and P. Colonna, “Assessment of fully-turbulent steady and unsteady adjoint
sensitivities for stator-rotor interaction in turbomachinery”, GPPS-NA-2018-130.
276 Stephan, B., Gallus, H., and Niehuis, R.. “Experimental investigations of tip clearance flow and its influence on
secondary flows in a 1-1/2 stage axial turbine”, ASME Turbo Expo 2000: Power for Land, Sea, and Air, pages
V001T03A099--V001T03A099. American Society of Mechanical Engineers, 2000.
219
and steady simulations, non-reflective boundary conditions are imposed277 at the stator inlet and at
the rotor outlet sections. The stator-rotor interface is resolved for the unsteady simulations using a
sliding mesh approach, whereas the steady simulations are based on a conservative mixing-plane
(MP) method278. The κ-ω SST turbulent model is considered with fully resolution of the viscous
sublayer. An unstructured grid is used to discretize the 2D computational domain with about 30000
triangular elements for each blade row and 10000 quad elements over each blade surface in order to
ensure y+ ≈ 1. In this work, the selected objective function (OF) for the calculation of the design
sensitivities is the non-dimensional entropy generation of the stage, defined as
Figure 9.25 Non-Dimensional Entropy Generation Using Unsteady (HB) vs Steady (MP)
277 Giles, M. B.. “Nonreflecting boundary conditions for euler equation calculations”. AIAA journal, 1990.
278 Giles, M.. ”A numerical method for the calculation of unsteady flow in turbomachinery”. Technical report,
Cambridge, Mass.: Gas Turbine Laboratory, Massachusetts Institute of Technology, 1991.
279 Saxer, A. P. “A numerical analysis of 3-D inviscid stator/rotor interactions using non-reflecting boundary
conditions”. Technical report, Cambridge, Mass.: Gas Turbine Laboratory, Massachusetts Institute of
Technology, 1992.
220
Figure 9.26 Case2: Non-Dimensional Pressure Contours for the Mixing Plane (MP) simulation (a) ,
and Harmonic Balance (HB) at Different Time Instances (bcd)
280A. Rubino, S. Vitale, M. Pini and P. Colonna, “Assessment of fully-turbulent steady and unsteady adjoint
sensitivities for stator-rotor interaction in turbomachinery”, GPPS-NA-2018-130.
221
to-static efficiency is only about 0:3%. As opposed to Case1, in this test case shocks occur on both
stator and rotor. The harmonic-balance simulation proved to be able to reproduce the non-linear
flow characteristics associated to the different stator-rotor mutual positions in time. When
accounting for unsteady effects, both stator and rotor exhibit a stronger shock wave intensity
compared to the mixing plane simulation, with the flow discontinuity appearing at a location different
from that provided by the time-average HB solution. Figure 9.26 reports the non-dimensional
pressure contour plot, for both MP and HB simulations. Further details can be obtained from281.
9.5.6.3 Design Sensitivity
As mentioned in the CASE STUDY section, the entropy generation Sgen(U,a) defined by Eq. 9.22 is
selected as objective function for the optimization problem. A set of design variables corresponding
to the control points of a Free-Form Deformation (FFD) box encapsulating the blade profiles, as
shown in [Rubino et al.]282 for a set of twelve variables on the stator surface. For both mixing-plane
and harmonic balance method, the design sensitivities are retrieved from the adjoint solution
according to
N−1
dO ∂O ∂Fn
= + ∑ λTn
dα ∂α ∂αn
n=0
Eq. 9.23
Where O is the objective function corresponding to α as design variables vector, F is the fixed point
iteration operator, and λn is the adjoint solution. This sensitivities correspond to the gradient given
by the total derivative of the entropy generation with respect to the FFD control points, dsgen /da . The
validation between the objective function gradients obtained with the reverse mode of Algorithmic
Differentiation (AD) and the gradients calculated with second-order finite differences (FD), for Case1
is obtained283. The AD and FD gradients, calculated with respect to a representative ensemble of 24
FFD control points enclosing the stage blade rows, are well in agreement. The Root Mean Square
Error (RMSE) lower than 0:004 for both Mixing Plane and Harmonic Balance method. The same level
of accuracy was achieved when validating AD vs FD gradients for Case2. The computational cost and
memory requirements associated with the HB-based adjoint sensitivities are about 2K+1 higher than
the MP-based gradients computations. The CPU time linearly increases with the number of resolved
number of frequencies K because, for the time domain HB method adopted in this work, the system
of equations is solved in a segregated manner for each time instance. From the details about the
derivation of the HB operator it can be deduced that, for K input frequencies, 2K +1 time instances
must be resolved. Here an odd formulation of the time domain HB method is adopted to preserve
numerical stability284. Since, in terms of CPU time, the mixing plane computation can be regarded
approximately as a single time instance resolution, the associated computational cost is 2K+1 lower
when compared to the HB-based method. The memory requirements follow the same considerations
as for the CPU time: for K resolved harmonics, 2K+1 computational domains, relative to each of the
associated 2K+1 time instances, must be considered. As a result, the memory burden attained by the
HB method is 2K+1 higher than the steady-state calculation.
multi-stage turbomachinery simulations using the harmonic balance technique”. In 45th AIAA Aerospace
Sciences Meeting and Exhibit, page 892, 2007.
222
With the aim of preliminary assess whether accounting for unsteady effects can influence the optimal
design in stator rotor interaction problems, the adjoint-based design gradients obtained with the MP
and HB method are computed and analyzed for the two mentioned stage operating conditions, i.e.
Case1 and Case2 (see Table 9.3). Furthermore, in order to identify where the main differences
between MP and HB-based sensitivities occur on the computational domain, the absolute value of the
relative difference dsgen is introduced and defined as
dsgen dsgen
( ) −( )
dα HB dα MP
δsgen =
dsgen
( )
⌊ dα HB ⌋
Eq. 9.24
The values of δsgen are computed for each control point of the FFD box and interpolated, for the rest
of the domain, using a bi-cubic
polynomial response surface.
9.5.6.3.1 Case1 - Subsonic Stage
Figure 9.27 depicts the relative
differences between the MP and the
HB objective function gradients,
relative to the set of 24 design
variables. For the stator blade, this
difference is below 7% whereas, for
the rotor blade, it is as high as 63%. In
the stator Figure 9.27(a), the portion
of the domain close to the trailing edge
is the one showing the highest values
of δsgen. This is possibly related to two
main reasons: (a) Case 1 - Stator
• when compared to the sliding-
mesh interpolation used for
the HB method, the MP leads to
a different static pressure at
the stator-rotor interface,
hence at the stator outlet;
• the unsteady potential stator-
rotor interaction effects are
not taken into account by the
MP.
simplification introduced by the MP simulation, in which the stator wake interaction with the rotor
is neglected by resorting to a mixing process at the blade rows interface. Furthermore, at the rotor
inlet boundary, the MP imposes a pressure about 4% higher than that calculated by the HB method.
9.5.6.3.2 Case2 - Transonic Stage
The adjoint-based sensitivities, relative to the test case configuration characterized by a transonic
flow, show an overall outcome comparable with the subsonic stage: the main gradient differences are
associated to the rotor cascade285. Also in this case, the relative difference on the stator are about
one order of magnitude lower than those on the rotor. From a closer inspection, the deviations
between MP and HB-based sensitivities are lower in magnitude when compared to Case1 but,
differently from the subsonic stage of Case1, they also show a sensible contribution by the control
points located near the suction side and the rear part of the rotor blade. This difference can be
explained by recalling that, for Case2, a shock wave pattern crosses the stator-rotor interface. The
resulting flow discontinuity interacts with the stator wake dissipating the associated velocity defects
before reaching the rotor leading edge. However, the shock interacts with the rotor altering its
pressure distribution along the stream-wise direction. The non-linear variation of the shock location,
related to the unsteady position of the rotor in time, is not captured by the mixing plane steady-state
simulation. This results in high values of δsgen not only near the rotor leading edge, as opposed to
Case2.
9.5.6.4 Conclusions
This work documents an assessment between steady and harmonic-balance unsteady adjoint
sensitivities for turbomachinery design problems involving unsteady effects. In this study, the open
source code SU2 was extended in order to deal with unsteady HB multi-row simulations and the
calculation of the corresponding adjoint-based sensitivities. An exemplary axial turbine stage,
operating at subsonic and transonic conditions, was considered. The adjoint-based gradients were
successfully validated against second order finite differences. Results showed that, compared to
steady state calculations, the harmonic balance sensitivities are about 2K +1 more costly, with K the
number of resolved input frequencies. Memory requirements exhibit the same trend, with higher
allocation needed for the harmonic balance computation. Although the steady calculations accurately
predicted the time-average stage performance, the design gradients, as computed by the mixing
plane and the harmonic balance method, were found to be significantly different for both subsonic
and transonic flow conditions. The areas in which this difference was predominant are located in the
proximity of the stator-rotor interface. Possible reasons for such divergence are:
➢ Different pressure imposed by the mixing-plane method at the stator-rotor interface;
➢ The unsteady calculations are able to capture potential and wake-rotor interaction effects.
The assessment conducted in this work indicates that accounting for unsteady effects in the design
process may lead to a different optimal configuration. In order to confirm the present results, current
efforts are devoted to extend the adjoint-based method presented in this study to the shape
optimization of turbomachinery problems involving stator-rotor interactions.
285A. Rubino, S. Vitale, M. Pini and P. Colonna, “Assessment of fully-turbulent steady and unsteady adjoint
sensitivities for stator-rotor interaction in turbomachinery”, GPPS-NA-2018-130.
224
9.6 Case Study - Stator-Rotor Interaction in Axial Turbine: Flow Physics and Design
Perspective286
Citation : Gaetani, P. (2018). Stator-Rotor Interaction in Axial Turbine: Flow Physics and Design
Perspective
The stator-rotor interaction is an important issue in turbomachinery design when the highest
performances are targeted. Different characters mark the interaction process in high-pressure or
low-pressure turbines depending both on the blade height and on the Reynolds number. For small
blade heights, being the stator secondary lows more important, a more complex interaction is found
with respect to the high blades, where the stator blade wake dominates. In low-pressure turbines,
the stator wake promotes the transition to turbulent boundary layer, allowing for an efficient
application of ultra-high lift blades. First, a detailed discussion of the low physics is proposed for
high- and low-pressure turbines. Some of-design conditions are also commented. Then, a design
perspective is given by discussing the effect of the axial gap between the stator and the rotor and by
commenting the effects of three-dimensional design on the interaction.
9.6.1 Introduction
The design of high efficiency axial low turbine stages has to face many challenging problems, and one
of these is connected to the interaction between the stationary and the rotating rows of the machine.
In high-pressure gas turbines, additional issues related to the combustor turbine interaction take
laces leading to further complexity in the design process. The overall context for the design space is,
in fact, an unsteady and three-dimensional low field, where the Mach and the Reynolds numbers vary
along the machine. High-pressure stages typically operate in high-subsonic or transonic regimes and
are normally affected by shock-induced separation on the rotor crown and unsteady stator rear
loading287. Moreover, the high-loading, combined to the low aspect ratio of the first stage blading,
drives the generation of wide swirling structures, whose mixing contributes significantly to the loss
budget288. These secondary lows also affect the low angle distribution and momentum redistribution
inside the blade channel and their accurate prediction is fundamental for the designer of the gas
turbine cooling system.
All of these low structures affect the blade cascade where they are generated and the adjacent ones
in the so-called stator-rotor interaction process. To make clear such a complex low feature, all of
them will be recalled and schematized according to what are available in the open literature. The
primary low structures involved in the interaction process are the wake and the secondary lows.
Many research studies have been proposed in the open literature discussing the wake and the
secondary low evolution and their parametric dependence on the typical turbomachinery
parameters.
The interaction process has been addressed in the last 20 years by many authors both for the high-
pressure stages and for the low-pressure ones. Differences between high- and low-pressure stages
arise for the dependence of the boundary layer and its transition on the Reynolds number. When the
high-pressure stages are of concern, the interaction takes place mainly in terms of shock wave, wake
and secondary lows, leading to the so-called wake-blade and vortex-blade interaction. Thanks to the
high Reynolds number and high inlet turbulence levels, the blade boundary layer state is less
influenced by the incoming viscous structures. It has to be taken into account that also the inlet
286 Paolo Gaetani, “Stator-Rotor Interaction in Axial Turbine: Flow Physics and Design Perspective”, Chapter 5,
http://dx.doi.org/10.5772/intechopen.76009.
287 Denos R, Arts T, Paniagua G, Michelassi V, Martelli F. Investigation of the unsteady rotor aerodynamics in a
boundary layer properties may cause some pressure fluctuation on the cascade loading, as discussed
in289.
Low-pressure stages, on the contrary, are very sensitive to Reynolds number effects. The wakes
coming from the upstream cascade periodically act as a trigger for the boundary layer transition from
laminar to turbulent conditions. Such periodic transition, possibly re-laminarization, is beneficial in
preventing the boundary layer separation and this allows for higher loading. In this context, ultra-
high lift blade can be proficiently applied either to reduce the aero-engine weight or to power the fan
(among others290-291).
All these issues have been addressed both experimentally and by proper CFD simulations;
experiments require high promptness instrumentation like FRAPP (among others292) or LDV and PIV.
Simulation, as well, requires high performance codes and schemes able to face the sliding of rotors
with respect to the stationary components.
In order to gain a general perspective and to quote the importance of the interaction on the cascade
aerodynamics, the reduced frequency concept has been introduced. It refers to the ratio between the
time scale of the unsteadiness (typically: Ss/U, where Ss is the stator pitch and U is the rotor peripheral
speed) and the one related to the transport of the mass low across the device (i.e. b/Vax where b =
axial chord and Vax = the mean axial velocity component).
The reduced frequency definition then is: f = (bU)/(Ss Vax). When f <<1, the process can be considered
as steady and its time variation related to U can be approximated as a sequence of steady state. When
f >> 1, the process is dominated by the unsteadiness. Finally, when f ≈ 1, the unsteady and quasi steady
processes have the same order of magnitude and importance. In many cases, turbomachinery work
is in the range of f ≈ 1 while for example the combustor-1° stage interaction lies in the quasi steady
conditions293.
In the present contribution, the focus is given mainly to the gas turbines geometries and operating
conditions, even though the same mechanism can be applied to steam stages. As already introduced
by the title, the core of this contribution is devoted to the general discussion of the low physics, rather
than on the quantification and on the detailed description of the specific issues: this way, in author’s
opinion, once the general aspects are acknowledged, the detailed issues; as discussed in papers here
referenced, can be properly understood.
Finally, the discussion will be on a single stage, constituted by a stator and a rotor, taken as a
representative for the whole machine. In the case of multistage turbomachines, the low field
discharge by the rotor will affect, with the same mechanics described in the following, the subsequent
stator. Additionally, there could be some “clocking” features between stators and rotors that may
alter the single stage performance.
Experimental results have been taken by means of a steady five holes probe and fast response
aerodynamic pressure probe (FRAPP) on the high-pressure axial turbine located at the laboratory of
fluid-machinery (LFM) of the Politecnico di Milano. More information on the rig and measurement
techniques reported in various papers. It is important to stress that the FRAPP is applied in a
stationary frame and gives the phase resolved total and static pressure (and hence the Mach number)
289 Hu B, Ouyang H, Jin G-Y, Du Z-H. The influence of the circumferential skew on the unsteady pressure fluctuation
and the low angle; then, by assuming a negligible effect of the temperature fluctuations, the relative
Mach number and, by this, the relative total pressure are calculated. CFD results have been obtained
on the same HP turbine geometry by means of Fluent® code.
9.6.2 Stator-Rotor Interaction in Axial Stages
The stator-rotor interaction features different characters if occurred in high-pressure or low-
pressure stages. In low-pressure stages, thanks to the high blade height, the main interaction element
is the wake in a general low-Reynolds environment. On the contrary, high-pressure stages are
typically characterized by small blade heights, due to the high mean density and by a stream with
high Mach and Reynolds numbers and high mean temperatures. As in all stages, the wake generated
by the stator impinges on the rotor blades being an important source of interaction, but due to the
specific features of HP stages–other sources of interaction are present.
The small blade height has the primary impact of powering the effects of the secondary and clearance
lows; in fact, they cannot be considered as negligible and modifies the potential low pattern for a
large amount of the blade span. From the stator-rotor interaction perspective, this feature makes the
problem much more complex as an additional source of interaction takes place.
A common feature of the different kind of secondary lows is to be connected to loss cores, as found
for wakes. However, secondary lows are also vortical structures and hence characterized by vorticity
whose sense of rotation is different among the different vortices. Therefore, in the analysis of the
interaction mechanism, this last feature has to be properly taken into account.
Mach number typically modulates the intensity and position of the swirling cores and, if supersonic,
sets the shock wave pattern discharged by the cascades. The Reynolds number, typically high and for
this the low can be regarded as turbulent, mainly sets the interaction between the incoming
structures and the rotor blades boundary layers. As mentioned earlier, to aid the reader in the
comprehension, the different kinds of interaction are discussed separately.
9.6.2.1 Stator Wake-Rotor Blade
Interaction
The stator wake can be regarded either
as a velocity defect or a loss filament.
According to the first approach, the
velocity triangle composition shows a
very different direction and magnitude
for the relative velocity. Figure 9.28
shows the triangles for the free stream
and for the wake low; it is evident how
the relative velocity of the wake flow
(WW) heads towards the blade suction
side, featuring also a negative
incidence on the rotor blade.
According to the second approach, the
wake has no streamwise vorticity
associated to it, being the only vorticity
present related to the Von Karman
street, whose axis is parallel to the
blade span. Once the wake interacts
with the downstream rotor blade, it is
Figure 9.28 Velocity triangles for the free stream
bowed and then chopped by the rotor (subscript FS) and the wake (subscript W) lows. V =
leading edge. Later on, it is transported absolute velocity, W = relative velocity, U = peripheral
inside the rotor channel, being velocity
smeared and showing two separate
227
legs: one close to the suction side and one to the pressure side. Globally, the wake is pushed towards
the rotor suction side by the cross-passage pressure field and, possibly, its suction side leg may
interact with the blade boundary layer, this feature depends mainly on the rotor loading. Figure
9.29 shows the wake in terms of entropy filament, as computed by CFD in 2D – 1 × 1 case.
Downstream of the rotor blade, the wake typically appears as a distinct loss core close to the rotor
wake or as a part of
the rotor wake; this
option is strictly
dependent also on
the axial position
downstream of the
rotor where the
analysis is done. For
this reason, in some
papers this
mechanism is
acknowledged as
“wake-wake”
interaction.
Being the rotor blade
different in number
with respect to the
stator one, different
rotor channels
experience the
interaction in Figure 9.29 Pattern of entropy evolution (bowing, chopping and transport) of
different time even the stator wake in the rotor channel, as foreseen by CFD
though the basic
mechanism does not
differ. The rate of the interaction depends on the stator wake intensity, that is, on the stator loading,
on the blade trailing edge thickness, on the axial stator-rotor gap, on the Reynolds numbers and, for
cooled blade, on the kind of cooling applied.
In case of low pressure turbines, where typically the Reynolds number is low, as for the aeroengine
cases, the wake – wake interaction is in fact the only effective mechanism. Its importance grows as
the Reynolds number decreases and specifically, the incoming wakes, once interacting with the
suction side boundary layer, promotes the laminar to turbulent transition. Such a transition, on one
hand increases losses but on the other hand increases the boundary layer capability to face adverse
pressure gradient and for this delaying the boundary layer separation and hence the blade stall.
Thanks to this mechanism, aero-engines low-pressure stages have seen an increase of loading and
for this a reduction of weight, either for a reduction of solidity or overall number of rows 294-295. It
has to be recalled that this mechanism constitutes an aerodynamic forcing on the rotor blade whose
frequency depends on the stator blade passing frequency that is in the rotating frame of reference
the frequency which the rotor sees the stator wake passing ahead.
9.6.2.2 Stator Secondary Lows-Rotor Blade Interaction
The basic mechanism for this kind of interaction is the same of the wake-blade one, the vorticial
294 Ravindranah A, Lakshminarayana B. Mean velocity and decay characteristics of the near and far-wake of a
compressor rotor blade of moderate loading. ASME Journal of Engineering for Power. 1980;102(3):535-548
295 Hodson HP, Howell RJ. The role of transition in high lift low pressure turbines. Effects of Aerodynamic
filament is bowed, chopped and hence transported in the rotor channel. Notwithstanding such
similarity, two main differences can be acknowledged. The vortical structure has its own streamwise
vorticity in terms of magnitude and sense of rotation and for this a different interaction and impact
with the rotor can be expected depending on the entering position in the rotor channel. Moreover,
the vortex entering in the rotor channel is a low structure specifically localized along the blade span
and pitch, whereas the wake is distributed along the span.
It has to be recalled, without aiming at being exhaustive, that different swirling structures can be
acknowledged downstream of the stator, as depicted in Figure 9.30 and discussed in296. The main
ones are the passage vortices, located symmetrically at tip and hub, activated by the pressure
gradient across the passage and hence directed from the pressure side to the suction side. These
vortices have a wide extension but typically low intensities (i.e. vorticities). At the same time, the
presence of the inlet boundary layer activates also the horseshoe vortices, two legs per end wall.
Coupled to each passage vortex, the shed vortex can be found, activated by the interaction between
the passage vortex and the low momentum fluid belonging to the blade wake. The two passage
vortices have opposite sense of rotation. The two horseshoe vortex legs have opposite sense of
rotation between them and the pressure side leg is co-rotating to the corresponding passage vortex.
Figure 9.30 Simplified Schematic of the Secondary Flows System Downstream of a Rotor
Passage and horseshoe vortices start their growing at the stator leading edge and continue it along
the stator channel, possibly merging among them or smearing depending on the stator loading and
on the inlet boundary layer thickness.
