Stephan Geerts

Department of Fluid Mechanics and Thermodynamics Faculty of Applied Sciences Vrije Universiteit Brussel Pleinlaan 2, B-1050 Brussel Belgium
Experimental and Numerical Study of An Axial Flow Pump
Stephan Geerts
Ph. D. Thesis
May 2006
Promoter: Prof. Dr. Ir. Chris Lacor Co-promoter: Prof. Dr. Ir. Charles Hirsch
A dissertation submitted to obtain the degree of Doctor of Engineering Sciences
Acknowledgments
In the first place, I highly thank Prof. Ch. Hirsch who gave me the opportunity to start with my research in the fascinating experimental and numerical world of turbomachinery. During this research, he also gave me the opportunity to improve my knowledge in other disciplines like wind engineering, development of numerical codes for dynamic response of wind turbines and their electrical output, and numerical and experimental research of confined double annular jet. I became also acquainted with all measuring facilities that are used in fluid mechanics. I very much appreciate Prof. Hirsch for his critical advice, remarks and ideas, as well as for the academicals freedom he gave me during my career at university. Further, I would like to thank Prof. Ch. Lacor that he continued promoting me. He motivated me, and helped me improve myself in this fascinating scientific environment. Thanks to Prof. J. De Ruyck, the right priorities were made, and the test facility got completely finalised, giving me the chances to perform all necessary measurements needed for this thesis. His guidance has been of great value for my research. Without the aids of the technical staff, it was not possible set up a test facility of this scale. Therefore, I would like to thank Michel Desees, Andre Plasschaert, and Daniel Debondt for their help and creativity. Also the staffs of NUMECA International are thanked for the use of their software tools. My thanks extend to all present and former colleagues at the department for providing an open and stimulating atmosphere. Special thanks to Alain Wery for his computer and electronics support, and to Mrs. Jenny Dhaes for helping me with the administration. Finally, without the encouragement, support and patience of my girlfriend Raquel, this work would not be the same. Whatever useful results this thesis may contain, they wouldnt be effective without her constant and patient guidance. I am grateful to the jury members for having shouldered the burden of examining the present manuscript and offering valuable suggestions for improvements, namely Prof. J. De Ruyck,, Prof. E. Dick, Prof. P. Hendrick, Prof. T. Thienpont, Prof. J. Vereecken and of course, my two promoters, Prof. Ch. Hirsch and Prof. Ch. Lacor.
Table of Contents
Table of Contents
Nomenclature Introduction Chapter 1: Literature Review
1.1 Introduction.......................................................................................................... 1.1 1.2 CFD of Axial Flow Pumps .................................................................................. 1.1 1.3 Flow Analysis of Axial Flow Pumps ................................................................... 1.2 1.3.1 Two-dimensional Analysis ......................................................................... 1.2 1.3.2 Three-Dimensional Analysis ...................................................................... 1.5 1.3.2.1 Theory of Radial Equilibrium......................................................... 1.6 1.3.2.2 Computer-Aided Methods for Solving the Through-Flow Problem1.8 1.3.2.3 Computational Fluid Dynamics ...................................................... 1.9 1.4 Global Performance Characteristic .................................................................... 1.11 1.5 Characteristics of the 3D Flow Field ................................................................. 1.14 1.5.1 Forces Active on the Flow ........................................................................ 1.14 1.5.2 Prerotation................................................................................................. 1.15 1.5.3 Secondary Flow in Axial Flow Pumps ..................................................... 1.16 1.5.4 Review on Tip Leakage Flow ................................................................... 1.21 1.5.5 Losses in Boundary Layers....................................................................... 1.22 1.5.6 Cavitation.................................................................................................. 1.23 1.5.7 Unsteady Phenomena................................................................................ 1.24 1.6 Summary ............................................................................................................ 1.24
Chapter 2: Experimental Test Facility

2.1 Introduction.......................................................................................................... 2.1 2.2 Experimental Test Facility................................................................................... 2.1 2.2.1 Infrastructure and Test Rig ......................................................................... 2.1 2.2.2 Description of Single-Stage Axial Flow ..................................................... 2.5 2.2.2.1 Rotor and Stator Design.................................................................. 2.8 2.2.2.2 Hub and Shroud Design .................................................................. 2.9 Experimental and Numerical Study of an Axial Flow Pump 1
Table of Contents 2.3 Measurement Equipment for the Performance of the Pump................................ 2.9 2.3.1 Flow Measurements .................................................................................. 2.10 2.3.2 Wall Pressure Measurement ..................................................................... 2.10 2.3.3 Rotor Shaft Torque and Rotation Speed ................................................... 2.11 2.4 Performance Uncertainty Analysis .................................................................... 2.13 2.5 Summary ............................................................................................................ 2.14
Chapter 3: Applications of the Laser Doppler Anemometry Equipment

3.1 Introduction.......................................................................................................... 3.1 3.2 Technical Specification of LDA .......................................................................... 3.1 3.3 Measurement Procedure....................................................................................... 3.2 3.4 Triggering ............................................................................................................ 3.5 3.5 Special Difficulties in Flow Study....................................................................... 3.8 3.6 Optical Correction.............................................................................................. 3.10 3.6.1 Displacement between Measurement Volumes of a 2D-probes ............... 3.11 3.6.2 Fringe Spacing of the Measurement Volume and Velocity Corrections .. 3.14 3.6.3 Direction of Measured Flow Velocity ...................................................... 3.16 3.6.4 Proportion of the Displacements between Measurement Volume and Probe ........................................................................................................ 3.16 3.6.5 Effect of Astigmatism on the Data Rate ................................................... 3.18 3.7 Axial Velocity Measurement ............................................................................. 3.18 3.8 Tangential Velocities Measurement .................................................................. 3.19 3.9 Seeding Particle ................................................................................................. 3.21 3.10 Choice of Number of Particles and Estimation of the Errors .......................... 3.22 3.11 Conclusions...................................................................................................... 3.23
Chapter 4: Experimental Results of Axial Flow Pump

4.1 Introduction.......................................................................................................... 4.1 4.2 Working Regime .................................................................................................. 4.1 4.3 Global Performance of the Axial Flow Pump...................................................... 4.1 4.3.1 Results and Discussion of the Global Performance of the Propeller Pump 4.3 4.4 Measured Flow Field at Design Point.................................................................. 4.7 4.4.1 Measured Data .......................................................................................... 4.10 4.4.2 Contour Plots of the Raw Flow Quantities of the Sections ...................... 4.10 Experimental and Numerical Study of an Axial Flow Pump 2
Table of Contents 4.4.3 Pitch-averaged Velocity Profiles of the Measured Locations................... 4.16 4.4.3 Blade-to-blade Velocity Profiles of the Measured Locations................... 4.16 4.5 Consistency Check............................................................................................. 4.21 4.6 Discussion .......................................................................................................... 4.22 4.7 Conclusion ......................................................................................................... 4.26
Chapter 5: Numerical simulations of the Axial Pump Flow

5.1 Introduction.......................................................................................................... 5.1 5.2 Prediction Method................................................................................................ 5.1 5.3 Turbulence Models .............................................................................................. 5.2 5.3.1 Turbulence Model of Baldwin-Lomax ....................................................... 5.2 5.3.2 Turbulence Model of Spalart-Allmaras ...................................................... 5.3 5.4 Solving Algorithms.............................................................................................. 5.5 5.5 Numerical Model of Axial Pump......................................................................... 5.5 5.5.1 Computational Domain ............................................................................... 5.5 5.5.2 Boundary Conditions .................................................................................. 5.6 5.5.3 Discretisation and Grid Resolution............................................................. 5.8 5.5.4 Physical Modeling .................................................................................... 5.13 5.5.5 Initial Solution .......................................................................................... 5.14 5.6 Discussion of Results and Observed Flow Field ............................................... 5.14 5.7 Discussion of the Different Flow Structure Caused by the Adoption of Window in the Casing ................................................................................... 5.41 5.8 Summary and Conclusions ................................................................................ 5.42
Chapter 6: Evaluation-Comparison of test data and CFD

6.1 Introduction.......................................................................................................... 6.1 6.2 Comparing Experimental Performance versus Numerical Results...................... 6.1 6.3 Comparing Experimental versus Numerical Flow Field ..................................... 6.8 6.3.1 Contours plot of Normalized Velocities at 40% Chord Axially Upstream of Rotor Tip Leading Edge....................................................... 6.10 6.3.2 Contours plot of Normalized Velocities at 20% Chord Axially Upstream of Rotor Tip Leading Edge....................................................... 6.11 6.3.3 Contours plot of Normalized Velocities at Rotor Leading Edge.............. 6.12 6.3.4 Contours plot of Normalized Velocities at 25% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.13 Experimental and Numerical Study of an Axial Flow Pump 3
Table of Contents 6.3.5 Contours plot of Normalized Velocities at 50% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.14 6.3.6 Contours plot of Normalized Velocities at 75% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.15 6.3.7 Contours plot of Normalized Velocities at Rotor Trailing Edge .............. 6.16 6.3.8 Contours plot of the Normalized Velocities at 20% Chord Axially Downstream of Rotor Tip Trailing Edge .................................................. 6.17 6.3.9 Contours plot of the Normalized Velocities at 40% Chord Axially Downstream of Rotor Tip Trailing Edge .................................................. 6.18 6.3.10 Pitch-averaged Velocity Profiles of the Measured Locations................. 6.19 6.3.11 Blade-to-blade Axial Velocities at 40% Chord Axially Upstream of Rotor Tip Leading Edge....................................................... 6.22 6.3.12 Blade-to-blade Axial Velocities at 20% Chord Axially Upstream of Rotor Tip Leading Edge....................................................... 6.23 6.3.13 Blade-to-blade Axial Velocities at Rotor Leading Edge ........................ 6.24 6.3.14 Blade-to-blade Axial Velocities at 25% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.25 6.3.15 Blade-to-blade Axial Velocities at 50% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.26 6.3.16 Blade-to-blade Axial Velocities at 75% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.27 6.3.17 Blade-to-blade Axial Velocities at Rotor Trailing Edge......................... 6.28 6.3.18 Blade-to-blade Axial Velocities at 20% Chord Axially Downstream of Rotor Tip Trailing Edge .................................................. 6.29 6.3.19 Blade-to-blade Axial Velocities at 40% Chord Axially Downstream of Rotor Tip Trailing Edge .................................................. 6.30 6.3.20 Blade-to-blade Tangential Velocities at 40% Chord Axially Upstream of Rotor Tip Leading Edge....................................................... 6.31 6.3.21 Blade-to-blade Tangential Velocities at 20% Chord Axially Upstream of Rotor Tip Leading Edge....................................................... 6.32 6.3.22 Blade-to-blade Tangential Velocities at Rotor Leading Edge ................ 6.33 6.3.23 Blade-to-blade Tangential Velocities at 25% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.34 6.3.24 Blade-to-blade Tangential Velocities at 50% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.35 Experimental and Numerical Study of an Axial Flow Pump 4
Table of Contents 6.3.25 Blade-to-blade Tangential Velocities at 75% Chord Axially Downstream of Rotor Tip Leading Edge.................................................. 6.36 6.3.26 Blade-to-blade Tangential Velocities at Rotor Trailing Edge ................ 6.37 6.3.27 Blade-to-blade Axial Velocities at 20% Chord Axially Downstream of Rotor Tip Trailing Edge .................................................. 6.38 6.3.28 Blade-to-blade Axial Velocities at 40% Chord Axially Downstream of Rotor Tip Trailing Edge .................................................. 6.39 6.3.29 Blade-to-blade Velocity in tip (95% Span)............................................. 6.40 6.4 Discussion: Numerical versus Experimental Results ........................................ 6.42 6.5 Conclusions........................................................................................................ 6.45
Chapter 7: 3D Flow Structure

7.1 Introduction.......................................................................................................... 7.1 7.2 Evolution of the Flow Structure in the Impeller .................................................. 7.1 7.3 Flow Structure of the Tip Clearance Flow......................................................... 7.10 7.4 Casing Wall Stream Lines near the Impeller ..................................................... 7.15 7.5 Conclusions........................................................................................................ 7.17
Summary and Conclusions Bibliography Appendix A: Laser Doppler Anemometry

A.1 Introduction.........................................................................................................A.1 A.2 Characteristics of LDA .......................................................................................A.1 A.2.1 Non-Contact Optical Measurement ...........................................................A.1 A.2.2 No Calibration - No Drift...........................................................................A.1 A.2.3 Well-Defined Directional Response ..........................................................A.1 A.2.4 High Spatial and Temporal Resolution......................................................A.1 A.2.5 Multi-Component Bi-Directional Measurements ......................................A.2 A.3 Principles of LDA ...............................................................................................A.2 A.3.1 Laser Beam ................................................................................................A.2 A.3.2 Doppler Effect............................................................................................A.3 A.3.3 Calibration..................................................................................................A.5 A.4 Implementation ...................................................................................................A.5 Experimental and Numerical Study of an Axial Flow Pump 5
Table of Contents A.4.1 The Fringe Model.......................................................................................A.5 A.4.2 Measuring Volume.....................................................................................A.6 A.4.3 Backscatter Versus Forward Scatter ..........................................................A.7 A.4.4 Optics .........................................................................................................A.9 A.4.5 Frequency Shift ........................................................................................A.10 A.4.6 Signal Processing ....................................................................................A.12 A.4.7 Data Analysis ...........................................................................................A.13 A.5 Making Measurements......................................................................................A.14 A.5.1 Dealing With Multiple Probes (3D Setup) ..............................................A.14 A.5.2 Calculating Moments ..............................................................................A.15 A.5.3 Velocity Bias and Weighting Factor .......................................................A.15
Appendix B: Experimental Uncertainties

B.1 Introduction .........................................................................................................B.1 B.2 Sources of Errors.................................................................................................B.1 B.2.1 Errors in the Geometry of the Propeller Pump...........................................B.1 B.2.2 Errors of the Rotation Speed of the Axial Pump........................................B.1 B.2.3 Error of the Mean Velocities by the LDA System .....................................B.2 B.2.3.1 Calibration Uncertainly ..................................................................B.3 B.2.3.2 Data Acquisition Uncertainty.........................................................B.4 B.2.4 Sources of the LDV Frequency Bias..........................................................B.6 B.2.4.1 Processor Bias ................................................................................B.6 B.2.4.2 Fringe Bias .....................................................................................B.7 B.2.4.3 Velocity Bias ..................................................................................B.7 B.2.4.4 Velocity Gradient Bias ...................................................................B.8 B.2.4.5 Seed Particles and Polydispersie-bias ............................................B.9 B.2.4.6 Coincidentie Bias .........................................................................B.10 B.2.5 Data Reduction Uncertainty .....................................................................B.11 B.3 Conclusions .......................................................................................................B.12
Experimental and Numerical Study of an Axial Flow Pump
Nomenclature
Nomenclature
Roman Symbols
r r r 1x ,1y ,1z
r ar
Unit vector along the x, y, z directions Acceleration vector Relative acceleration vector Cross sectional area of fluid flow Arrival time Bias error Torque Correction factor for fluid velocity Power coefficient Velocity of light, chord Impeller diameter Initial beam thickness Distance to closest wall, distance from the beam waist, Numerical dissipation term Distance between fringes Particle size Total internal energy per unit mass Edge intensity Unit vectors of incoming light Unit vector of scattered light Resulting force, Flux vector Surface force Frictional force per unit mass
A ai B C Cf Cp c D DL d df dp E e ei es
r F r Fs
r f
Nomenclature.1
Nomenclature
f fi fs G g Hideal Hstat Htot i,j,k K k L M m r n N Nf Ns Nw n nf ni nw P Pe
Component of the flux vector Frequency of incoming light Frequency of scattered light Refraction correction factor Gravitation Ideal Head Static head Total head Mesh points location in x, y, z directions Scaling constant Turbulence kinetic energy, Correction factor Length Number of data samples within a given discrete storage window or bin mass Cell face normal vector Number of particles Number of fringes Rotational speed Discrete Storage Window Normal or normal component, Normal distance to the wall, rotational speed Refraction index of fluid Index of refraction of medium i (i=0; air, i=1; wall, i=2; fluid) Refraction index of cylinder wall Shaft Power, production term, precision error Effective power Nomenclature.2
Nomenclature
p pt q
Q
Static pressure Total pressure Heat flux Source term Mass Flow, Source term Heat flux component Vector from the rotation axis Residual of iterative scheme, Specific gas constant Radius of curvature Reynolds number Inner radius of cylinder wall, Inner diameter of the tube wall Outer radius of cylinder wall, Outer diameter of the tube wall Experimental result, radius from the axis of rotation Radius of virtual position of beam intersection without refraction True radius position Characteristic surface, Standard deviation, Vortices Strain rate tensor Tip clearance Torque, Temperature Turbulence density Time Transition time Time step Velocity vector Particle velocity, tip blade speed
Q q
r r
R Rcm Re Ri Ro r ra rf S Sij s T Tu t ti
t
U U
Nomenclature.3
Nomenclature
Ui UR Utip Uxi ui
r V
Mean velocity component Uncertainty of the result Tip velocity Uncertainty in the measured value Xi to 95% coonfidence Fluctuating velocity component Absolute velocity vector Mean absolute velocity Velocity Discretised velocity Corrected tangential velocity Component of the velocity in the plane of the laser beams Component of fluid velocity calculated using half angle of unrefracted beams and water length in air Corrected component of fluid velocity Absolute axial velocity Absolute tangential velocity Relative velocity vector Average relative velocity Relative axial velocity Relative tangential velocity Cartesian coordinates, Components in x,y,z directions Experimental variables from I=1 to J Law of the wall coordinate =
V V(t) Vn Vt,cor Vx Va Vf vz vt r w W wz wt x,y,x Xi y+
u y
Greek Symbols
Beam divergence, off-axis angle, angle of absolute velocity Nomenclature.4
Nomenclature
Angle of relative velocity Total uncertainty Scattering volume dimensions in the x,y,z directions Normalized rms error, Effective diffusion coefficient Refraction angles of the laser beam in media 1 and 2 Refraction angle of the optical axis in media i Kronecker symbol Flow coefficient Efficiency Angle between vector U and the direction of measurement Wavelength of laser light Turbulent eddy viscosity Eddy viscosity Effective kinematic viscosity Laminar kinematic viscosity Turbulent kinematic viscosity Angle between normal to the interface and the laser beam, Beam crossing angle
x,y,z
A1, A2, B1, B2 i
ijk t t eff lam turb
Density of water Average axial square fluctuations, Integral time scale Average tangential fluctuations Rotation vector Angular velocity of rotation Head coefficient Angular rotation speed
1 2
Nomenclature.5
Nomenclature
Superscripts
~ =
Periodic or deterministic quantity Random of nondeterministic quantity Time averaged
Subscripts
d h s t ref 1 2 3
Design Hub Shroud Tip Reference At the inlet of the rotor At the outlet of the rotor At the outlet of the rotor
Nomenclature.6
Introduction
Introduction
Flows in hydropumps are three-dimensional, complex and unsteady. Several flow phenomena dominate the performance, efficiency, noise and vibrations of these machines, and as a result must be accurately predicted. Flow analysers are used to give a deeper insight into the underlying physics and enables one to identify and understand the predominant dynamic mechanisms, both steady and unsteady, which must be controlled and manipulated to enhance the performance design. In an analysis, an accurate description of both the local spatial and temporal organisation of the flow is essential. Numerical simulation techniques for studying flow phenomena have been evolved from the so-called one-dimensional (main line or critical path line) calculations and the basic two- and three-dimensional methods (Quasi 3D flow calculations) to the advanced (viscous) 3D Computational Fluid Dynamics. In CFD, the continuity and Navier-Stokes equations, which describe the motion of a fluid flow, are solved numerically. When predicting flow fields, CFD simulations provide a fast and cheap way of gaining insight on the distribution of local variables, such as pressure and velocity, at different operating conditions. Knowledge about local variables, which depend on the geometrical layout and the initial configuration, allows for predicting integral variables such as pressure rise and efficiency. This information is of great importance for engineers in design situations with the goal to produce a safe, reliable, efficient and economically competitive system that meets design point pressure ratio and flows with adequate stall margin and good efficiency potential. Computational Fluid Dynamics (CFD) in industry has come to play a crucial role in predicting and analysing fluid flows. This development has been driven by the availability of robust in-house and commercial CFD codes and by the massive increase in affordable computer speed and memory. This leads to a steady reduction in the costs of simulations compared to prototyping and model experiments. The challenge of CFD is thus to accurately predict the flow yield so that the testing of a new design can be done numerically and hence to minimise experimental testing. This reduces development time and costs considerably. The fundamental problem of CFD is that the accuracy of the solution depends on the assumptions made when modelling physical phenomena such as turbulence and unsteady phenomena. For modelling turbulence, most CFD codes are based on the RANS family (Reynolds Averaged Navier-Stokes). Through a time averaging of the Navier-Stokes equations, a statistical description of the evolution of the mean quantities is given and a model is applied in order to take into account the effect of the temporal variations due to turbulence. A wide range of such turbulence models is available, ranging from simple algebraic models (e.g. Baldwin-Lomax) to two equations models (k- and k- models) and more expensive algebraic or full Reynolds stress models. Due to the statistical approach the standard turbulence models also lack the ability to describe accurately the temporal variation of the flow, especially if the times scales of the unsteadiness is not much larger than that of the turbulence-scales. These are all phenomena characterising flows in turbomachinery. Despite the inadequacy of current turbulence models, designers are becoming increasingly dependent on (un)steady viscous three-dimensional CFD methods for the design of the machines. In order to extend the range of confidence in these numerical codes, some means of validation by comparing the numerical results with data from Experimental and Numerical Study of an Axial Flow Pump Intro.1
Introduction well-controlled experiments have to be performed. These data will not only improve the accuracy of the numerical simulations, but also the understanding of the complex physics of flows in hydropumps. These subjects will be handled in this thesis. There are not a lot of experiments that deal with the incompressible flow of water in a one-stage axial pump. In order to extend the range of our numerical prediction codes, flow models, and physical understanding of this incompressible flow problem, we need to perform a new experiment. A propeller pump was designed for internal flow studies and performance investigations. An axial flow pump facility at the department of Fluids Mechanics and Thermodynamics at the Vrije Universiteit Brussel has been developed to perform a scaled experiment of the incompressible flow of water through a two-blade-row propeller pump. This propeller pump with a specific speed Ns of 353, and a blade chord Reynolds numbers of 6.4.104, consists of a rotor-stator and is used for large volumes water distribution at relatively low heads. The facility is accommodated with a variety of instrumentation for the measurement of the performance, and is also provided with an optical access to the impeller for LDA measurements. A two-component laser Doppler anemometer (LDA) survey at design flow was done in nine different cross-sections from 40% axially chord upstream the tip rotor leading edge until 40 % axially chord downstream the tip rotor trailing edge with each survey having a dense distribution of measurement points. These experiments resulted in a large quantity of flow data throughout the impeller of the propeller pump. It includes information on the secondary flow in the impeller, tip and the wakes downstream of the rotor blades. These results of the pump performance and measured velocity field are validated with RANS simulations that used the algebraic turbulence model of Baldwin-Lomax and the one-equation turbulence model of Spalart-Allmaras. Chapter 1 gives an introduction to the subject of pump flow analysis and the current knowledge on the flow field present in axial flow pumps, as well as the numerical methods used in industry for simulating the flows in hydropumps. Chapter 2 describes the experimental test facility with his measurement instrumentation, and its uncertainties. Also the experimental propeller pump will be introduced. The flow field in the propeller pump is measured with a Laser Doppler Anemometry (LDA) and his technical specifications, together with the measurement procedure, will be set up in chapter 3. The working principles of the LDA will be explained in appendix A. Due to the curved casing of the propeller pump, a solution for the optical access for the laser Doppler anemometer is researched. Even the uncertainties of the LDA are broadly outlined in appendix B. The objectives designed into this test facility are the hydraulic performance testing of the axial flow pump and the measured flow field through the impeller of the axial flow pump. The measured quantities of the flow pump include the axial and tangential velocities, and also the turbulence Reynolds stress components. These measured quantities and performance of the axial flow pump will be discussed in chapter 4. The details of the measurement matrix will be handled and also the consistency of the measurements will be checked. In order to investigate the global pump performance and to make a numerical flow field analysis in the impeller of the axial flow pump, a numerical model has been created and the time marching finite volume based Navier-Stokes solver Euranus of FineTM Turbo was applied with the turbulence models of Baldwin-Lomax and SpalartAllmaras. The set-up of the model will be outlined and its numerical results will be Experimental and Numerical Study of an Axial Flow Pump Intro.2
Introduction first discussed in chapter 5, before the results of the experimental performance, together with the measured flow field in the impeller of the axial flow pump at design flow are compared and validated with the numerical CFD simulations in chapter 6. The comparison of the measured and the calculated results provides sufficient evidence of that the numerical methods predicts the complex flow field reasonable well. An attempt was made in chapter 7 to use the numerical results to shed new light into the complex flow structure inside the impeller. The numerical results of the turbulence models of Spalart-Allmaras will be further discussed using a topological analyse to clarify many of the (secondary) flow phenomena. On the basis of the achievements of this study, conclusions are drawn at the end of this thesis. Hereby, perspectives for the future use of this experimental test facility are discussed. The investigations reviewed here are mainly concerned with the effects of viscosity, and not with the effects of cavitations.
Intro.3
Chapter 1: Literature Review

1.1 Introduction
The aim of this chapter is to provide an overview of the published work that is relevant to the current investigation. Given the scope of the present research, a comprehensive review of all of the contributing research is not possible. This chapter briefly surveys some of the models used to synthesize the (noncavitating) performance and the characteristics of the flow field in axial flow pumps, provides a brief discussion of the sources of unsteady flow, loss generation mechanisms and secondary flow development in these machines, and ends with an insight into the mechanisms of interaction between the rotor and stator of axial pumps. First, the deal of Computational Fluids Dynamics (CFD) will be situated in the industry.
1.2 CFD of Hydropumps

Flows in axial flow pumps are three-dimensional, complex and unsteady. These flow phenomena dominate the performance, efficiency, noise and vibrations of these machines. In the design phase of axial flow pumps, the performance requirements are often expressed in terms of pressure rise, efficiency and power consumption and must be accurately predicted. Before the advent of CFD analysis methods for turbulent flows, designers were able to achieve high efficiency levels using experimental, empirical, and semi-empirical methods. Drawings were handmade and the design consisted of long series of operations with long loops of experimentation/redesign. Nowadays, the drawing board has been replaced by 3D CAD, which leads to great improvements in cost, quality and delay, allowing quick prototyping and numerical plants definition. Quickly generated grids of axial flow pumps are used with increasingly predictive CFD codes, with three main objectives: Comparison between different technical solutions or concepts for product optimisation. Some experimentation work for validation is sometimes necessary in this case. Advanced understanding of some key physical and technological topics in complex systems. As the results are usually not sufficiently predictive, the computation is then done in parallel with experimental work, which can be considered as a complementary tool to computation. The high levels of performance that have always been economically important for practically all turbomachinery applications.
The numerical simulations techniques for studying these phenomena have found widespread use, the state of the art with a perspective to hydromachinery, and more general to turbomachinery, has been reviewed at regular intervals, for example by Japikse (1976), McNally and Sockol (1985), Lakshminarayana (1991), Hirsch (1994), and by Denton and Dawes (1999). Further discussions of the different mathematical models applicable to the calculation of the flow in axial flow pumps, spanning the range of dynamic levels of approximations from the full time-dependent NavierStokes equations to linearized potential flow models and singularity methods, can be found in Cumpsty (1989). A more comprehensive discussion of flow models is given Experimental and Numerical Study of an Axial Flow Pump 1.1
Chapter 1: Literature Review by Hirsch (1988). Denton (1985), in an introduction to the application of the Euler equations to turbomachinery flows, gives a concise but very clear summary of the classical flow models and solution methods (streamlines curvature, streamfunction and potential flow), discussing the properties, advantages, drawbacks, and limitations of each (as applied to 2D flow). Even the first important application of CFD methods around fifteen years ago, the efficiency has increased further by 1% to 2%. Nevertheless, such small steps in efficiency represent quite large reductions in the remaining sources of loss. It is probable that unsteady 3D CFD methods with more accurate turbulence models suitable for the flow structures found in the turbomachinery flows will be essential to make these next small steps. The Computational Fluid Dynamics (CFD) in industry has become to play a crucial role in predicting and analysing fluid flows. This development has been driven by the availability of robust in-house and commercial CFD codes and by the massive increase in affordable computer speed and memory leading to a steady reduction in the costs of simulations compared to prototyping and model experiments. The challenge of CFD is thus to accurately predict the flow yield so that the testing of a new design can be done numerically and hence minimise experimental testing. This reduces development time and costs considerably. The inclusion of numerical testing makes the design process more cost-efficient and is thus an essential competition parameter.
1.3 Flow Analysis of Axial Flow Pump

In order to introduce the subject of pump analysis, this chapter emphasizes the numerical approach towards the prediction of the flow field in pumps. To understand the importance of employing advanced numerical methods for analysing pump flows a thorough discussion of the general characteristics of the flow field is provided.
1.3.1 Two-dimensional Analysis

In principle an axial flow pump is a relatively simple machine consisting of a rotating impeller with a set of stator blades enclosed within a stationary housing (see figure 1.1).
Figure 1.1: Axial flow pump
1.2
Chapter 1: Literature Review In an axial flow pump, the impeller pushes the liquid in a direction parallel to the pump shaft and adds momentum to the fluid flow through the unit by transfer of energy between the fluid and the rotating propeller blades. It results in a total pressure increases. Axial flow pumps are sometimes called propeller pumps, because they operate essentially the same as the propeller of a ship. Though simple in concept, axial pumps are very complicated due to the complex geometry. In order to make an analytical approach to predict the pump flows, the flow field in an axial pump can be approximated as quasi two-dimensional with streamlines following the geometrical layout of the hub, shroud and impeller blades. The energy exchange over the impeller can be estimated from a so-called one-dimensional approach analysing the idealized velocity polygons at the entry and exit of the impeller. The simplest approach to the study of axial flow compressors is to assume that the flow conditions prevailing at the mean radius fully represent the flow at all other radii. This two-dimensional analysis at the pitch line can provide a reasonable approximation to the actual flow, if the ratio of blade height to mean radius is small. When this ratio is large, however, as in the first stage of a compressor, a three-dimensional analysis is required. Some important aspects of three-dimensional flows in axial flow pumps are discussed later. Two further assumptions are that radial velocities are zero, and that the flow is invariant along the circumferential direction (i.e. there are no blade-toblade flow variations). Figure 1.2 shows a single stage axial-flow. The fluid essentially passes almost axially through a row of moving rotor blades and a row of fixed stator blades. The incompressible flow assumption is used because the pressure rise of the stage is usually small. The simplified vector diagram analysis assumes that the flow is onedimensional and leaves each blade row at relative velocity exactly parallel to the exit blade angle. The velocity diagram of the axial propeller pump comprises a rotor row followed by a stator row, as shown in figure 1.1. Fluid enters the rotor with absolute velocity v1 at angle 1 (Fig. 1.2). All angles are measured from the axial direction. The sign convention is such that the angles and velocities as drawn in figure 1.2 will be taken as positive throughout this chapter. The result of vectorially subtracting the blade speed U with the inlet velocity v1 is the inlet relative velocity w1 at angle 1, which ideally should be parallel to the rotor leading edge (the axial direction is the datum for all angles). Relative to the blades of the rotor, the flow is turned to the direction 2 at the outlet with a relative velocity w1. Clearly, by adding vectorially the blade speed U onto w2 gives the absolute velocity from the rotor v2 at angle 2. The stator blades deflect the flow towards the axis and the exit velocity is v3 at angle 3. Since there is no radial flow, the inlet and exit rotor speeds are equal, and the onedimensional continuity equation for uniform, steady flow requires that the axialvelocity component remains constant. The mass flow Q through the duct is described by A i viz , where is the density of water, Ai is the cross-sectional area of fluid flow cross section i and viz is the absolute axial flow velocity through the cross-section i. A1v1z =A 2 v 2z =A3 v3z =Q=cte 1.1
1.3
Figure 1.2: Velocity diagrams for a compressor stage
From the geometry of the velocity diagram, the normal velocity (or volume flow) can be directly related to the blade rotational speed u: u = .r = v1z .(tg1 +tg1 ) = v 2z .(tg 2 +tg 2 ) 1.2
Thus the flow rate can be predicted from the rotational speed and the blade angles (1,2). Applying the angular momentum theorem to the axial flow pump, the power P delivered to the fluid is thus:
P = .C =.
or H ideal =
d r r ( rmv ) =.Q. ( u 2 v2t -u1v1t ) dt
1.3
P 1 u = ( u 2 v 2t -u1v1t ) = ( v t ) gQ g g
1.4
r Where m is the mass, is the angular velocity, C is the torque, r is the radius from r the axis of rotation and v is the absolute velocity vector, vit is the velocity component mutually perpendicular to both the rotation axis and the radius vector, Hideal is the total head and g is the gravity acceleration.
This is the Eulers turbomachine equation, showing that the torque C, power P and total head Hideal are functions only of the rotor velocities u1,2 and the absolute fluid tangential velocities v1t,2t, independent of the axial velocities through the machine. Experimental and Numerical Study of an Axial Flow Pump 1.4
Chapter 1: Literature Review Meanwhile, since u=w1t+v1t and u=w2t+v2t, Eulers relation for the pump head becomes
gH ideal =u. ( w t )
1.5
Strictly speaking, equation 1.5 applies only to a single streamtube of radius r, but it is a good approximation for very short blades if r denotes the average radius.
1.3.2 Three-Dimensional Analysis