The shed vortex, being activated by the viscous transport, starts growing at the stator trailing edge
at the expense of the passage vortex swirling energy, reaches its highest intensity in about half chord
296Langston LS. Secondary flows in axial turbines; A review. 2006. Annals of the New York Academy of Science.
htps://doi.org/10.1111/j.1749-6632.2001.tb05839.x
229
and then weakens due to the viscous stress that smoothens the velocity gradients. Its sense of
rotation is opposite to the one of the corresponding passage vortex.
Tip clearance vortex may be present depending on the sealing geometries of the stator and of the
rotor. Typically, it is located at the hub in stators while it is at the tip in rotors, this later case being
much more important and frequent. Its sense of rotation is opposite to the passage vortex, being
directed from the pressure side towards the suction side across the blade.
It is important to underline that all these swirling lows are present both in stators and in rotors, but
with opposite sense of rotation as a consequence of the different cross pressure gradient versus in
the two channels.
The secondary flow magnitude and position, besides the difference related to the tip clearance, is
different between the hub and the tip. In fact, the radial equilibrium, that onsets due to the tangential
component downstream of the stator, makes the static pressure at the tip higher than at the hub and
for this a higher Mach number at the hub. The effect of the Mach number is well known and primarily
documented by [Perdichizzi]297. Moreover, the pressure gradient acts to diffuse and to shift
centripetally the vortical structures at the tip and to confine close to the end wall the hub ones.
Figure 9.31 Total pressure loss (Y%), streamwise vorticity (Ωs) and absolute Mach number (M)
downstream of the stator. Experiments at the Fluid machinery Lab. at Politecnico di Milano (Italy).
Possible incidence angles to the stator additionally modulate the secondary lows. Positive incidence
angle strengthens secondary lows, as well as lower solidities, as a consequence of the higher blade
297Perdichizzi A. Mach number effects on secondary low development of a turbine cascade. Journal of
Turbomachinery. 1990;112:643-651. DOI: 10.1115/1.2927705
230
Figure 9.32 Rotor inlet low field in the rotating frame of reference. Frame (a) Yloss = total pressure
loss. Frame (b) CPT,R: relative total pressure coefficient. Experiments at the Fluid machinery Lab. at
Politecnico di Milano (Italy).
Once the swirling filament is bended in the rotor channel, the pressure side leg sense of rotation
changes while the suction side leg preserves the original one (Figure 9.33). Moreover, being the
suction side leg accelerated by the overspeed on the rotor section side, its vorticity increases; on the
contrary, the pressure side leg decreases and it is smeared out along its transport.
231
Once the vortical structures enter the rotor channel, they interact both with the passage pressure
field and with the rising vortical structure of
the rotor itself. So far, the stator tip passage
vortices, being opposite to the rotor one, tend
to weaken it (and the same occurs for the hub
ones). On the contrary, the stator shed vortex,
being co-rotating with the rotor passage one
will strengthen it. Swirling lows structure
entering in the rotor close to the end walls will
have stronger effects on the rotor secondary
lows generations; on the contrary, the ones
entering far from end walls will interact in the
downstream portion of the channel.
Moreover, the pressure side legs, as their sense
of rotation is opposite with respect the original
one, will undergo the opposite interaction
features.
As the stator vorticial structures enter the
rotor periodically, with a frequency in the
rotor frame equal to the stator blade passing
one, the interaction process takes places
periodically and this generates a pulsation of
Figure 9.33 Schematics of the stator vortical
the rotor field.
structure transport in the rotor passage
Before discussing in detail the different time
frames, it is straightforward to consider first
the mean flow (Figure 9.34 (a–b)): the CPT,R coefficient is in fact the total pressure in the relative
frame and this evidence the loss cores generated in the rotor and in the stator. The wide low CPT,R
region is mainly due to the rotor wake with some strengthening and enlargement due to the rotor
secondary vortices (tip clearance and tip/hub passage vortex). The vorticial structures can be
acknowledged by making use of the Rankine vortex model applied to the deviation angle map and
Figure 9.34 Time mean flow field downstream of the rotor for a subsonic operating condition (expansion
ratio 1.4, reaction degree at midspan 0.3 and incidence angle close to zero). Frame (a) relative total
pressure coefficient (CPT,R); frame (b) deviation angle (δ). Experiments at the Fluid machinery Lab. at
Politecnico di Milano (Italy).
232
reported in Figure 9.34 b. The clearance low experiences a positive deviation angle as it is less
deflected by the blade than the main low. At the same time, the cross low activated at the hub by the
transversal pressure gradient, generates higher low deflection and for this a negative deviation angle
is found. However, the time mean low field differs from the instantaneous one due to the interaction
process.
The full rotor crown has been calculated by applying a phase leg technique to the experimental
results, measured downstream of the rotor for different stator/rotor phases theory one stator pitch.
It is clearly shown in Figure 9.35 how the 25 channels of the turbine rotor experience different low
conditions, each of them different with respect to the time mean one.
The tip region, being dominated by the tip clearance vortex is weakly sensitive to the periodic low
evolution. On the contrary, the midspan/hub region is strongly periodically pulsating. Being the
stator (n° 21) and rotor blades (n° 25) prime numbers and given that the closest periodicity is around
one-thirds, the pattern evidences a periodicity every 120°.
Figure 9.35 Relative total pressure coefficient on the whole rotor crown. Experiments at the Fluid
machinery Lab. At Politecnico di Milano (Italy).
By considering the total pressure unresolved unsteadiness, calculated as the standard deviation of
the total pressure for each phase and position in the measuring plane , the turbulent structure can be
acknowledged: for this it will be considered as the turbulence (Tu). Some of them are rotor
dependent, like the rotor wake, clearance lows and rotor secondary lows; other structures, on the
233
contrary have a clear periodic evolution with some instant where they do not exist.
Figure 9.36 reports different instants of the rotor evolution for three quantities: the relative total
pressure coefficient, the deviation angle and the unresolved unsteadiness. With respect to the time
averaged low reported in Figure 9.34 (for the same operating condition), the relative total pressure
coefficient shows a fluctuating loss region, with the widest extension at t/BPP = 0.83 and the smallest
one at t/BPP = 0.25, mainly in the hub region. For this later time instant, the deviation angle shows
the smallest gradient in the hub region and the unresolved unsteadiness the lowest intensity.
Figure 9.36 Relative total pressure coefficient (CPT,R), deviation angle (δ) and turbulence (Tu) for 4
interaction phases. Experiments at the Fluid machinery Lab. at Politecnico di Milano (Italy).
234
At t/BPP = 0.37, a vortical structure, evidenced by a high deviation angle gradient, appears in the hub
region and magnifies up to t/BPP = 0.83 where its intensity is the largest. Unresolved unsteadiness
and relative total pressure, mark this structure as a loss core periodically impacting on the rotor
channel. The sense of rotation allows accounting this phenomenon as the impact of the stator hub
shed vortex, strong enough at the stator exit, on the rotor hub passage vortex. The tip region on the
contrary experiences an opposite trend with the strongest vortical structure at t/BPP = 0.25, when
likely the higher inlet total pressure is found.
The highest turbulent and loss contents are found in the tip clearance region due to the high
dissipation related to the clearance low. The interaction here briefly described is unfortunately for
the designer case dependent where, as discussed later one, the axial gap and the loading are
important issues.
To get a comprehensive perspective on the importance and on the region where the interaction takes
place, the standard deviation among the different time fames is straightforward and reported in
Figure 9.37. With reference to the relative total pressure coefficient, the wider fluctuations are, as
qualitatively expected by Figure 9.35 and Figure 9.36 in the midspan-hub region. The more
sensitive region at midspan is located on the wake suction side border that is the place where the
stator wake interacts with the rotor one; in that region, the deviation angle evidences coherently
small fluctuations. The tip region is in fact steady as the clearance low dominates over all other
structures. The hub regions, experience both variation on the total pressure coefficient and deviation
angle, being the seat of the vortex/wake and vortex/vortex interaction. With reference to the
deviation angle map (Figure 9.37 b), the deviation angle experiences the highest fluctuation in the
interface between the tip and hub passage vortex; moreover, all along the rotor wake, it fluctuates,
showing the low turning to be highly sensitive to the periodic interaction.
Figure 9.37 Standard deviation of the Cptr and δ for the different time frames. Experiments at the Fluid
machinery Lab. at Politecnico di Milano (Italy).
calculated among the different time instants, for a negative incidence conditions (Figure 9.38 a,
incidence at midspan = −10°) and a positive one (Figure 9.38 b, incidence at midspan = +10°). When
the overall effect is of concern, the different interaction intensity leaves a trace on the total to total
efficiency that can be summarized by stating that the higher the interaction, the lower is the
efficiency. When the fluid-dynamic forcing on the rotor blade is under study, the frequency of the
forcing event is the one of the stator passing frequency multiplied by the number of swirling
structures found along the pitch. It has to be brought to the attention of the reader that the stator-
rotor interaction is fundamental for the analysis of the interconnection frames, as discussed in and
of the turbine acoustic behavior298.
Figure 9.38 Rotor loading effects on the stator-rotor interaction. Experiments at the Fluid machinery
Lab. at Politecnico di Milano (Italy).
298Knobloch K, Holewa A, Guérin S, Mahmoudi Y, Hynes T, Bake F. Noise transmission characteristics of a high
pressure turbine stage. 22nd AIAA/CEAS Aeroacoustics Conference; Lyon, France; 2016.
236
suction side shock is stronger than the pressure side one. The suction side shock propagates
downstream and interacts with the following row, while the pressure side one impinges on the
adjacent blade, specifically on the suction side, being further reflected downstream.
Across the shock, the low experiences a steep and opposite pressure gradient that, if applied to the
boundary layer, acts to de-stabiles it, leading to separated low bubbles. Thanks to the high Reynolds
number, whose action is to promote the momentum exchange in the boundary layer, the effect is not
that critical in high-pressure stages; it has to be recalled that rarely the outlet Mach number exceed
1.5, value where the entropy rise due to shock starts to be important.
As the stator shock sweep the rotor leading edge region, unsteadiness in the static pressure is found
and for this in the boundary layer evolution; luckily, this happens where the boundary layer
momentum deficit is close to be the smallest at the very beginning of the boundary layer evolution.
As reported by [1, 36–38], the rotor trailing edge region is slightly affected, at least in term of static
pressure and for this the boundary layer and the rotor wake are expected to be almost steady. The
highest interaction is found in the leading edge/suction side region as clearly reported in Figure
9.39; the shock sweeping on the rotor leading edge first interact with the suction side of the blade
(approx. in the location of measuring point n° 6, in Figure 9.39) and then reached the leading edge
(measuring point n° 2).
Figure 9.39 Vane shock-rotor interaction in axial turbine blades. Red: computation, black: experiments.
Adapted from [Denos et al.]
The pressure side is less affected by the interaction being overshadowed by the leading edge. It is
clear how the blade shape, in terms of camber/stagger angles and front/rear loading as well, it is a
key parameter for this class of interaction. The magnitude of the stator shock impinging on the rotor
is strictly dependent on the axial gap; the wider it is, the weaker is the shock effect, being the shock
decay rather fast.
From a mechanical perspective, the forcing induced by the stator shock on the rotor is at the stator
passing frequency multiplied by the number of shocks impinging on the rotor per each stator
passage, even though typically only one is important. Other interesting studies on the interaction in
transonic turbine are, where different conditions and geometries are discussed.
237
Figure 9.40 Standard deviation for the different instants of the interaction phases. (A) axial gap: x/bs =
16%; (B) axial gap: x/bs = 35%, nominal; C) axial gap = 50%. Experiments at the Fluid machinery Lab. at
Politecnico di Milano (Italy).
299Gaetani P, Persico G, Osnaghi C. Effects of axial gap on the vane-rotor interaction in a low aspect ratio turbine
stage. Journal of Propulsion and Power. 2010;26:325-334
238
one. At the same time, the lean change the blade loading along the span by amplifying the tip one and
reducing the hub one: overall such a feature makes the secondary low at the hub less intense than
the case of prismatic blades.
Figure 9.31 shows the vorticity field downstream of the lean annular cascade, characterized by a
positive lean of 10°. When the lean is applied symmetrically at hub and tip, this benefit is gained also
at tip. Overall, the final effect in the frame of the stator/rotor interaction process is the reduction of
the secondary lows and their segregation at the end walls. This design methodology leads to an
overall benefit even though the single cascade does not improve significantly its performance.
9.6.3.3 Cascades Clocking
Cascades clocking refers to the design option related to the proper alignment of blades belonging to
different cascades in the same frame of reference (stator/stator or rotor/rotor) in the context of
multistage machines. In fact, downstream of each stage the “wake avenue’ is found, that is the global
effect of the stator wake and secondary lows on the rotor outlet low field in the time mean context.
This concept can be applied also to the rotor wake, when a multistage environment is considered. To
ease the understanding, let us refer to a two stages machine and specifically to the impact of the 1°
stator wake avenue on the 2° stator. Depending on the kind of stages and their loading,
the impact of the wake avenue can be proficiently used for increasing the following cascade efficiency.
In order to clock the two different rows, cascades should have number of blades that are multiple
each other and for this the design assumption of prime number have to be abandoned (it can be kept
for the stator and rotor of the single stage). So far, the highest efficiency is therefore gained by using
the same blade numbers between the two stators (or rotors for rotor-clocking).
In LP turbines the clocking is directly linked to the wakes, while in transonic HP turbines the effect is
mainly related to the interaction of wakes and secondary lows, that is the total pressure and total
temperature fields on the whole, downstream the first stage with the second stator.
According to the early work from300, the efficiency is achieved when the segments of the first vane
wake avenue, released by the rotor, impinge on the leading edge of the second vane. The basic reason
for this result is that the low momentum fluid coming from the first stage, collapse in the boundary
layer of the second vane and for this do not affect the passage, studied in detail the clocking effects
driven by the stator secondary lows in a two stage subsonic and transonic turbines.
According to [Schennach et al.]301, the interaction with secondary vortices is highly complex due to
the different kind and intensity of the vortical structure itself. When the rotor structures dominate,
as can happen in the tip region due to the tip clearance or for the hub secondary vortex, the clocking
effect is somehow shadowed. The outer part of the channel, being typically the rotor tip passage
vortex highly sensitive to the stator-rotor interaction in the upstream stage, is the place with the
highest potential for the clocking. This result makes the proper alignment choice complex for the
designer as it is not really general.
In case of transonic stages, the hub region, being the seat of the 1° stator shock wave and hence of
the highest stator-rotor modulation, has a high potential for clocking.
As a general conclusion, when the low momentum fluid enters on the 2° stator leading edge or close
to the pressure side, the highest efficiency is found. On the contrary, when the low momentum fluid
coming from the 1° vane enters close to the 2° vane suction side, the lowest efficiency is found, as a
consequence of the destabilizing effects on the suction side boundary layer and the lowest expansion
ratio there available and for this lowest suction side overspeed and for this blade lift.
Low-pressure turbines behavior, where an increase of 0.7% in the efficiency is found by numerical
simulations.
300 Jiang JP, Li JW, Cai GB, Wang J. Effects of axial gap on aerodynamic force and response of shrouded and
unshrouded blade. Science China Technological Sciences. 2017;60(4):491-500
301 Schennach O, Woisetschläger J, Paradiso B, Persico G, Gaetani P. Three dimensional clocking effects in a one
and a half stage transonic turbine. Journal of Turbomachinery. 2010;132(1):011019. 1-10. Inglese
240
As a conclusive comment, the benefit achievable by clocking the cascades can be of the order of 1%
in the 2° stator efficiency, being anyway highly depending on the stage features.
241
Phase 1
field. This should give the same
solution as the original analysis at the
design point. However, the results of
performance testing and, in particular,
measured velocity and pressure
profiles in the vicinity of the blade row
may be used to test the theoretical
analysis and design. Where
differences are found, corrections to
the mathematical models or to the
hardware may be required.
In summary, the design and analysis
of a turbomachine requires an
Phase 2
accurate description of the internal
flow field. A numerical simulation of
the turbomachine, including the
various components of the
equations of motion in either exact
or approximate (empirical) form,
must be constructed. This
simulation consists of two basic
parts, a meridional plane or
through-flow solution and a blade
design method that includes an
analytical blade-to-blade flow
Phase 3
solution. Each of these solutions
affects the other and must be
developed iteratively to produce a
consistent model of the flow. Once a
model is generated, a check against
the desi, “an requirements must be
made to assure that no part of it fails
the design requirements. Once Figure 10.2 Turbomachine Design Process
complete, the model must be
compatible with mechanical limitations and manufacturing capabilities. An additional iteration
with the design constraints may be necessary to finally arrive at an acceptable engineering
solution to the design of the turbomachine. Figure 10.2 illustrated these steps require in
turbomachinery design and analysis302.
302M. V. Casey, “Computational Methods for Preliminary Design and Geometry Definition in Turbomachinery”,
Fluid Dynamics Laboratory, Sulzer Innotec AG, CH-8401, Winterthur, Switzerland.
243
Traditional
Gradient Method
Adjoint Variable
(AV)
Gradient Based
Algorithms
Conjugate Gradient
Method
Quasi-Newton
Techniques
Direct Methods
Genetic Algorithms
Turbomachinery
Evolutionary
Optimization Algorithms
Process Stochastic
Algorithms
Simulated
Annealing (SA)
Inverse Methods Set of Equations
Particle Swarm
Optimization (PSO)
303Zhihui Li, Xinqian Zheng, “Review of design optimization methods for turbomachinery aerodynamics”,
Progress in Aerospace Sciences, July 2017.
244
or adjoint formulations. These methods are efficient and can find a true optimum as long as the
objective function is differentiable and convex. However, the optimization process can sometimes
lead to a local, not necessarily a global, optimum close to the starting point. Furthermore, such
computations can easily get bogged down when many constraints are considered.
Genetic Algorithms and Evolutionary algorithms are typical stochastic optimization algorithms.
These methods are robust optimization algorithms that can cope with noisy, multimodal functions,
but are also computationally expensive in terms of the necessary number of flow analyses required
for convergence. They start with multiple points sprinkled over the entire design space and search
for true optimums based on the objective function instead of the local gradient information by using
selection, recombination, and mutation operations. Figure 10.3 shows a typical optimization
process for turbomachinery application.
10.2.1 Wu’s Pioneering (S1 and S2) Scheme
In the early 1950's, Wu304, as previously recognized, documented these problems and formulated a
set of equations which had the possibility of a solution. He broke the problem of 3D flow into a set of
coupled 2D solutions. Figure 7.1 helps to explain Wu's analysis in which he broke the problem into
two planes (S1 and S2) generally perpendicular to each other. One, the meridional plane, describes
the flow on hub-to-tip stream surfaces. The other, the blade-to-blade solution, describes the flow on
planes generally parallel to the hub surface of the machine and perpendicular to the blading. A
complete solution by Wu's method would require a number of both parallel meridional and parallel
blade-to-blade solutions. The solutions are coupled and must be solved iteratively to simultaneously
satisfy the equations on all of the solution planes (Quasi-3D). At the time of formulation of Wu's
analysis, computational methods and machines were not large or fast enough to give a
comprehensive solution. As a result, many approximate methods evolved. [Wislicenus]305
summarized many of the design techniques in use at the time. Most of these techniques relied heavily
on experimental data to be useful. Smith 306 rearranged the equations of motion in the meridional
plane to give a time and spatially averaged picture of the flow in a blade row. At the same time,
additional computerized techniques were developed to solve the through-flow problem. [Novak]307
formulated a solution (Streamline Curvature Method) that solved for the velocities and streamlines
rather than the stream functions where solution was basically inviscid and non-turbulent. The
problem of losses due to viscosity and turbulence was addressed by Bosman and Marsh308, but in
general, experimental data are always required to adequately model the real fluid effects
encountered in a turbo machine.
10.2.2 Concept of Streamline Curvature Method
The Streamline Curvature Method (SCM) offers an advantage in that the equations and solution are
in terms of physical variables of velocity and pressure rather than those of a stream function, which
previously mentioned. Additionally, viscous and turbulence effects are much easier to incorporate
into the (SCM) because their models are developed in terms of physical variables. Where the 3D
nature of the flow field is required, determination effects of the blading on the meridional flow
requires flow field solutions on the blade-to-blade surfaces. The design of blade sections also requires
304 Wu, C. H., "A General Theory of Three Dimensional Flow in Subsonic and Supersonic Turbomachines of Axial,
Radial, and Mixed Flow Types," National Advisory Committee on Aeronautics, NACA TN 2604, 1952.
305 Wislicenus, C. F., Fluid Mechanics of Turbomachinery, New York, N. Y., Dover Publications Inc., 1965.
306 Smith, L. H., Jr., "The Radial Equilibrium Equation of Turbomachinery," Trans. ASME, J. Eng. Power, 1966.
307 Novak, R. A., "Streamline Curvature Computing Procedures for Fluid Flow Problems," Trans. ASME, J. Eng.
an accurate analysis technique. [Kansan’s]309 was one first to successfully compute the velocity and
pressure distribution on the blade-to-blade plane. None of the above methods incorporates sufficient
modeling of the turbulent boundary layer flow associated with turbomachine blade rows. [Raj and
Lakshminarayana]310 conducted experiments which gave insight into the nature of the blade
boundary layers and the structure of the wake shed from the trailing edge of a blade. These data will
help in the formulation of more accurate models of this flow phenomenon. The availability of the
various through-flow and blade-to-blade solutions leads to the possibility of synthesizing a three-
dimensional model of the turbomachine flow field. The interaction of the flow on the meridional
plane and on the blade-to-blade planes becomes important in this case. The result of blade-to-blade
analysis is that forces due to the geometry of the blading may be determined. [Novak and Hearsey]311
utilized the Streamline Curvature Method in a similar manner to generate a quasi-three-dimensional
analysis. It should be stressed that the above techniques are for analysis of already designed blade
rows and do not apply to the actual determination of a blade shape, i.e., the design problem. The aim
here to address the design problem by using the Streamline Curvature Method to construct an
averaged through flow picture that satisfies the general design requirements. Then two methods, the
Mean Streamline Method and the Streamline Curvature Method, will be used to actually define
blading that generates the flow field prescribed by the through-flow analysis. Combining the
through-flow and the blade-to-blade analysis, a quasi-three-dimensional analytical
representation of the flow field is generated.