In the previous paragraph, the fluid motion through the blade row of axial flow pumps was assumed to be two-dimensional in the sense that radial (i.e. spanwise) velocities did not exist. The flow in the meridional plane was essentially two-dimensional, and that the effects of the velocities (and the gradients in the velocity or pressure) normal to the meridional surface were negligible. Moreover, it was tacitly assumed that the flow in a real axial flow pump could be calculated using a series of two-dimensional solutions for each radius. In doing so it is implicitly assumed that each annulus corresponds to a streamtube such as depicted in figure 1.3 and that the geometric relations between the inlet location, r1, and thickness, dr1, and the discharge thickness, dn, and location, r2, are known a priori.
Figure 1.3: Geometry of a meridional streamtube in a pump impeller (Brennen)
In practice this is not the case and quasi-three-dimensional methods have been developed in order to determine the geometrical relation, r2(r1). These methods continue to assume that the streamsurfaces are axisymmetric, and, therefore, neglect the more complicated three-dimensional aspects of the flow (secondary flows) as will be discussed below (section 1.5.3: Secondary Flow in Axial Flow Pumps). Nevertheless, these methods allow the calculation of useful axial flow pump performance characteristics, particularly under circumstances in which the complex secondary flows are of less importance, such as close to the design condition. When the axial flow pump is operating far from the design condition, the flow within a blade passage may have streamsurfaces that are far from axisymmetric. This two-dimensional supposition, in the sense that radial (spanwise) velocities did not exists, is also not an unreasonable assumption for axial flow pumps of high hubtip ratio. However, with hub-tip ratio less than about 4/5, radial velocities through a blade row may become appreciable, the consequent redistribution of mass flow (with
1.5
Chapter 1: Literature Review respect to radius) seriously affecting the outlet velocity profile (and flow angle distribution). 1.3.2.1 Theory of Radial Equilibrium Several approximate methods have been employed in order to determine r2(r1) as a part of a quasi-three-dimensional solution to the flow. Most of these are based on some application of the condition of radial equilibrium. In its simplest form, the radial equilibrium condition assumes that all of the terms in the radial momentum equations are negligible, except the pressure gradient and the centrifugal acceleration terms. Consider a small element of fluid of mass dm, shown in figure 1.4, of unit depth and subtending an angle d at the axis, rotating around the axis with tangential velocity, vt
Figure 1.4: A fluid element in radial equilibrium (Dixon, 1975)
at radius r. The element is in radial equilibrium so that the pressure forces balance the centrifugal forces:
( p + dp )( r + dr ) d - prd - p+
1 dp drd = dmvt2 / r 2
1.6
Writing dm = rd dr and ignoring terms of the second order of smallness the above equation reduces to, 1 dp vt2 = dr r 1.7
This assumption is differently embedded in several approaches to the solution of the flow. All of these flows in axial flow machines use a condition like equation 1.7 (or some more accurate version) to relate the pressures in the different streamtubes upstream of the rotor (or stator), and a similar condition to connect the pressures in the streamtubes downstream of the rotor (or stator). When these relations are combined with the normal continuity and energy equations for each streamtube (that connect the conditions upstream with those at the downstream location), a complete set of equations is generated, and a solution to the flow can be obtained. Notable
1.6
Chapter 1: Literature Review examples of this class of quasi-three-dimensional solutions are those devised by Hirsch (1988) and Denton (1985). The following example will illustrate one use of the ``radial equilibrium'' condition. We shall assume that the inlet flow is in radial equilibrium. This inlet flow is then divided into axisymmetric streamtubes, each with a specific radial location, r1. Some initial estimate is made of the radial location of each of the streamtubes at discharge (in other words an estimate of the function r2(r1)). Then an iterative numerical method is employed, in which the total pressure rise through each streamtube is evaluated. Hence, the pressure distribution at discharge can be obtained. Then the width of each tube at discharge is adjusted (r2(r1) is adjusted) in order to obtain the required radial pressure gradient between each pair of adjacent streamtubes. Subsequently, the process is repeated until a converged solution is reached. In some simple cases, analytical rather than numerical results can be obtained. More generally, it should be noted that quasi-three-dimensional analyses of this kind are often used for the design of axial axial flow pumps. A common objective is to achieve a design in which the total pressure pt is increasing (or decreasing) with axial position at the same rate at all radii, and, therefore, should be invariant with radial position. Combining the total 1 2 + vt2 with the condition for radial equilibrium (equation 1.7) pressure pt = p + vz 2 and that the flow is incompressible lead to:
dv v d 1 dpt = vz z + t ( rvt ) dr dr r dr
1.8
If the axial velocity vz is constant with radius, then equation 1.8 implies that the circumferential velocity, vt, must vary like 1/r. Such an objective is termed a ``free vortex'' design. Another basic approach is the ``forced vortex'' design in which the circumferential velocity, vt, is proportional to the radius r; then, according to the above equations, the axial velocity must decrease with r. More general designs in which vt=a.r+b/r (a and b being constants) are utilized in practice for the design of axial compressors and turbines, with the objective of producing relatively uniform head rise and velocity at different radii. However, in the context of pumps, most of the designs are of the ``forced vortex'' type. Stepanoff (1948) lists a number of reasons for this historical development. Radial equilibrium of the discharge flow may be an accurate assumption in some machines but not in others. When the blade passage is narrow (in both directions) relative to its length, the flow has adequate opportunity to adjust within the impeller or rotor passage, and the condition of radial equilibrium at discharge is usually reasonable. This is approximately the case in all pumps except propeller pumps of low solidity. However, in many compressors and turbines, the blade height is large compared with the chord and a radial equilibrium assumption at discharge is not appropriate. Under these circumstances, a very different approach utilizing an ``actuator disc'' has been successfully employed. The axial width of the blade row is modelled by an infinite small thin blade row, while the space-chord ratio, the blade angles and overall length of the machine are maintained constant. As the deflection through each blade row for a given incidence is fixed by the cascade geometry, a blade row of reduced width may be considered to affect the flow in exactly the same way as the original row. While the tangential velocity undergoes an abrupt change in direction, the axial and radial velocities are continues across the disk. An approximate solution to the velocity fields upstream and downstream of the
1.7
Chapter 1: Literature Review actuator can be found in terms of the axial velocity distributions far upstream and downstream of the blade row. The detailed analysis of the "actuator disk" exceeds the scope of this thesis, involving the solution of the equations of motions, the equation of continuity and the satisfaction of boundary conditions at the wall and disc. The form of the approximate solution is of considerable interest and is quoted in Dixon (1975). 1.3.2.2 Computer-Aided Methods for Solving the Through-flow Problem Although actuator disc theory has given a better understanding of the complicated meridional (the radial-axial plane) through-flow problem in axial flow pumps of simple geometry and flow conditions, its application to the design of axial flowcompressors has been rather limited. The extensions of actuator disc theory to the solution of the complex three-dimensional, compressible flow in compressors with varying hub and tip radii and non-uniform total pressure distributions were found to have become unwieldy in practice. In recent years advanced computational methods have been successfully evolved for predicting the meridional compressible flow in axial flow pumps with flared annulus walls. The two different methods are: 1. Streamline curvature 2. Matrix through-flow analysis. Both methods solve the same equations of fluid motion, energy and state for axisymmetric flow through a turbomachine with varying hub and tip radii and therefore lead to the same solution. In the first method the equation for the meridional
2 velocity cm = vr2 + vz
in a plane (at x=xa) contains terms involving both the slope
and curvature of the meridional streamlines which are estimated by using a polynomial curve fitting procedure through points of equal stream function on neighbouring planes at (xa-dx) and (xa+dx). The major sources of difficulty are the accurately estimation of the curvature of the streamlines. In the second method a grid of calculating points is formed on which the stream function is expressed as a quasilinear equation. A set of corresponding finite difference equations are formed which are then solved at all mesh points of the grid. A more detailed description of these methods is rather beyond the scope of the present work. Though the one-dimensional and the streamline curvature approach have found widespread application in the design process they are of little value in predicting local flow properties as well as secondary flows or separation and recirculation. From the above discussion it is evident that these phenomena largely influence the physics of the flow. It is therefore clear that the analytical and empirical methods have only a limited accuracy, especially at off-design operating conditions, and it appears unlikely that the loss coefficients provide more than an indication of the influence on the integral parameters and the general performance. Analysing the relation between the local flow behaviour, e.g. velocities and pressure, and the geometrical layout of the pump requires a profound insight into the complexity of the flow. Computational Fluid Dynamics (CFD) based on solving the three dimensional Reynolds averaged Navier-Stokes (RANS) equations with the simple turbulence models has therefore become a fully integrated tool in the early stages of industrial design today.
1.8
Chapter 1: Literature Review 1.3.2.3 Computational Fluid Dynamics A thorough knowledge of all significant inviscid effects (blade blockage, flow turning, finite hub/tip ratio, etc.) and viscous effects (boundary-layer growth, energy dissipation, etc.) is essential in the accurate prediction of the flow in all turbomachinery. Relevant to this, the availability of computer with large storage capacities and fast computation times greatly enhance the possibility of numerically solving the complete equations of motion. It is widely accepted that the incompressible time-averaged Navier-Stokes equation (equation. 1.10) together with the continuity equation (equation 1.9) comprise a closed set of equations, and which should be supplemented by appropriate boundary and initial conditions to form a r r complete description of the fluid velocity u( x, t ) and pressure field p( x, t ) in flow pumps.
U j x j =0
1.9
DU i U i = Dt xi x j
ui' u 'j 1 p D = +U j with x j Dt t x j xi
1.10
r Here p , , and u i denote to the pressure, density, kinematic viscosity and the Reynolds-averaged velocity based on Cartesian tensor notation, respectively. The time averaged Navier-Stokes equations have more unknowns that the number of equations. In order to solve this closure problem, it is necessary to model the Reynoldsstress tensor, u 'i u ' j , which appears in the time averaged momentum equation. To determine the unknown Reynoldsstress components in equation 1.10, different turbulence models have been adopted. The turbulence models used in turbomachinery could be summarised as follows: Zero-equation models such as Prandtls mixing length model. The eddy viscosity is proportional to the square of a length scale and to the rate of flow deformation. The length scale must be prescribed empirically and depends on the characteristics of the flow. This model is applicable when the flow has a privileged direction, such as boundary layer, shear stress, etc. The model is unsuitable for most complex flows in turbomachinery applications because it is very difficult to estimate the distribution of the mixing length by rotation, curvature, three-dimensionality, separation, free stream turbulence, large scales unsteadiness and other complex strain effects. Even this model is still widely used in both internal and external flows. The Baldwin Lomax turbulence equation is a zero-equation model and will be used in this work. In the chapter of the numerical simulations of the axial pump flow, this model will explained more in detail. One-equation models: A suitable transport equation is added, usually for the turbulent kinetic energy (such as the one which has been proposed by Kolmogorov in 1942 and Prandtl in 1945). The main shortcoming concerns the length scale. Its transport is not considered, which is important in separated flows. This length scale must be prescribed empirically. In practice, this model gives good results for the same cases as Experimental and Numerical Study of an Axial Flow Pump 1.9
Chapter 1: Literature Review Prandtls mixing length model but with higher computational costs. The turbulence model of Spalart-Allmaras will be used in this work and will be more explained in chapter 5. Two equation models: there are many possible choices for the second variable. The most popular is the k- model, which is a good compromise between a general description and economy of use for many problems. The standard model is valid for high turbulent Reynolds numbers, but is not valid in the near-wall viscosity affected region. However there are a few versions of low Reynolds number models that need a very fine mesh. In addition, the use of the eddy-viscosity concepts imposes isotropy. This assumption is too simple when there is rotation, and in flows with streamlines curvature (swirl), which is the case in most of engine applications. The consequence of these limitations is that a larger number of k- model variants involving a modified equation have been developed. Models using more than two equations: Reynolds-stress models. Despite the fact that this type of model provides a more rigorous and realistic approach and that it captures anisotropy effects automatically, the results are disappointing compared with the classical eddy-viscosity models. They are much more complex, computationally more expensive and less stable. These equations with the Navier-Stokes equation are a set of coupled, non-linear, mixed elliptic-parabolic partial differential equations that require special techniques for numerical solution. It is possible, in principle, to simulate any turbulent flow by solving the foregoing equations with appropriate boundary conditions using a suitable numerical procedure. This thesis will not review the major characteristics of these equations and difficulties associated with solving them numerically. The manual of the used CFD code (FineTM/Turbo User Manual) summarizes the major issues arising from solving these flow problems by the algorithms used for this thesis. All turbomachinery specialists recognise that the ultimate limit in predictive capability in the performance estimation of turbomachinery is related to the uncertainties arising from the turbulence modelling. Turbomachinery internal flows are among the most complex to encounter in fluid dynamic practice, as follows: The geometry of the internal flow passage is highly complex, with stationary vanes and rotating blades and all manner of curved passages, flow leakage paths, obstructions and steps in the flow channels. The flow is generally three-dimensional, with laminar, transitional and turbulent regions, Extremely complex flow details occur in the turbulent regimes due to the appreciable pressure gradients, rotation, curvature, high turbulence, heat transfer, separation and interaction of wakes and boundary layers. Strong vortices often dominate the flow, such as secondary flow vortices and tip clearance vortices. The flows are generally also unsteady in both the rotating and absolute frames of reference.
1.10
Chapter 1: Literature Review As a result of the complexity of the flow and the geometry, it is hardly surprising that most technical papers on turbulence in turbomachinery applications are unable to provide clear recommendations on the most suitable turbulence model to be used. Due to the shortcomings associated with the assumption of isotropy, the models fail for flows with rotation, curvature, strong three-dimensionality and separations. More complex turbulence models do not necessary improve the modelling of the flow in turbomachinery and is often not universally applicable to the many types of flow structures to be found in axial flow pumps. Despite the inadequacy of current turbulence models, designers are becoming increasingly dependent on steady viscous three-dimensional CFD methods for the design of the machines. The key to the extensive use of CFD is that it helps the designer: To understand and control global feature of the flows (such as incidence, blade loading, velocity levels, mean flow direction, secondary flows and unsteady effects) To produce new designs that achieve high performance levels by copying good flow features from earlier machines with less need for large experimental testing programs to complete new design in a shorter time and at a lower cost. The drive to reduce the costs and to increase performance ensures that further improvements in simulation methods are given high priority in industry and in universities, including the validation of appropriate CFD tools and testing of turbulence models. One alternative way to solve these equations could be the LES approach. Nevertheless, the application of LES still presents a number of issues such as: excessive computational effort at high Reynolds number, difficulties in specifying initial and boundary conditions and the necessity to perform 3D time dependant simulations for all applications. This thesis will use only the RANS equations with a zero - and one-equation turbulence models, which numerical results will be validated with the measured data of the propeller pump.
1.4 Global Performance Characteristic

The performance characteristic represents the overall performance of a hydropump in the form of total pressure ratio and efficiency for a series of constants speeds, plotted to a base of a mass or volume flow (see figure 1.5). Though pumps are designed to operate in the best efficiency point (design point), they are expected to have a stable and efficient performance over a wide range of flows. For reasons of control, they are even expected to have very specific performance characteristics at off-design. It is common practice to illustrate the off-design performance of a pump through a so-called Q-H curve where Q is the volume flow rate and H is the total pressure head. A typical curve is shown qualitatively in figure 1.5 and compared to the ideal curve predicted form.
1.11
Figure 1.5: Q-H curve is obtained by subtracting of hydraulic losses from ideal head Hi (Stepanoff, 1948).
Figure 1.6: Schematic of noncavitating performance of an axial flow pump
The head Hs seems to be significantly lower than the idealized prediction Hi. This is primarily due to recirculation at low flow rates, friction and leakage losses, which increases with the flow rate. The influence of these phenomena on the performance can be approximated through numerous empirical loss coefficients, see e.g. Stepanoff (1948) or Dixon (1975). As the losses largely depend on the type and individual design of the pump, the tuning of these loss coefficients depends on engineering experience of experimentally or numerical testing and an unified guideline can therefore not be provided. Figure 1.6 plotted the performance characteristic of an axial flow pump. The noncavitation performance consists of the head coefficient, , as a function of the flow coefficient , where the design conditions can be identified as a particular point on the () curve. The () curve terminates at the left side in a point, which belong to the stall line (or surge line). This line marks the limit of stability of axisymmetric flow. To the left of this line large oscillations of the mass flow rate may occur (called surge) or severe self-induced circumferential flow distortions may rotate around the annulus (rotating stall), or a combination of both phenomena may appear. Rotating stall induces large vibratory stresses in the blades of the impeller and is therefore often unacceptable for structural reasons. Surge can be intolerable from the point of view of system operation and can also lead to high blade and casing stress levels. Thus, no matter which type of instability appears when the stall line is crossed,
1.12
Chapter 1: Literature Review the stall line generally represents a limit to the useful operation of the machine and is therefore to be avoided. The instabilities in the type of pumping system under consideration arise in general due to the presence of stall. Looked at from the point of view of the individual diffusing passages in the rotor, stall generally implies separation of the flow from one or more of the passage walls. However, impeller blade rows consist of many of these diffusing passages in parallel so that phenomena can occur which do not happen with a single airfoil or diffusing passage. One of the most striking of these is rotating stall. This is a flow regime in which one or more stall cells propagate around the circumference of the rotor with a constant rotational speed, which is usually between twenty and seventy percent of the rotation speed. In the cells the blades are severely stalled. Typically, in these regions, there is negligible net through-flow, with areas of local reverse flow. The cells can range from covering only part of the span (either at the root or at the tip) and being only a few blades in angular width, to covering the full span and extending over more than 180 degrees of the compressor annulus. The basic explanation of the mechanism associated with the onset of stall propagation was first given by Emmons et al. (1955) and can be summarized as follows. Consider a row of axial impeller blades operating at a high angle of attack, such as is shown in Figure 1.7. Suppose that there is non-uniformity in the inlet flow such that a locally higher angle of attack is produced on blade B, which is enough to stall it. If this happens, the flow can separate from the suction surface of the blade so that a substantial flow blockage occurs in the channel between B and C. This blockage causes a diversion of the inlet flow away from blade B and towards C and A (as shown by the arrows), resulting in an increased angle of attack on C and reduced angle of attack on A. Since C was on the verge of stall before, it will now tend to stall, whereas the reduced angle of attack on A will inhibit its tendencies to stall. The stall will thus propagate along the blade row in the direction shown, and under suitable conditions it can grow to a fully developed cell covering half the flow annulus or more. In this fully developed regime, the flow at any position is quite unsteady; however, the annulus averaged mass flow is steady, with the stall cells serving only to redistribute this flow. The onset of rotating stall is thus associated with an instability, which arises due to the stall of the impeller blade passages. As far as the overall system is concerned, this can be regarded as a localized instability. In contrast to the behaviour during rotating stall, the annulus averaged mass flow and the system pressure rise during surge undergo large amplitude oscillations. The stall point of hydro pump is essentially set by the geometrical parameters of the machine, but the most important of these is the inlet distortion.
1.13
Figure 1.7: Physical mechanism for inception of rotating stall (Brennen).
1.5 Characteristics of the 3D Flow Field

The flow in an axial flow pump is three-dimensional and should be treated as such in order to achieve a solid knowledge of the flow field and understand the mechanisms causing the performance do diverge from the ideal case. An insight into the physics of the flow can be derived through a discussion with special attention to secondary and partial load phenomena. When looking into the flow pattern in axial pump impellers, it is essential first to understand the influence of the forces acting upon the flow.
1.5.1 Forces Active On the Flow

The Navier-Stokes equation written in the rotating frame of reference is given by:
r r r r r r r p W r r + W .W = 2 W ( r ) + 2 W 1.11 t r r r r r Where 2 W is the Coriolis acceleration and ( r ) is the centripetal
( )
acceleration vector, which is exerted on the particle because of the rotation of the relative system. In contrast to most other applications the Coriolis acceleration r r r r 2 W , is of primerimportant in the field of hydromachinery. In a rotor, W , is r zero only if and W are parallel. However, if the relative velocities are everywhere parallel with the axis the corresponding rotor will not be able to change the energy level of the fluid (see equation 1.4). The existence of a Coriolis acceleration is Experimental and Numerical Study of an Axial Flow Pump 1.14
Chapter 1: Literature Review necessary for producing energy changes in rotors; in fact, large energy changes require large Coriolis accelerations. These accelerations are also responsible for the fundamentally different flow patterns in rotor and stator. The inertia forces (centrifugal and Coriolis forces) have a significant influence on the flow field. The most obvious effect is the setting up of a radial pressure field due to the centrifugal force acting on the rotating fluid. As indicated implicitly in the name, this is the main source of the pressure rise obtained in centrifugal pumps. The Coriolis force is perpendicular to the flow acting from suction to pressure side and causes a build up of pressure at the pressure side of the impeller blade making a transfer of torque possible.
1.5.2 Prerotation
Perhaps no aspect of hydromachinery flow is more misrepresented and misunderstood than the phenomenon of prerotation. While this belongs within the larger category of secondary flows (dealt in the section), it is appropriate to address the issue of prerotation separately, not only because of its importance for the hydraulic performance, but also because of its interaction with cavitation and stall. It is first essential to distinguish between two separate phenomena, which both lead to a swirling flow entering the pump. These two phenomena have very different fluid mechanical origins. Here, we shall distinguish them by the separate terms, backflowinduced swirl and inlet prerotation. Both imply a swirl component of the flow entering the pump. Improper entrance conditions and inadequate suction approach shapes may cause the flow in the suction pipe to spiral from some distance ahead of the actual impeller entrance. This phenomenon is called prerotation, and it is attributed to various operational and design factors. The flow has axial vorticity (if the axis of rotation is parallel with the axis of the inlet duct) with a magnitude equal to twice the rate of angular rotation of the swirl motion. Moreover, there are some basic properties of such swirling flows that are important to the understanding of prerotation. These are derived from the vorticity transport theorem. In the context of the steady flow in an inlet duct, this theorem tells us that the vorticity will only change with axial location for two reasons: (a) because vorticity is diffused into the flow by the action of viscosity, or (b) because the flow is accelerated or decelerated as a result of a change in the cross-sectional area of the flow. The second mechanism results in an increase in the swirl velocity due to the stretching of the vortex line, and is similar to the increase in rotation experienced by figure skaters when they draw their arms closer to their body. When the moment of inertia is decreased, conservation of angular momentum results in an increase in the rotation rate. Thus, for example, a nozzle in the inlet line would increase the magnitude of any pre-existing swirl. The backflow is caused by the leakage flow between the tip of the blades of an impeller and the pump casing. The circumstances are depicted in figure 1.8. Below a certain critical flow coefficient, the pressure difference driving the leakage flow becomes sufficiently large that the tip leakage jet penetrates upstream of the inlet plane of the impeller, and thus forms an annular region of backflow in the inlet duct (Predin, 2003). After penetrating upstream a certain distance, the fluid of this jet is then entrained back into the main inlet flow. The upstream penetration distance increases with decreasing flow coefficient, and can reach many diameters upstream of the inlet plane.
1.15
Figure 1.8: Lateral view of impeller inlet flow showing tip leakage flow leading to backflow (Brennen).
Obviously the backflow has a high swirl velocity imparted to it by the impeller blades. But what is also remarkable is that this vorticity is rapidly spread to the core of the main inlet flow, so that almost the entire inlet flow has a nonzero swirl velocity. The rapidity with which the swirl vorticity is diffused to the core of the incoming flow remains something of a mystery, for it is much too rapid to be caused by normal viscous diffusion. It seems likely that the inherent unsteadiness of the backflow (with a strong blade passing frequency component) creates extensive mixing which effects this rapid diffusion. However it is clear that this backflow-induced swirl, or prerotation, will affect the incidence angles and, therefore, the performance of the pump.
1.5.3 Secondary Flow in Axial Flow Pumps

The flow in the close vicinity of the blade-tip region of ducted propellers and similar axial flow pumps can be quite complex due to the presence and dynamic interactions of the tip-leakage vortex, the blade trailing edge vortex, the gap shear flow, the wall (casing) boundary layer, and the wake from the blade boundary layer. This tip region flow is important as it has the potential to contribute to a substantial loss in total efficiency and pumping head. Denton (1993) gave an extensive review of the loss generating mechanisms in turbomachinery. The losses in the propeller pump can be mainly classified as 1. Profile losses due to blade boundary layers and their separations, possibly including shock/boundary layer interaction in high speed condition, and wake mixing 2. Endwall boundary layer losses, including secondary flow losses and tip clearance losses
1.16
Chapter 1: Literature Review 3. Mixing losses due to the mixing of various secondary flows, such as the passage secondary flow (passage vortex) with leakage flow (or tip leakage vortex) Secondary flows in the propeller pump are defined normally as the difference between the real flow (including small-scale turbulent fluctuations) and a primary flow. The primary flow can be referred to, for example, as idealized axisymmetric flow or midspan flow. The secondary flow arises from the presence of endwall boundary layer and depend mainly on blade to blade and radial pressure gradients, centrifugal force effects, blade tip clearance, and the relative motion between the blade ends and the annulus walls. Although the secondary flow structure in a blade passage has a strong dependence on the incidence angle or flow inlet angle, Reynolds number and blade profile, their qualitative features are quite general. Normally, the secondary flow relates directly to the generation and evolution of various concentrated vortices, such as passage vortex, leading edge horseshoe vortex, corner vortex, tip leading vortex, scraping vortex in an unshrouded rotor, blade trailing edge vortex filament and shed vortex inside wakes. Hence the following reviews are sectioned in the vortex terminology. Secondary flow pattern have been presented by many authors in the literature, Hawthorne (1955), Vavra (1960), Lakshminarayana and Horlock (1963), Salvage (1974), Inoue and Kuroumarou (1984), Kang and Hirsch (1993) and Zierke et al. (1994). In this review, only the models given by Hawthorne (1955), Lakshminarayana and Horlock (1963) and Inoue and Kuroumarou (1984) are shown in Figs. 1.9, 1.10 and 1.11.
Figure 1.9: Secondary low pattern, after Hawthorne (1955)
Hawthornes model describes the classical secondary flow vortex system for the first time (see figure 1.9). This system presents the components of the vorticity in the flow direction when a flow with inlet velocity is deflected through a cascade. The passage vortex represents the distribution of secondary circulation. The vortex sheet at the trailing edge is composed of the trailing filament vortices and the trailing shed vorticity.
1.17
Figure 1.10: Secondary flow patterns, after Lakshminarayana and Horlock (1963).
In the model of Lakshminarayana and Horlock (1963), figure 1.10, tip leakage flow and relative motion influence are presented with the tip leakage vortex and scraping vortex. The concentration of the trailing vortex filament is described. Based on the measurement behind a rotor blade row, Inoue and Kuroumarou (1984), figure 1.11, proposed a three-dimensional structure of the vortices inside and behind a compressor rotor passage. In the following, losses in boundary layers and all the secondary flow phenomena will be described one by one.
Figure 1.11: Secondary flow patterns, after Inoue and Kuroumarou (1984)
Passage Vortex In order to accomplish either work or energy conversion in an axial flow pump, the streamlines passing trough a blade row must experience deflection. This deflection in the inviscid flow region would be balanced by the centrifugal force along the streamlines and the static pressure gradients across the passage. From the assumption of boundary layer theory that the potential in the inviscid region spreads its gradient Experimental and Numerical Study of an Axial Flow Pump 1.18
Chapter 1: Literature Review into the endwall boundary layer, a balance of the centrifugal force on the low momentum fluid within the boundary layer causes a movement of the fluid from the high pressure side (the pressure side) to the low pressure side (the suction side). The fluid outside the boundary layer moves from the suction side to the pressure side, so that the fluid in the vicinity of the wall will be turned as a vortex motion and spirally stretch downstream (see figure 1.12).
Figure 1.12: Secondary flow in bend (Vavra, 1960)
This vortex is the so-called classical passage vortex. A twin vortex with the opposite rotation sense will form near the opposite wall in the passage. With the existence of the passage vortex, the flow close to the wall is overturned, while the flow far away from the wall is underturned. Hence the passage vortex directly influences the energy conversion of flows passing through the cascade. Its mechanism has been investigated at first by Carter and Cohen (1946) for uniform flows with thin boundary layers and by Squire and Winter (1951) for nonuniform flows. The initial model describing the passage vortex was proposed by Hawthorne (1955) as show in figure 1.9, including the trailing vortex sheet that will be reviewed in the following. Including real flow effects, the picture will change due to the interactions of the low energy boundary layer material flowing into the corner formed by the side wall and the suction side of the blade and the main stream, resulting in the formation of corner vortices. Trailing Vortex The trailing vortex, stretching from blade trailing and existing inside the wake, can be divided into two parts with different origins, as trailing shed vortex and trailing filament vortex. The trailing shed vortex is due to the spanwise change of blade circulation and is analogue to that of an isolated wing with finite span. The trailing filaments vortex arises due to the stretching of the inlet vortex filaments when passing through cascade with different velocities between the suction and pressure sides. The two trailing vortices have the same sense but are opposite to the passage vortex.
1.19
Chapter 1: Literature Review Tip Leakage Vortex In unshrouded axial flow pumps a gap exists between moving rotor blade tips and casing wall or between rotating hub and stator blade ends. Due to the sudden discharge of blade loading at the tips (or ends), a large pressure difference, directing from the pressure side to suction side and tending to equalize, remains across the tip gap. The pressure difference will drive the fluid in the pressure side / endwall corner to the suction side / endwall corner. This is the pressure driven leakage flow or simply leakage flow. The leakage flow, against the passage transverse pressure gradient, rolls up into a concentrated vortex, tip leakage vortex, lying in the corner of the endwall and suction surface and moves along the mainstream direction. Its rotation sense is opposite to the passage vortex. Scraping Vortex In the relative reference system, rotor blades are stationary but the casing wall is moving. Because of the action of the fluid viscosity, the moving wall induces a tangential shear flow or a vortex sheet inside the wall boundary layer. The vortex sheet will separate from the wall and roll up into a concentrated vortex, so called scrapping vortex. Its sense is the same as the passage vortex in turbine rotors but in opposite to the passage vortex in compressor rotor. In a compressor, the casing wall rotates in the direction from the pressure side to suction side of the same blade, and the scraping vortex will occur in the pressure side corner. Correspondingly, the scraping vortex in a turbine rotor will occur in the suction side corner. Positions of the scrapping vortex created within turbine or compressor stators near hub walls are completely analogue. Horseshoe Vortex When a shear flow along a plate approaches a blade standing on the plate, the shear flow will be separated from the plate and roll up into a vortex in front of the blade leading edge. This vortex is called horseshoe vortex due to its particular shape, flowing around the blade in both sides. The two legs, pressure side and suction side leg, of the horseshoe vortex have different evolutions on the way to downstream and are significantly influenced by the traverse pressure gradient and other vortices. The pressure side leg, with the same sense as the passage vortex, will be, depending upon the blade turning, merged with the passage vortex, or develops individually towards downstream when the turning angle is small. The synchronous evolution of horseshoe and the passage vortex has been described by, for example, Moore and Smith (1984), Sieverding and Van de Bosch (1983) and Inuoe and Kuroumarou (1984). The suction side leg, however, may appear as a corner vortex or counter vortex (Langston et all, 1977). In the passage of compressor blade rows, evolution of horseshoe vortex will also be significantly influenced by the severe streamwise adverse pressure gradient. In rotor tip regions, if the tip gap is small, the horseshoe vortex will still appear but with less strength than in zero tip clearance case (Zierke et al, 1994). Corner vortex The corner vortex rotates in the opposite sense to the passage vortex and lies invariably right in the endwall corners. Normally its size is too small to be visualised and measured. A corner vortex in the pressure side corner may be similarly generated due to the spanwise secondary flow.
1.20
Chapter 1: Literature Review Other Secondary flows Except for the secondary flows correlated with the various vortices mentioned above, there also exist two other kinds of secondary flows, the radial flows near blade surface and radial mixing flows in wakes. From the equilibrium equations, the centrifugal forces will balance the radial pressure gradients along the flow paths in the potential flow region. As in the case of the cross flow in the endwall boundary layer, the low energy fluid inside the blade surface boundary will also be pushed by the radial pressure gradient. The volume of the radial flow certainly depends on the thickness of the blade surface boundary layers. All the reviewed vortices lie mainly in the streamwise direction. But there are also other vortices, existing in an axial compressor blade passage, lying in other directions. For example, both separation bubbles on blade surfaces and trailing edge separations may form vortices lying in the radial direction. Furthermore, Von Karman vortex streets in blade wakes also stretch in the radial direction. It should be addressed before closing the section that the secondary flows and the associated vortices cannot always be described separately from the experimental results. Some of them may be mixed, merged, or interacted with each other.
1.5.4 Review on Tip Leakage Flow

Over the life span of a turbomachine, the machine efficiency and operating margin deteriorate due to the increase in tip gap spacing. The engineering importance of tip gap flows and the complexity of the physics they contain, has stimulated numerous previous studies of these phenomena. Recent studies of tip-leakage flow have involved various turbomachinery configurations, including linear cascades, compressors, turbines and pumps. Experimental studies of tip-gap flows in cascades have been reported by Kang and Hirsch (1989,1993), while Laksminarayana et al (1986), Inoue et al (1986,1989), Suder and Celestina (1994), and Foley an Ivey (1994) have conducted measurements in compressors. Pump flow experiments have been reported by Graham (1986) and Zierke et al (1994a, 1994b). Depending on the parameters and the geometries involved, either of these may be the primary driver in a given application. When the gap is large and the blade loading is high, the tip-gap flows are generally dominated by pressure effects, while very narrows gaps or low blade loading lead to gap flows that are more dependent on the relative movement between the blade tip and the casing. In general, however, inertial forces in this narrow gap region are sufficient to generate a tip leakage vortex on the suction side of the blade that not only produces substantial losses, but may also initiate blade surface cavitation. Beginning in the pressure surface boundary layer near the rotor blade tip of a turbomachine, a vortex sheet passes through the clearance. Under the influence of the induced velocity field of the vortex sheet, the free edge of the vortex sheet curls over and takes the form of a spiral with a continually increasing number of turns: the roll up of the tip leakage vortex. As opposed to rotor blade tip without an endwall, a rotor blade tip with an endwall clearance results in a higher blade loading that produces a larger pressure difference between the pressure and the suction sides of the blade. This pressure difference produces a jet of fluid that caries the vortex sheet through the clearance. Storer and Cumpsty (1991) show that this distinct jet of low-loss fluid occurs downstream of the minimum pressure location of the suction surface at about Experimental and Numerical Study of an Axial Flow Pump 1.21
Chapter 1: Literature Review 15% of the chord at the design clearance. This location is also in good agreement with that observed by Zierke et al. (1994a). With his high Reynolds number pump facility, they were able to visualize this vortex as it moved away from the suction face by reducing the tunnel pressure to the point that cavitation appeared at the core of the vortexes. Photographs and videos showed the existence of some cavitating bubbles originating in the clearance near the rotor blade leading edge. Most of the cavitation, however, appears to begin near the suction surface at 15 percent chord, a position very close to the measured minimum pressure point on the suction surface at 90 percent span. Inuoe and Kuroumaru (1989) and Storer and Cumpsty (1991) contend that the onset of the vortex sheet rolling-up into the tip leakage vortex occurs near the point of minimum pressure on the suction surface. At this location, the pressure difference across the clearance reaches a maximum, creating a larger leakage jet. The vortex remains quite close to the suction surface until near 80 percent chord, where it begins to migrate away from the blade in the circumferential direction as the blade rotates in the other direction. The circumferential position of the vortex within the passage depended weakly on the tip clearance. For large clearance, the vortex was positioned along the suction surface, for smaller clearance, the vortex was approximately the maximum blade thickness away from the suction surface at the trailing edge. There experiment shows also that the trailing edge separation vortex lies closer to the casing that the tip leakage vortex, with both vortices rotating with the same sense. These vortices pass one another, at different radii, near the trailing-edge plane and the interaction of these two vortices may affect the vortex core trajectory. This interaction should occur for any blades where the centrifugal effects create a radial flow on the suction surface with some trailing edge separation. While the trailing edge separation vortex is smaller in strength than the tip leakage vortex, it can induce an additional downstream velocity component onto the tip leakage vortex that may help the trajectory move back into the relative flow direction. Also, since the two vortices rotate with the same sense, the trailing edge separation vortex will rotate around the stronger tip leakage vortex if the two vortices are close enough together. As the tip leakage vortex convects downstream, the position of the vortex core moves radially. The flow visualization also showed the spatial tip leakage vortex wandering and kinking as it convected downstream, probably due to the inlet flow distortions, ingested free stream turbulence, and/or the instability of the vortex core.
1.5.5 Losses in Boundary Layers

The efficiency of turbomachine bladings and consequently the overall performance of the machine are strongly dependent on the boundary layer development on the blades and their separations (if their is any), tip clearance flows and wakes, which are responsible for energy losses existing in blade passage. For these reasons the knowledge of the boundary layer and its development plays an important role for the better understanding and further improvement of axial flow pumps. The flow in axial flow pumps is highly unsteady and turbulent because of the interaction between the rotor and the stator blade rows due to the wakes and the effect of the potential flow yield. In particular the periodic influence of the passing wakes plays an important role for the unsteady boundary layer development on the blades. Generally, there are three fundamental modes for the boundary layer transition from the laminar to the turbulent state:
1.22
Chapter 1: Literature Review The first one is denoted as natural transition. It is observed for low levels of free stream turbulence. The transition process starts with a weak instability in the laminar boundary layer, which develops through various stages to a fully turbulent boundary layer. The second mode is known as the bypass transition. In this case transition is initiated by disturbances in external flow (e.g. high free stream turbulence, wakes) and bypasses the natural transition. This is the predominated route of transition in axial flow pumps, where the natural transition process is periodically disturbed by the incoming wakes (wake-induced transition). A third node occurs if the boundary layer separates. It is therefore known as separated flow transition and can be found in compressors and low-pressure turbines. There are several parameters influencing the boundary layer development on the blades. These are the properties of the incoming wakes; free stream turbulence, blade loading, Reynolds number, the profile pressure distribution and others. Depending on these parameter and as a result of the periodic wake influence, unsteady, highly complex boundary layer behaviour can be observed, where a combination of different forms of transition can be found (multimode transition). Numerous investigations of the boundary layer for steady and unsteady incoming flow is known from literature.
1.5.6 Cavitation
Cavitation in vortical structures is a common, albeit complex, problem in engineering applications. Cavitation vortical structures can be found on the blade surfaces, in the clearance passages, and at the hubs of various types of turbomachinery. It is sufficient at this juncture to observe that cavitation is generally a malevolent process, and that the deleterious consequences can be divided into three categories. First, cavitation can cause damage to the material surfaces close to the area where the bubbles collapse when they are convected into regions of higher pressure. Cavitation damage can be very expensive, and very difficult to eliminate. For most designers of hydraulic machinery, it is the preeminent problem associated with cavitation. Frequently, one begins with the objective of eliminating cavitation completely. However, there are many circumstances in which this proves to be impossible, and the effort must be redirected into minimizing the adverse consequences of the phenomenon. The second adverse effect of cavitation is that the performance of the pump, or other hydraulic device, may be significantly degraded. In the case of pumps, there is generally a level of inlet pressure at which the performance will decline dramatically, a phenomenon termed cavitation breakdown. This adverse effect has naturally given rise to changes in the design of a pump so as to minimize the degradation of the performance, or to put it another way, to optimise the performance in the presence of cavitation. One such design modification is the addition of a cavitating inducer upstream of the inlet to a centrifugal or mixed flow pump impeller. Another example is manifest in the blade profiles used for supercavitating propellers. These supercavitating hydrofoil sections have a sharp leading edge, and are shaped like curved wedges with a thick, blunt trailing edge. The third adverse effect of cavitation is less well known, and is a consequence of the fact that cavitation affects not only the steady state fluid flow, but also the unsteady or dynamic response of the flow. This change in the dynamic performance leads to Experimental and Numerical Study of an Axial Flow Pump 1.23
Chapter 1: Literature Review instabilities in the flow that do not occur in the absence of cavitation. Examples of these instabilities are ``rotating cavitation," which is somewhat similar to the phenomenon of rotating stall in a compressor, and ``auto-oscillation," which is somewhat similar to compressor surge. These instabilities can give rise to oscillating flow rates and pressures that can threaten the structural integrity of the pump or its inlet or discharge ducts. While a complete classification of the various types of unsteady flow arising from cavitation has yet to be constructed, we can, nevertheless, identify a number of specific types of instability. The study of cavitation phenomena exceeds the scope of this thesis. However, cavitation is also used as an important flow visualization tool for the structure, location, and relative strength of vortices, information that is paramount in the understanding of tip leakage flow phenomena (Zierke et all, 1994)
1.5.7 Unsteady Phenomena

While it is true that cavitation introduces a special set of fluid-structure interaction issues, it is also true that there are many such unsteady flow problems, which can arise even in the absence of cavitation. One reason these issues may be more critical in a liquid hydropump is that the large density of a liquid implies much larger fluid dynamic forces. Typically, fluid dynamic forces scale like 2 D4 where is the fluid density, and and D are the typical frequency of rotation and the typical length, such as the span or chord of the impeller blades or the diameter of the impeller. These forces are applied to blades whose typical thickness is denoted by . It follows that the typical structural stresses in the blades are given by 2 D4/2, and, to minimize structural problems, this quantity will have an upper bound, which will depend on the material. Clearly this limit will be more stringent when the density of the fluid is larger. In many pumps and liquid turbines it requires thicker blades (larger ) than would be advisable from a purely hydrodynamic point of view. This monograph presents a number of different unsteady flow problems that are of concern in the design of hydraulic pumps and turbines. For example, when a rotor blade passes through the wake of a stator blade (or vice versa), it will encounter an unsteady load, which is endemic to all axial flow pumps. This rotor-stator interaction problem is an example of a local unsteady flow phenomenon. There also exist global unsteady flow problems, such as the auto-oscillation problem mentioned earlier. Other global unsteady flow problems are caused by the fluid-induced radial loads on an impeller due to flow asymmetries, or the fluid-induced rotor dynamic loads that may increase or decrease the critical whirling speeds of the shaft system.
1.6 Summary
To understand the importance of employing advanced numerical methods for analysing pump flows, a thorough discussion of the general characteristics of the flow field was provided in this chapter. The fundamentals of pumps have been explained and the general performance characteristics discussed. Special attention was paid to the development of secondary flows due to the action of inertial and pressure forces and the flow phenomena occurring at partial load. Based on the revealed complexity of the flow it is argued that empirical methods for pump flow predictions employed for design purposes provide only an indication of the influence of the complex
1.24
Chapter 1: Literature Review phenomena on the pump performance. A profound insight into local flow properties and an understanding of the physics of the flow phenomena in general is essential for controlling the physical phenomena and requires a high accuracy in the modelling of the flow.
1.25

2.1 Introduction
This chapter introduces the experimental test facility that will be used for the measurement of the performance of the axial pump, and the experimental investigation of the flow in the axial pump using the Laser Doppler Velocimetry (LDV). The infrastructure and the test rig will be explained in detail, and the measuring instruments for the performance including their accuracy will be discussed. Also this chapter gives the description of the experimental axial flow pump.
2.2 Experimental Test Facility

2.2.1 Infrastructure and Test Rig
The general arrangement of the pump test facility is shown schematically in figure 2.2. The test rig is a closed circuit existing of a 23 m long 500 DIN pipe, which the pipe is supported over a 12 m water reservoir. The water reservoir is underground and is not shown in figure 2.2. The experimental axial pump is mounted in a horizontal test unit installed in a simple closed hydraulic circuit using rainwater as the working fluid. This rig is located in the laboratory of Fluids Mechanics at the Vrije Universiteit Brussel. Following the direction of flow in figure 2.2 (from left to right and clockwise), the next components are mounted in the test bench: A DC motor with variable speed drives the feed pump with a maximum flow of 1200 l/s. It is installed at the same floor level of the water reservoir. The feed pump is not shown in figure 2.2 but indicated with the text box. It is installed at the same level of the bottom of the water reservoir. The pump is connected with the butterfly valve (2 in figure 2.2). The flow rate of the test rig can be regulated by the rotation speed of the booster pump, and also with the butterfly valve mounted in the bypass of the booster pump (6, figure 2.2). The flow is measured with an electromagnetic flow meter that is mounted in the pressure pipe of the booster pump (9, figure 2.2). After the flow meter, a nozzle with a surface ratio of 0.9 (11, figure 2.2) connects the experimental axial flow pump with the closed circuit (14, figure 2.2).
Figure 2.1: Detail view of the axial pump
2.1
Booster pump
Figure 2.2: Experimental Test Facility
Pressure taps Driving console 2.2
Chapter 2: Experimental Test Facility Figure 2.1 and 2.2 shows the profile of the axial flow pump unit with its key components.
Figure 2.3: Single-stage Rotor-Stator axial pump: (a) Upstream view, (b) side view
Water is conducted from the inlet pipe to the 3-blade backward swept impeller via the inlet bowl (see figure 2.3 and 2.4). Upon leaving the impeller, the flow enters the 8vane stator that is slightly radial to fit into a conical diffuser to compensate the flow disturbances created by the spherical hub of the propeller pump (see figure 2.4). The stator is supposed to restore the flow to axial direction. A conical diffuser follows the stator, and discharges the flow into the outlet pipe. The dimensions (in mm) of these components are detailed in the next paragraph Description of Single-Stage Axial Flow Pump.
Figure 2.4: View of 3-blade impeller of pump
Figure 2.5: View of stator of pump
An induction motor with a transmission shaft drives the axial pump. Bearings in the stator house of the propeller pump support the shaft that is mounted at the induction motor with a universal joint. This driving console (see figure 2.2, indicated with a text box and figure 2.6) provides also a torque - and frequency converter so that the power absorbed by the pump or produced by the turbine can be directly read from the converter. A frequency converter that powers the 755 kW AC motor regulates the rotation speed of the propeller pump. In figure 2.6 is the assembly of the power console shown. Experimental and Numerical Study of an Axial Flow Pump 2.3
Figure 2.6: View of the power console.
An encoder attached to the shaft of the pump provides the clock signal necessary to analyse the measurements of the flow (see figure 2.6, on the right side of the power console). This encoder delivers a reference impulse per one runner rotation to the data acquisition of the Laser Doppler Velocimetry. The static head of the test pump is measured with 4 wall pressure taps combined in a ring tube at the inlet and outlet of the pump unit (see figure 2.2). Downstream the axial pump, a diffuser (15, figure 2.2) mounts the axial pump in the circuit. At the outlet of the hydraulic circuit, a butterfly valve is fitted to simulate the pipe characteristics (21, figure 2.2). All butterfly valves operate hydraulic and potentiometers read the angles of the valves for data acquisition. A computer console operates the valves and reads the instrumentations that are needed for the measurement of the performance of the test pump. This code is assigned with Labview and figure 2.7 shows the interface of this console.
2.4
Figure 2.7: Interface of the control console
All pipe work including elbows and volutes are made of stainless steel, except the 8 stationary blades, 3-blade impeller and the shaft of the axial pomp that are cast in bronze and fixed in a casing of Ertalon. The test unit may be disassembled and removed in about 2 workdays. However access to the rotor and stator blades is possible in less time by removing the cover at the outlet of the axial pump (14, figure 2.2).
2.2.2 Description of Single-Stage Axial Flow Pump

The performance tests and the flow yield investigations were carried out on a high specific speed propeller pump with a high specific speed Ns of 353, consisting of a rotor-stator stage. The rotor was tested without any inlet guide vanes. The pump is designed to operate at a flow rate of Qd = 302 l/s (or at a flow coefficient = 0.088, see equation 4.8) with water at T = 293 K as working fluid. In the design point the stage provides a pressure rise corresponding to a total head at Htot,d = 1.24 m (or at a head coefficient = 0.016, see equation 4.5). At the design point, the rotor blades rotated counter clockwise (looking downstream) at 755 rpm which yields a rotor tip speed of Utip = 13.9 m/s (using the nominal tip radius of 0.1757 m). During our LDV measurements, the impeller ran at a constant speed of 755 rpm, and the feed pump and the butterfly valve adjusted the flow rate until a prescribed flow and head was obtained. In all measurements, inlet pressure at the axial pump is maintained around 1.25 atm. The Reynolds number, Re=Utip.D1/, based on the outer diameter D1 at the leading edge of the rotor and the rotor tip speed Utip, is of the order 2.4.106. The operating conditions are summarized in Table 2.1 and the measured performance curves are illustrated in figure 4.1, 4.2 and 4.3 of chapter 4 (Experimental Results of Axial Flow Experimental and Numerical Study of an Axial Flow Pump 2.5
Chapter 2: Experimental Test Facility Pump). In the present work, velocity measurements are done at design flow Q = Qd and several off-design conditions are performed for the validation of the numerical performance with the experimental one. Rotational Speed Flow Rate Head Power Reynolds number Specific speed 1 n Qd Hd P Re Ns 755 302 1.24 4.4 2,4.106 353 rpm l/s m kW -
Table 2.1: Design point operation conditions for pump
The rotor comprises 3 blades and it has a variable chord length of 161.7 mm at the tip and a chord length of 109.9 mm at 25% span. The blade angle of the rotor at the tip leading edge is 76.83 and at the hub 55.24. The blade stagger angle varies from 14.45 at the hub to 6.26 at the tip. The blade angle can be changed using a blade rotating mechanism contained within the hub of the rotor. Blade angle variation is made possible using the spherical hub and tip contours. In this work, the pitch angle is fixed. In order for the rotating blades to avoid rubbing the casing endwall, a clearance exists between the rotor tip and the casing. The rotor tip clearance to chord ratio is s/c = 0.93% with the tip clearance of about 1.5 mm and a chord c = 161.7 mm at the tip of the impeller. An 8-vane stator completes the stage. It is fitted in a diffuser and compensates the flow disturbances created by the spherical hub. The blade row of the rotor and the stator are separated at the zero span by nearly 98% of the axial tip rotor chord from rotor trailing edge to the stator leading edge and 67% axial tip rotor chord at full span height. It has a variable chord length of 86.6 mm at the tip and a chord length of 59.1 mm near the hub. The blade angle of the stator at the tip leading edge is 13.31 and at the hub 45.67. The blade stagger angle varies from 45.67 at the hub to 13.31 at the tip Other absent geometrical dimensions of the impeller and stator are shown in table2.2.
Description Impeller Stator
Rotation Speed (rpm) Radius at tip of leading edge (mm) Radius at tip 25% downstream leading edge (mm) Radius at tip 50% downstream leading edge (mm)
755 175.7 178.7 179.7
154.7 184.2 188.0
The specific speed is defined as N s =
n.Qd Hd
3
where n [rpm], Qd [m3/s] and Hd [m].