309 Katsanis, T., "Use of Arbitrary Quasi-Orthogonal for Calculating Flow Distribution on a Blade-to-Blade
Surface in a Turbomachine." NASA TN D-2809, May 1965.
310 Raj, R., and B. Lakshminarayana, "Characteristics of the Wake Behind a Cascade of Airfoils," J. Fluid
Turbomachinery," Part I and TI, Trans. ASIT4, J. Fluids Eng., March 1977.
312 Benini, E., “Advances in Aerodynamic Design of Gas Turbines Compressors “, University of Padova Italy.
246
➢ Objectives: Maximize Adiabatic Efficiency (η), Maximize stall margin (SM), both at nominal
condition.
➢ Boundary conditions: inlet conditions (pressure P, temperature T), flight Mach number, M,
(in the case of an aero-engine).
➢ Decision (Design) variables: number of stages, compressor and stage geometry parameters.
➢ Functional constraints: mass flow rate, m, (based on engine Power or Thrust requirements),
Pressure Ratio, PR, (from Cycle analysis), correct compressor component matching (i.e.
intake-compressor, compressor-combustor, and above all compressor-turbine) as
determined by a Matching Index (MI).
➢ Side Constraints: each decision variables must be chosen within feasible lower and upper
bounds (sides).
➢ Multi-disciplinary constraints: structural and vibrational, weight, costs, manufacturability,
accessibility, reliability.
attributed to reaching the best trade-off between performance and weight, objectives which are
intuitively conflicting each other, while cost function is inevitably different to the one assigned to the
civil application.
Finally, a constraint based on the static thrust to be delivered by the overall engine at sea level is set
which inevitably influences compressor design. From the problem formulations given above,
remarkable importance is attributed to maximize or minimize some compressor performance
indexes or figures of merit. Therefore, before examining how to deal with such problems, it is worth
analyzing how performance can be significantly affected by the choice in the design variables. For
instance, maximizing adiabatic efficiency requires a deep understanding of the physics governing
stage losses, which have to be minimized either in design and off-design conditions. This, in turn, will
have an important impact on the choice of stage geometrical and functional variables. On the other
hand, maximizing stall margin involves acquiring a proper insight of stall physics and minimizing
stall losses. Again, such problem can be tackled if proper stage geometry is foreseen. Lastly,
minimizing compressor weight (at least from the aerodynamic point of view) implicates reducing the
number of compressor stages and increasing individual stage loading, a fact which ultimately affects
the choice of the blade shape, particularly cascade parameters. Based on the arguments above, in the
following a brief summary of basic and advanced compressor aerodynamics is given313.
Figure 10.4 Sketch of a Compressor Stage (left) and Cascade of Geometries at Mid- Span (right)
313 Benini, E., “Advances in Aerodynamic Design of Gas Turbines Compressors “, University of Padova Italy.
314 Same source – see above.
248
this process, which will be described later on, technological and process constraints, as well as
restrictions on weight and cost, play an important role that must be properly accounted for. In this
framework, some early choices could be revisited and subject to aerodynamic criteria checking, so
that an iterative process occurs until a satisfactory preliminary design is obtained. A second
preliminary step, distinct from the 1D procedure, is the two-dimensional design (2D), which include
both cascade and through flow models, from which a characterization of both design and off-design
multi-stage compressor performance can be carried out after some iterations, if necessary. In this
case, both direct and inverse design methodologies have been successfully applied.
Numerical optimization strategies may be of great help in this case as the models involved are
relatively simple to run on a computer. Often an optimization involves coupling a prediction tool, e.g.
a blade to blade solver and/or a through flow code, and an optimization algorithm which assists the
designer to explore the search space with the aim of obtaining the desired objectives. Finally, a fully
three-dimensional (3D) design is carried out including all the details necessary to build the
aerodynamic parts of the compressor. In this phase, some design intervention is needed to account
for the real three-dimensional, viscous flows in the stages, especially tip clearance, secondary flows
and casing treatment for stall delay. This is usually carried out using CFD models, where the running
blade is modeled in its actual deformed shape, analyzed and, if necessary optimized. While
traditional 3D analyses are aimed at evaluating and improving compressor performance of a single
stage, the recent availability of powerful computers makes the analyses of multistage compressors
an affordable task for most industries. Most advanced CFD computations include evaluation of
complex unsteady effects due to successive full-span rotor-stator interaction315.
315 Benini, E., “Advances in Aerodynamic Design of Gas Turbines Compressors “, University of Padova Italy.
249
blade thickness and distributed frictional force representing the entropy increase due to viscous
stresses and heat conduction can be incorporated. Three methods are basically used for this purpose:
The SCM has the advantage of simulating individual streamlines, making it easier to be implemented
because properties are conserved along each streamline but is typically lower compared to the other
methods. On the other hand, MTFM uses a fixed geometrical grid, so that streamline conservation
properties cannot be applied. However, despite stream function values must be interpolated
throughout the grid, the MTFM is numerically more stable than SCM. Finally STFM is a hybrid
approach which combines advantages of accuracy of SCM with stability of MTFM. These methods
have recently been made more realistic by taking account of end-wall effects and span wise mixing
by four aerodynamic mechanisms: turbulent diffusion, turbulent convection by secondary flows,
span wise migration of airfoil boundary layer fluid and span wise convection of fluid in blade wakes.
As a result of the application of through flow codes, the compressor map in both design and off design
operation can be obtained exhibiting high accuracy319. Remarkable developments in the design
techniques have been obtained using such codes. Among others, [Massardo] described a technique
for the design optimization of an axial-flow compressor stage. The procedure allowed for
optimization of the complete radial distribution of the geometry, being the objective function
obtained using a through flow calculation (see Figure 10.7). Some examples were given of the
possibility to use the procedure both for redesign and the complete design of axial-flow compressor
stages.
10.3.6 Detailed Design Techniques (3D)
10.3.6.1 Direct Methods
Advanced optimization techniques can be of great help in the design of 3D compressor blades when
direct methods are used320. These are usually very expensive procedures in terms of computational
cost such that they can be profitably used in the final stages of the design, when a good starting
solution, obtained using a combination of 1D and/or 2D methods, is already available. Moreover,
large computational resources are necessary to obtain results within reasonable industrial times.
Examples of 3D designs of both subsonic and transonic compressor blading’s are today numerous in
the open literature. For numerical optimization, searching direction was found by the steepest decent
and conjugate direction methods, and it was used to determine optimum moving distance along the
searching direction. The object of present optimization was to maximize efficiency. An optimum
stacking line was also found to design a custom-tailored 3D blade for maximum efficiency with the
other parameters fixed. The method combined a parametric geometry definition method, a powerful
blade-to-blade flow solver and an optimization technique (breeder genetic algorithm) with an
appropriate fitness function. Particular effort has been devoted to the design of the fitness function
for this application which includes non-dimensional terms related to the required performance at
design and off-design operating points. It has been found that essential aspects of the design (such as
the required flow turning, or mechanical constraints) should not be part of the fitness function, but
need to be treated as so-called "killer" criteria in the genetic algorithm. Finally, it has been found
worthwhile to examine the effect of the weighting factors of the fitness function to identify how these
affect the performance of the sections321.
A multi-objective design optimization method for 3D compressor rotor blades was developed by
[Benini, 2004], where the optimization problem was to maximize the isentropic efficiency of the rotor
and to maximize its pressure ratio at the design point, using a constraint on the mass flow rate. Direct
objective function calculation was performed iteratively using the 3D Navier-Stokes equations and a
multi-objective evolutionary algorithm featuring a special genetic diversity preserving method was
used for handling the optimization problem. In this work, blade geometry was parameterized using
three profiles along the span (hub, mid span and tip profiles), each of which was described by camber
and thickness distributions, both defined using Bezier polynomials. The blade surface was then
obtained by interpolating profile coordinates in the span direction using spline curves. By specifying
a proper value of the tangential coordinate of the first mid span and the tip profiles control point with
respect to the hub profile, the effect of blade lean was achieved. Results of tip profiles control point
with respect to the hub profile and the effect of blade lean was achieved. Performance enhancement
severe shock losses (Figure 10.7). Computational time was enormous, involving about 2000 CPU
hours on a 4-processor machine.
10.3.6.2 Inverse Methods
In the last two decades, 3D inverse design methods have emerged and been applied successfully for
a wide range of designs, involving both radial/mixed flow turbo-machinery blades and wings. Quite
a new approach to the 3D design of axial compressor blading has been recently proposed by [Tiow
2002]. In this work, an inverse method was presented which is based on the flow governed by the
Euler equations of motion and improved with viscous effects modeled using a body force model.
However, contrary to the methods cited above, the methodology is capable of providing designs
directly for a specific work rotor blading using the mass-averaged swirl velocity distribution.
Moreover, the methodology proposed by [Tiow], joins the capabilities of an inverse design with the
search potential of an optimization tool, in this case the simulated annealing algorithm. The entire
computation required minimal human intervention except during initial set-up where constraints
based on existing knowledge may be imposed to restrict the search for the optimal performance to a
specified domain of interest.
Two generic transonic designs have been presented, one of which referred to compressor rotor,
where loss reductions in the region of 20 per cent have been achieved by imposing a proper target
surface Mach number which resulted in a modified blade shape. Figure 10.8 comparison of blade
loading distributions of an original supersonic blade, a new design (prescribed by inverse mode), and
a reference blade (R2-56 blade) for a given pressure ratio (left); comparison of passage Mach number
distributions at 95% span. Results
showed that an optimum combination
of pressure-loading tailoring with
surface objective can lead to a
minimization of the amount of sucked
flow required for a net performance
improvement at design and off-design
operations. By prescribing a desired
loading distribution over the blade the
placement of the passage shock in the
new design was about the same as the
original blade. However, the passage
shock was weakened in the tip region
where the relative Mach number is
high.
10.3.7 Concluding Remarks
Continuous effort is currently being
spent in building advanced design
techniques able to tackle the problem Figure 10.8 Comparison of Blade Loading Prescribed by
efficiently, cost-effectively and Inverse Mode
accurately. Plenty of design
optimization techniques has been and are being developed including standard trial and-error 1D
procedures up to the most sophisticated methods, such as direct or indirect methods driven by
advanced optimization algorithms and CFD. Advanced techniques can be used in all stages of the
design. In the field of 1D, or mean line methods, correlation-based prediction tools for loss and
deviation estimation can be calibrated and profitably used for the preliminary design of multistage
compressors. 2D methods supported by either through-flow or blade-to-blade codes in both a direct
and an indirect approach, can be used afterwards, thus leading to a more accurate definition of the
flow path of both meridional and cascade geometry. To enhance the potentialities of such methods,
optimization algorithms can be quite easily used to drive the search toward optimal compressor
configurations with a reasonable computational effort. Detailed 3D aerodynamic design remains
peculiar of single stage analyses, although several works have described computations of multistage
configurations, either in steady and unsteady operations. However, the latter is an approach suitable
for verification and analysis purposes, thus with a limited design applicability. The 3D design
optimization techniques can realistically be used if local refinement of a relatively good starting point
is searched for. On the other hand, if more general results are expected, simplified design methods
are mandatory, such as those based on supervised learning procedures, where surrogate models of
the objective functions are constructed. Other very promising techniques include adjoin methods,
where the number of design iteration can be potentially reduced by an order of magnitude if local
derivatives of physical quantities with respect to the decision variables are carefully computed322.
322 Benini, E.,” Three-Dimensional Multi-Objective Design Optimization of a Transonic Compressor Rotor”.
Journal of Propulsion and Power, Vol. 20, No. 3 (May/June), pp.559-565, ISSN: 0748-4658-2004.
253
10.4 Case Study 2 – Turbine Airfoil Optimization using Quasi 3D Analysis Codes
Turbine airfoil design has long been a domain of expert designers who use their knowledge and
experience along with analysis codes to make design decisions. The turbine aerodynamic design is a
three-step process that is pitch line analysis, through-flow analysis, and blade-to-blade analysis, as
depicted in Figure 10.9. In the pitch line analysis, flow equations are solved at the blade pitch, and
a free vortex assumption is used to get flow parameters at the hub and the tip. Using this analysis the
flow path of the turbine is optimized, and number of stages, work distribution across stages, stage
reaction, and number of airfoils in each blade row are determined. In the through-flow analysis, the
calculation is carried out on a series of meridional planes where the flow is assumed to be
axisymmetric and the boundary conditions of each stage are determined. The axisymmetric through-
flow method allows for variation in flow parameters in the radial direction without using the free
vortex assumption and accounts for interactions between multiple stages. In the blade-to-blade
analysis, airfoil profiles are designed on quasi-3D surfaces using a computational fluid dynamics
code.
The design of airfoil profiles involves slicing the blade on quasi-3D surfaces, designing each section
separately, and stacking the sections together to obtain a sooth radial geometry. The objective of
airfoil design is to define the airfoil shape so as to ensure structural integrity and minimize losses.
The primary sources of losses in an airfoil are profile loss, shock loss, secondary flow loss, tip
clearance loss, and end-wall loss. Profile loss is associated with boundary layer growth over the blade
profile causing viscous and turbulent dissipation. This also includes loss due to boundary layer
separation because of conditions such as extreme angles of incidence and high inlet Mach number.
Shock losses arise due to viscous dissipation within the shock wave which results in increase in static
pressure and subsequent thickening of the boundary layer, which may lead to flow separation
downstream of the shock. End-wall loss is associated with boundary layer growth on the inner and
outer walls on the annulus. Secondary flow losses arise from flows, which are present when a wall
boundary layer is turned through an angle by an adjacent curved surface. Tip clearance loss is caused
by leakage flows in the tip clearance region of the rotor blade, where the leaked flow fails to
254
contribute to the work output and also interacts with the end-wall boundary layer. The objective of
the design is to create the most efficient airfoil by minimizing these losses. This often requires
trading-off one loss versus another such that the overall loss is minimized.
To compute all these losses a 3D viscous analysis is required; however, due to the computational load
of such a code, a quasi-3D analysis code is often used in the design process. Thus the impact of the
blade geometry on 3D losses cannot be determined and only 2D losses can be minimized, that is,
profile and shock losses. A viscous quasi-3D analysis though less computationally intense is still too
expensive for use in design optimization, and an inviscid quasi-3D code is used instead. Consequently,
viscous losses are not computed from the analysis code and airfoil performance is gauged by the
characteristics of the Mach number distribution on the blade surface. The most practical formulation
for low-speed turbine airfoil designs still remains the direct optimization formulation based on 2D
inviscid blade-to-blade solvers. This work automates the direct design process as described in the
next section323.
10.4.1 Parametric Representation of Airfoil Design Process
The parametric representations of the airfoils used in this work are based on the standard design
tools and practices. There are separate
models for the high-pressure and low
pressure turbine blades. The high-
pressure turbine blades are subject to
very high temperatures and need to be
cooled. The parametric representations
of the airfoils used in this work are based
on the standard design tools and
practices. There are separate models for
the high-pressure and low pressure
turbine blades. The high-pressure
turbine blades are subject to very high
temperatures and need to be cooled. As
a result, these airfoils are made thick to (A) High Pressure Airfoil
accommodate cooling passages inside
the blades. For such thick airfoils,
suction and pressure surfaces need to be
manipulated independently of each
other. So the airfoil is represented as a
combination of two separate curves, one
for the pressure side and the other for
the suction side (see Figure 10.10 (A))
Bezier curves are well suited for these
airfoils. Low- pressure airfoils, on the
other hand, have lower thermal stresses,
are much longer, and have a lower speed
of rotation compared to high-pressure
airfoils. These airfoils are usually very
thin and the two-surface model does not (B) Low Pressure Airfoil
work very well as it is very difficult to
vary the pressure and suction surfaces Figure 10.10 Parametric Representation of an
independently and still maintain a Airfoil
323
Goel, Sanjay,” Turbine Airfoil Optimization Using Quasi-3D Analysis Codes”, University at Albany, USA.
255
are minimized. Peak-location is the normalized location of the peak Mach number on the suction side.
It is desirable to have an increasing Mach number as far along on the suction side as possible to
prevent a thickening of the boundary layer. Imposing a constraint on which allows the peak to occur
after 65% of the blade width guards against upstream diffusion and helps in achieving a smooth
accelerating Mach number on the suction side. Inlet-valley-ratio is the ratio of the Mach number at
the inlet of the airfoil on the pressure side, to the minimum Mach number on the pressure side. This
constraint controls the diffusion near the inlet on the pressure side and restricts the thickening of the
boundary layer, reducing chances of flow separation. Constraints are also imposed on:
• The curvature change on the unguided portion of the airfoil (unguided turning)
• The difference between the blades mean line angle and the flow angle at the trailing edge
(over turning), and
• The difference between the inlet angle and the metal angle at the inlet (Δ1).
These additional checks further ensure that the designed airfoil stays within design practice
guidelines324. To ensure mechanical and structural feasibility, constraints are imposed on the blade
geometry. The primary geometry parameters are cross section area, maximum thickness of airfoil,
wedge angle, and nose radius. In cooled airfoils, the constraints on the geometry stem from the
necessity to construct cooling channels in the airfoil; these constraints are dictated by manufacturing
requirements. In low-pressure airfoils,
Design Lower Upper Initial Final
these constraints are primarily driven by
s reassess and manufacturing limitations. Variables Bound Bound Value Value
Most of these constraints have soft limits C 1 0.2 0.5 0.35 0.35
on them; that is, it is best to have the C2 0.25 0.75 0.5 0.632
responses within a given range, beyond a C3 0.25 0.75 0.5 0.569
threshold of the range a penalty that Tmaxx 0.05 0.15 0.139 0.139
increases nonlinearly with increased
Stagger 8 40 31.643 39.464
violation of the constraint is added to the
objective function. The objective function Pcttle -0.25 0.25 0 0
includes the performance metrics and Ratl 0 4 1.25 2.703
constraints where the violations are Ratu 0 4 2.59442 2.727
included via penalty functions. These Ti 0 1 0.5 1
factors can vary for different problems E 1 5 3 2.044
based on the requirements of the specific
problem. The design variables and typical Table 10.3 Airfoil Design Variables
range of variations are listed
Table 10.2 in and the constraints imposed on the problem are listed in Table 10.3. During the
design of an airfoil, multiple sections are designed concurrently, and the objective function is a sum
of the objective functions of all the cross sections being designed. Constraints for all the sections are
also included in the problem formulation. Polynomial fits are used to represent the radial variation
of the design variables; thus the objective function becomes a combination of the coefficients of the
fits across multiple sections rather than individual parameters for each section. A second-order
polynomial fit is used in the formulation; so corresponding to each metric we have three coefficients.
The solution to the problem can be attempted using a variety of optimization techniques including
numerical optimization, genetic algorithms, simulated annealing, and heuristic search. In the current
investigation, the BFGS variable metric method implemented in an optimization code ADS was used.
324 Goel, Sanjay,” Turbine Airfoil Optimization Using Quasi-3D Analysis Codes”, University at Albany, USA.
257
A one-dimensional search technique was used in which the search was bounded followed by use of
polynomial interpolation.
10.4.3 Quasi-3D CFD Analysis and Results
A quasi 3D CFD solver is used in the current investigation to analyze the flow on the airfoil, which is
an isentropic that uses the streamline curvature method that computes the Mach Number/Pressure
distribution on the airfoil surface 325. In the absence of a viscous code, designers usually estimate the
quality of the airfoil by visually examining the Mach number distribution obtained from an in-viscid
quasi 3D CFD solution. Since optimization techniques are driven by a numerical value of the objective
function, and the visual perspective of the designer is the only proven metric available, it must be
captured in a suitable numerical algorithm to provide a measure of quality of an airfoil. The current
work employs curve fitting coupled with design heuristics to compute quality metrics from the Mach
number distribution and the airfoil geometry. These metrics are weighted for different designs based
on individual designer preferences. Primary evaluation metrics that have been defined are diffusion,
deviation, incidence deviation, and leading edge crossover. A physical interpretation of these metrics
is presented below. Diffusion is defined as the deceleration of the flow along the blade surface. It is
measured as the cumulative aggregate of all flow diffusions at each point along the airfoil surface. As
the flow diffuses, the boundary layer thickens, and the momentum loss in the boundary layer
increases. In this case, the increased drag causes a significant loss of momentum; flow separation
may result, causing much larger losses. Thus, the objective of the design is to minimize the diffusion
effect. Since the impact of diffusion on the pressure and suction sides is different, separate terms are
defined for the suction and pressure sides. In the test case presented here, a low-pressure turbine
nozzle is optimized. The flow-path of the low-pressure turbine used in the investigation is shown in
Figure 10.12. The radial distances in the figure are measured with reference to the centerline of
the engine and the axial distances are measured with reference to a point upstream of the first stage
of the turbine. The horizontal lines in the figure represent the streamlines of the flow. Thirteen
streamlines are shown, the top and bottom of which coincide with the casing and the hub
respectively. The vertical lines represent the edges of the blade rows and the location of the frame.
The turbine has six stages, each stage composed of two blade rows. The first blade row consists of
325 Goel, Sanjay,” Turbine Airfoil Optimization Using Quasi-3D Analysis Codes”, University at Albany, USA.
258
nozzles and the second blade row consists of buckets. The stages are numbered from 1 to 6 in the
Figure 5.5.
In the current investigation, stage 5 nozzle was designed using sections from five streamlines equally
spaced along the blade span (hub to
tip). Figure 10.13 Shows the
approximate locations of the
streamlines for an airfoil in which the
first and the last streamlines are shown
at the hub and tip. In reality however
streamlines at 5% and 95% span were
used instead of streamlines directly on
the hub and tip because Mach number
distributions very close to the end walls
are distorted by the end wall effects and
not representative of the flow away
from the walls. The starting solution for
the test case was obtained by
estimating the airfoil shape based on Figure 10.13 Schematics of an airfoil showing
shapes of similar airfoils designed in stream lines along the radial direction
the past. All the Mach number and
airfoil geometry plots use the same reference radial and axial locations as shown in Figure 10.14.
To ensure slope and curvature smoothness of the geometry, second- order polynomials were used to
represent the radial distribution of geometry parameters.
Thus there are three design variables for each geometry
parameter, that is, C0, C1, and C2. These are the coefficients of the
2nd polynomial representing the geometry parameter. The
efficient of the fit match well with the starting design since the
design is based on a previously designed airfoil. Subsequently the
smoothness is maintained since the parameters are not changed
directly but rather the coefficients of the polynomials are varied.
The geometry parameters which describe the low-pressure
turbine airfoil geometry are Stagger, Tmaxx, C1, C2, C3, Ratu, Ratl,
Pcttle, ti, and E. These geometry parameters are varied within
limits typically prescribed in design practice and on the basis of
prior experience and manufacturing limitations. The limits for
these parameters are described along with the results for each Figure 10.14 3D model of
specific test case. an airfoil showing the
passage between adjacent
10.4.4 Concluding Remarks
airfoils
Here we presented a mathematical formulation for design of
turbine airfoils using 2D geometry models and 2D inviscid
analysis codes. The reduced computational complexity of the new formulation compared to 3D
viscous analysis makes the airfoil design problem amenable to the use of formal optimization
methods. The paper presents results from design of a low-pressure turbine nozzle. There are three
primary contributions of this work:
• A numerical metric for emulating designer judgment in evaluation of airfoil Mach number
Distribution.
• An optimization formulation for design of airfoil sections.
259
Designer heuristics are computed using curve fits and error norms. A set of penalty functions has
been defined which allows for flexible constraint boundaries and influence constrained variables
even within constraint limits. In the new approach multiple two-dimensional sections of the airfoil
are designed with constraints on radial smoothness using polynomial fits on the parametric
geometry variables in the radial direction (Figure 10.14)326. Airfoil design is a labor intensive,
repetitive, and cumbersome task for the designers and is a bottleneck for both the design cycle and
rapid generation of inputs for complex multistage analyses. Automating the design process
significantly cuts down the design cycle time and facilitates the task of running multistage analysis
by rapidly generating airfoil geometries. While designing an airfoil, it is hard to establish the
existence of a unique optimum. Multiple evaluation criteria which are weighted together to define
the objective function and the relative importance of these are determined based on designer
experience. Furthermore, the analysis codes are not exact, and even with precisely defined quality
metrics, a significant margin of error remains. In manual design the evaluation criteria are implicitly
considered by the designer, with weighting factors based on past experience and individual biases.
Subjectivity is introduced into the design process since the evaluation criteria for the design are
partially based on heuristics abstracted from designer experiences. Thus in order to completely
understand the results of airfoil optimization, an evaluation of the qualitative changes to the design
is essential after the optimization is completed. Over time as the metrics to evaluate airfoil design
become more acceptable, a standard metric will emerge, till such time designers will need to tinker
with the weights to suit their own preferences 327
326 Goel, Sanjay,” Turbine Airfoil Optimization Using Quasi-3D Analysis Codes”, University at Albany, USA.
327 Goel, Sanjay,” Turbine Airfoil Optimization Using Quasi-3D Analysis Codes”, University at Albany, USA.
328 A. Rubinoa, M. Pini, P.Colonna, T. Albring, S. Nimmagadda, T. Economon, J. Alonso, “ Adjoint-based fluid
dynamic design optimization in quasi-periodic unsteady flow problems using a harmonic balance method”,
https://doi.org/10.1016/j.jcp.2018.06.023.
329 P. Stadtmüller, L. Fottner, “A test case for the numerical investigation of wake passing effects on a highly
loaded LP turbine cascade blade”, ASME Turbo Expo 2001, Power for Land, Sea, and Air, ASME, 2001.
260
conditions, employing the SST turbulence model330. In order to calculate the cascade performance,
the total pressure loss coefficient is evaluated a
330 F. Menter, M. Kuntz, R. Langtry, “Ten years of industrial experience with the SST turbulence model, in:
Turbulence, Heat and Mass Transfer”, vol. 4 (1), 2003, pp.625–632.
331 A. Prasad, “Calculation of the mixed-out state in turbomachine flows”, J. Turbomachinery. 127(3) (2004).
332 A. Rubinoa, M. Pini, P.Colonna, T. Albring, S. Nimmagadda, T. Economon, J. Alonso, “ Adjoint-based fluid
dynamic design optimization in quasi-periodic unsteady flow problems using a harmonic balance method”,
https://doi.org/10.1016/j.jcp.2018.06.023.
261
To verify the HB solution, a second-order time-accurate URANS simulation using the dual time
stepping method is per-formed with a time-step 150xsmaller than the lowest period (1/f1). The total
pressure loss coefficient from this simulation is compared and HB solution obtained with 3, 5, and 7
time instances. The selected time instances correspond to the solution for the frequency vectors ωN3
= [0, ± ω0], ωN5 = [0, ± ω0, ± 2ω0] and ωN7 = [0, ± ω0 , ± 2ω0, ± 3ω0]. The resolved frequencies are,
therefore, multiples of the fundamental blade passing frequency only. The total pressure loss
coefficient, defined in Eq. 10.2, as function of time, and is obtained by spectral interpolation of the
harmonic balance result. The RMSE of the total pressure loss coefficient for the solution obtained
with 5 time instances is equal to 0.010. The harmonic balance solution obtained with 5 time instances
is about 9x faster than the time-accurate solution calculated over a total simulation time of five
periods, which includes the initial transient before reaching convergence to a periodic flow field
solution. 5 time instances are used for shape optimization, as a trade-off between accuracy and
computational cost. In Figure 10.18 - a , b, and c, the Mach number contours from the HB simulation
are reported for 3 different time instances with the simulation period given by T = 1/f 1. The results
show the bar wakes entering the cascade and a separation area occurring at about x/l = 0.7.
10.5.2 Optimization Problem and Results
The shape optimization problem of the cascade configuration is considered. It can be expressed as
⏟
Minimize ζP (𝐔𝐧 , 𝐗 𝐧 , 𝛂)
α
Subject to α out < α out,0 + 4 and δt = δt0
𝐔n < Gn n = 1 , 2, , , , , , N and 𝐗 𝑛 = M𝑛
Eq. 10.4
where the time-averaged total pressure loss coefficient ζP, obtained from Eq. 10.2, is selected as
objective function. Inequality constraints on the absolute exit flow angle (α out) and trailing edge
thickness (δt) are imposed. The optimization is performed using an ensemble of 16 geometrical
Figure 10.16 Shape Optimization History of the Total Pressure Loss Coefficient and Comparison
Between Baseline and Optimized Blade Profile (OptC1)
262
Figure 10.18 Mach Number Contours calculated at Three Different Time Instances with the HB
Method, Based on the OptC1 test case, for both the Baseline (a), (b), (c) and the Optimized (d), (e), (f)
Blade Profile
333 J. Samareh, Aerodynamic shape optimization based on free-form deformation, in: 10th AIAA/ISSMO
Multidisciplinary Analysis and Optimization Conference, 2004.
334 A. Rubinoa, M. Pini, P.Colonna, T. Albring, S. Nimmagadda, T. Economon, J. Alonso, “ Adjoint-based fluid
dynamic design optimization in quasi-periodic unsteady flow problems using a harmonic balance method”,
https://doi.org/10.1016/j.jcp.2018.06.023.
263
convergence of the optimization to the minimum objective is nearly reached after only 7 evaluations,
although satisfying the constraint requires more evaluations. Figure 10.17-b highlights that the
performance of the optimized blade is significantly improved, as the total pressure loss coefficient is
approximately 38%lower, while the constraint on the absolute outlet flow angle is satisfied. The
separation area, as seen in Figure 10.18, is considerably smaller with the optimized blade shape.
The unsteady optimization leads to a decrease in the peak of the total pressure loss coefficient of 44%
and a reduction of 54% of the signal amplitude in Figure 10.17. Furthermore, the objective function
spectrum obtained from a URANS simulation of the optimized blade (Figure 10.17) does not contain
additional frequencies when compared with the baseline configuration.
10.6 Case Study 4 - Using Shock Control Bumps To Improve Transonic Compressor
Blade Performance335
Citation : John, A., Qin, N., and Shahpar, S. (March 2, 2019). "Using Shock Control Bumps to Improve
Transonic Fan/Compressor Blade Performance." ASME. J. Turbomach. August 2019; 141(8):
081003. https://doi.org/10.1115/1.4042891
Shock control bumps can help to delay and weaken shocks, reducing loss generation and shock-
induced separation and delaying stall inception for transonic turbomachinery components, as
described by [John et al.]336. The use of shock control bumps on turbomachinery blades is
investigated here for the first time using 3D analysis. The aerodynamic optimization of a modern
research fan blade and a highly loaded compressor blade are carried out using shock control bumps
Figure 10.19 Tip Speed With Pronounced Negative Camber. From Prince – Courtesy of [Prince]
335 Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
Fan/Compressor Blade Performance”, GT2018-77065.
336 See Previous.
264
to improve their performance. Both the efficiency and stall margin of transonic fan and compressor
blades may be increased through the addition of shock control bumps to the geometry. It is shown
how shock induced separation can be delayed and reduced for both cases. A significant efficiency
improvement is shown for the compressor blade across its characteristic, and the stall margin of the
fan blade is increased by designing bumps that reduce shock-induced separation near to stall. Adjoint
surface sensitivities are used to highlight the critical regions of the blade geometries, and it is shown
how adding bumps in these regions improves blade performance. Finally, the performance of the
optimized geometries at conditions away from where they are designed is analyzed in detail.
10.6.1 Introduction and Motivation
Shocks are a major source of loss for transonic fans and compressors. They cause entropy generation,
boundary layer thickening and shock induced separation. The impingement of the shock on the blade
suction surface (and the resulting, strong, adverse pressure gradient) can cause the boundary layer
to detach, leading to larger blade wakes, reduced efficiency, lower blade stability and reduced stall
margin. Any method that can be used to alleviate shock strength (and the associated negative effects)
therefore has the potential to significantly improve transonic fan/compressor performance.
10.6.2 Shock Control for Turbomachinery & Literature Survey
Relatively little work on designing geometries directly to weaken the shock waves in transonic
turbomachinery components can be found in the literature, though it has been known for some time
that reducing the pre-shock Mach number of transonic compressors can improve their efficiency337.
It was clear to transonic compressor designers in the 70s and 80s that shock strength was increased
by the amount of convex curvature on the suction side between the leading edge and the shock338.
Nearly flat suction surfaces that minimized the expansion were therefore favored, with the next step
to try designs with concave curvature (often referred to as negative camber). Geometries with
negative camber result in gradual compression along the suction surface which weakens the shock.
The concave curvature of the blade surface and the reduction of flow area in the flow direction leads
to a deceleration of the supersonic flow through compression waves, and therefore a weaker shock.
[Prince]339 designed a rotor with pronounced negative camber. This lead to a rise in static pressure
along the suction surface prior to the shock as intended, but the resulting efficiency was
disappointing due to the strong shock on the pressure surface. [Ginder and Calvert]340 had more
success in designing a rotor with negative camber. With negative camber, the Mach number ahead of
the shock was reduced to 1.4 (compared to 1.5 for the traditionally designed blade) which drastically
reduced the amount of boundary layer separation and loss.
Recently, it was demonstrated by [John et al.]341 how the freeform shaping of a compressor blade can
improve blade efficiency by delaying and weakening the shock and reducing separation. The flexible
parameterization method used allowed an s-shaped design to be generated that included a pre-
compression geometry around mid-span. This s-shaped, pre-compression geometry is similar to the
negative camber designs described above. The effect of the pre-compression geometry on the shock
and separation is described in Figure 10.20. The current work proposes the use of shock control
bumps as an alternative method to reduce shock related loss to those described above. Shock control
bumps have the benefit that relatively small modifications to the original geometry are required to
337 Ginder, R., and Calvert, W., 1987. “The design of an advanced civil fan rotor”. Journal of turbomachinery,
109(3), pp. 340–345.
338 Cumpsty, N. A., 1989. Compressor aerodynamics. Longman Scientific & Technical.
339 Prince, D. C., 1980. “Three-dimensional shock structures for transonic/supersonic compressor rotors”.
Figure 10.20 Schematic Of Shock Structures (A) Datum, (B) S-Shaped Design. Courtesy of [John et
al.]
342 Ashill, P., and Fulker, J., 1992. “92-01-022 a novel technique for controlling shock strength of laminar-flow
aerofoil sections”. DGLR BERICHT, pp. 175–175.
343 Drela, M., and Giles, M. B., 1987. “Viscous-inviscid analysis of transonic and low Reynolds number airfoils”.
[Qinet al.]346 first proposed 3D shock control bumps with a finite width, allowing additional design
complexity. They showed that 3D bump configurations were more robust than 2D bump designs
(where a 2D bump is extended continuously along the span).
The only use of a shock control bump on turbomachinery blades found in the literature is by
[Mazaheri and Khatibirad]347, who tested a 2D shock control bump on a (mid-span) section of the
NASA rotor 67 geometry. They added a bump modelled using the Hicks-Henne function. It was shown
how the interaction of the bump with the original wave structure resulted in a more desirable
pressure gradient, with a weaker compression wave fan and a more isentropic compression field.
The bump design was optimized and was shown to reduce the separation area at an off-design
condition. They describe how this may have the potential to improve the stall properties of the blade
section. Two optimizations were carried out, one at the design condition and another at 4% higher
rotational speed. Optimal bumps were produced for each condition, with an increase in efficiency of
0.67% for the on-design case and 2.9% in the off design case reported. The optimized geometry for
the design condition is shown in Figure 10.21.
The work by [Mazaheri & Khatibirad]348 demonstrated the benefit that bumps may provide at both
on and off-design conditions, and their potential to improve stall margin. The simplified 2D analysis
lacks accuracy
however as the
complex
behavior of
radial and
separated flow
cannot be
predicted. For a
thorough
understanding
of the potential
for the use of
shock control
bumps, 3D
analysis and the
design of 3D
Figure 10.21 Datum Geometry and Optimized Shock Control Bumps on The Mid-
bumps is
Section of Nasa Rotor 67- From Mazaheri et al..
needed to truly
assess their effect.
10.6.4 Test Case - NASA Rotor 37
The case studied here is NASA Rotor 37. This has a very strong shock wave (with a relative tip Mach
number of nearly 1.5) which causes large separation, decreasing the blade efficiency. It is a well-
documented case, having been extensively tested and simulated as part of a turbomachinery
validation study. It is a transonic rotor with inlet hub-to-tip ratio 0.7, blade aspect ratio 1.19, rotor
tip relative inlet Mach number 1.48 and rotor tip solidity 1.29. It has historically been a challenge for
CFD simulation. The very high pressure ratio, strong shock wave-boundary layer interaction, large
tip leakage vortex and highly separated flow mean that it poses challenges for turbomachinery
346 Qin, N., Wong, W., and Le Moigne, A., 2008. “Three dimensional contour bumps for transonic wing drag
reduction”. Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering,
222(5), pp. 619–629.
347 Mazaheri, K., and Khatibirad, S., 2017. “Using a shock control bump to improve the performance of an axial
solvers. Rotor 37 has been the subject of review articles that highlight the complexity of matching
experimental and computational measurements and the associated uncertainties. The CFD setup is
shown in Figure 10.22. At the inlet, a radial distribution of total pressure and temperature (based
on the original experimental values) is specified. The inlet turbulence intensity is 1%. At the outlet, a
value for circumferentially mixed-out and radially mean-mass capacity (non-dimensional mass flow)
is used. Periodic boundaries are used to represent full annulus flow. Stationary walls are treated as
adiabatic viscous walls and the rotational speed of the non-stationary portions of the domain is
Figure 10.22 The R37 CFD Domain Used – Courtesy of [John et al.]
1800.01rads1, as specified in the experiment. Rolls-Royce CFD solver Hydra is used for all of the
simulations presented here, using the Spalart-Allmaras turbulence model (fully turbulent). The 4.27
M cell mesh is generated by PADRAM, has y+ of the order of one on all surfaces with 30 cells in the tip
gap.
10.6.5 Validation
As previously alluded to, many studies have struggled when matching simulations of Rotor 37 to the
experiment. A wide range of work has been undertaken to investigate the discrepancy found between
simulation and experiment, with the primary work being the 1994 ASME/IGTI blind test case study
in which a range of codes were used to simulate the rotor, with no knowledge of the experimental
values. A large variation was seen between the different predictions, prompting analysis by
268
[Denton]349. Recent work has also been carried out by [Chima]350 and [Hah]351. The differences are
usually attributed to uncertainty in the experimental measurements, the lack of real geometry in the
simulations (e.g. the upstream hub cavity is usually missing) and also the difficulty in fully resolving
the complex flows. The pressure ratio agreement is reasonable across the characteristic, but the
efficiency prediction is about 2% below the experimental value at the design point (98% choke). This
matches the trend of previous results, where the better the PR prediction, the worse the efficiency
match. This ’trade-off’ has been seen in a range of previous simulations. Figure 10.23 gives the
radial profiles of total PR and efficiency at 98% of simulated choke compared to the experimental
values at 98% experimental choke. The radial trends have been captured fairly well, although there
is an offset from the experiment for both. The choke mass flow found in the simulations was
20.91kg/s, matching quite closely the experimental of 20.93kg/s.
349 Denton, J., 1997. “Lessons from rotor 37”. Journal of Thermal Science, 6(1), pp. 1–13.
350 Chima, R., 2009. “Swift code assessment for two similar transonic compressors”. In 47th AIAA Aerospace
Sciences Meeting including The New Horizons Forum and Aerospace Exposition, p. 1058.
351 Hah, C., 2009. “Large eddy simulation of transonic flow field in NASA rotor 37”. 47th AIAA Aerospace
Sciences Meeting including The New Horizons Forum and Aerospace Exposition, p. 1061.
269
separate and a large wake forms. It is at this design point that Rotor 37 will be optimized, as a
reduction in this separation could significantly increase blade efficiency.
10.6.7 Validation
Due to experimental data for this geometry not being available, simulation validation was carried out
using a similar fan blade geometry that has experimental data available. The related blade has very
similar performance parameters, and the simulation set up is identical. The results are given here. A
comparison of the simulations of this related blade against experimental data can be seen in 352. Both
the pressure ratio and efficiency curves match the experimental data well, though there is a slight
offset to the overall values and stall margin. The radial curves show good comparison to experimental
data, although the radial variation in efficiency is under predicted compared to the experiment.
Overall, the simulation compares well, lying within 1% across the range of flow rates.
Figure 10.24 (a) 3d Separation (Orange) On The R37 Geometry (Flow Right To Left), (b) Rel. Mach
No. Contour At 60% Span – Courtesy of [John et al.]
352 Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
Fan/Compressor Blade Performance”, GT2018-77065.
353 Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
area to investigate the benefit of shock control bumps. It is the shock-induced separation that is
responsible for limiting the operating range of the blade, and if this separation can be reduced then
it is expected that this will extend the stable working range of this fan.
Figure 10.25 Shock Region Flow Features For RR-Fan At Points A) A, B) B, C) C, D) D, E) E, F) F. Flow
Direction – Courtesy of [John et al.]
354 Duta, M. C., Shahpar, S., and Giles, M. B., 2007. “Turbomachinery design optimization using automatic
differentiated adjoint code”. ASME Turbo Expo 2007: Power for Land, Sea, and Air, American Society of
Mechanical Engineers, pp. 1435–1444.
271
efficiency, and therefore if shock control bumps are applied here some benefit should be found. For
complete details, please consult the [John et al.]355
10.6.10 Shock Bump Parameterization & Optimization
The CST (Class Shape Transformation) method is used in this work to define the bump geometries.
The CST method uses Bernstein polynomials to create smooth (second derivative continuous)
contour bumps. For this project 3rd order CST bumps are used, constructed from four Bernstein
polynomials.
Controlling the
weighting (amplitude)
of these polynomials
modifies the bump
height and asymmetry.
Figure 10.26 shows
how the bump
geometry (solid black
line) to be added to the
blade surface is a sum of
the four Bernstein
polynomials (colored
dashed lines). The CST
bump parameterization
provides a high degree
of flexibility, enabling
the generation of Figure 10.26 Example 2d CST Bump (Solid Line) And The Four
Polynomials Used To Construct It (Dashed Lines) – Courtesy of [John et al.]
smooth, asymmetric
bumps in 2D and 3D.
The CST bump parameterization
technique was implemented inside of
the PADRAM geometry and meshing
software. The technique modifies each
2D radial section of the blade geometry,
adding a bump. The properties of these
2D bumps are smoothly interpolated in
the radial direction from control
sections. The resulting geometry is
controlled by the bump start and end
positions, the four Bernstein
polynomial amplitudes and the span-
wise distribution. This allows 3D
variation of the bumps in the radial
direction. Both continuous (where Figure 10.27 a) Example Individual Bump Geometry
bump amplitudes are smoothly And B) Example Continuous Bump Geometry – Courtesy of
interpolated radially) and individual [John et al.]