4
2.6
Radius at tip 75% downstream leading edge (mm) Radius at tip of trailing edge (mm) Hub to tip radius at leading edge Hub to tip radius 25% downstream leading edge Hub to tip radius 50% downstream leading edge Hub to tip radius 75% downstream leading edge Hub to tip radius at trailing edge No. of rotor blades Blade angle at tip leading edge (degree) Blade angle at hub leading edge (degree) Blade angle at tip trailing edge (degree) Blade angle at hub trailing edge (degree) Blade chord at tip (mm) Blade chord at 75% Span (mm) Blade chord at 50% Span (mm) Blade chord at 25% Span (mm) Blade maximum thickness-to-chord at tip Blade maximum thickness-to-chord at 75% Span Blade maximum thickness-to-chord at 50% Span Blade maximum thickness-to-chord at 25%Span
Data of Rotor-Stator
180.0 180.0 0.3263 0.3517 0.3619 0.3650 0.3569 3 76.83 55.24 70.57 40.79 161.7 144.4 127.2 109.9 0.040 0.069 0.099 0.142
191.9 195.7 0.3834 0.3118 0.3297 0.3644 0.3955 8 -14.95 -46.51 1.636 -0.843 86.6 80.4 69.6 59.1 0.0859 0.0716 0.0735 0.0835
Distance tip trailing edge rotorleading edge stator (mm) Distance hub trailing edge rotorleading edge stator (mm)
28.7 43.7
Table 2.2: Geometry parameters of axial pump unit.
The Ganz Mavag Company in Budapest manufactured the stage. The pump in the test rig is a scaled version of an industry-size one, that is now manufactured by Flygt. The geometric scale of 1:1.5 has been chosen as big as possible to have high measuring accuracy in flow and power. Typical application areas for this pump include stormwater stations, sewage-treatment plants, land drainage, irrigation, aquaculture and water attractions. Both the stator and rotor blades have been machined with CAD/CAM techniques using the Flygt Computer vision computer connected to a 3-axis NC-milling machine. The maximum geometrical inaccuracy of the blade shapes is probably less than 0.1 mm. The inaccuracy of the blade general position is less than 0.7 mm. The suction and pressure side corner were rounded slightly to prevent the sharp corner from Experimental and Numerical Study of an Axial Flow Pump 2.7
Chapter 2: Experimental Test Facility breaking. Gearhart (1966) found that this rounded corner prevented local separation of the flow through the tip clearance and thus improved the gap cavitation performance.
2.2.2.1 Rotor and Stator Design
The geometry of the rotor and stator in 3 different spans are plotted in figure 2.8. The edges of the rotor blades and the stator blades are blunted. The plots of the blades on this page are orientated so that the flow run from left to right of the page and the impeller rotates from down to up.
Geometry of rotor at full span
Geometry of stator at full span
Geometry of rotor at midpan
Geometry of stator at midpan
Geometry of rotor at zeropan
Geometry of stator at zerospan
Figure 2.8: Geometry of the blades (left rotor, right stator)
Their dimensions are shown and written in table 2.3. Experimental and Numerical Study of an Axial Flow Pump 2.8

2.2.2.2 Hub and Shroud Design
The geometry of the hub and the casing of the propeller pump are complicated. The meridional geometry of the shroud and hub are plotted in figure 2.9. The hub radius at the leading edge of the rotor is 53 mm and 59 mm at the trailing edge. The radius of the shroud is 177 mm at the leading edge and 180 mm at the trailing edge of the impeller. The nose of the hub is 57 mm from the leading edge of the rotor. The hub radius at the leading edge of the stator is 60 mm and increases to 78 mm at the trailing edge. The radius of the shroud varies from 181 mm at the leading edge of the stator to 196 mm at the trailing edge. The shape and the dimensions of the hub and shroud are shown in figure 2.9 and table 2.3. nr. 1 2 3 4 D [mm] 57 73 44 52 R [mm] 124 180 181 118 nr. 5 6 7 8 D [mm] 86 30 45 64 R [mm] 53 59 60 78
Table 2.3: Dimensions belong to figure 2.9. D8 D7 D6 D5
Stator R1 R2 Rotor R3
R4
R5 D1 D2
R6 D3
R7 D4
R8
Figure 2.9: Dimension of the blades
2.3 Measurement Equipment for the Performance of the Pump

The performance of the pump depends on the flow rate, pressure and power. The measurement equipment for the performance of the pump will be discussed in this paragraph. Also calibration and accuracy will be handled.
2.9
2.3.1 Flow Rate

The flow rate in the experimental test facility is measured by an electromagnetic flow meter. Electromagnetic flow meters operate on Faraday's law of electromagnetic induction that states that a voltage will be induced when a conductor moves through a magnetic field. The liquid serves as the conductor. Energized coils outside the flow tube create the magnetic field. The amount of voltage produced is directly proportional to the flow rate. The flow tube mounts directly in the pipe. Pressure drop across the meter is the same as it is through an equivalent length of pipe because there are no moving parts or obstructions to the flow. The voltmeter can be attached directly to the flow tube or can be mounted remotely and connected to it by a shielded cable. The manufacturer calibrated the magnetic flow meter and its signal converter to an overall accuracy of 0.5 %. In Table 2.4 contains the important technical data of the used magnetic flow meter. Trade Mark Instrument Full Scale Display Accuracy BBC Veriflux Electromagnetic Flow meter 0-12 m/s 10 % - 50 % FS: linear from 1 % -0.5 % 50 % - 100 % FS: 0.5 %
Table 2.4: Technical data of Magnetic Flow Meter
2.3.2 Wall Pressure Measurement

In order to measure the time-averaged static head of the axial pump, we machined pressure taps in 4 different circumferential positions behind the stator and at the impeller inlet (332 mm upstream of the tip of the rotor and 305 mm downstream the trailing edge of the stator, see figure 2.2). The 4 static pressure taps that are combined in a ring tube are connected with a differential transducer for measuring the head of the pump, and a gage pressure transducer that measure the pressure at the inlet of the pump with the atmospheric pressure as its reference pressure. The transducers are typical variable reluctance pressure transducers consisting of a diaphragm of magnetically permeable stainless steel clamped between two blocks of stainless steels. Embedded in each block is an inductance coil on an E-shaped core. This coil assembly, covered by an Inconel disc, has a corrosion resistant surface. In the undeflected position, the diaphragm is centred with equal gaps between it and the legs of each E-core to provide equal reluctances for the magnetic flux paths of each coil. A pressure difference applied through the pressure ports deflects the diaphragm toward the cavity with the lower pressure, decreasing one gap and increasing the other. As the magnetic reluctance varies with the gap and determines the inductance value of each coil, the diaphragm deflection increases the inductance of one coil and decreases that of the other. The transducer connected in an AC bridge circuit is able to take advantage of the inductance variations in the transducer coils. Since the Experimental and Numerical Study of an Axial Flow Pump 2.10
Chapter 2: Experimental Test Facility diaphragm displacement is linear with the pressure applied, the bridge output is also linear with the pressure. The pressure transducers have to be calibrated and this is done regular with a manometer. The transducers and the manometer are connected with a reservoir of water. The transducer is calibrated for different water heights, varying between 20 cm H2O until 1,5 m H2O. The height of the column of a liquid of known density acted upon the force gravity can be measured with the manometer. The overall pressure-system accuracy is verified with a manometer and is maintained to within 0.25% F.S. Trade Mark Type Measuring range Accuracy Validyne DP15 0.88 m 1.4 m H2O
0.25 % FS
Table 2.5: Technical Data of torque and rotation speed
The total head of the pump (Htot) is estimated with the next equation
H tot = H +
2 v2 v2 1 2g 2g
2.1
With H is the difference of the measured static head of the wall pressure taps at the inlet (Hstat1) and outlet (Hstat2) of the axial flow pump
H = H stat2 H stat1
2.2
And where v1 and v2 are the averaged absolute velocities at the inlet and the outlet of the pump. These averaged velocity speed is the ratio of the measured flow Q and the cross-sectional area of fluid flow (Ai) where the static wall pressure are mounted (equation 2.3)
vi =
Q Ai
2.3
2.3.3 Rotor Shaft Torque and Rotation Speed

Force measurements are important for determining the mechanical power required to operate the pump at various flow conditions. In order to measure the rotor shaft torque, we mounted the torque transducer type T32 FNA from HBM, which is without slip rings and bearings that provide torque sensing as well as speed measurement. The torque transducer consists of a measurement body, which is screwed to a transmission shaft by means of two flanges. The hollow shaft, which forms the measurement body, is equipped with strain gauges for torque measurement mounted on the rotor in the direction of the main stress. It is connected as a Wheatstone bridge so that only torque will unbalance the bridge. Temperature-effects are also compensated by means of further balancing elements. Experimental and Numerical Study of an Axial Flow Pump 2.11
Chapter 2: Experimental Test Facility Furthermore, the rotor carries a toothed wheel with 15 teeth for the determination of speed. On the stator, there are evenly distributed, four inductive pick-ups. This produces pulses, which are proportional to speed. Data were taken with an amplifier for torque and speed rotation. This amplifier converts the pulse frequency into a voltage, which is proportional to the torque, and another module forms a voltage, which is proportional to the speed of rotation from a variable frequency rectangular pulse frequency. In both cases the output is read by the data acquisition. Trade Mark Instrument Measuring Amplifier Torque Measuring Amplifier Rotation Speed Nominal Torque Maximum Torque Maximum Rotation Speed Accuracy Rotation Speed Accuracy Torque HBM T32FN MD 2188 N 2119 A 2 kNm 2.5 kNm 15000 rpm 2 rpm
0.03 % at 50 % Nominal Torque
Table 2.6: Technical Data of torque and rotation speed
With this instrument the effective power Pe is directly measured on the shaft of the propeller pump as the product of the torque C and the rotation speed :
Pe = C.
2.4
The torque transducer is calibrated with a calibrated dead weight loading network attached to the end of a stiff beam acting as the moment arm. If the exact length of the arm from the centre of rotation to the suspension point of the weight is known, the resulting torque can be computed easily by multiplying the load force by the length of the moment arm. The moment arm is set approximately horizontal and the calibrating weights are suspended from it. The output of the measuring is observed. Knowing the shaft power into the rotor of the propeller pump and the energy net hydraulic power, equals to .Q.g.h, the overall pump efficiency of the propeller pump will be defined as
Q g H tot C
2.5
where is the density, Q is the flow rate, g is the gravity, Htot is the total head, C is the torque and is the rotation speed.
2.12
Chapter 2: Experimental Test Facility The power input to the rotor is always less than the power supplied at the coupling because of external energy losses in the bearings. This efficiency will be different than the hydraulic efficiency of the pump that is broadly defined as in the numerical code.
2.4 Performance Uncertainty Analysis

All experimental results contain a certain degree of uncertainty that one must consider when analysing these results. Here, we shall consider the uncertainty associated with the performance of our test bench, using a standard uncertainty analysis, such as the one outlined by Coleman and Steele (1989). For a general case of an experimental result, r, as function of J variables Xi,
r = r ( X 1 , X 2 ,..., X J )
the uncertainty of the result is given by
2 2 2 2 r r r U r = UX + X U X 2 + ... + X U X j X 1 1 j 2 1
2.6
2.7
where Uxi are the uncertainties in the measured values Xi. The uncertainties correspond to our estimates of the combination of bias and precision error. For this uncertainty analysis, we attempted to estimate the uncertainties to 95% confidence. For the total head measurement, equations 2.1 to 2.3 are used. The uncertainties of the static pressure (H) and flow (Q) are given in table 2.4 and 2.5. In order to compute the uncertainty of the total head (Htot) we need to estimate the uncertainty of the velocity at the inlet (v1) and outlet of axial pump (v2)
R Q v1 = + 4. 1 v1 Q R1
2 2
2.8
(R22s + R22h ).R Q v 2 + 4 . = Q v2 (R22s R22h ) 2
2.9
where R1 is the radius of the inlet pipe, R2h is the inner radius of the outlet pipe, R2s is the outer radius of the outlet pipe and Ri=1,2 is the uncertainty of the dimensions of the pipe. The uncertainty of the total pressure Htot can be computed now as
H tot
2 v2 v2 v1 v12 2 2 = H + v 2 + g v1 + g 2 2 2 2
g g
2.10
2.13
Chapter 2: Experimental Test Facility The power of the axial flow pump is calculated with equation 2.4. The uncertainty of the power P becomes
P C = + P C
2 2
2.11
where C and are the uncertainties of the torque and the rotation speed given in table 2.6. The efficiency of the propeller pump is defined as equation 2.5. The uncertainty of the efficiency can be calculated as
Q H tot = + Q + H tot
g C + + + g C
2
2.12
During the test, we use the 95 % confidence level of the data. We focussed the number of data giving a precision error (P) less than 2 %. The total uncertainty is a combination of the bias error (B) with an estimate of the precision error as
= B 2 + (t v ,95 P) 2
2.13
Using the precision error of 2% and the bias errors of the technical tables of the measurement instruments, the maximum total uncertainties of the different physics quantities, which defined the performance of the pump, are resumed in table 2.7. Measured Quantity Uncertainty Measured Quantity
P (% ) P
Uncertainty
Q (% ) Q
H tot (% ) H tot
2.0
2.0
2.1
eff
eff
(% )
2.4
Table 2.7: Uncertainties of the measured quantities to 95 % confidence.
2.5 Summary
This chapter presented the infrastructure of the experimental test facility with its instrumentation, and introduced the experimental single stage axial flow pump, also called as propeller pump. The calibration methods of the instrumentation and their accuracy were discussed. The uncertainties of the measured quantities for the performance of the axial flow pump are given in table 2.7. These are conforming to ISO 3555.
2.14

3.1 Introduction
The flow field in the propeller pump is measured with the Laser Doppler Anemometry (LDA). The theory of the Laser Doppler Anemometry can be found in the literature. This will not be repeated here, but appendix A gives a short summary of the working principle of the laser Doppler anemometer. The aim of this paragraph is to introduce the technical specifications of the used LDA system and the measurement procedure. Optical difficulties due to the curved casing of the propeller will be investigated and the consequences of the curved rotor blades and the blade shadow.
3.2 Technical Specification of LDA

The power source is an Argon Ion Laser with rated output of 4 watts, of which 1.7 watt is emitted in the green colour at the wavelength 514.5 nm and 1.3 watt is emitted in the blue colour at the wavelength 488.0 nm. The beam exits from the laser tube with a diameter of 1.34 mm and the beam divergence is 0.03 degree. The LDA system used for the purpose is based on fibre optics, with back scatter method and has processors based on spectrum analysis of the back scattered signals. This gives the processor ability to process signals with very poor (very low signal to noise ratio) quality. The measurements are operated using data acquisition software (Burstware, from Dantec). A series of measurement parameters are fixed using this software. We precise some of these parameters: All the measurements were taken with the 300 mm focal length front lens. The two beams were then focused at a point within the impeller, the measuring point. The angle between the two converging beams was 7.40. The crossing of the beams generates an ellipsoidal probe volume, of radius approximately 0.12 mm and length 0.9 mm. The back-scattered light from particles present in the water was collected by the focusing lens and produced an image of the measuring volume on the EMI model 9813 photo multiplier. The Bragg cell introduces frequency shift driven by a 40 MHz signal from one of the processors. A PC drives the controllers and the processors. The LDA processors are connected through the IEEE 488 bus and the traverse controller is connected to RS232 port. A digital storage oscilloscope is used as the on-line monitor of the Doppler signals. In all the studies conducted in this project high intensity laser beams are used. The measurement point can be chosen and moved using a three-axis traverse system. The entire laser-Doppler system could be traversed relative to the pump impeller via a large horizontal traversing table. The precision of this system in the x and y directions (horizontal plane) are 0.0125 mm and in the z direction (vertical) is 0.00625 mm. The green and blue beams belong to 2 planes whose intersection is called the LDA probe axis. This axis was always chosen horizontal and vertical, so that the axial and tangential velocities are measured. The measurements were made on the test section centreline through a specially designed window. For each particles, velocity, transit time across the probe volume, and arrival time, are recorded.
3.1
Chapter 3: Applications of the Laser Doppler Anemometry Equipment 3.3
Measurement Procedure
The LDA does not measure the instantaneous velocity V(t) continuously in the manner that a fast-response device such as a hot-wire probe would. Instead, the LDA measures discrete samples of V(t). At each radial position, the LDA measure N particles samples, with each sample n given a value of depending on the decodermeasured circumferential location of the rotating shaft. In the stationary frame of reference, the values corresponding to rotor rotation can represent values of time. Therefore, the instantaneous velocity V(t) is represented by the discretized velocity Vn. This unsteady velocity can be decomposed into a time-averaged velocity V , a ~ periodic or deterministic unsteady velocity, V , and a random or non-deterministic unsteady velocity, V: ~ V (t ) = Vn = V + V + V '
3.1
Figure 3.1 shows this velocity decomposition. The entire rotor rotation through 2 can be divided into Nw discrete storage window or bins, as shown for one period T in figure 3.1. Any storage window contains M particle samples with each sample m having a value within the bounds of that particular window. Statistically, the velocity samples within each storage window yield an ensemble-averaged velocity,
V =
V
i =1
3.2
A random or non-deterministic unsteady velocity,