(where the bump amplitude returns to
zero periodically in the radial direction) CST bumps were tested. Examples of the blade with
individual and continuous bumps added is shown in Figure 10.27.
Alistair John, Ning Qin, and Shahrokh Shahpar, “Using Shock Control Bumps To Improve Transonic
355
During this work, a study was carried out (not detailed here for brevity) to compare the benefit of
using individual bumps (where a series of discrete bumps is added to the datum geometry in the
radial direction) with a continuous bump (note continuous bumps are still ’3D’ and their shape,
position and amplitude can vary in the radial direction). It was concluded that, for these cases, the
individual bumps needed to have greater amplitude than the continuous bumps to offer the same
benefit, leading to increased separation downstream of the bump position. The continuous bumps
tested offered greater benefit, and therefore only results using the ’continuous’ bump geometry
approach are presented here.
10.6.11 Optimization Method
In this work the Multi-point
Approximation Method (MAM) is
used for the optimization studies. It is
a gradient based method that uses
localized Design of Experiments (DoE)
and trust regions to efficiently search
through the design space. When using
MAM, an initial generation of
simulations (chosen by DoE) is carried
out around the start point. A response
surface is constructed for this region
and the sub-optimal point found. The
search is then moved to this point,
where a new generation is Figure 10.28 Spanwise Slice Of The Datum And Optimized
constructed and the process repeated R37 Geometries At 60% Span – Courtesy of [John et al.]
until the search converges on the
optimal design. The MAM method has
been shown to be an efficient and consistent approach for a wide range of highly constrained
optimization problems, working successfully for design
spaces made up of hundreds of parameters.
10.6.12 Rotor 37 Bump Optimization
For the Rotor 37 optimization, the bump geometry was
controlled at 5 radial heights (to allow radial variation of
the parameters) with the geometry smoothly interpolated
between the control stations using a cubic B-spline.
Towards the tip the bump placement and movement range
are increased in chord-wise position as the shock is sat
further downstream at the tip. The initial design used at the
start of the optimization process had bumps positioned
with approximately 60% of the bump downstream of the
datum shock, as is known to be beneficial from previous
work. The objective function for the optimization was blade
efficiency and the simulations were carried out at 98%
simulated choke. The optimizations were carried out on the
Rolls-Royce CFMS cluster using the MAM method. The Figure 10.29 Optimized R37 Bump
geometry of the optimized shock bump can be seen in (Blue) Added To The Datum Blade
Figure 10.28. A slice at 60% span is shown. The 3D Geometry (Grey)
geometry compared to the datum is shown in Figure
10.29.
The bump applied to the datum geometry varies radially, with the maximum bump amplitude and
273
width localized between 40 and 60% span. This makes sense as the strongest shock location, largest
separation and maximum adjoint sensitivity occur around mid-span for Rotor 37, and therefore
greater shock control is needed in this region. The resulting variation from hub to tip of the geometry
demonstrates the benefit provided by optimizing the geometry. Without optimization it would be
difficult to manually specify the bump position, width, amplitude and asymmetry, which would result
in reduced benefit.
10.6.13 Analysis of the R37 Optimized Bump Design
The flow features for the resulting, optimized, continuous bump design is compared to the datum in
Figure 10.30. The datum shock position is shown via a white line on the optimized geometry. It can
be seen how the use of bumps has delayed the shock. The reduction in separation for the optimized
design can be seen in Figure 10.32. The delay of the shock position has reduced the separation
initiation point and the volume of separated flow. The performance of this geometry is compared to
the best individual bumps geometry (not described in detail here) and the datum in Table 10.4. It
can be seen that the efficiency benefit is greatest for the continuous bump design. The efficiency is
increased by 1.48%, while the pressure ratio is also increased. A summary of previous optimization
results for Rotor 37 by various researchers is given by [ John et al.]356. The maximum efficiency
benefit achieved by those studies was around 1.7-1.9% (without decreasing PR). These optimizations
were able to modify parameters such as blade camber, thickness, lean and sweep though, so had
Figure 10.30 Datum (Left) And Optimized (Right) Rotor 37 Static Pressure Contours. Flow Direction
Right To Left – Courtesy of [John et al.]
356John, A., Shahpar, S., and Qin, N., 2017. “Novel compressor blade shaping through a free-form method”.
Journal of Turbomachinery, 139(8).
274
Figure 10.32 Datum (Left) And Optimized (Right) Rotor 37 Separated Flow Contours (Orange). Flow
Direction Right To Left.
275
357John, A., Shahpar, S., and Qin, N., 2017. “Novel compressor blade shaping through a free-form method”.
Journal of Turbomachinery, 139(8).
276
blade designs. The capability of the code was validated by comparing the computed results to
experimental data available in the open literature from probe traverses up and downstream of the
rotor. A multi-objective evolutionary algorithm was used for handling the optimization problem
that makes use of Pareto optimality concepts and implements a novel genetic diversity evaluation
method to establish a criterion for fitness assignment. The optimal rotor configurations, which
correspond to the maximum pressure ratio and maximum efficiency, were obtained and compared
to the original design.
10.7.1 Introduction
In modern high-performance aircraft engines, compressor stages are required to operate with the
highest values of both the efficiency and the compression ratio. This helps minimize the fuel
consumption and decrease the engine weight and size due to the reduction in the number of stages
and their cross-sectional area. Transonic compressors have been developed during the past three
decades to achieve this goal. In these machines, the inner portion of the rotor blade operates at
subsonic relative speeds, whereas in the outer part the flow is supersonic. This is a direct
consequence of the high relative Mach numbers of the flow approaching the blade needed to obtain
a high-compression ratio per single stage.
The main drawback is that the rotors experience the existence of intense shock waves that are
generated close to blade tip and over part of the span. It is well known that shock waves are
responsible for high aerodynamic losses and entropy generation that negatively influence overall
efficiency. At the same time, however, the shock determines a significant increment in the blade
diffusion and, thus, contributes to achieving a high-compression ratio.
The presence of complex shock structures that interact with the main flow and wall boundary layers
explains why the design of such machines is so challenging. In fact, the design principle is not to avoid
shocks (as is the case in subsonic compressors), but to control their locations and strengths to achieve
maximum performance.
Computational fluid dynamics (CFD) is of practical importance for this purpose because it helps the
designer to better understand the features of the flowfield and establish the way the geometry of the
blade has to be designed and, if necessary, modified. Examples of CFD calculations on transonic
compressor blades can be easily found in the open literature 359-360. Navier–Stokes analyses, in
particular, make it possible to deal with the real three-dimensional design problem and to study the
effect of changes in the three-dimensional blade shape. These include both the application of high-
performance airfoils for transonic bladings, as well the use of sweep and lean in blade radial stacking
358 Ernesto Benini, “Three-Dimensional Multi-Objective Design Optimization of a Transonic Compressor Rotor”,
Journal of Propulsion and Power Vol. 20, No. 3, May–June 2004.
359 Miller,D. P., and Bryans, A. C., “The Relative Merits of Inviscid Euler3D and Quasi-3D Analysis for Design of
Transonic Rotors,” American Society of Mechanical Engineers, ASME Paper 88-GT-69, 1988.
360 2Hah, C., and Wennerstrom, A. J., “Three-Dimensional Flow Fields inside a Transonic Compressor with Swept
361-362.
CFD alone can be very useful in the framework of a conventional trial-and-error procedure of a
transonic blade, but it becomes even more powerful when combined with an optimization technique.
This aspect has recently been investigated by several researchers and the results documented in
relevant publications. [Lee and Kim]363 used a gradient-based method to optimize the blades of a
compressor using the three-dimensional Navier–Stokes equations.
[Ahn and Kim]364 developed an optimization technique for the NASA rotor 37 based on the response
surface method. [Oyama et al.]365 used evolutionary algorithms in the redesign of the blade of NASA
rotor 67. The latter was also adopted by [Tiow and Zangeneh]366 as a test case for the application of
a three-dimensional inverse design methodology. [Tiow et al.]367 developed and applied an inverse
design technique coupled with a simulated annealing optimization algorithm in the design of
transonic axial blade cascades.
In all of the cited works, optimization and inverse methodologies were adopted to minimize
aerodynamic losses. In the present work, a multi-objective optimization algorithm is applied with the
aim of optimizing a transonic rotor blade with respect to both the aerodynamic efficiency and the
pressure ratio. To achieve this goal, evolutionary algorithms16 are preferred to deterministic
algorithms in view of their ability to capture global optima and because of their intrinsic capability
to support an optimization problem involving multiple objectives. The paper is organized in the
following way:
First, a brief summary on the influence of three-dimensional blade geometry on performance of
transonic rotors is presented. This is done with the aim of giving a justification about the choice of
the design parameters. Then, the multi-objective optimization problem with regard to NASA rotor 37
is formulated, and the optimization tools used to deal with it are described. Finally, the results of the
optimization are presented and discussed.
10.7.2 Influence of Blade Shape on Transonic Compressor Flows
10.7.2.1 Effect of Blade Profiles
At high-subsonic relative Mach numbers, some areas of supersonic flow appear on both the suction
and the pressure surfaces. These areas are usually followed by shock waves, which interact with the
profile boundary layer and often lead to its separation. Since the development of the first and second
generation of controlled diffusion airfoils368, the knowledge regarding the design of shock-free or
shock-controlled profiles has dramatically improved, and this has led to successful increments in
stage efficiency, loading, and stall margin.
In blade profiles with a supersonic inlet flow, the greatest attention is directed toward the inlet region
361 Prince, D. C. J., “Three-Dimensional Shock Structures for Transonic/Supersonic Compressor Rotors,” Journal
of Aircraft, Vol. 17, No. 1, 1980, pp. 28–37.
362 Wennerstrom, A. J., “Experimental Study of a High-Through-Flow Transonic Axial Compressor Stage,” Journal
of Engineering for Gas Turbines and Power, Vol. 106, No. 3, 1984, pp. 552–560.
363 Lee, S. Y., and Kim, K. Y., “Design Optimization of Axial Flow Compressor Blades with Three-Dimensional
Navier–Stokes Solver,” American Society of Mechanical Engineers, ASME Paper 2000-GT-0488, 2000.
364 Ahn, C.-S., and Kim, K.-Y., “Aerodynamic Design Optimization of an Axial Compressor Rotor,” American
NASA Rotor 67,” Journal of Power and Energy, Vol. 216, No. 3, 2002, pp. 243–255.
367 Tiou, W. T., Yiu, K. F. C., and Zangeneh, M., “Application of Simulated Annealing to Inverse Design of
Transonic Turbomachinery Cascades,” Journal of Power and Energy, Vol. 216, No. 1, 2002, pp. 59–73.
368 Hobbs, D. E., and Weingold, H. D., “Development of Controlled Diffusion Airfoils for Multistage Compressor
Applications,” Journal of Engineering for Gas Turbines and Power, Vol. 106, No. 1, 1984, pp. 271–278.
278
because it fixes the maximum mass flow capacity and produces the majority of the pressure rise 369.
Downstream of the inlet region, the flow is generally subsonic as a result of the presence of a passage
shock, and therefore, the criteria usually employed for subsonic blading may be appropriate there to
guide the designer toward achieving best performance. For a given operating point defined by the
inlet Mach number and incidence angle, the behavior of the inlet region is defined by the loading
distribution, particularly the position and magnitude of the peak load. Recently, this aspect has been
very well described by [Tiow et al.,]370 who demonstrated how a reduction in the peak load and the
shift in its position from the rear to the front result in diminishing the intensity of the shock patterns.
In the rear of the profile, the loading distribution is still regarded as very important and has to be
carefully specified371.
Some recent achievements in the field of transonic compressor bladings, however, have shown that
shock structure and strength cannot be controlled only by using a proper airfoil shape, but also by
taking advantage of the shape of the radial stacking curve, in particular including three-dimensional
lean and sweep.
10.7.2.2 Effect of 3-D Sweep and Lean
The concept of incorporating blade sweep for controlling shock structures and secondary flows has
been largely documented in the literature. In absence of shocks, sweep and lean are known to be
effective in reducing the onset and development of secondary flows within a blade row. [Yamaguchi
et al.]372 tested, in particular, the effect of forward sweep and described a qualitative mechanism for
suppressing secondary flows near the endwall region. They found that in forwardswept rotor blades
the accumulation of low-momentumfluid near the endwall tip region is much lower than in
conventional radial blades due to the decreasing radial migration of fluid particles within the
boundary layer, a phenomenon that follows the imbalance between the centrifugal force and the
pressure gradient.
In transonic compressors, the presence of shock structures further complicates the flow within the
blades. Since the experience of [Prince]373,6 it has been recognized that a three-dimensional-shaped
shock structure, as occurs in swept rotors, may be responsible for reduced shock losses. This would
be the consequence of the formation of a less strong oblique shock wave in the spanwise direction
compared to a high-strength normal shock, which is usually found in conventional transonic blades.
Recently, [Hah et al.]374 carried out an extensive experimental and numerical study on three transonic
rotors (unswept, aft swept and forward swept) and found that the forward-swept rotor had a higher
peak efficiency than the baseline unswept configuration, as well as 30% larger stall margin. However,
the aft-swept rotor showed almost the same peak efficiency as the baseline rotor, but a reduction of
about 40% in the stall margin.
The influence of blade lean in transonic compressor rotors has not been extensively described in the
literature. Results of both experimental and numerical studies refer, in fact, to the effect of lean (often
369 Cumspty, N. A., Compressor Aerodynamics, Longman Group UK, London, 1989.
370 Tiou, W. T., Yiu, K. F. C., and Zangeneh, M., “Application of Simulated Annealing to Inverse Design of
Transonic Turbomachinery Cascades,” Journal of Power and Energy, Vol. 216, No. 1, 2002, pp. 59–73.
371 Benini, E., and Toffolo, A., “Development of High-Performance Airfoils for Axial Flow Compressors Using
Evolutionary Computation,” Journal of Propulsion and Power, Vol. 18, No. 3, 2002, pp. 544–554.
372 Yamaguchi, N., Tominaga, T., Hattori, S., and Mitsubishi, T., “Secondary-Loss Reduction by Forward-Skewing
of Axial Compressor Rotor Blading,” Proceedings of 1991 Yokohama International Gas Turbine Congress, 1991.
373 Prince, D. C. J., “Three-Dimensional Shock Structures for Transonic/Supersonic Compressor Rotors,”
Transonic Compressor Rotors Through Aerodynamic Sweep,” American Society of Mechanical Engineers, ASME
Paper 98-GT-561, 1998.
279
called dihedral) on subsonic linear cascades22 and subsonic annular cascades (bowed stators)375.
However, the use of lean is currently exploited in axial-flowfan rotors. The general guidelines that
result from these studies are difficult to apply to transonic rotors because of the strong interaction
between low-momentum fluid near the endwall and the shock waves. Recently, the use of negative
tip dihedral (where the profile stacking curve is skewed toward the direction of rotation) was
investigated numerically in a transonic rotor and the results showed a positive influence on the
overall rotor efficiency376.
10.7.3 Optimization of NASA Rotor 37
Rotor 37 designed at NASA John H. Glenn Research Center at Lewis Field, was selected for optimization
because it is a welldocumented test case, where strong interactions (such as corner stall,
shock/boundary-layer, tip–vortex, and tip leakage secondary interactions) occur. Even though it was
developed more than 20 years ago, rotor 37 is representative of design and performance levels
of the most advanced transonic
blades used today in gas turbine
inlet stages. Some design
information and overall stage
performance came from Reid and
[Moore]377, and detailed
measurement data were provided
by [Moore and Reid]. The main
dimensions of the rotor are given in
Figure 10.35.
The rotor has 36 multiple-circular-
src (MCA) blades, inlet hub–tip
diameter ratio of 0.7, blade aspect
ratio of 1.19, and a tip solidity of
1.29. The running tip clearance is
0.0356 cm (0.45% of the blade
span). The inner diameter increases
in the mean flow direction, whereas
the outer diameter decreases, and
the blades are stacked in the radial Figure 10.35 Meridional geometry of NASA rotor 37, from
direction in such a way that the flow AGARD
shows typical three-dimensional
features. Design performance at the nominal rotating speed of 17,188 rpm (at International
Standards Organization conditions) as estimated during the design computations are the following:
mass flow rate 20.19 kg/s, total pressure ratio 2.106, total temperature ratio 1.27, and adiabatic
efficiency 0.877. The experimental maximum mass flow rate at the choking condition is 20.93 kg/s.
The flow surveys were placed in stations 1 and 4 [Figure 10.36, (see378 Ref. 30)]. Cobra probes were
used for total pressure and flow angle measurements, with a thermocouple for total temperature
data; wedge probes were used for static pressure, and wall static pressure taps were employed on
the hub and tip walls.
375 Wisler, D. C., “Loss Reduction in Axial-Flow Compressors Through Low-Speed Model Testing,” Journal of
Engineering for Gas Turbines and Power, Vol. 107, No. 2, 1985, pp. 354–363.
376 Ahn, C.-S., and Kim, K.-Y., “Aerodynamic Design Optimization of an Axial Compressor Rotor,” American
Figure 10.36 Measurement Stations within NASA Rotor 37, from AGARD
The purpose of the optimization considered herewas to maximize the two-objective function:
p04
𝐅 = (ηis , )
p01
Eq. 10.5
where ηis is the adiabatic efficiency and p04/p01 the total-tototal pressure ratio, both referred to
stations 1 and 4 of Figure 10.36. The optimization was conducted for one mass flow condition (m˙
/m˙ choke =0.98), this being one of the cases for which many of the experimental and computational
results are available to the public. The inlet total pressure and total temperature were fixed at p01
=101325 Pa and T01 =288.15 K. The optimization was carried out using an optimization method that
integrates a code for three-dimensional blade geometry parameterization, a Navier–Stokes solver,
and an optimization algorithm. A detailed description of this method, along with examples of its
application to design optimization of turbomachinery blades may be found elsewhere 379-380-381.
10.7.3.1 Blade Geometry Definition
To make the results of the optimization comparable to those regarding the baseline configuration,
the meridional contours of the hub and casing were not modified. Actually, a change in the meridional
area would have had a strong impact on the aerodynamic blockage and, therefore, on the compressor
379 Benini, E., and Toffolo, A., “Development of High-Performance Airfoils for Axial Flow Compressors Using
Evolutionary Computation,” Journal of Propulsion and Power, Vol. 18, No. 3, 2002, pp. 544–554.
380 Benini, E., “Optimal Navier Stokes Design of Compressor Impellers Using Evolutionary Computation,”
of Power and Energy, Vol. 215, No. A4, 2001, pp. 465–473.
281
flow capacity.
The rotor blade geometry was
parameterized using three profiles
along the span (hub, midspan, and tip
profiles), each of which was described
by camber and thickness distributions
(Figure 10.37). These were defined
by fourthorder Bezier polynomials,
where only the values of the ordinates
of the control points (θ for the camber
line and t the thickness) were allowed
to vary as independent design
variables.
The blade surfacewas then obtained
by the interpolation of profile
coordinates in the span direction by
use of spline curves. When a proper
value of the θ coordinate of the first
midspan and the tip profiles’ control
point with respect to the hub profile
were specified the effect of blade lean
was achieved. The use of blade sweep,
on the other hand, which could be
obtained in principle by giving a
different z coordinate to the three
profiles,was not investigated. The
results of preliminary calculations
indicated that, because the casing
contour is fixed here, a forward-swept
rotor would necessarily have a greater
diameter and, therefore, would lead to Figure 10.37 Parameterization of a Compressor Airfoil
higher aerodynamic losses caused by
higher incidence tip Mach numbers.
For the same reason, the chord distribution along the span was not changed compared to the original
design.
A total of 14 parameters for the camber lines plus 9 parameters for the thickness distributions (the
leading-edge and trailing-edge radii were not changed), that is, 23 parameters in total, were used to
describe the three-dimensional shape of the rotor. Each parameter was given a range of variation. To
avoid the creation of rotors having very different mass flow rates from that of the baseline
configuration, the code that handled the blade parameterization calculated the geometric throat area
between adjacent blades: Only the geometries that gave throat areas in the range of ±0.2% with
respect the original geometry were then simulated, and the others were disregarded and eliminated
before processing.
10.7.3.2 Flow Solver: Description and Validation
The CFD code CFX-TASCflow® was used to calculate the flowfield around the rotor, where the 3D
Reynolds averaged Navier–Stokes equations (RANS) are solved using a finite-elementbased finite
volume method. An algebraic multigrid method based on the additive correction multigrid strategy
was used along with the second order skew upwind differencing scheme with physical advection
correction.
282
The code was first validated against experimental data provided by [Moore and Reid]. Amultiblock
structured grid of about 240 K nodes per single passage (Figure 10.38) was adopted following the
guidelines provided in a recent AGARD report. The flow region close to the profile wall was
discretized using an
O-type grid, whereas
the outer part was
meshed using an H-
type grid. The effect of
the tip clearance was
also modeled. The κ-ε
turbulence model35
along with
standardwall
functions were
employed. Thewalls
were treated as
smooth and adiabatic.
The boundary
conditions were fixed
as follows: At the
domain inlet, the total
pressure, total Figure 10.38 Multiblock Grid Used in the Simulations
temperature, and flow
angle were imposed; at the outlet, average static pressure was applied for both near-stall and near-
choke conditions; periodic boundary conditions were imposed on the lateral faces of the flow domain.
An angular velocity corresponding to the nominal rotational speed was applied to the rotor. The rotor
performance was calculated
over the entire operating range.
For each simulation, the
convergence criterion was
established when normalized
rms residuals were less than
5×10−7. Each simulation took
about 4 hours to complete on a
Workstation AlphaServer ES40.
Results of the code validation are
given in [Benini] 382. With regard
to overall performance, both the
adiabatic efficiency and pressure
ratio were slightly
underestimated in all of the
operating conditions. In
particular, the pressure ratio Figure 10.39 Comparison Between Calculated and Experimental
seemed to have a dominant Mach Number Contours at 90% Span and m˙ /m˙ choke = 0.98
effect on the overall efficiency, as :Maximum Mach = 1.576, Minimum Mach = 0, and Contour Interval
= 0.031.
suggested by the radial plots,
based on pitch-averaged data at
Evolutionary algorithms384 are optimization techniques that use both stochastic and deterministic
elements, where an artificial evolution process that imitates the natural evolution of biological
organisms is implemented. The evolution process starts with a randomly initialized population of
individuals (a set of points in the search space) that evolves following the Darwinian principle of the
survival of the fittest. According to this approach, new generations of solutions are created using
some simulated evolutionary operators, such as crossover and mutation: The probability of survival
for each individual depends on its fitness, that is, on how well it performs with respect to the
objective(s) of the optimization problem.
As a result, evolutionary algorithms are very well suited to deal with multi-objective problems
because they make use of an evolving population of solutions that is driven toward the set of the true
tradeoff among the objectives, the Pareto optimal set. For the same reason, they can be beneficially
applied to highly multidimensional problems, where an effective exploration of the design space can
be carried out only using populations with several individuals.
Here, the structure of the evolutionary algorithm follows the main steps of an evolution strategy and
has been described in detail in previous papers. After the parents are selected, reproduction using
uniform crossover and mutation are performed to generate offspring that are then evaluated. The
evaluation step includes reconstruction of rotor geometry from actual decision variables, CFD
analysis with postprocessing, and ranking according to the usual Pareto concepts385.
Then a genetic diversity evaluation method (GeDEM) is applied to establish a criterion for fitness
assignment and to build the next population of parents. In short, the GeDEM preserves genetic
diversity of the best-so-far population of candidate solutions to the optimization problem by
performing an additional evaluation after the common measure of objective fitness. This evaluation
ranks the solutions according to their fitness value and their reciprocal distance as a way to give more
reproduction chances to both highly fit and highly distant individuals. The loop starts again until the
predetermined number of generations has elapsed.
10.7.4 Results
During the optimization run, a population of 20 individuals evolved for a total of 100 generations.
Computations were performed in parallel on a four-processor Workstation AlphaServer ES40 and
the overall turn around time was about 2000 h. Because of time constraints, it was not possible to
continue the computations further. A mass flow rate boundary condition at the outlet was applied to
each configuration examined during the evolutionary process to fix the condition m˙/m˙ choke = 0.98.
Not shown here (see [Benini]386), the results of the optimization, that is, the performance of
optimized configurations with respect to both maximum efficiency and maximum pressure ratio.
These configurations correspond to the borders of the final Pareto front obtained after the
optimization. In fact, in the objective function space, the performance of the optimized individuals
was very much clustered around the original as a result of the tight constraint imposed on the mass
flowrate, which prevented the search algorithm from generating and evaluating geometries very
different from the original design.
At the chosen optimization point, an improvement in the adiabatic efficiency was achieved (+1.5%
with respect to the original geometry) without modifying the pressure ratio. On the other hand, an
optimal individual was obtained that showed a higher pressure ratio (+5.5%) with a slightly smaller
efficiency (−0.8%) compared to the original design. This behavior was observed over the entire
operating range of the compressor. In fact, for nearly the same pressure ratio, the efficiency-
optimized (E-O) rotor definitely showed a superior efficiency close to the choking condition (ηis =
+2%). Moreover, the operating range was very similar to the original one. The pressure ratio-
384 Deb, K., Multi-Objective Optimization Using Evolutionary Algorithms, Wiley, Chichester, England, U.K., 2001.
385 Deb, K., Multi-Objective Optimization Using Evolutionary Algorithms, Wiley, Chichester, England, U.K., 2001.
386 Ernesto Benini, “Three-Dimensional Multi-Objective Design Optimization of a Transonic Compressor Rotor”,
optimized (PR-O) rotor had instead a smaller operating range compared to the original one. The last
computation for which the CFD code was able to reach convergence occurred at m˙/m˙ choke = 0.951,
which perhaps denotes the stall limit.
The geometries of the optimized configurations are compared with the original design in Figure
10.40. The main differences could be found in both profile shape and radial stacking. Changes in the
profile shape concerned the region close to the trailing edge, as one might expect, because the
constraint on the geometrical
throat area led to cascade
geometries having very
similar inlet regions.
The E-O rotor had profiles
(particularly the one located
close to blade tip) with an
increased thickness toward
the rear and a different
curvature. Furthermore, the
blade profiles leaned
significantly in the direction
of rotation. This result
confirms the one obtained by
Ahn and Kim,12 even if the
amount of leaning was
considerably higher here
(+1.5 deg compared to +0.22
deg). The impact of this lean
on the structural strength of
the blade should, therefore,
be checked carefully. The PR-
O blade was characterized by
higher cambered profiles
toward the rear, with no
noticeable changes in the
thickness distribution
compared to the original
design. (An exception was the
profile at midspan, where the Figure 10.41 Mach Number Contours at 95 and 50 % Span of
maximum thickness is Baseline and Optimized Geometries
slightly lower). Again the
blade leaned substantially in the direction of rotation, even if in a less apparent way compared to the
E-O blade.
The effect of blade shape changes on rotor performance can be better understood by examining the
contours of the Mach number reported in Figure 10.41. From Figure 10.41, it appears that in the
E-O blade the shock intensity was reduced close to blade tip as the shock wave moved from a nearly
normal to a much more oblique pattern.
Actually, the normal shock wave at the tip was substituted by two oblique shocks of lesser intensity,
whereas at midspan the shock still weakened and became more oblique to the incoming flow. This is
also confirmed by Figure 10.42, where a comparison of the Mach number contours near the suction
surface for the three blades is given.
286
In the original configuration, a strong shock wave occurred within the blade passage that turned
normal to the casing and led to high aerodynamic losses and severe shock/boundary-layer
interaction.
Figure 10.42 Mach Number Contours on The Suction Surface of the Baseline and Optimized Blades
In both the E-O and PR-O blades, the shock bifurcated into two lesssevere branches and almost
vanished into two shock/boundary-layer interaction zones. In the E-O blade, this led to higher
efficiency, whereas in the PR-O blade the concomitant effect of an increased blade profile curvature,
which helped to achieve a high-pressure ratio, resulted in a more evident boundary-layer separation
toward the rear and, therefore, in a lesser overall efficiency. In addition, this is the reason why the
PR-O blade achieved a smaller operating range: The blade was no longer capable of effectively
withstanding flow deviation at reduced incidence angles without incurring in massive separations
(stall) toward the rear.
10.7.5 Conclusions
Amethod for three-dimensional multi-objective optimization of a transonic rotor blade was
developed and tested which was based on an evolutionary algorithm and a Navier–Stokes code. The
method was applied to the design optimization of NASA rotor 37 with the aim of achieving maximum
efficiency and maximum pressure ratio with a constraint on the mass flow rate. The rotor blade was
287
described using three profiles along the span, each of which was defined using parametric curves.
The effect of blade lean was considered by changing the mutual tangential coordinates of the three
profiles.
The optimization run was carried out on a multi-processor computer and demonstrated that the
overall adiabatic efficiency can be improved by approximately 1.5% (without changing the pressure
ratio in a significant way) by giving the blade a proper lean toward the direction of rotation and by
slightly changing the profile shape, especially toward the tip. This improvement followed from a
drastic modification in the shock structure within the blade passage.
The results also showed that the improvement in the overall efficiency, achieved in one operating
point, is maintained at off-design conditions. The results also showed that the pressure ratio can be
improved by about 5.5% by paying for a small efficiency drop (−0.8 %). This was achieved by leaning
the blade in the direction of rotation and by slightly increasing the profile curvature toward the rear
to assure a subsonic diffusion. In this case, however, the presence of a shock wave, although less
intense, accentuated the interaction between the shock and the boundary layer on the rear of the
suction surface, a phenomenon that possibly determined a reduction in the operating range of the
compressor.
387Tariq Alsawy and Ahmed Farouk AbdelGawad, “Effect of Twist of a Wide-Chord Fan-Blade on the
Aerodynamic Performance of the Fan of a High-Bypass Turbofan Engine”, ICFD13-EG-6S01.
288
showed that both forward and backward sweeps have different effects on the secondary flow pattern
and shock wave structures, depending on the baseline case. [El–Sayed and Ibrahim] studied the effect
of different tip clearance shape on performance of an axial-flow compressor. They found that
decreasing tip clearance positively affects engine performance and surge margin.
Moreover, they found that linear-expanding gap leads to increase in both pressure ratio and
efficiency of the compressor stage compared with the linear-shrinking gap. [Banks et al.] investigated
the effect of tip clearance on the performance of a high-pressure shrouded turbine rotor blade. They
stated that shroud significantly affects the blade-tip flow field. Moreover, the integrated shroud
increases the turbine stage efficiency and decreases its sensitivity to clearance. [Karrabi and
Rezasoltani] investigated the effect of lean, twist, and bow of the blade on the performance of axial
turbines. Their results showed that turbine blade-twist clearly affects the performance, while sweep
and bow have minor effects on the performance.
However, there is little information in the literature regarding the three-dimensional flow features
of fan blades, especially, the studies that involve the influence of blade geometry on the fan
performance. [Frohnapfel et al.] experimentally measured the response of a fan rotor operating with
continuous inlet swirl distortion. Worst case scenario flow pattern associated with aggressive
aircraft maneuvers was tested. They demonstrated that the swirl distortion continues after
interaction with the fan
rotor which operates the
downstream cascade on
their off-design conditions.
[Han et al.] studied the
effect of casing shape on a
transonic fan performance.
It was found that the
shrinking case improves
the peak efficiency.
Convex shrinking case
reduces losses in tip area
for all operating
conditions, whilst concave
shrinking case reduces
losses in tip area at large
mass flow points and has
the opposite effect at small
mass flow points. [Hisse et
al.] numerically simulated
the shock wave
propagation in turbofan
engines intakes. [Hall et al.]
used a genetic algorithm
and computational fluid
dynamics to optimize the
design of the turbofan inlet
duct. [Chai et al.] studied
the tip clearance of wide-
chord fan-blade of a high-
bypass turbofan engine. Figure 10.43 Different Cases of Blade Twist
Their results indicated
that, after certain fan
289
angular acceleration, blade deformation occurs, which can cause sever rubbing or impact between
blade tip and the casing.
Less research investigations are available regarding the geometry of the modern wide-chord fan
blade and its effect on fan performance.
Thus, this paper presents a 3D transient CFD study of a 3D modelled fan of a modern high-bypass
turbofan engine, Rolls Royce Trent XWB, rotating at its reference 100% speed (2700 rpm) at take-off
conditions [11]. The parameter of interest of this paper is the twist of the fan-blade and its effect on
the fan aerodynamics and performance which directly affect the engine performance.
10.8.2 Numerical Models and Technique
10.8.2.1 Model Construction
A CAD model of a wide-chord fan of a modern high-bypass turbofan engine, Rolls-Royce Trent XWB,
is constructed using SOLIDWORKS software. The CAD geometry is constructed similar to the realistic
3D model of Trent XWB engine released by Rolls-Royce plc [12]. Main dimensions of the blade (tip,
root and blade sections necessary to build the model) were obtained from the model using “ImageJ”
software. The profile of the constructed blade was approximated to match the realistic 3D model
using multiple blade sections between the root and tip of the constructed blade.
The different studied cases of blade twist are shown in Figure 10.43. The approximated CAD model
used in this work is shown in Figure 10.43c. It has a blade tip angle of 65° and will be used as a
baseline case for this work. Varying the twist, for the studied cases, is done by changing the tip blade
angle, while holding the blade chord length and the
general blade shape constant for all cases. At all blade Number of blades 22
sections, the blade angles are set as a ratio of the tip angle Fan diameter 3m
at the baseline case (65°). Fan speed 2700 rpm
Changing the tip angle by certain ratio, will change all Root chord length 0.47 m
blade section by the same ratio keeping the general blade Tip chord length 0.64 m
shape approximately the same. Using this method results
in isolating the blade twist from other geometric Table 10.5 Approximated model
confounding factors. So, the results obtained from the dimensions and specifications
study for different cases are proportional to the blade
twist only. Main parameters of the approximated blade model is shown in Table 10.5. In this work,
in addition to the baseline case studying, blade twist angles of 45°, 55°, and 75° are studied as shown
in Figure 10.43a, Figure 10.43b, and Figure 10.43d for the 45°, 55°, 75° twist, respectively.
10.8.2.2 Domain Construction and Meshing
Due to the similarity of the fan blades and to save computational effort and time, only one blade is
considered. ANSYS DesignModeler© is used to build the flow domain for a single blade passage. This
is achieved by implementing FlowPath and ExportPoints features in ANSYS DesignModeler. These
features allow the domain to be exported to TurboGrid to generate the computational grid (mesh). A
single blade passage with a structured H-type grid is generated by TurboGrid. Number of elements,
which is held constant for all cases, is 0.6 million elements for single blade passage. Mesh for the
baseline case is shown in Figure 10.44. As it is clear from Figure 10.44 b, careful consideration
was paid to achieve good meshing, with the finest possible resolution, for accurate results. The
technique of periodic domain is applied with the appropriate inlet and outlet surfaces, Figure 10.44
c, Figure 10.44 d.
10.8.2.3 Numerical Scheme
The CFD solver used to obtain the unsteady RANS solution is ANSYS FLUENT [3] [9] [13] [14]. Due to
the high rotational speed (2700 rpm) of the fan in this work, air is modelled as compressible low. k-ε
model is widely used for turbulence modelling in the numerical analyses of turbomachines [3] [7]
[15] [14]. Thus, for this study, the Realizable k-ε turbulence model is used. Realizable k-ε model
290
outperform both RNG and standard k-ε models in this type of numerical analysis as it provides
superior performance for flows involving round jets and flows under strong adverse pressure
gradients, separation, and recirculation [16]. Enhanced wall treatment is chosen to be the near-wall
model as it combines both the two-layer model with enhanced wall functions. Compressibility effects
correction is added to the used turbulence model.
Sliding mesh technique is used to simulate the rotation of the fan blade. This is mainly because sliding
mesh provides relatively accurate solution compared to other methods such as MRF (Multiple
Reference Frame) [13] [16]. Only one mesh zone exists in this work, which is a rotating mesh zone.
The rotational speed of this zone is set to be the same reference 100% speed of the Trent XWB which
is 2700 rpm.
A Pressure Inlet boundary condition is used at the inlet, where the total pressure is set to be
atmospheric pressure and total temperature is assumed to be 300 K at take-off conditions. As an
initial value, static pressure at inlet is set slightly less than atmospheric pressure to provide more
(c) Periodic domain isometric view, (d) Periodic domain meridional view,
showing inlet (green) and,outlet (red). showing inlet at left and outlet at right
stable initialization for the solution. The outlet of the domain is set to be Pressure Outlet which
requires static pressure input. For every studied case, the static pressure of the outlet is changed to
obtain different operating point for the blade. Both blade and hub are set as a moving wall with a
speed of 0.0 rpm relative to the rotating domain. Shroud wall is set to be a stationary wall, which act
as a counter rotating wall to the rotating domain. Rotating periodic boundary condition is applied at
both sides of the rotating zone.
10.8.2.4 Methodology and Time Stepping
Momentum, continuity, turbulence, and energy equations are discretized using Second-Order
scheme. However, Transient formulation is discretized using Bounded Second-Order scheme, as this
scheme provides the same accuracy of the second-order, but it is more stable [16]. To ensure stable
and accurate solution for this kind of complex flow domain, solution is done over gradual steps. All
equations are set to First-Order discretization scheme. Solution is initialized using a low speed (100
rpm) and gradually increased to the full speed (2700 rpm). When reaching the full speed,
discretization schemes are gradually changed from First to Second-Order.
Time step size is chosen to be small enough to attain acceptable accuracy while consuming
reasonable computational time. Time step size is set to be 2.5×10-5 s. So that, the rotor blade crosses
full blade passage in approximately 40 timesteps and perform full revolution in approximately 880
timesteps [14].
Solution monitors are monitoring total pressure and mass flow rate at inlet and outlet boundaries of
the domain. Solution is aborted when monitored variable reaches to an approximately constant value
and becomes approximately time independent. If, for different operating conditions, solution
fluctuation starts to occur in a small amplitude, the solution for this point is taken as the mean of this
fluctuation. Moreover, the point is marked as a limit of the operating range of the tested case on which
the blade starts to stall.
10.8.3 Numerical Results
10.8.3.1 Performance of All Cases
The isentropic efficiency is defined as the isentropic compression work divided by actual total work
and is calculated by [7]:
𝛾−1
p 𝛾
(p02 ) −1
01
η=
T
(T02 ) − 1
01
Eq. 10.6
All the obtained results are for one blade passage. Starting from case of 65o twist as a baseline case,
the performance of all the four cases is shown in Figure 10.45. It can be seen in Figure 10.45 a that
increasing the twist angle to 75° decreases both mass flow rate and pressure ratio limits of the fan.
However, decreasing the twist angle to 55° has positive effect on both mass flow rate and pressure
ratio. The fan blade achieves higher pressure ratio at higher mass flow rate before the fan blade
passage chokes. Nevertheless, the fan blade in this case has less operating range before the blade
stalls. Further decrease in the twist angle to 45° negatively affects the performance of the fan blade.
Moreover, in this case, the fan blade has more restricted operating range than 55° twist case.
Figure 10.45 b shows that, although the blade twist of 75° case has less mass flow rate and pressure
ratio limits, it is the most efficient case to operate at low mass flow rate conditions (54.5 kg/s)
compared to the other cases. The baseline case of 65° twist has the highest peak efficiency among the
tested cases when operating at higher mass flow rate conditions (65 kg/s) and pressure ratio of (1.4).
The 55° blade twist case has slightly less peak efficiency than the 65° twist case. However, the fan
292
blade in this 55° twist case can deliver more mass flow rate (67 kg/s) at higher pressure ratio (1.45).
The 45° twist case shows very poor efficiency compared to other cases
Further analysis is delivered in this paper to the flow field for the cases operating at high mass flow
rate conditions (blade twist of baseline case, 55° twist and 45° twist) at their highest efficiency. The
flow field for the cases operating in low mass flow rate conditions are analyzed at their highest
efficiency as well. Analyzed points are marked in blue color in Figure 10.45 b for high mass flow
cases and red color for low mass flow cases. The flow field analysis aims to investigate the reason
why some cases outperform the other cases in almost the same operating conditions.
waves impinging on the leading edge of the pressure surface [17]. This can be clearly observed in of
Figure 10.46 b and Figure 10.46 c.
For the 65° twist case shown in Figure 10.46 a, Mach number upstream the bow shock is the lowest
among the other cases, leading to reduced shock strength and shock losses. Shock waves cover the
entire suction surface, increasing the pressure on this surface and eliminating blade stall. This
explains the high efficiency of this case in these conditions. Further stretching in the passage, a
normal shock wave is developed due to the supersonic Mach number downstream the oblique shock.
The normal shock impinges on the pressure side of the blade leading to slight boundary layer
separation which reattaches to the blade before the trailing edge.
Figure 10.46 b shows Mach number distribution for the 55° twist case. A bow shock is developed
upstream the leading edge. Stretching downstream, the bow shock attaches and results in an attached
oblique shock. The downstream Mach number of the oblique shock is supersonic resulting in
developing a normal shock in the passage impinging on the suction surface.
The Mach number downstream the oblique shock is higher than the 65° twist case, which in turn
increases the shock strength and shock losses. The suction side/boundary layer interaction is more
obvious in this case. The normal shockwave impinging on the leading edge of the blade results in
growth and separation of the boundary layer. Combining the boundary layer separation with the
increase in the upstream Mach number can explain the decreased efficiency of this case compared to
the 65° twist case.
Figure 10.46 Mach Number Distribution for High Mass Flow Point
Figure 10.46c shows the Mach number distribution for the 45° twist case. The bow shock developed
upstream the leading edge turns to oblique shock on the leading edge of the blade. Mach number
downstream the oblique shock is the highest for the different cases. The suction surface
294
Figure 10.47 shows the variation of the total pressure along the blade span for high mass flow
points of different twist cases at domain outlet. Flow field entropy is shown in Figure 10.48 at 85%
span for 65°, 55°, and 45° twist cases, respectively. It can be noticed that the 65° twist case has lower
entropy values than the 55° twist case. This agrees with the slight increase in efficiency of the 65°
twist case. The shock losses are observed in all cases in Figure 10.48. However, entropy is
significantly raised by the boundary layer separation more than the shock losses. Accordingly, blade
stall is the dominant factor in reducing effeminacy in Figure 10.48 c of the 45° twist case that has
severe entropy rise due to the blade stall. This further agrees with the decreased efficiency of this
case.
10.8.3.3 Blade Twist Effect at Low Mass Flow Rate Conditions
At low mass flow rate points, Figure 10.49 shows Mach number distribution of the flow field
representing the baseline case of 65° twist and the 75° twist case, respectively. Figure 10.49 a
shows the Mach number distribution for the 65o baseline case. A bow shock upstream the leading
edge is developed and becomes closer to normal shock. Downstream the bow shock and stretching
into the suction surface, an oblique shock wave is developed. For this case, there is no obvious normal
shock developed in the passage compared to the large mass flow point, which may reduce the shock
losses. [Han et al.] obtained similar shock dynamics in the low mass flow point study. However, in the
present work, the suction-side boundary layer is separated due to the interaction with the suction-
side shock wave, which in turn leads to blade stall and efficiency decrease.