V ' = Vm V
3.3
And an ensemble-averaged variance,
V '2 =
(V
M m =1
V )
3.4
M 1
This variance shows the level of non-deterministic unsteadiness at an average value of within a particular discrete storage window- a level of unsteadiness is either partially of entirely due to the turbulence. The time-average of the entire unsteady velocity signal,
V (t ) =
1 V (t )dt T 0
3.5
Could be found by averaging over all N samples in the entire discretized velocity signal. However, the particles are more likely to be measured in the free stream regions of the flow field and less likely to be measured in regions involving boundary layers, wakes or vortices because of problems in entraining particles into these regions.
3.2
Figure 3.1: Unsteady Velocity Signal
In order to alleviate this bias problem (see more information in appendix B: Uncertainly LDA), one can obtain the time-average velocity by averaging all of the ensemble-averaged velocities in each storage window,
V =
V
i =1
Nw
3.6
Nw
Where Nw represents the total number of windows. This type of averaging is essentially an area averaging technique that gives the circumferential average value of the time-average velocity. Also, the level of non-deterministic unsteadiness of the entire velocity signal can be found by averaging the variances in each storage window,
V '2 =
V '
i =1
Nw
3.7
Nw
This total variance gives the circumferential-average level of non-deterministic unsteadiness. At this point, one can determine the periodic of deterministic unsteady velocity, ~ V =V-V 3.8
This deterministic unsteady velocity is a function of the average or t corresponding to a particular storage window. To show the level of deterministic unsteadiness (a
3.3
Chapter 3: Applications of the Laser Doppler Anemometry Equipment type of unsteadiness that correlates with rotor shaft speed), one can compute the variance
~ V2 =
n w =1
(V V )
Nw
3.9
Nw
For each component, also, one can compute the correlation
%V % V z t =
n =1
w
(Vz Vz ).(Vt Vt )
N
w
3.10
Nw
To see how well the axial Vz and tangential Vt components of the deterministic unsteady velocity correlate with one another. Finally, if one computes the variance of all N samples in the entire velocity signal,
~ + V ') (V
(V
N n =1
n V
3.11
One obtains information on the level of overall unsteadiness. The entire velocity signal varies because of both the deterministic and non-deterministic unsteadiness. This overall unsteadiness level can also be computed from the previously defined levels of deterministic and non-deterministic unsteadiness, if one assumes that the deterministic and non-deterministic unsteady velocities do not correlate with one another,
~ (V + V ')
2
~ = V 2 + V '2
3.12
The use of the above equations was possible due to the fundamental assumption that fluctuations have a Gaussian distribution even in non-isotropic turbulence. A typical run consisted of collecting 1.500.000 measurements yielding an average of 4.000 measurements at each angular position. Run times typically vary between 30 and 60 minutes for every position. Data were collected using a filed point measurement method. In this procedure, the measurement volume remains stationary, and each LDA measurement sample is tagged with the angular position of the rotor via the trigger (see paragraph 3.4. Triggering). The circuits was filled with rainwater that enough seeding particle contain. The data acquisition software includes some filters in order to keep only, among detected particles, those considered as valid. This aims to exclude too heavy particles, which are not passively transported by the flow, and also particles, which went through the probe volume, but with a too large angle. This validation filter is done automatically during the measurement acquisition. Raw valid data are recorded for each position: velocity, transit time and arrival time. After the acquisition, data can be
Experimental and Numerical Study of an Axial Flow Pump 3.4
Chapter 3: Applications of the Laser Doppler Anemometry Equipment processed. We will later indicate that the quantities and spatial positions of the measurements need an optical correction due to the distortion of the laser beams by the window and working fluid. This theme will be discussed more in the paragraph of optical correction. For each position, all the valid data have been stored. As indicated above, several statistical quantities have also been estimated for each position: first and second moments. The radial and axial positions of measurement were known to 0.0005 m and introduce no significant error. The circumferential position of measurement was evaluated to 0.2. The effect of the bandwidth of the spectrum analyser on the spectrum analyser was tested and found to be insignificant. The sweep rate of the analyser precluded the interpretation of an individual velocity to within 0.025 m/s for the axial velocity and 0.05 m/s for the circumferential component of velocity. 2D Laser Doppler Anemometer (LDA) system was used to record two components (vz, axial and vt, tangential) of the velocity of particles detected in a small volume. When seeding particles in the rainwater are illuminated with a laser beam, the frequency of the light scattered from the particles gives instantaneous values of the velocities. Then an average is performed over a large number (N) of particles, providing the following measured quantities: Mean value of axial velocity Vz = <vz> Mean value of radial velocity Vt = <vt> Average axial square fluctuations 1 =< (Vz vz )2 > Average radial square fluctuations 2 =< (Vt vt )2 > 3.13 3.14 3.15 3.16
3.4 Triggering
An encoder attached to the shaft of the axial pump provided the clock signal necessary to analyse the flow in the pump. This encoder delivers a reference impulse per one revolution and another with 360 pulses by revolutions. These signals are sent to the spectrum analyser and used for the synchrony analysis. The reference impulse serves to determine the phase angle of the velocity measurements. There are two different ways which encoded data can be collected. The first is called time stamped. Time stamped encoded data uses the arrival time clock to determine where (relative to a reset pulse) the sample occurs in phase-space. Usually phase-space refers to an actual special position on the tip of the rotor or between two blades of the impeller. But, in general, phase-space refers to angular position relative to a reset pulse. The arrival time clock is reset periodically and the angular positions of the velocity measurements are calculated using the rpm of the rotating impeller and the time since the last reset pulse (assuming constant angular velocity). The second way in which encoded data can be collected is referred to as space stamping. Space stamping uses the encoder count as a direct indication of angular position. The encoder counter is reset periodically (usually once per revolution) and the encoder count corresponds to an actual angular position. Space stamping requires a large number of pulses per revolution to obtain an angular resolution comparable to Experimental and Numerical Study of an Axial Flow Pump 3.5
Chapter 3: Applications of the Laser Doppler Anemometry Equipment that obtained using time stamping, however, space stamping has the advantage of being able to map velocity measurements onto angular position correctly in the case of applications with fluctuating rpm. The first method is used because the measuring time. The encoder fixed on the pump shaft gives the positions of the measuring point in the blade-to-blade plane with a resolution of 3600 angular readings for our shaft revolution. Using the field point measurement method, we obtained LDA data with a stationary measurement volume and tagged each LDA measurement sample with the angular position of the rotor via a shaft encoder. In analysing these data, we divided the entire blade rotation of 360 degrees into discrete angular storage windows or bins. These bins are used to statistically determine circumferential variations due to the blade rate, as well as any blade-to-blade variations. However, a trade-off arises when deciding how small one should make the bin size. On one hand, smaller bin sizes result in a finer circumferential resolution and decrease the chance of averaging important circumferential features. On the other hand, smaller bin sizes result in fewer data samples per bin in which to perform a statistical analysing- giving higher standard deviations.
Figure 3.2: Bin size dependence for a rotor blade wake (Tangential velocity at 76.2 % span and 75 % chord axially downstream of the rotor tip leading edge)
In analysing the LDA measurements in the experiment, we completed a trade-off study to determine the best bin sizes in which to reduce the data. As an example, figure 3.2 shows a rotor blade wake analysed as a function of rotor position for various bin-sizes- with the ensemble-averaged velocity plotted for each bin. For bin sizes greater than 0.5 degrees, figure 3.2 shows that over averaging smears out the wake deficit. For smaller bin sizes, such as 0.1 degrees, the number of data samples per bin decreases below 850 for a single component of velocity (using the typical 50,000 sample size survey per component). The figure shows that this reduced the
3.6
Chapter 3: Applications of the Laser Doppler Anemometry Equipment number of samples in performing the ensemble averaging within each bin leads to a very uneven wake profile when the bin size equalled 0.1 degrees. Even though the profile corresponding to a bin size of 1.0 degree appeared quite acceptable, some bines contained as few as 8 or 9 samples giving an unacceptable confidence in the data. Therefore, for LDA surveys requiring a small circumferential resolution- such as the details of wakes and vortices- we chose a bin size of 0.5 degrees. In other surveys, we chose a bin size of 1.0 degree in order to educe the standard deviations associated with the ensemble averaging within each bin. After deciding on the size of the storage window or bin, one still needs to consider the uncertainly in determining quantities such as velocity and turbulence intensity from the LDA measurements. This study is done in appendix B: Experimental Uncertainties. In figure 3.3 is the uncertainty of tangential velocity for a rotor blade waked (tangential velocity at 76.2 % span and 75 % chord axially downstream of the rotor leading edge, cfr figure 3.2). The uncertainty of the tangential velocity is calculated with equation B.23.
Figure 3.3: Uncertainty dependence the bin size for a rotor blade wake (Tangential velocity at 76.2 % and 75 % chord axially downstream of the rotor tip leading edge)
The figure 3.3 concludes that the uncertainty of the measurements is less that 2 % when the bin size is 1.0 deg, except in the rotor blade wakes due to the smaller number of data samples near the rotor blades. These mean values are recorded for as many different positions as needed. An important practical limitation is time, since the detection of N different particles can be long, depending on the zone, which is explored. Indeed, the acquisition rate is proportional to the density of particles times the velocity of the flow; assuming a nearly homogeneous density of seeding particles, the acquisition rate is low in low velocity regions, such as stagnation zones, reverse flows and near wall boundary.
3.7
Chapter 3: Applications of the Laser Doppler Anemometry Equipment With 360 intervals per blade passage, the interval length varies between 3.1 mm at the tip of the blade and 1.1 mm at the hub.
3.5 Special Difficulties in Flow Study

The use of LDA measurements in internal flows, such as those occurring in the passages of the propeller pump requires that transparent windows are available. Most measurements for this thesis were made on the test section centreline upstream, between and downstream the impeller. Figure 2.1 shows a detailed cross section of the original test bench. The blade of the impeller and the stator are marked with dark areas and the casing of the circular Plexiglas tube has variable inner radius and thickness. The optical distortion of the laser beams, both in transmission through the complex geometry of the casing and due to the index refraction changes associated with water, Plexiglas and air interfaces leads to unwanted displacement, rotation and distortion of the measurement volume, change in the intersection angle of the laser beams and misalignment of the laser beams with resulting in loss of Doppler signal. Investigations of those effects have been carried out by several authors, especially for the case with cylindrical and curved window surfaces (Parry et al. (1990), Doukelis et al. (1996), Zhang et al (1995), Zhang (2004) and Kehoe and Prateen (1987)). The results are typically only available in complicated computational forms and not sufficiently explicit to derive all of the effects due to the refraction of the laser beams and the alignment of the interception of the beams. In principle, some refractions effects may be restored by an inverse application of Snells law to the optical situation at hand. Examples of such correction procedures are given by Boadway and Katahan (1981). However, apart from traceable errors, refraction may impede measurements in certain regions due to total internal reflection by blades or hub. Furthermore, if due to refraction the laser beams fail to occupy a common volume, the application of LDA may be inhibited even in regions where the absolute amount of refraction is small. To minimize or prevent the problems related to the refraction of the beams by the pipe wall, a number of precautions may be taken:
Immersion in a square box. By mounting the transparent pipe in a liquid-filled box with flat windows, the largest optical transition, viz. that between air and the liquid, is made flat. This is a standard measure, which always should be taken even in combination with one of the following specific measures. Refractive index matching. In fluid flows, it may be possible to control the index of refraction of the fluid. By making the refractive indices of the transparent pipe and the fluid equal, refraction is eliminated. In this way, the experiment fixture disappears optically (and therefore has no influence on the laser beams). Refractive index matching has been performed in the past by matching the refractive indices of the fluid and solid boundaries in the case of wall-bounded flows (Liu et al., 1990), tube bundles or inside complicated passages of turbomachines. The paper of Durret et al (1985) presents techniques for refractive index matching in liquid flows. Others examples of this practice are given by Edwards and Dybbs (1987), who used various mixtures of two types of oils. Refractive index matching is generally limited to small flow rigs due to the cost of these fluids, the necessity of temperature control and the possible change of the optical properties due to ageing. Furthermore, it may be
3.8
Chapter 3: Applications of the Laser Doppler Anemometry Equipment necessary to seal the system, since fluids proper for refractive index matching are often volatile and noxious. A disadvantage from the fluid mechanical point of view is the relatively high kinematic viscosity of these fluids relative to standard fluids like water. This simple method of matching the refractive index of the test fluid to that of the pipe wall has often been applied in small-scale laboratory measurements. It is impossible in most industrial applications with big volume of fluid.
Corrective lenses. It is possible to anticipate for the effect of refraction by the use of corrective lenses that deflect the laser beams before they enter the flow system. An example of this approach is the work of Durret et al (1985), who proposes the use of simple spherical lenses, included machining circular tubes with rectangular outer walls.
The general disadvantage of the use of such strategies is, that each optical situation requires specific lens geometry. Often, rather complicated lenses are required. Furthermore, it may be necessary to realign the lenses between two measurement positions. Even when a complete correction for the measurement position is obtained, the beam intersection angle will vary from point to point, which further complicates the measurement procedure. Due to the complex geometry of the casing, it is not possible to use this method.
Thin pipe walls. It is obvious that a reduction of the pipe wall thickness leads a smaller amount of refraction, provided that the pipe is surrounded by a rectangular box as described above. Although it is possible to manufacture pipe sections with a wall thickness smaller than 100 m out of a thicker pipe, it is difficult to obtain a satisfactory optical quality in this way. A better result is obtained by the use of prefabricated flexible transparent films. Mizushina and Usui (1977) used short pipe sections made of polyester film with a thickness of 35 m and an internal diameter of 2.53 cm. Their test fluid was water with a dilute polyethylene oxide solution. Van Maanen and Fortuin (1983) removed a half part of 50 mm glass pipe over a length of 16 cm, while the complete inner wall of the pipe was covered with a thin transparent film (type unreported) with a thickness of 100 m. In this references, not attention is given to the complications encountered with transparent films, like motions generated by turbulent pressure fluctuations of the blade rotation, the optical behaviour of films, the technical set up of the test-section and handling procedures, the selection of sheet material, etc. The purpose of the present enumeration is to give the reader an impression of the practical possibilities and limitations of the method.
In other to allow the flow measurement in the propeller pump, the optical access within the impeller was provide through a flat polished polycarbonate window of 25 mm thickness that was contoured to match the inside wall of the casing. The window is 117 mm long in the axial direction and 51 mm width (16 deg arc) in the circumferential direction and is mounted in the casing, going halfway the chord of the stator blade until 110 % chord rotor blade upstream the leading edge of the rotor (see figure 3.4 and 3.5). The nominal design value of the rotor tip clearance to chord ratio was s/l = 0.92% with the tip clearance about 1.5 mm.
3.9
Figure 3.4: Detail view of window mounted in the casing of the propeller pump.
Figure 3.5: Window mounted in the casing of the propeller pump.
3.6 Optical Correction

Refraction of the laser beams, both in transmission through this flat window and through the flow leads to displacement of the measurement volume and changes in the intersection angle of the laser beams. These corrections will be discussed in the next paragraph, even the influence of the off-axis alignment of the probe relative to the internal flow. The off-axis alignment of an LDA probe to a plane window is often occurs in LDA measurement, in particular, when three velocity components in an external flow are measured simultaneously using two probes with three pairs of laser Experimental and Numerical Study of an Axial Flow Pump 3.10
Chapter 3: Applications of the Laser Doppler Anemometry Equipment beams. For this type of measurement, the probe with the third pair of laser beams (1Dprobe) must be mounted off-axis in order to resolve the third velocity component. There are two main experimental difficulties in the use of such a 2-probe system. First, it is necessary to ensure all the three measurement volumes are coincident. This is difficult because of the very small dimensions of the available measurement volume (typically only 0.12 x 0.12 x 0.9 mm), which necessitates the aid of a pinhole for accurate adjustment. Second, as will be shown in this paragraph, the coincidence of all three measurements volumes is only valid at the point of the adjustment. A twodimensional readjustment is than needed when the measurement volumes are traversed in the flow. Another case of off-axis alignment of the LDA probes is needed when measurements are done in turbomachines with strong staggered blades. Due to the blade twist, it cause shadowed region in the measurement region. To measure the velocity field over the complete blade passage, it is than necessary to mount the probe in off-axis alignment. These off-axis alignments of the LDA probe, even for the special case of a plane window due: Change in the fringe spacing, Change in the orientation of the measurement volume, Change in the path of the measurement volume when traversing, Effect of astigmatism 1 , Distortion of fringe patterns and Change to the maximum data rate. These effects can be sufficiently large to deviate the beams from their initial positions and to significantly influence the accuracy of LDA measurements. The next paragraph analyses these aspects of off-axis alignments of LDA probes.
3.6.1 Displacement Between Measurement Volumes of a 2D-Probes

LDA probes are usually laid out so that all laser beams converge at the focus of the focusing lens. A focusing light bundle can be imagined, which comprises these entire laser beams. The off-axis configuration of the probe to the flow causes astigmatism of this light bundle, both in passing through the window and in the flow. For a twocomponent probe this means that the four laser beams, which were originally focused on a single point, may no longer intersect after passing through the window. Even if the two-component probe with off-axis alignment is positioned such that the two beams pairs lie in meridional and radial plane of this light bundle respectively, the two intersection points will no longer be coincident. The spatial displacement between both intersection points depends on many parameters, including:
Astigmatism is an optical phenomenon and in the present context refers to loss of the unique focusing point of a light bundle after its refraction through a non-perpendicular interface. Following the refraction at the interface there are two separate converging points of the light bundle, according to meridional (tangential) and radial plane of a light bundle, respectively. The spatial distance between both converging points is called the astigmatic difference. The phenomenon of astigmatism is often considered in photography to help describe image aberration. However the appearance of astigmatism in LDA measurements, especially in configurations with off- axis alignments of the probe and thick glass windows, may decisively influence the measurement capability and accuracy.
1
3.11
Chapter 3: Applications of the Laser Doppler Anemometry Equipment The degree of the off-axis probe alignment, The beam intersection angle, The colour difference between the two pairs of laser beam, The thickness of glass window and The depth of the beam intersection points in flow. In fact the dispersion caused by different colours of the laser beams (514.5 and 488 nm) has only a negligible effect on beam refraction and is therefore not considered here. The consequent changes in the laser beam path may be determined from Snells law: n1.sin 1 = n2 .sin 2 3.17
Where ni is de index of refraction in medium i and i is the angle between the normal to the interface and the laser beam, in medium i. Figure 3.6 illustrates how index of refraction changes alter not only the half-angle between two laser beams but also the location at which the two beams intersect. For this particular example, Snells law and trigonometry may be used to determine the change in the half angle and the shift in the location of the probe volume.
Figure 3.6: Refraction of laser beams during velocity measurements (Durst, 1981).
According to detailed calculations using the theory of geometric optics (Zhang 1993), the displacement between both intersection points for a measurement arrangement according to figure 3.7 is given as:
xm = 1
(L1 1 + L2 2 )
3.18
where
m = tan A2 tan B 2
3.19
3.12
1 =
n n 1 cos 2 cos 2 / 2
2 1 2 0
(tan A1 tan B1 ) (tan A 2 tan B 2 )
3.20
2 =
n n 1 cos 2 cos 2 / 2
2 2 2 0
3.21
K = n0 cos cos( / 2 ) tan + tan 2 2
3.22
with A1 , A2 , B1 and B 2 as the refraction angles of the laser beams (A and B) in media 1 and 2 respectively, as defined in figure 3.7.
Figure 3.7: Beam transmittance through a glass window into the flow and formation of two intersection points (Zhang, 1995).
In the above equations m refers to the medium in which the intersection point is present (here medium 2). 1 and 2 refer to the media, through which the refracted laser beams are transmitted from the probe into the flow. In equation 3.19 xm is a linear function of L1 and L2, the thickness of the corresponding optical media before the intersection point m1 (fig 3.7). A similar linear equation, but with more than two terms can be derived, if there are more than two layers of media which refract the laser beams. It can also be demonstrated, that those terms in equation 3.19, denoted by i , disappear if the medium ni has the same refractive index as the medium no (ni = n0 ) . In particular if an internal airflow (n2 = 1) were to be measured by a probe in air n0 = 1 , then 2 = 0 and only the thickness of the glass window determines the displacement xm . The spatial displacement between both measurement volumes according to fig 3.7 is then
S = xm cos o 2
3.23
with o 2 as the refraction angle of the optical axis. In fact neither of the intersection points m1 and m2 lie on the refracted optical axis, but this deviation can be show to be negligibly small.
3.13
Figure 3.8: Displacement S between both intersection points of two pairs of laser beams form a 2Dprobe dependent on the depth of the measurement volume in the water L2 and on the off-axis alignment angle . The half intersection angle of beams in air is /2=3.70
Figure 3.8 presents the displacement between the intersection points of the two pairs of laser beams in a single probe when the probe is aligned off-axis. These have been calculated from Eqs. 3.18 to 3.23 for the beam transmittance of an air-glass-water system with a polycarbonate glass thickness of L1 = 25 mm. Values S at L2 = 0 show the effect of glass window. It can be seen that, because of the large displacement between two intersection points, it becomes impossible to carry out twocomponents coincident measurements with an off-axis configuration of a single probe. This determines that for a two-probe system (used for simultaneous measurements of three components of velocity) only the 1D-probe may be positioned off-axis, and, in fact, it must be realigned in the plane of both laser beams to ensure their intersection.
3.6.2 Fringe Spacing of the Measurement Volume and Velocity Corrections