Figure 10.49 Mach number distribution for low mass flow points
Figure 10.49 b represents the Mach number distribution for the 75° twist case. A bow shock forms
upstream the leading edge. Stretching downstream, the bow shock turns to an oblique shock. The
Mach number downstream the oblique shock is supersonic, leading to a normal shock occurrence in
the passage. However, shock waves cover the entire suction surface of the blade which diminishes
the suction surface stall.
Eliminating the suction surface stall is key to reduce losses and increase efficiency. This can interpret
the increased efficiency for this case in the low mass flow conditions. Both the boundary layer of the
pressure surface and the normal shock interact leading to growth and detachment of the boundary
layer, which reattaches again due to the increased pressure on this surface.
The flow field entropy distribution is shown in Figure 10.50. The baseline case in Figure 10.50 a
shows increased entropy due to the suction-surface shock/boundary layer interaction leading to
blade stall. This explains the low efficiency of the baseline case in low mass flow rate conditions.
Figure 10.50 b shows the entropy distribution for the 75° twist case. There is far less entropy
296
increase in the flow field, which mainly because of the attached boundary layer due to shock waves
on the suction surface as mentioned before.
10.8.4 Conclusion
A three-dimensional unsteady
numerical analysis is used to analyze
the influence of a transonic fan blade
twist on the fan performance. A
modern fan blade of a high-bypass
turbofan engine “Rolls-Royce Trent
XWB” is approximately modelled to
perform the analysis. Twist of the
blade is changed by changing the tip
blade angle. Studied cases are a
baseline case of 65° twist, 45°, 55°,
and 75° twist. The rotating speed is
held constant at 2700 rpm through
all cases. One blade passage is
studied with rotating periodic
boundary condition. Performance
parameters are obtained for one
blade passage. The major findings
are as follows:
(1) Generally, the present
computational results compare well
with the results of others with
consideration of the operating
differences.
(2) The baseline case of 65° twist has
wider range of operation. Mass flow
ranges from 53 to 67 kg/s. Pressure
ratio ranges from 1.45 to 1.225,
respectively with mass flow points.
The best efficiency point is at 65 kg/s
and pressure ratio of 1.34. For this
case the increased efficiency is due Figure 10.50 Entropy Distribution for Low Mass Flow Points
to shock waves in the passage
covering the entire suction surface
eliminating blade stall.
(3) Increasing the blade twist to 75° reduces the operating limits of the fan. Mass flow ranges from
51 to 59.5 kg/s. Pressure ratio ranges from 1.35 to 1.175. However, it has the highest efficiency of the
studied cases while operating in low mass flow rate conditions (at mass flow of 55 kg/s and 1.35
pressure ratio). This is mainly because the shock waves cover the entire suction surface and diminish
the suction surface boundary layer separation. While the baseline case has slight boundary layer
growth and separation due to the suction surface shock/ boundary layer interaction.
(4) Reducing the blade twist to 55° increases the maximum mass flow limit to 68.2 kg/s. However,
the minimum mass flow before blade stall is reduced to 61 Kg/s, which reduces the overall
performance range of the fan. However, the blade has higher total pressure ratio. In high mass flow
operating conditions, this case has slight blade stall due to suction surface shock/boundary layer
297
interaction which slightly reduces efficiency compared to the baseline case. But it produces higher
mass flow and total pressure at its best efficiency point (mass flow of 68 kg/s and pressure ratio of
1.44).
(5) Further reducing the blade twist to 45° results in more decrease of the operating limits of the fan.
Moreover, it greatly decreases the efficiency due to increased shock waves strength and blade stall
driving this case out of competence with other cases.
(6) According to this blade design, for low mass flow operating conditions (54 to 58 kg/s), the 75°
blade twist is more efficient. At high mass flow operating conditions (65 to 68 kg/s), both 55° and 65°
blade twist are more efficient. The 65° twist case has wider operating range before the blade stalls.
However, the 55° twist case can produce more mass flow rate and pressure ratio but with limited
operating range before the blade stalls. The 45° twist case stalls, at all operating points, results in a
greatly decreased efficiency compared to other cases.
10.8.5 References
[1] R. Bhaskaran, T. Wood, U. Paliath, and A. Breeze-String fellow, “Towards Large Eddy Simulation
of a 3D Transonic Fan,” 46th AIAA Fluid Dynamics Conference, Washington, D.C., USA, 2016.
[2] R. S. Amano, and C. Xu, “Blade Sweep Effects of Turbomachinery,” 43rd AIAA Aerospace Sciences
Meeting and Exhibit, Nevada, USA, 2005.
[3] A. A. A. El-Sayed, and M. M. Ibrahim, “Numerical Investigation of Different Tip Clearance Shape
Effects on Performance of an Axial Flow Compressor Stage”, The Online Journal on Power and Energy
Engineering (OJPEE), Vol. 1, No. 2, 2010.
[4] W. V. Banks, A. A. Ameri, and J. P. Bons, “Numerical Investigation of Effects of Tip Clearance in
Shrouded Turbine Rotor Blade Performance,” 2018 AIAA Joint Propulsion Conference, Cincinnati,
Ohio, USA, 2018.
[5] H. Karrabi, and M. Rezasoltani, “The Effect of Blade Lean, Twist and Bow on the Performance of
Axial Turbine at Design Point,” ASME 2011 International Mechanical Engineering Congress and
Exposition, Colorado, USA, 2011.
[6] D. J. Frohnapfel, W. F. O’Brien, and K. T. Lowe, “Fan Rotor Flow Measurements in a Turbofan Engine
operating with Inlet Swirl Distortion,” 55th AIAA Aerospace Sciences Meeting, Texas, USA, 2017.
[7] L. Han, Y. Wang, X. Zhang, X. Zhang, and D. Wei, “Effect of Casing Shape Surrounded Rotor on
Aerodynamic Performance of a Transonic Fan,” 53rd AIAA/SAE/ASEE Joint Propulsion Conference,
Atlanta, Georgia, USA, 2017.
[8] J. Thisse, C. Polacsek, and J. Mayeur, “Numerical Simulations of Shock-Wave Propagation in
Turbofan Intakes,” 22nd AIAA/CEAS Aeroacoustics Conference, Lyon, France, 2016.
[9] Z. M. Hall, V. Ahuja, R. Hartfield, A. Shelton, and A. Ahmed, “Optimization of a Turbofan Inlet Duct
using a Genetic Algorithm and CFD,” 27th AIAA Applied Aerodynamics Conference, USA, 2009.
[10] X. Chai, P. Han, T. Shi, and Z. Wang, “The Study of Tip Clearance of a Wide-Chord Fan Blade of a
High Bypass Ratio Turbo-Fan Engine,” ASME Turbo Expo 2014: Turbine Technical Conference and
Exposition, Düsseldorf, Germany, 2014.
[11] “EASA.E.111 Rolls-Royce plc Trent XWB Series Engines,” European Aviation Safety Agency, 2018.
[12] “World's Most Efficient Large Aero-Engine,” Rolls-Royce plc, 2018. [Online]. Available:
www.rollsroyce.com/products-and-services/civilaerospace/airlines/trent-xwb.aspx.
[13] P. Gullberg, and R. Sengupta, “Axial Fan Performance Predictions in CFD, Comparison of MRF and
Sliding Mesh with Experiments,” SAE International, 2011.
[14] D. Brzozowski, O. Uzol, Y.-C. Chow, J. Katz, and C. Meneveau, “A Comparison of Unsteady RANS
Simulations with PIV Data in an Axial Turbomachine,” ASME Fluids Engineering Division Summer
Meeting and Exhibiti, Houston, TX, USA, 2005.
[15] J. Åhman, CFD Validation of Tip Clearance Flows in an Axial Compressor, Luleå, Sweden: Luleå
University of Technology, 2016.
[16] “ANSYS Fluent User's Guide Release 15.0,” Canonsburg, Pennsylvania, USA, ANSYS, Inc, 2013.
298
[17] P. K. Ray, and W. N. Dawes, “Detached-Eddy Simulation of Transonic Flow Past a Fan-Blade
Section,” 15th AIAA/CEAS Aeroacoustics Conference (30th AIAA Aeroacoustics Conference), Miami,
Florida, USA, 2009.
[18] S.-M. Li, and R. Ramakrishnan, “A CFD Study and Performance Evaluation of Service-Run Variable
Vanes in a High-Pressure Compressor of a Turbofan Engine,” 54th AIAA Aerospace Sciences Meeting,
San Diego, California, USA, 2011.
299
11 Radial Flow
Up to now we were mostly concern with Axial flows. Now we pay homage to Radial flows which
designed in many everyday life tools. According to dictionary, in radial flows, the working fluid is
flowing mainly along the radii of rotation.
388 Shepard, Dennis G. “Principles of Turbomachinery”, McMillan. ISBN 0-471-85546-4. LCCN 56002849, 1956.
389 Lakshminarayana, B. (1996). “Fluid Dynamics and Heat Transfer of Turbomachinery”, New York: John Wiley
& Sons Inc. ISBN 0-471-85546-4.
390 Japikse, David & Baines, Nicholas C., “Introduction to Turbomachinery”, Oxford: Oxford University press.
change in radius from inlet to exit of the rotor/impeller. The 1940s-era German Heinkel HeS 011
experimental aviation turbojet engine was the first aviation turbojet design to have any sort of
"mixed compressor" design in its fore-sections, as it had a single-stage "diagonal flow" main
compressor ahead of a triple-stage axial unit, driven by a twin-stage turbine.
11.1.3 Components of a simple Centrifugal Compressor
A simple centrifugal compressor has four components: inlet, impeller/rotor, diffuser, and collector.
Figure 11.2 shows each of the components of the flow path, with the flow (working gas) entering
the centrifugal impeller axially from right to left (blue). As a result of the impeller rotating clockwise
when looking downstream into the compressor, the flow will pass through the volute's discharge
cone moving away from the figure's viewer. The inlet to a centrifugal compressor is typically a simple
pipe. It may include features such as a valve, stationary vanes/airfoils (used to help swirl the flow)
and both pressure and temperature instrumentation. All of these additional devices have important
uses in the control of the centrifugal compressor.
11.1.3.1 Inlet
The inlet to a centrifugal compressor is typically a simple pipe. It may include features such as a valve,
stationary vanes/airfoils (used to help swirl the flow) and both pressure and temperature
instrumentation. All of these additional devices have important uses in the control of the centrifugal
compressor.
11.1.3.2 Centrifugal Impeller
The key component that makes a compressor centrifugal is the centrifugal impeller, Figure 11.1,
which contains a rotating set of vanes (or blades) that gradually raises the energy of the working gas.
This is identical to an axial compressor with the exception that the gases can reach higher velocities
301
and energy levels through the impeller's increasing radius. In many modern high-efficiency
centrifugal compressors the gas exiting the impeller is traveling near the speed of sound. Impellers
are designed in many configurations including "open" (visible blades), "covered or shrouded", "with
splitters" (every other inducer removed) and "w/o splitters" (all full blades). Figure 11.2 show open
impellers with splitters. Most modern high efficiency impellers use "back sweep" in the blade
shape391-392. Euler’s pump and turbine equation plays an important role in understanding impeller
performance.
11.1.3.3 Diffuser
The next key component to the simple centrifugal compressor is the diffuser. Downstream of the
impeller in the flow path, it is the diffuser's responsibility to convert the kinetic energy (high velocity)
of the gas into pressure by gradually slowing (diffusing) the gas velocity. Diffusers can be vaneless,
vane or an alternating combination. High efficiency vane diffusers are also designed over a wide
range of solidities from less than 1 to over 4. Hybrid versions of vane diffusers include: wedge,
channel, and pipe diffusers. There are turbocharger applications that benefit by incorporating no
diffuser. Bernoulli's fluid dynamic principle plays an important role in understanding diffuser
performance.
Figure 11.3 Jet Engine Cutaway Showing the Centrifugal Compressor among others
11.1.3.4 Collector
The collector of a centrifugal compressor can take many shapes and forms. When the diffuser
discharges into a large empty chamber, the collector may be termed a Plenum. When the diffuser
discharges into a device that looks somewhat like a snail shell, bull's horn or a French horn, the
391Japikse, David. “Centrifugal Compressor Design and Performance”. Concepts ETI . ISBN 0-933283-03-2.
392 Aungier, Ronald H. (2000). “Centrifugal Compressors, A Strategy for Aerodynamic Design and Analysis”. ASME
Press. ISBN 0-7918-0093-8.
302
collector is likely to be termed a volute or scroll. As the name implies, a collector’s purpose is to gather
the flow from the diffuser discharge annulus and deliver this flow to a downstream pipe. Either the
collector or the pipe may also contain valves and instrumentation to control the compressor.
11.1.4 Applications
Below, is a partial list of centrifugal compressor applications each with a brief description of some of
the general characteristics possessed by those compressors. To start this list two of the most well-
known centrifugal compressor applications are listed; gas turbines and turbochargers.
11.1.4.1 Gas Turbines and Auxiliary Power Units
In their simple form, modern gas turbines operate on the Brayton cycle. Either or both axial and
centrifugal compressors are used to provide compression. The types of gas turbines that most often
include centrifugal compressors include turboshaft, turboprop, auxiliary power units, and micro-
turbines. The industry standards applied to all of the centrifugal compressors used in aircraft
applications are set by the FAA and the military to maximize both safety and durability under severe
conditions. Centrifugal impellers used in gas turbines are commonly made from titanium alloy
forgings. Their flow-path blades are commonly flank milled or point milled on 5-axis milling
machines. When tolerances and clearances are the tightest, these designs are completed as hot
operational geometry and deflected back into the cold geometry as required for manufacturing. This
need arises from the impeller's deflections experienced from start-up to full speed/full temperature
which can be 100 times larger than the expected hot running clearance of the impeller.
11.1.4.2 Automotive and Diesel Engines Turbochargers and Superchargers
Centrifugal compressors used in conjunction with reciprocating internal combustion engines are
known as turbochargers if driven by the engine’s exhaust gas and turbo-superchargers if
mechanically driven by the engine. Ideal gas properties often work well for the design, test and
analysis of turbocharger centrifugal compressor performance.
11.1.4.3 Natural Gas to Move the Gas from the Production site to the Consumer
Centrifugal compressors for such uses may be one or multi-stage and driven by large gas turbines.
The impellers are often if not always of the covered style which makes them look much like pump
impellers. This type of compressor is also often termed an API-style. The power needed to drive these
compressors is most often in the thousands of horsepower (HP). Use of real gas properties is needed
to properly design, test and analyze the performance of natural gas pipeline centrifugal compressors.
11.1.4.4 Oil Refineries, Natural Gas Processing, Petrochemical and Chemical Plants
Centrifugal compressors for such uses are often one-shaft multi-stage and driven by large steam or
gas turbines. Their casings are often termed horizontally split or barrel. Standards set by the
industry (ANSI/API, ASME) for these compressors result in large thick casings to maximize safety.
The impellers are often if not always of the covered style which makes them look much like pump
impellers. Use of real gas properties is needed to properly design, test and analyze their performance.
11.1.4.5 Air-Conditioning and Refrigeration and HVAC
Centrifugal compressors quite often supply the compression in water chillers cycles. Because of the
wide variety of vapor compression cycles (thermodynamic cycle, thermodynamics) and the wide
variety of workings gases (refrigerants), centrifugal compressors are used in a wide range of sizes
and configurations. Use of real gas properties is needed to properly design, test and analyze the
performance of these machines.
11.1.4.6 Industry and Manufacturing to Supply Compressed Air
Centrifugal compressors for such uses are often multistage and driven by electric motors. Inter-
cooling is often needed between stages to control air temperature. Note that the road repair crew
and the local automobile repair garage find screw compressors better adapt to their needs Ideal gas
303
relationships are often used to properly design, test and analyze the performance of these machines.
Carrier’s equation is often used to deal with humidity.
11.1.4.7 Air Separation Plants to Manufacture Purified End Product Gases
Centrifugal compressors for such uses are often multistage using inter-cooling to control air
temperature. Ideal gas relationships are often used to properly design, test and analyze the
performance of these machines when the working gas is air or nitrogen. Other gases require real gas
properties.
11.1.4.8 Oil Field Re-Injection of High Pressure Natural Gas to Improve Oil Recovery
Centrifugal compressors for such uses are often one-shaft multi-stage and driven by gas turbines.
With discharge pressures approaching 700 bar, casing are of the barrel style. The impellers are often
if not always of the covered style which makes them look much like pump impellers. This type of
compressor is also often termed API-style. Use of real gas properties is needed to properly design,
test and analyze their performance.
inherently unsteady and involves the full-annulus flow field. Also, at part speeds, the predicted
efficiency values are noticeably lower than the measured values.
12.2.2 Multi-Stage Analysis
12.2.2.1 Steady Mixing-Plane Simulations
Since the mixing-plane method was first introduced in [Denton & Singh 1979] it has become the
industry standard type of rotor-stator simulations. A mixing-plane simulation is steady and only
requires one rotor blade and one stator blade per stage. Between the rotating blade passage and the
steady vane passage the flow properties are circumferentially averaged in a so-called mixing-plane
interface. This will of course remove all transient rotor-stator interactions, but it still gives fairly
representative results. In some commercial codes (CFX for example) mixing-plane interfaces are also
called stage-interfaces.
12.2.2.2 Steady Frozen Rotor Simulations
In a frozen rotor simulation the rotating and the stationary parts have a fixed relative position. A
frame transformation is done to include the rotating effect on the rotating sections. This will give a
steady flow and no transient effects are included. With a frozen-rotor simulation rotating wakes,
secondary flows, leading edge pressure increases etc. will always stay in exactly the same positions.
This makes a frozen rotor simulation very dependent on exactly how the rotors and the stators are
positioned. Most often a mixing-plane simulation gives better results. Frozen rotor simulations are
mainly performed to obtain a good starting flow-field before doing a transient sliding-mesh
simulation.
12.2.2.3 Unsteady Sliding Mesh Stator-Rotor Simulations
This is the most complete type of stator-rotor simulation, and very CPU intensive. In most engines
the number of stators and rotors do not have a common denominator (to avoid instabilities caused
by resonance between different rings). Hence, to make a full unsteady sliding-mesh computation it is
necessary to have a mesh which includes the full wheel with all stators vanes and all rotor blades.
This is often not possible, instead it is necessary to reduce the number of vanes and blades by finding
a denominator that is almost common and then scales the geometry slightly circumferentially. Here
is an example: Real engine: 36 stator vanes, 41 rotor blades Approximated engine: 41 stator vanes,
41 rotor blades, making it possible to simulate only 1 stator vane and 1 rotor blade Scaling of stator:
All stator vanes are scaled by 36/41 = 0.8780 circumferentially.
12.2.2.4 Unsteady Harmonic Balance Simulations
To overcome the computational costs associated with sliding mesh, a technique called Harmonic
Balance is used. The analysis exploits the fact that many unsteady flows of interest in turbomachinery
are periodic in time. Thus, the unsteady flow conservation variables may be represented by a Fourier
series in time with spatially varying coefficients. This assumption leads to a harmonic balance form
of the Euler or Navier–Stokes equations, which, in turn, can be solved efficiently as a steady problem
using conventional computational fluid dynamic (CFD) methods, including pseudo time marching
with local time stepping and multigrid acceleration. Figure 12.2 displays a full analyses blade
solution using a harmonic balanced techniques, courtesy of (CD-Adapco.com). To relax the
fundamental linear assumption while taking advantage of the high solution efficiency, a nonlinear
harmonic method was proposed. Similarly to the time-domain Fourier model, the unsteadiness is
represented by the Fourier series. But now each harmonic will be balanced (‘harmonic balancing’)
respectively in the nonlinear flow equations. Consequently, for a Fourier series retaining N
harmonics, we will have 2N equations for the complex harmonics. In addition, the time-averaged flow
will now be different from the steady flow due to the added deterministic stresses. So in total we have
2N+1 steady-like flow equations, which are solved simultaneously to reflect the interactions between
the unsteady harmonics and the time mean flows. The interactions among the harmonics are
included in a more complete nonlinear harmonic formulation by Hall’s harmonic balance
309
formulations. The nonlinear harmonic approach have been extended to effectively solve rotor-
rotor/stator-stator interactions in multistage turbomachines 395.
12.2.2.5 Hybrid Steady-Unsteady Stator-Rotor Simulations
Hybrid steady-unsteady methods have been proposed in literature in order to have an unsteady
simulation embedded in a multistage steady study. There are several advantages related to this
method: mainly grid size and number of iterations.
12.2.2.6 Other Advanced Multi-Stage Methods
Time-inclined, Adamczyk stresses, etc.
Figure 12.2 Full Blade Simulation using Harmonic Balanced Method (Courtesy of CD-adapco)
(with mixing plane or similar approaches) or an unsteady one. Although steady simulations with
mixing plane have been extensively performed during the 90s, it must be underlined that the
assumption of a smeared-out field on the rotor/stator interface is too strong for the current request
of accuracy. In fact, different authors396 has shown that the stagnation pressure is representative of
the losses in a steady environment only: in a steady adiabatic case an entropy rise on the streamline
is always associated to a total pressure decrease, while considering an unsteady but inviscid case, the
pressure variations in time influence the stagnation enthalpy. He demonstrated that, in an unsteady
viscous situation, the total pressure variations can provide some information on the global losses but
are also affected by the Euler flow field far from the blade surfaces. Furthermore, Payne et al. 397
individuated a large fluctuation of the time resolved stage efficiency, underlying the importance of
the vane phase on the unsteady losses entity. In addition, [Pullan]398 demonstrated that a steady
simulation generates 10 % less losses compared with the unsteady one.
Another classical error caused by a steady simulation is the analysis of the redistribution of a hot spot
in the rotor row. It was demonstrated that in an axial machine the hot fluid tends to accumulate on
the pressure side of rotor blades. This result can be explained considering that for a steady isentropic
flow without body forces, for a prescribed geometry with a uniform total pressure inlet field, the
streamlines, both the Mach number and the static pressure fields at the vane outlet are not influenced
by the total temperature inlet field. It means that at the stator exit section the hot fluid has a higher
velocity than the surrounding one. Considering the velocity triangles at the rotor inlet, the typical
mechanism of the segregation effect399 is obtained. The so-called “positive jet effect” is an inherently
unsteady phenomenon that interacts with passage vortex: the secondary redistribution brings hot
fluid from suction to pressure side circumferentially across the vane, thus spreading hot fluid over
the entire pressure surface of the blade 400. As a result, the heat load on the blade pressure side is
increased and the life time of the blade reduced by the increased rate of creep. A steady calculation
with mixing plane is not able to reproduce such kind of phenomenon since tangential non-
uniformities at the vane exit section are neglected401. It can be concluded that an accurate unsteady
simulation of the turbine stage should be always done as a support to the steady simulation results.
The unsteady analysis allows to model several important phenomena:
• Unsteady inlet distortions when boundary conditions affect the performances of the gas
turbine;
• Potential interaction caused by the pressure waves travelling (and reflecting) across the
stator/rotor gap;
• Rotating stall: typical of the compressors, is caused by the blockage of some vanes due to the
wrong incidence which causes flow separation;
396 He, L. VKI Lecture Series Part I: Modelling issues for computations of unsteady turbomachinery flows. VKI
Lecture Series on “Unsteady Flows in Turbomachines”, Von Karman Institute for Fluid Dynamics, (1996).
397 Payne, S. J., Ainsworth, R. W., Miller, R. J., Moss, R. W., & Harvey, N. W.,”Unsteady loss in a high pressure
turbine stage: Interaction effects”. International Journal of Heat and Fluid Flow, 26, 695–708, 2005.
398 Pullan, G. (2006). “Secondary flows and loss caused by blade row interaction in a turbine stage”. ASME Journal
All said, because of inherent difficulty, today most turbo-machinery simulations are performed as
steady-state simulations. Transient simulations are done when some kind of transient flow behavior
has a strong influence on the global flow field. Examples of transient simulations are detailed
simulations of rotor-stator interaction effects, simulations of large unsteady separations etc.
Sometimes when you perform
a steady stationary simulation
you can see tendencies of
unsteady behavior like for
example periodic vortex
shedding behind blunt trailing
edges. This is often first seen
as periodical variations of the
residuals. If the unsteady
tendencies are judged to not
affect the overall simulation
results it might be necessary
to coarsen the mesh close to
the vortex shedding or run a
different turbulence model in
order to make the simulation
converge. Sometimes you are
still forced to run a transient Figure 12.3 Transient Blade Row Extensions Enable Efficient Multi-
simulation and average the Stage CFD Simulation (Courtesy of ANSYS.com)
results if you don't obtain a
converged steady solution. Figure 12.3 shows the transient blade row extensions enable efficient
multi-stage CFD simulation (courtesy’s of ANAYS.com).
12.5 Meshing
In turbomachinery applications structured multi-block hexahedral meshes are most often used for
flow-path simulations. In most solvers a structured grid requires less memory, provides superior
accuracy and allows a better boundary-layer resolution than an unstructured grid. By having cells
with a large aspect ratio around sharp leading and trailing edges a structured grid also provides a
better resolution of these areas. Many companies have automatic meshing tools that automatically
mesh blade sections with a structured mesh without much user intervention. Unstructured meshes
are used for more complex and odd geometries where a structured mesh is difficult to create. Typical
examples where unstructured meshes are often used are blade tip regions, areas involving leakage
flows and secondary air systems, film cooling ducts etc. When meshing avoid to create large jumps
in cell sizes. Typically the cell size should not change with more than a factor of 1.25 between
neighboring cells. For structured meshes also try to create fairly continues mesh lines and avoid
discontinuities where the cell directions suddenly change. For multi-block structured meshes avoid
placing the singular points where blocks meet in regions with strong flow gradients since most
schemes have a lower accuracy in these singular points. Figure 12.4 shows a typical meshing for a
312
402F. Montomoli et al., “Uncertainty Quantification in Computational Fluid Dynamics and Aircraft Engines”,
SpringerBriefs in Applied Sciences and Technology, DOI 10.1007/978-3-319-14681-2_2. Chapter 2-
Limitations in Turbomachinery CFD , 2015.
313
about 100 cells in the stream wise direction. In the radial direction a good first approach is to use
something like 30 cells for a wall-function mesh and 100 cells for a low-Re mesh. It is important to
resolve leading and trailing edges well. Typically at least 10 cells, preferably 20 should be used
around the leading and trailing edges. For very blunt and large leading edges, like those commonly
found on HP turbine blades, 30 or more cells can be necessary. Cases which are difficult to converge
with a steady simulation and which show tendencies of periodic vortex shedding from the trailing
edge, can sometimes be "tamed" by using a coarse mesh around the trailing edge. This, of course,
reduces the accuracy but can be a trick to obtain a converged solution if time and computer resources
do not allow a transient simulation to be performed. Figure 12.5 shows a multi-block grid for the
space shuttle main engine fuel turbine (AIAA 98-0968). As an example, several schemes for steady
analysis of a two-dimensional profile could lead to a non-converged solution when the spatial
resolution in the trailing edge region is too fine, since an unstable base region could occur despite the
steady assumption. At the same time, it is wrong to perform a large-eddy simulation with a coarse
mesh, since the sub-grid scale model would try to account for the vortex structures. It is worth
mentioning also the evaluation of the boundary layer development, which is strongly dependent on
both the selected model and the near wall mesh resolution403.
12.5.2 Case Study - Mesh Resolution Effect on 3D RANS Turbomachinery Flow Simulations
Over the past twenty years, guidelines for choosing the mesh resolution for the numerical simulation
of turbomachinery viscous flows using the RANS models have changed several times: from 100-200
K cells per one blade in 90s to 0.5 -1.0 M cells per one blade now. Usually, a mathematical basis of
such recommendations is not clear, requirements to the mesh refinement are often not well-founded
(perhaps, with the only exception for y+) and the question of the solution convergence remains open.
Recently, a new investigation by [Yershov & Yakovlev]404 presented the study of the effect of a mesh
refinement on numerical results of 3D RANS computations of turbomachinery flows. The CFD solver
based on the second-order accurate scheme. The simplified multigrid algorithm and local time
stepping permit decreasing computational time. The flow computations are performed for a number
of turbine and compressor cascades and stages. In all flow cases, the successively refined meshes of
H-type with an approximate orthogonality near the solid walls were generated. The results obtained
are compared in order to estimate their both mesh convergence and ability to resolve the transonic
flow pattern. It is concluded that for thorough studying the fine phenomena of the 3D
turbomachinery flows, it makes sense to use the computational meshes with the number of cells from
several millions up to several hundred millions per a single turbomachinery blade channel, while for
industrial computations, a mesh of about or less than one million cells per a single turbomachinery
blade channel could be sufficient under certain conditions.
12.5.2.1 Formulation of Problems
We have performed the RANS simulation of the 3D turbulent compressible viscous flow through
several turbomachinery stages and cascades where the k–ω SST turbulence model was used. The
main objective was a qualitative study of the numerical solution convergence without being tied to
the experimental data. It is evident that both the insufficient adequacy of the mathematical model as
well as approximation and experimental errors could lead to the fact that in some cases the
differences between the numerical results and the experimental data may increase as the mesh is
refined. The meshes considered in this study were conventionally divided into five groups based on
the number of cells per one blade channel:
403 F. Montomoli et al., “Uncertainty Quantification in Computational Fluid Dynamics and Aircraft Engines”,
SpringerBriefs in Applied Sciences and Technology, DOI 10.1007/978-3-319-14681-2_2. Chapter 2-
Limitations in Turbomachinery CFD , 2015.
404 Sergiy Yershov, Viktor Yakovlev, “Mesh Resolution Effect On 3d Rans Turbomachinery Flow Simulations”,
405Sergiy Yershov, Viktor Yakovlev, “Mesh Resolution Effect On 3d Rans Turbomachinery Flow Simulations”,
Institute For Mechanical Engineering Problems Of Nasu, Kharkiv, Ukraine, 2016.
315
maximum 1.24. For a low-Re mesh this usually gives around 40 cells in the boundary layer whereas
a wall-function mesh does not require more than 10 cells in the boundary layer. If you are uncertain
of which wall distance to mesh with you can use a y+ estimation tool to estimate the distance needed
to obtain the desired y+. These estimation tools are very handy if you have not done any previous
similar simulations. As a rule of thumb a wall-function mesh typically requires around 5 to 10 cells
in the boundary layer whereas a resolved low-Re mesh requires about 40 cells in the boundary layer.
12.5.4 Periodic Meshing
To reduce the time, efforts and complexity of meshing the rotational periodicity of the impeller
geometry is taken advantage. Axial machines and rotating fluid zone of radial & mixed flow machines
are meshed using this approach. Choosing a single periodic flow passage is the first step in this
approach. The periodic angle of the flow passage is decided by the number of vanes/blades present.
For example, Periodic angle or Angle of Rotational Periodicity = 360°/number of blades.
• For a radial turbine with 16 blades, Angle of rotational periodicity → 360°/16 =22.5° (single
blade passage)
• For a pump with 4 blades, Angle of rotational periodicity → 360°/4 = 90° (single blade
passage)
This periodic geometric sector can be chosen in two different ways.
• Flow passage between two blades (suction side of first blade to the pressure side of next
blade).
• To have one complete blade inside the periodic flow passage.
There are two different scenarios based on the flow physics. If the flow physics is also periodic (most
axial flow machines), the mesh is generated only for a single blade fluid passage (ϴ), regardless of the
number of blades and is directly used for simulation. But if the flow physics is not periodic (radial &
mixed flow machines with volute), the mesh is generated for the single periodic flow passage / sector
and is copy rotated to get mesh for the complete geometry (360°). Meshing software provides an
option for periodic meshing to ensure both sides of periodic passage has same number of nodes and
same node location with a rotational offset of ϴ.
rows. [Giles]407 presented a unified theory for the construction of Non-reflecting boundary conditions
for the Euler equations (NRBC’s). The boundary conditions are based on the linearized Euler
equations written in terms of perturbations of primitive variables about some mean flow. Wave-like
solutions are substituted into the flow equations, and the solution is circumferentially decomposed
into Fourier modes. The zeroth mode corresponds to the mean flow and is treated according to one-
dimensional characteristic theory. This allows average changes in incoming characteristic variables
to be specified at the boundaries. Simply put, since the numerical solution is calculated on a truncated
finite domain, and one must prevent any nonphysical reflections of outgoing waves at the far-field
boundaries that could contaminate the numerical solution.
This becomes essential in turbomachinery applications in which the boundaries are often not very
far from the blades, because the physical spacing between the blade rows can be quite small. It
therefore becomes highly important for an accurate simulation to construct nonreflecting boundary
conditions (NRBCs). Preventing
spurious reflections that would
corrupt the solution is not only
important to get an accurate
prediction of the flow field, but
also to get more efficient
computations; convergence rate
is enhanced due to an
improvement of the transmission
of outgoing waves, allowing
smaller meshes to be used408.
Figure 12.6 compares a contour
plot the pressure Vs. y-
coordinate for both Riemann BC
and NRBC’s in a supersonic
cascade for a 2-D flow [F. De
Raedt, 2015]. The most notable
observation is that at the outflow,
when Riemann BC are applied,
the pressure lines diverge from
the boundary and rarely cross
that boundary. This is a direct Figure 12.6 Pressure contour plot, 2nd order spatial
result of the reflectivity of the discretization scheme
boundary conditions. When the
boundary is far away from the airfoil, the effectiveness of these reflections on the airfoil flow-field is
minimal, as observed by comparing the long flow-field simulations of the Riemann BC and NRBC.
However when the boundary is close to the airfoil, the simulations using Riemann BC become
completely inaccurate. In contrast the short flow-field simulations using the NRBC result in very
similar pressure contours to those of the long flow-field. This clearly demonstrates the effect of the
NRBC implementation. One can have a closer look at the boundary itself to further clarify this
comparison409.
407 Giles, Michael B., “Nonreflecting Boundary Conditions for Euler Equation Calculations,” AIAA Journal, Vol. 28,
No. 12, Dec. 1990, pp. 2050-2058.
408 “Three-Dimensional Nonreflecting Boundary Conditions for Swirling Flow in Turbomachinery”, Journal of
The exact knowledge of boundary conditions for numerical simulations is probably one of the most
challenging problems in CFD and it is crucial in turbomachinery. turbomachinery components are
subjected to non-uniform conditions whose distributions have to be determined with high accuracy.
A typical example of this kind of problems is the simulation of a high-pressure stage with realistic
inlet conditions. Salvadori et al410 demonstrated that a non-uniform inlet temperature profile,
including hot streak migration, generates 10% variation in blade suction side static pressure
distribution at mid-span, a 60 % variation of Nusselt number value on blade pressure side and a 19
% variation in the peak total temperature at mid-span at the stage exit section with respect to cases
with uniform inlet. Considering that the distribution of turbine entry temperature is not measured
directly and that an error of more than 50 K is common in real gas turbines, it is clear the impact of
such parameter411.
12.8 Aero-Mechanics
Now let’s look at the challenges of aeromechanics. Whereas the aerodynamicist generally prefers
designs with very thin blades, the structural engineer prefers thick blades to minimize stress and
optimize vibration characteristics. Those interested in material cost and weight would no doubt side
with the aerodynamicist, whereas those responsible for honoring the machine warranty would favor
the structural viewpoint. Achieving agreement requires a balance, and that is where the field of
aeromechanics comes in. Aeromechanics is by no means new. What is new is the fidelity with which
engineers can practically consider both the fluid mechanics and the structural aspects of the solution.
The real behavior of rotating blades is indeed very complex, and the mechanical loads are very high.
410 Salvadori, S., Montomoli, F., Martelli, F., Chana, K. S., Qureshi, I., & Povey, T. “Analysis on the effect of a
nonuniform inlet profile on heat transfer and fluid flow in turbine stages”. Journal of Turbomachinery, 134(1),
011012-1-14. doi:10.1115/1.4003233, (2012).
411 F. Montomoli et al., “Uncertainty Quantification in Computational Fluid Dynamics and Aircraft Engines”,
For example, a single low-pressure steam turbine blade rotating at operating speed generates a load
of several hundred tons! Long, thin blades are susceptible to vibration. Engineers strive to design
blades whose natural frequencies do not coincide with the disturbances that arise due to operating
speed, etc. That is complicated enough, but there are also periodic disturbances that can originate
from more distant blade rows or aerodynamic effects412.
In the past, analysis of fluids and structural dynamics was mostly separate and simplified. But for
some time, at least in principle, the ability to perform high-fidelity coupled analysis has been
available. In reality, solving for time-dependent, three-dimensional fluid-structure interaction is very
time-consuming and expensive, even on today’s high performance computing systems. Engineers
have opted for more practical, usually disconnected and often lower-fidelity analysis methods.
Recently, practical yet high-fidelity multiple physics solution methods have emerged.
Prediction of aerodynamic blade damping, or “flutter,” is one such method. The procedure is to first
solve for the mechanical modes of vibration, and then feed that information to the CFD simulation.
The unsteady CFD simulation deforms the blade in the presence of the flow field and predicts
whether the blade is aerodynamically damped, and hence stable, or not. This high-fidelity approach
is practical because it provides a solution to the full wheel (all of the many blades in a given row) by
solving only for one or at most a few blades in the blade row of interest. Cyclic symmetry is the
enabling structural technology here, while the Fourier Transformation method is key on the CFD side.
Tightly coupled these two efficient methods provides great advances in computing fidelity and speed.
Predicting forced response is essentially the inverse workflow to flutter. Figure 12.7 shows where
analysis provides the mechanical modes of blade vibration required for flutter analysis. First, the
unsteady fluid dynamic loads are predicted, and made available to the structural solver. After a
mechanical harmonic response simulation, the engineer evaluates the results for acceptable levels of
blade displacement, strain and stress. The concept of Nodal Diameter is explained next.
disturbances like incoming wakes and vortices from previous stages, surface roughness effects and
small steps or gaps in the surfaces play a significant role.
The turbomachinery codes that have transition prediction models often use old ad-hoc 414 models like
the Abu-Ghannam and Shaw model or the Mayle model. These models can be quite reliable if they
have been validated and tuned for a similar application. Do not trust your transition predictions
without having some form of experimental validation. Menter has also recently developed a new form
of transition model that might work fairly well, but it is still too new and untested. For some turbo
machinery applications, like modern high-lift low-pressure turbines, transition is critical. For these
applications a CFD code with a transition model that has been tuned for this type of applications
should be used.
medium) changes the temperature field along the flow path. Be it a steady or transient run, a fluid
which enters with a given temperature will usually experience temperature variations, for instance,
caused by convective boundary conditions, prescribed temperature profiles at flow obstacles etc.
Mesh consists of fluid and solid domain(s): additionally to the above, you want info too w.r.t .the
(spatial, temporal or even spectral) temperature field distribution in surrounding, confining or
immersed solids like channel walls or heat exchanger tubes. This is also called conjugate heat transfer
CHT in the CFD context. CHT requires a good boundary layer resolution; usually the wall mesh needs
to be rather refined, to obtain realistic heat flux results at the fluid/solid interface. Flow and heat
transfer convergence require different time step settings, to properly capture changes in flow and
heat quantities respectively. In either case, verify (strict necessity depends on CFD code used, CFX
for instance checks and assists in regard) that model dimensions, boundary conditions and
properties are in consistent units, hold appropriate values. Check temperature-dependence of
properties and other numbers before the run. In heat transfer predictions (depending on the CFD
code in use) besides the flow solver, you may have to activate the thermal solver too, as a job
specification.
12.14.1 Keeping it Cool in Gas Turbine
One of the key problems in the design of advanced gas turbine engines is the development of effective
cooling methods for the turbine vanes and blades415. Due to competition and continuous
improvement an increased complexity of cooling technology is required in the design of turbine
engine parts. In view of the material and time costs for experimental research, CFD has been accepted
by turbomachinery companies as one of the main methods for evaluating the performance of new
designs. Industrial CFD applications range from classical single and multi-blade-row simulations in
steady and transient mode to heat transfer and combustion chamber simulations. Depending on the
type of machine, physical and geometrical effects have to be taken into account. A complicating factor
is that it is necessary to carry out parametric studies considering several geometric options in the
process of designing the cooling systems. This normally takes a lot of time to generate mesh models
due to the mesh resolution required in the boundary layer.
415 A.V. Rubekina, A.V. Ivanov, G.E. Dumnov, A.A. Sobachkin (Mentor Graphics Corp., Russia); K.V. Otryahina
(PAO NPO Saturn, Russia), “Keeping it Cool in Gas Turbines”, Mentor paper.
416 Runa Nivea Pinto , Asif Afzal , Loyan Vinson D'Souza , Zahid Ansari , Mohammed Samee A. D., “Computational
Fluid Dynamics in Turbomachinery: A Review of State of the Art”, Archives of Computational Methods in
Engineering , DOI: 10.1007/s11831-016-9175-2, April 2016.
322
• The results obtained from unsteady flow can be employed to generate the unsteady blade
loading which indeed helps in prediction of mechanical issues of blading.
• The unsteady flow predicts the loss generation which is currently in use and need to be
explored further.
• With CFD, the prediction of generation of entropy when vortices and wakes mix in unsteady
flow has become definitely very easy than the experiment; but for validation of CFD results,
high class experimental results are required and to interpret the results from CFD.
• Numerical and modeling errors need to be reduced further.
• The accuracy of CFD results obtained for turbulence flow need to be increased.
• The assumptions and approximations made in CFD analysis of turbomachinery has to be
developed for every flow and geometric conditions.
Another important aspect understood from this survey is that the unsteady flow computations need
huge time to obtain results417. This is due to the enormous computational time required by the CFD
software to generate converged results. In order to reduce the computational time of the codes are
parallelized to run on various multi-core/multi-processor architectures. The major parallel
computing tool popularly used for the parallelization of CFD codes in turbomachinery is MPI. Various
parallel computing tools that can be employed further to attempt parallelization of these codes are
provided below:
So far, very less effort have been reported in literature regarding parallelization of turbomachinery
related CFD codes employing parallel computing tools mentioned above. These tools have been
proved to be of great computational performance enhancers for the CFD software used in other fields.
More specifically Open MP+MPI hybrid parallel computing approach has shown highly improved
computational performance.
417F. Montomoli et al., “Uncertainty Quantification in Computational Fluid Dynamics and Aircraft Engines”,
SpringerBriefs in Applied Sciences and Technology, DOI 10.1007/978-3-319-14681-2_2. Chapter 2-
Limitations in Turbomachinery CFD , 2015.
323
simulation with increasing accuracy and computational costs. It can be also underlined that in
presence of multiphase flows, i.e. for cavitation, and combustion, the selected methodology will
introduce a specific limitation on the obtained result whose entity is hard to be quantified. Most of
the cited problems are related to the stochastic uncertainty, also called reducible uncertainty because
it is a function of the knowledge of the problem physics and of the complexity of the algorithm. Then,
numerical accuracy can rise with an improved knowledge of the physics and with the computational
resources. However, user’s knowledge represents the most important assurance for a good CFD,
while uncertainty quantification is a strong support in the analysis and design of turbomachinery418.
418F. Montomoli et al., “Uncertainty Quantification in Computational Fluid Dynamics and Aircraft Engines”,
Springer Briefs in Applied Sciences and Technology, DOI 10.1007/978-3-319-14681-2_2. Chapter 2-
Limitations in Turbomachinery CFD , 2015.