The fringe spacing of the measurement volume in the flow for the on-axis probe alignment is determined by
x0 =
n 2 sin ( / 2 )on
3.24
where the wavelength n and the half intersection angle of the laser beams / 2 are for the refraction index n . The fringe spacing in this case is the same as the initial fringe spacing in air.
3.14
Figure 3.9: Refraction and intersection of laser beams (Zhang, 1995)
Considering the off-axis probe alignment to the flow, the half intersection angle of the laser beams refracted in the flow should be determined by the application of the law of refraction to each of the laser beams (Fig. 3.9):
n1 sin A1 = n2 sin A2 n1 sin B1 = n2 sin B 2
3.25 3.26 3.27
( / 2)off = ( A 2 B 2 ) / 2
The fringe spacing of the measurement volume in this case then becomes:
x =
n sin ( / 2 )on = x0 2 sin ( / 2 )off sin ( / 2 )off
3.28
where ( / 2)on and ( / 2 )off are for the flow measured.
Figure 3.10: Distortion of the fringe spacing of an off-axis probe compared to that of an on-axis probe, as function of the off-axis alignment angle. Calculated from equation 3.28 for the water flow (n=1.34). o/2 is the initial half intersection angle of laser beams
In Fig. 3.10 the distortion of the fringe spacing x/x0, according to equation 3.28, is show as a function of the off-axis alignment of the probe () for the initial half
3.15
Chapter 3: Applications of the Laser Doppler Anemometry Equipment intersection angles (0/2) of laser beams. As can be seen, the off-axis alignment of the probe results in an increase in the fringe spacing, which usually cannot be neglected.
3.6.3 Direction of Measured Flow Velocity

After the refraction of the laser beams, the bisector of the beams is neither parallel to the optical axis of the probe nor the same as the refracted optical axis. In fact, the measurement volume does not lie on the refracted optical axis. However, it can be shown (to a first approximation using the first order of the Taylor series) that for small beam intersection angles / 2 the bisector of both laser beams in the flow is precisely the refracted optical axis. So it can be assumed, that the direction of the bisector of both laser beams in the flow is equal to that of refracted optical axis. The flow velocity measured is then perpendicular to this axis.
3.6.4 Proportion of the Displacements Between Measurement Volume and Probe

Most applications of LDA probes have been used to carry out measurements of the flow distribution in a flow field by traversing the probe. The probe has to be moved in order to shift the measurement volume. The relationship between the shift of the measurement volume and the probe movement can be established according to Fig. 3.9 as described below. In Fig. 3.9 the virtual and actual intersection points of laser beams are marked by m and m respectively. Considering the distance between both laser beams on the interface this gives: xm (tan A 2 tan B 2 ) = xm (tan A1 tan B1 ) = h Thus:
dxm tan A1 tan B1 = dxm tan A 2 tan B 2
3.29
3.30
Because the displacement of the probe is the same as the displacement of the virtual intersection point m, Equation 3.30 gives the proportion of the x-displacement between the measurement volume and the probe. The measurement volume m will also move in the y-direction, even though the probe is moving merely in the x-direction. According to Fig 3.9, the y-coordinate of the measurement volume can be written as
ym = xm tan A 2 x p tan A1
3.31
Differentiation of equation (3.31) yields
dx dym = tan A 2 p tan A1 dxm dxm
3.32
3.16
Chapter 3: Applications of the Laser Doppler Anemometry Equipment This describes the path, which the measurement volume follows, when the probe is moving in the x-direction, dx p / dxm in Equation 3.31 is simply the reciprocal value of Equation 3.30. It should be noted that in Equation 3.31 only the beam A is considered. If the beam B is also considered, the statement for dym/dxm will be different to Equation 3.32. However the result for dym/dxm is the same in both cases. In fact, considering any point on both beams refracted in the flow will yield the same result. As seen in Equation 3.32 dym / dxm is normally not equal to zero, except for the special cases of on-axis alignment, of off-axis alignment with a single laser beam perpendicular to the flow interface, or with the flow medium the same as the medium in which the probe is present. Therefore the path of the measurement volume is normally two-dimensional.
Figure 3.11: Two-dimensional shift of the measurement volume in water (n=1.34) dependent on the off-axis alignment angle
Figure 3.11 presents this two-dimensional displacement of the measurement volume according to Eqs. 3.30 and 3.32. Although dym / dxm is small, it should not be neglected if the measuring points in the flow have to be traversed across a large distance. This two-dimensional behaviour of the shift of the measurement volume was derived by Zhang (1993) and considered by Eisele et al. (1994) in LDA measurements in a pump. It is of special importance in cases where simultaneous three-component velocity measurements are made using two probes. Normally both probes are fixed to the same traversing system and aligned for obtaining coincidence of all threemeasurement volumes. The movement of the traversing system in the x-direction will then lead to separation measurement volumes. The separation is not only in xdirection but also in y-direction. This means that the coincidence of measurement volumes is valid only at a single point in the flow, and measurement of three separate velocity components during a traverse is not possible without realignment.
3.17
3.6.5 Effect of Astigmatism on the Data Rate

An LDA measurement with an off-axis probe is rather different to that of an on-axis probe, as shown above. In addition to the known effects of the astigmatism, dynamical measurement criteria such as the data rate will also be considerably affected. It is shown below that this is true even if single component measurements are executed and there is no measurement volume coincidence problem. It is known that for a backscatter fibre LDA system, the scatted light from the measurement volume is received by the front focusing lens of the probe. The light will then be parallel and is focused into the fibre for further signal processing by a second lens is not parallel to the lens axis, then it will not be sufficient focused into the fibre by the second lens. The scattered light will then be weaker in intensity, and the data rate will be considerably lower (see appendix A). This undesirable situation always occurs when the LDA-probe is aligned off-axis to the flow. We can analyse this by assuming that a small beam is sent from each elementary segment of the probe-focusing lens. Because of the astigmatism the majority of these beams do not pass through the measurement volume, which is formed by the two transmitting beams. Conversely, the scatter light received by the focusing lens from the measurement volume will only be parallel for elementary segments from which the assumed beam pass through the measurement volume. Because these elementary segments are only a small part of the lens surface, the great part of the scattered light received by the probe cannot be sufficient focused by the second lens into the fibre. Thus the measuring data rate will be very low.
Figure 3.12: Data rate dependence on the off-axis alignment angle . Measured in a pump by Zhang (1993)
Fig 3.12 presents an example of the dependence of the data rate on the off-axis alignment of the LDA probe relative to the normal of the water flow in the pump (Zhang 1993). The data rate is here strongly dependent on the off-axis alignment angle ( ) of the probe, as mentioned above, because of the effect of astigmatism. To minimize the effect of astigmatism on the data rate, the off-axis alignment angle of the probe should not be too large.
3.7 Axial Velocity Measurement

Measurements of axial velocities require that the two-laser beam lie in a plane parallel to the pipe axis. By arranging the plane containing the two laser beams (designated as the optical plane) to go through the pipe axis, the laser beam refraction on the internal wall of the circular pipe is comparable to that on a perpendicular plane surface. As Experimental and Numerical Study of an Axial Flow Pump 3.18
Chapter 3: Applications of the Laser Doppler Anemometry Equipment long as the optical plane is kept parallel to the pipe axis, without having to go through it, the intersection angle between the two refracted laser beams remains unchanged and the intersection of the point will move as equation 3.24. When deviations of the optical plane from the pipe axis exist, the effect of astigmatism becomes significant and the quality of velocity signals will be considerably deteriorated. Since the deviation of the LDA optical plane from the pipe axis brings about undesirable features, large deviations should be avoided. For moderate deviations and as long as the signal rate is sufficiently high, no particular attention needs to be paid to this. The fringe distortion in the measurement volume, which results from the dislocation of laser beam waists and which is dependent on the deviation of LDA optical plane form the pipe axis, is anyhow not evident and will not be treated (Zhang, 2004)
3.8 Measurement of Tangential Velocities

For the measurement of the tangential velocities, the laser beams are aligner, so that the optical plane is perpendicular to the pipe axis. In regarding the laser beam refractions on the circular surface, two main aspects need to be considered:
The shift of the measurement volume in the flow is no longer proportional to the shift of the LDA unit. The intersection angle between the two refracted laser beams and thus the property of the measurement volume depend on the local position of the measurement volume in the flow.
Detailed laser beam calculations should therefore be conducted to ensure the correct LDA measurements. The curvature of the cylinder will make it act as a convex lens thus increasing the angel of intersection of beams and moving the real position of the intersection nearer to the optical system. Assuming that for small angles the sine and tangent are equal, the equation 3.33 was derived for a refraction factor G (Boadway and Karahan (1981), Bicen (1982), Zhang (2004): n2 1 ra (1 n1 ) Ro n1 G = n2 1 + . + . R0 n1 n2 RI
3.33
Where G is the refraction correction factor, n2 is the refractive index of water, n1 is the refractive index of polycarbonate glass, ra is the radius position of the virtual intersection of beams with no refraction (Note that ra is negative for positions beyond the axis from the optical source), R0 is the outer radius of the tube wall and RI is the inner radius of the tube wall. The true radius position of the beam intersection may be calculated in a similar by the equation below:
rf = ra G
3.34
The maximal error arising from the approximations leading to equation 3.34 is below 0.1% and really negligible (Zhang, 2004)
3.19
Chapter 3: Applications of the Laser Doppler Anemometry Equipment In using the laser beam intersection angle in fluid, the fringe spacing in the measurement volume is calculated as
x =
2 2sin 2
3.35
Because 1 2 = n2 n1 , this equation is rewritten as

x = n1 sin 1 1 n sin 1 = 1 x0 n2 sin 2 2sin 1 n2 sin 2
3.36
Whereby xo represents the initial fringe spacing, which is the same for laser beams both in the air and in the pipe wall. Equation 3.36 point out that the measured tangential velocity Vt has to be corrected by a factor equal to
k= n1 sin 1 n2 sin 2
3.37
Or with respect of equations 3.33 and 3.34

Vt ,cor = Vt G
3.38
Where Vt is the measured velocity with equation A.7 and Vtcor is the corrected velocity of fluid. The true position of the intersection of the point for the tangential velocity measurement is different than those of the intersection point for the axial velocity measurement. Hereby, it is not possible to measure the one component of the Reynolds stress ( < (u U )(v V ) > ).
LDA
Figure 3.13: Situation sketch of the blade shadow
3.20
Chapter 3: Applications of the Laser Doppler Anemometry Equipment Despite the blade twist causes shadowed region in the measurement region. The laser beams for the boundary layer measurement at the hub and the rotor blades near the hub will be blocked due to blade twist at all sections (cfr. see figure 3.13). Another important phenomenon that is not to neglect are the vibrations of the window. The vibrations are descending from the rotating machinery and the hydraulic noise, and carry by the star construction to the window. Owing to these vibrations, the geometrical optics will move too. Except due to the turbulence behaviour of the flow, the instantaneous measured velocity changes around the average value through the vibrations of the geometrical optics, because the fluid velocity is connected with the half angle of the laser beams, as given in equation A.7. Because the vibrations of the window move around equilibrium point, more samples of the instantaneous measured velocities are needed in environment with vibrated geometrical optics than in environment with free-vibration wall. Measurements near the endwall and the clearance region themselves were difficult due to the reflections off the blade tip gave erroneous data. Measurements near the endwalls will be possible by using a fluorescent dye seeding material (Strazisar and Powell, 1981) and an optical filter. Without the fluorescent seeding particles, reflected light from the endwall surfaces prevents measurements at radial positions less than 5 mm from either endwall.
3.9 Seeding Particle

The most important elements in laser velocimetry, yet the most neglected, is the small particle embedded in the flow field that scatters the light necessary to make velocity measurements. Since the laser velocimeter measures the velocity of small particles embedded in the flow and not the flow itself, measurement accuracy directly depends on the ability of these particles to faithfully follow the fluid flow. The characteristics of this lowly particle are often ignored in the effort to obtain data. This seems strange since it is the primary cause of measurement error. If the particle is too large, it will not follow the flow resulting in an inaccurate representation of the fluid velocity. If the particle is too small, it will not scatter sufficient light to provide the signal-to-noise necessary to minimize measurement uncertainty in the signal processing electronics. Choosing the proper seeding particle for laser velocimetry applications is a classic case of compromise. A smaller particle will more faithfully follow the fluid flow increasing measurement accuracy, while a larger particle will scatter more light increasing signal strength resulting in greater measurement precision. The chosen particle size is often determined by the ability of the optical system to see that particle and inaccuracies accepted. The sensitivity of the optical system is in turn constrained by facility, optical, and financial limitations. Several particle-seeding techniques for laser velocimetry applications have been presented by Meyers (1991). These techniques include atomizers and vaporization/condensation generators using liquids, and fluidised beds and liquid carriers for solid particles. The best seeding particles are polystyrene because they are spherical, light (specific gravity of 1.05) with a high index of refraction (1.59), mono disperse and, when injected with water-ethanol, do not agglomerate. These technique are interesting when the needed volume of seeding particle are small because the cost of the particles (In our case, we needed seeding particle for a volume
3.21
Chapter 3: Applications of the Laser Doppler Anemometry Equipment of 23.000 m3, and knowing that the price of the seeding particle is very expensive). The test bench is filled with rainwater. It has enough polydisperse seeding particle to scatter the light. The error is corrected simply by subtracting the extra variance from the measured value. The number of sample is studied in the following paragraph.
3.10 Choice of Number of Particles and Estimation of the Errors

After deciding on the size of the storage window or bin, one still needs to consider the uncertainty in determining quantities such as velocity and turbulence intensity from the LDV measurements. Appendix B developed a procedure to quantify this uncertainty. The uncertainty analysis includes considerations for bias (fixed) and precision (random) errors. Uncertainty in LDV measurement include bias errors from laser beam geometry, the counter processors, and the seeding, as well as precision errors from data processing. Data errors result from averaging a finite number of measured samples per data point. The laser beam geometric errors include, among others, the position uncertainty of the measurement volume, probe volume angular sensitivity, fringe spacing uncertainty, beam orientation bias, and finite probe volume bias. The processor bias results from errors in clock synchronization, quantizing errors, threshold limit errors, pedestal removal filter errors and electronic induced noise. Finally, the seeding bias includes errors associated with seeding injection low distortion, the arrival rate of seed passing through the probe volume, and partial lag in accelerating flows. Appendix B shows that the total uncertainty in LDV measurement can be determined by combining the precision and bias errors,
= B 2 + ( tv,95 P )
3.39
Where B is the total root-sum-square values of the elemental bias errors, t0.95 is the value from the Students t test for a 95% confidence level (which approaches 1.96 for large sample sizes) and P is the precision error of the measurement. The data processing errors are random errors, which result from the averaging of a finite number of data samples to obtain mean velocity and turbulence intensity measurements. In LDV measurements, the measured velocity does not remain constant during the elapsed sampling time because the fluctuation of the turbulence. As a result, the precision error of the mean velocity for any finite number of measurements of a fixed quantity can be calculated by
P= S M 1
3.40
with
S= 1 M (Vm V ) M 1 m =1
2
3.41
Where
V = 1 M
V
m =1
3.42
3.22
Chapter 3: Applications of the Laser Doppler Anemometry Equipment The precision error calculated by S is an estimate of the rms turbulence level. For the present investigation, the precision error of the mean velocity has to be less than 2%. With this criterion, the number of particles was chosen so that the relative precision error given by equation 3.39 is less than 2%.
t0.95 .P < 2% V
3.43
We fixed the choice at 8000 particles for the axial velocity and the tangential velocity per measurement point. Of course, this is valid only for the specific locations we considered: more turbulent zones need more particles to reach 2% error, whereas calmer zones need less. A systematic check of the number of particles necessary in each zone would be useful, but is not realistic. Therefore, in the following, we chose a fixed number of particles for all the positions.
3.11 Conclusions
A 2D laser Doppler anemometer of Dantec will measure the axial and tangential components of the flow field in the impeller of the propeller pump. After analysing the different possibilities of the optical access within the impeller, we decided to mount a polished polycarbonate window in the casing of the impeller. Refraction of the laser beams, both in the transmission through this window and through the flow leads to displacement of the measurement volume and changes in the intersection angle of the laser beams. After a profound research of the optical correction, we concludes that measurement of the velocity field in an internal flow using LDA though a transparent window with an off-axis alignment can result in quite different measurement parameters when compared to that using an on-axis probe alignment. Because of astigmatism, the intersection points of a 2D-probe with off-axis alignment are not coincident, so that coincident measurements become impossible. The path of the measurement volume is generally two-dimensional when the probe is traversed in one dimension. This traversing behaviour of the measurement volume influences the coincidence of the measurement volumes and needs special consideration, if all three components of a flow field will be measured simultaneously using a two-probe system and a traversing facility. Due to the difficult of the off-axis alignment, only a 2D flow field in the propeller pump is measured by moving the probe in on-axis alignment. Because the aberration arising from the curved interface of the circular pipe, measurements for axial and tangential must be carried out separately. Due to the refraction of laser beams at curved interfaces during the LDA measurements in flows, the position of beam intersection and the half angle may undergo non-linear changes in wavelength, with respect to the ones without refraction. Therefore, a guideline for the corrections of the measurements point of the two velocity components and the correction for the tangential velocities with respect to the shift and the optical properties of the measurement volumes has been presented. Time stamped encoded data were collected in bin size of 1 degree (or 0.5 degrees in the wakes region), so that the total average error with a number of 8000 samples is around 5,5 % (see appendix B).
3.23

4.1 Introduction
The primary objective designed into this test facility is the hydraulic performance testing of the axial flow (or propeller) pump. The facility is accommodated with the necessary instrumentation: Flow, wall pressure transmitters, rotor shaft torque and rotation speed measurements. This measurement equipment with its accuracy was discussed in detail in chapter 2. In this chapter, the measurement procedure, results and their errors will be treated. In the second phase of this chapter, the measured flow field through the impeller of the axial flow pump will be discussed. The measured quantities of the flow field include the axial and tangential velocities, and also the turbulence Reynolds stress components. The details of the measurement matrix will be handled and also the consistency of the measurement will be checked.
4.2 Working Regime

The pump was operating at the design flow of 302 l/s with a constant rotation speed of 750 rpm and a total head of 1.24 m, or at specific values of = 0.0886 and = 0.01597, except for the measurement of the performance of the propeller pump. The temperature of the water within the flow field generally increased during a day of testing, with a maximum increase of about 10C. For the experiment, the temperature ranged from 20 C to 30C, with an average of about 25C. For the numerical code the averaged flow temperature was used and the values of the density is = 1001 kg/m3 and the kinematics viscosity = 1.01.10-6 m2/s.
4.3 Global Performance of the Axial Flow Pump

To characterize the performance of the axial flow pump, the head, power and efficiency in function of the flow were measured for different rotation speeds of the pump. In running each constant speed curve, it was best to start at the lowest pressure ratio and take a full set of measurements. The flow decrease with the rotation speed of the booster pump and the outlet butterfly valve should be closed more slowly until the surge point is approached. The pressure at the inlet of the axial pump was keep always around 1.25 atm. Having located the surge point, suitable intervals of pressure ratio or mass flow can be determined for making full sets of measurements, including a set as close as possible to the surge point. Surge was detected audibly and visual. A sensitive test for surge is the oscillation of the manometer or pressure gauge attached to the outlet pressure tapping. At high speeds, the pressure ratio measurement is often more important as the mass flow variation is small. At low speeds, the mass flow measurement may be the more important as the curve of pressure ratio against mass flow often tends to be relative flat towards surge.
4.1
Chapter 4: Experimental Results of Axial Flow Pump The performance test has been performed at a model size rotation speed of about 750 rpm. The rotation speed of the propeller pump is kept around the design speed to keep the Reynolds numbers as constant as possible between the design and test conditions. Otherwise the variations of the Reynolds number has to be include in the performance scaling. The dimensionless pump performances become depending on the viscous or roughness effects, and Reynolds number corrections are recommended. There is no particular order in which the different speed curves must be run, but it may be convenient to run them in either ascending or descending order. Table 4.1 gives the limit of the flow for each rotation speed of the propeller pump. The negative head limits the maximum flow rate, and the minimum flow rate is limited by the surging. Rotation speed (rpm) 720 730 740 750 760 770 Min. Flow (l/s) 218 221 224 228 230 234 Max. Flow (l/s) 350 355 360 364 370 375
Table 4.1: Limit of flow in function of rotation speed
All experiments results contain a certain degree of uncertainly that one must consider when analysing these results. Here, we shall consider the uncertainly associated with the performance of the pump using a standard uncertainly analysis, such as outlined in paragraph 2.3. The uncertainties correspond to our estimates of the combination of bias and precision error. For the most of this uncertainty analysis, we attempted to estimate the uncertainties to 95% confidence. The numbers of data samples of the quantities are chosen so that the precision error will be less than 2 %. The data scatters between the 95 % confidence levels during the test. The bias errors of the measurement instruments are discussed in chapter 2. Using the root-sum square technique, we can combine the bias error and precision errors to obtain overall uncertainties in measuring the performance of the axial pump. Table 2.7 shows the uncertainty of the performance measurements. To eliminate errors due to fluctuations in the circuit, all instruments took 500 samples during the time of about 2,5 minutes for each duty point and are analysed with the data acquisition described in the previous chapter. The average M of the random variable X of N different samples is calculated with the next equations:
M= 1 N Xi N i =1
4.1
The statistical uncertainly that effects the rms error of the average M has been evaluated by the following expression with t0.95 = 1.96 (for a 95 % confidence level):
=
t0.95 .S N 1
4.2
4.2
Chapter 4: Experimental Results of Axial Flow Pump where S is the standard deviation;
S2 =
( X i M )2
i =1
4.3
The number of samples is chosen so that the rms error of the variable is less than 2 percent, except for the measurement in the surging area. The measurement of performance is conformed to the code ISO 3555. It is convenient to conduct the testing at predetermined values of the non-dimensional parameters flow coefficient , head coefficient , power coefficient Cp and efficiency as defined in the following equations 4.4, 4.5, 4.6 and 4.7:
=
=
CP =
Cm U
4.4
g.Htot U2
4.5
P C. = 3 5 .n .D .n 3 .D5
4.6
.Q.g.H tot C.
4.7
Where vm is the averaged flow velocity at the inlet of the pump, U is the tip blade speed, Htot is the total head over the pump, P is the shaft power as defined in equation 2.4, is the density of water, n is the rotation speed, D is the impeller diameter (0.3514 m) and g is the gravitation. The flow and the head coefficient are replaced by equation 4.8 and 4.9, because the averaged flow speed varies during the passage in the stage through the variable cross surface of the pump (see figure 2.1).
= =
with Q is the flow rate.
Q n.D3 g.Htot n 2 .D 2
4.8
4.9
The calculations of the different physics quantities are explained in paragraph 2.3.
4.3.1 Results and Discussion of the Global Performance of the Propeller Pump
The results of the performance tests for the axial flow pump in function of the rotation speed of the impeller are shown in figure 4.1 and 4.2. The efficiency characteristic is plotted in figure 4.5.
Figure 4.1: Head-flow rate characteristic of the axial flow pump in function of the rotation speed
Figure 4.2: Shaft Power-flow rate characteristic of the axial flow pump in function of the rotation speed
These results of the performance tests are dimensionless scaled with the flow, head and shaft power coefficients, defined by equations 4.8, 4.9, 4.6 and 4.7, and are shown in figures 4.3, 4.4 and 4.5 Figure 4.3 plots the results of the non-dimensional total head versus flow. The total headflow curve of the axial pump is convex as was to be expected for axial flow pumps. The losses increase quadratically with increasing flow, which leads to a Experimental and Numerical Study of an Axial Flow Pump 4.4
Chapter 4: Experimental Results of Axial Flow Pump decrease of the total head. The left point at which the curve is terminated is referred to as the stall point and marks the limit of stability. (Stall appears by flow smaller than the flow coefficient = 0.06655). To the left of this point, either oscillations of the mass flow rate may occur (stall), severe self-induced circumferential flow distortions rotate around the annulus (rotating stall) or a combination of both phenomena appears. Rotating stall induces large vibratory stresses in the blades of the axial pump and is therefore unacceptable for structural reasons. Surge can be intolerable from the point of view of system operation and can also lead to high blade and stress levels. Thus, no matter which type of instability appears when the stall point is crossed, the stall point generally represents a limit to the useful operation of the machine and is therefore to be avoided. During the stall, the annulus averaged mass flow and the system pressure rise during surge undergo large amplitude oscillations and were detected by the oscillation of the flow and the manometer. Even the rotation speed of the propeller pump oscillated more to the surging point.
Figure 4.3: Performance curve: Flow-Head Coefficient
The measurements of the pump performance were more stable when going father from the surging point. Greitzer (1981) reviewed very well the types of instabilities, which are encountered in pumping systems Figure 4.4 plots the results of the non-dimensional power versus flow. The curve of the shaft power coefficient has a declining curve with increasing flow coefficient. The shaft torque has to decrease for a constant rotation speed when the flow increases. Figure 4.5 plots the results of the non-dimensional efficiency versus flow. The maximum efficiency = 74.8 % is obtained at the flow coefficient = 0.0799 (or a flow rate Qd =274 l/s at a rotation speed of 755 pm). This point indicates also the working point of the axial pump (design point), even when the maximum head coefficient is at a lower flow coefficient where the total losses are minimum. The Experimental and Numerical Study of an Axial Flow Pump 4.5
Chapter 4: Experimental Results of Axial Flow Pump pump was designed by the Flygt company to operate at a flow coefficient = 0.088, or at a flow rate Qd = 302 l/s with a rotation speed of 755 rpm. Flygt designed the axial flow pump with the quasi-three dimensional flow analysis HEDS of Flygt. More details about HEDS are not known.
Figure 4.4: Comparison performance data: Flow-Power Coefficient
4.6
4.4 Measured Flow Field at Design Point

Two-component time-averaged LDV velocities surveys were conducted in 9 stations; 2 stations before the leading edge of the impeller, 5 stations in the rotor blade and 2 behind the trailing edge. These surveys map the range of the rotor influence upstream of the rotor and the continuous wake formation downstream of the rotor. Figure 4.6 shows the axial planes where measurements were taken, with each survey given a corresponding number from A to I. Table 4.2 provides further information about the surveys corresponding to these letters, including the locations of the LDV measurements and the number of points in the radial directions. In every survey, we move the LDV in the centreline of the propeller pump. Figure 4.7 shows the locations of the measurement point in the axial planes of the measurements. In the tip region, a denser grid is chosen than the one outside this region because of the high density of turbulence. The radial distance between the measurement points is 2,7 mm in the tip region and outside this region 6,7 mm. This is equal to moving the traverse table every 2 mm in the region of the tip and every 5 mm in the region outside the tip. The exact locations of the measurement volumes are calculated with equation 3.23 for the axial velocity and equation 3.34 for the tangential velocity. The total number of spanwise en circumferential locations in every survey plane change, because the cross sections in all the planes are not the same. The diameters of the hub and the shroud are not constant for all the surveys. As shown in figure 4.6 and table 4.2, LDV surveys A and B were performed to determine the level and the decay of the potential effects upstream of the rotor blades. Surveys C to G were taken to measure the details of the wakes in the blades at 5 different axial positions. Surveys H and I were taken behind the rotor blade tip trailing edge to measure the level and the decay of the wakes downstream the trailing edge of the rotor blades and complete the measurement over the rotor blade passage.
A B C D E F G H I
Figure 4.6: Axial planes of LDA measurement
For the surveys A and B the full 360 degrees of the rotor rotation is measured, while in the other surveys this was not possible because of the thickness of the blade and the blade shadow (see chapter 3: Applications of the Laser Doppler Anemometry). The
4.7
Chapter 4: Experimental Results of Axial Flow Pump numbers of the radial measured position increase with the spanwise position as show in table 4.2. The number of measurements in the tip region changes with the cross sections because of the variable gap between the tip and the shroud. The blades passages were identical so that the information obtained during complete revolutions could be used to calculate the mean velocities and the turbulence intensities for the propeller pump. The blade symmetry is shown in figure 4.8, which represents the measured mean axial and tangential velocity profiles of a measurement point at 25 % chord axially downstream the tip leading edge of the impeller and 74.6 % span for the complete revolution of the impeller. The velocity profiles are symmetric, which indicates the geometrical geometry of the impeller blades.
Survey
Axial Measurement Locations
Min. Nb. of Radial Points per Cycle Vt Vz 120 120 93 94 100 102 100 105 120
Nb. of Radial Points
A B C D E F G H I
40% ARC 1 upstream leading edge rotor 20% ARC upstream leading edge rotor Leading edge rotor 25 % ARC downstream leading edge rotor 50% ARC downstream leading edge rotor 75% ARC downstream leading edge rotor Trailing edge rotor 20% ARC downstream trailing edge rotor 40% ARC downstream trailing edge rotor
120 120 103 100 105 111 105 120 120
27 26 24 23 23 23 22 23 23
Table 4.2: LDV Surveys corresponding to locations shown in figure 4.4
ARC = Axial Rotor Chord
4.8
Figure 4.7: Grid of measurements
Figure 4.8: Mean blade-to-blade velocity profile to check blade symmetry
The zero-values of the velocity in figure 4.8 are the position of the blades. The axial velocity profile has less measurement points due to the blade shadow. The laser beams are interrupted through the rotation blades. By this, there will be less measurement points near the hub.
4.9
4.4.1 Measured Data

Since the LDA measures the absolute components of velocity perpendicular to the fringes, the data include the axial and tangential mean velocities and the root-meansquare values of the turbulent fluctuating velocities. The next quantities are calculated with the numerical results: Mean value of axial velocity Vz = <vz> Mean value of radial velocity Vt = <vt> Average axial square fluctuations 2 =< (Vz vz ) 2 > Average tangential square fluctuations 2 =< (Vt vt ) 2 > 4.10 4.11 4.12 4.13
The mean turbulent kinetic energy is the summations of the normal stress components as calculated using:
k= 3 3 2 2 = ( 1 + 2 ) (Vz vz ) + (Vt vt ) 4 4 4.14
The coefficient for k accounts for the variance of the out-of-plane velocity component, assuming that it is an average of the measured in-plane components.
4.4.2 Contour Plots of the Raw Flow Quantities of the Sections

In this section, the contours of the velocity measurements and the turbulent kinetic energy are plotted for the nine selected sections axially distributed inside and outside the rotor. The velocities were non-dimensionalized by dividing them with the blade tip speed. The time averaging is performed in the relative frame of reference, rotating with the rotor blades. These averages are carried out to remove any turbulent part in the velocity signal. In the contour plots, the rotor turns counter clock-wise. The right side is the pressure side and the left side is the suction side of the blade.
4.10
Contour Plots of Normalized Axial Velocities in the Sections
a) 40 % chord axially upstream of the tip rotor leading edge
b) 20 % chord axially upstream of the tip rotor leading edge
c) Rotor leading edge
d) 25 % chord axially downstream of the tip rotor leading edge
e) 50 % chord axially downstream of the tip rotor leading edge
f) 75 % chord axially downstream of the tip rotor leading edge
Figure 4.9: Contour plot of the normalized axial velocities
4.11
g) Trailing edge
h) 20 % chord axially downstream
i) 40 % chord axially downstream Figure 4.10: Contour plot of the normalized axial velocities (continue)
4.12
Contour Plots of the Normalized Tangential Velocities in the Sections
c) tip rotor leading edge
Figure 4.11: Contour plot of the normalized tangential velocities
4.13
g) trailing edge
h) 40 % chord axially downstream
i) 40 % chord axially downstream Figure 4.12: Contour plot of the normalized tangential velocities (continue)
4.14
Contour Plots of the Normalized Turbulence Kinetic Energy in the Sections
c) tip rotor leading edge
Figure 4.13: Contour plot of the normalized turbulence kinetic energy
4.15
g) trailing edge
h) 20 % chord axially downstream of the tip rotor trailing edge
Figure 4.14: Contour plot of the normalized turbulence kinetic energy (continue)
4.4.3 Pitch-averaged Velocity Profiles of the Measured Locations

The pitch-averaged velocity profiles of the measured locations are plot in figures 6.20 to 6.22 of chapter 6 (Evaluation of test data of CFD) for all cross sections (and compared with the numerical results that will be more explained in that chapter). Due to the blade shadow, the pitch-averaged of the measured velocities is not always available for the complete blade-to-blade profile, because of the missing information of the quantities in the blade shadow. Only in surveys A and B, 40% and 20 % upstream of the tip leading edge of the rotor, it was possible to pitch-averaged the quantities of the complete revolution. In the other surveys, the pitch-averaging of the experimental data is also done, but keep in mind that the points of the blade shadow are not accounted. The velocities were non-dimensionalized by dividing them with the blade tip speed and the blade spanwise distance is defined between the hub and the casing: r rh H= 6.15 rt rh where rh is the radius of the hub, rt is the radius of the casing and r is the radius of the blade-to-blade sections.
4.4.4 Blade-to-blade Velocity Profiles of the Measured Locations

In the next pictures (figures 4.15 to 4.19), the distribution over a blade-to-blade azimuth of the averaged axial, tangential velocities and the turbulence kinetic energy in the tip region, upstream, inside and immediately downstream of the rotor row are plotted for different blade spanwises H. The tip region is important as it has the potential to contribute to a substantial loss in total efficiency and pumping head. The pitchwise positions of each graph indicate the distance from the suction to the pressure side i.e. 10 % pitch is near the suction side, while 90 % pitch is near the pressure side. In chapter 6, more distribution over the blade-to-blade of the averaged axial (see figures 6.23 to 6.31) and tangential velocities (see figures 6.32 to 6.40) for five different radial positions in the impeller are compared with the numerical turbulence models.
4.16
Plots of the Normalized Velocities in the Tip
Figure 4.15: Plots of the Normalized velocity profiles in the tip region
4.17
4.18
4.19
Plots of the Normalized Turbulent Kinetic Energy in the Tip
Figure 4.18: Plots of the Normalized turbulent kinetic energy in the tip region
4.20
Figure 4.19: Plots of the Normalized turbulent kinetic energy in the tip region (continue)
4.5 Consistency Check

Before validating the numerical models with the experimental data, a consistency check of the experimental data is necessary. The axial velocity field can be checked by the continuity equation (see equation 1.1). The mass flow through the duct has to be constant. The measured mass flow is calculated with equation (4.16) as
2
Q=
r
0
shroud
vax (r , ).dS
4.16
hub
with v(r, ) being the measured axial velocities in a survey and dS=r.dr.d. This consistency check for all surveys is resumed in table 4.3. Survey A B C D E F G H I Flow mass based on experimental data (l/s) 294 302 304 301 300 307 306 298 297 Flow rate (l/s) 300 300 300 300 300 300 300 300 300 Deviation (%) 2.0 0.7 1.3 0.4 0.0 2.3 2.0 0.7 1.0
Table 4.3: Consistency check of axial velocity
The flow rates calculated with equation 4.16 approach quite well the measured flow rates during the LDA measurements. Differences are due to the numerical discretization of equation 4.16 over the cross-section of the survey, the absent of the
4.21
Chapter 4: Experimental Results of Axial Flow Pump velocity measurement in the shadow blades, the light variation of the flow rate and rotation speed during the measurement, and/or the accuracy of the measurements. The consistency check of the tangential velocity field is more difficult due to the lack of the design procedure of the propeller pump. One check can be done with the Euler equation (see 1.4). Suppose that the tangential velocity at the inlet of the propeller pump is neglected. The Eulers equation becomes
g.H tot = u2 .vt 2
4.17
whereby Htot is the total head [m], u2 is the rotation speed at the averaged radius of the outlet of the impeller [m/s], and vt2 is the tangential velocity at the same position of the average rotation vector [m/s]. Mathematically the tangential velocity vt2:
vt 2 = g .H tot g .H tot = = 1.24m / s u2 .raverage
4.18
where raverage is the average radius at the outlet of the impeller (m) and is the rotation speed (rad/s). The pitch averaged velocity profiles at the outlet of the impeller (see figure 6.16, normalized pitch-averaged velocity at rotor leading edge and at 20 % chord axial downstream of the rotor leading edge) have the same order as these non-dimensional tangential velocities vt2/Vtip =0.09. Knowing the tangential velocity at the averaged radius and assuming of the free vortex flow (r.vt = cte), the tangential velocity profile at hub of the outlet of the impeller can be checked as
ut =
hub
raverage .U t rhub
midspan
4.19
and at shroud of the outlet of the impeller as ut =

shroud
raverage .U t rshroud
midspan
4.20
The non-dimensionalized tangential velocity at the shroud is than vtshroud=0.06 and at the hub vthub=0.15, which approximates the values of graphic 6.22. However, the value of the measured tangential velocity near the hub is underestimated. Since the consistency check is done with a good agreement of the flow rate and the tangential velocity, the experimental data can now be used for validation of the numerical results.
4.6 Discussion
At first glance, one might observe that the flow in the propeller pump has produced a rather two-dimensional flow field. However, further analysis of the data shows that three-dimensional effects are significant. The LDA data provide information of the wakes and the tip leakage vortices in the casing endwall region.
4.22
Chapter 4: Experimental Results of Axial Flow Pump Note that in some pictures of the blade-to-blade velocity profiles, measurement points are kept away because of large uncertainties that are due to the problems in focusing the laser beams far from the tunnel window, and/or due to the important noise in the scattered light near the solid boundary, and/or simply because of the shadow blades. For these reasons, velocity profiles for lower spans inside the rotor row have less measurement points. Within the clearance regions themselves, reflections of the blade tips gave erroneous data. However, while the LDA measurement volume was located outside of the clearance regions and within the blade passages, data was measured. In all the contour plots (figures 4.9 to 4.14), there are zones with zero speeds, which indicate the cross section and the measurement shadow of the blade. The measured axial contour plots (figures 4.9 and 4.10) highlight that there is a reverse flow near the casing from the leading edge to the cross section 40% downstream the trailing edge of the impeller. Upstream of the impeller blades, a minimum of axial velocity is shown due to the blockage of the rotor blades, which causes a depression in the velocity field. Between the impeller, the axial velocity has a higher value in the middle of the passage than in the boundary of the blades. The axial velocity is higher at the suction side than at the pressure side of the blade. Downstream the midchord of the impeller, in the suction side near the tip, a region with high axial velocity is shown. Downstream of the trailing edges of the impeller, the wakes of the impeller are visualized by the minimum axial velocities. From the contour plots of the tangential velocity (figures 4.11 and 4.12), it is seen that the flow within the passage is dominated by the motion of passage vortex and the tip leakage vortex. The motion of the passage vortex is moving from the pressure side to the suction side near the hub and from the suction side to the pressure side near the shroud. The passage vortex flow becomes weaker from the entrance of the impeller to the end passage. The tip leakage vortex, in the suction side of the blade near the tip, moves in the contra direction of the passage vortex. The contour plots of the turbulent kinetic energy (see figures 4.13 and 4.14), which is calculated with equation 4.14, show the generation and development of the tip leakage vortex in the suction side corner in all the cross sections downstream the leading edge. It moves towards the midpassage over the casing during the passage through the impeller. The leading edge vortexes are also visualized in the cross section of 25% chord axially downstream, but disappeared in the next contours plots because these vortexes are weak to travel through the passage. The horseshoe vortex near the hub cannot be obtained by the lack of measurement points, because of the high reflection of the laser light by the hub. The horseshoe vortex in front of the tip of the impeller is also not developed because of the high tip gap. One might anticipate that some of the other vortex structures (indicated in the paragraph of literature review) from this lightly loaded rotor may be too weak to show up in the measurements. Based on the turbulent kinetic energy, the tip-leakage vortex seems to have a larger strength than the leading. The trailing edge vortex was also not visualized. The pitch-averaged axial and tangential velocity profiles are shown in figures 6.20 to 6.22. The axial velocity profile at the entrance of the impeller is not symmetric due to the convex form of the casing in front of the impeller. Before entering in the impeller, the flow near the casing will first accelerate, and later on near the hub, because of the strong concave form of the hub. The boundary layer near the casing grows when entering the impeller, stay more or less constant in the tip of the impeller and continue growing when leaving the impeller. The boundary layer in the casing is much thicker than at the hub, because of the development of the tip leakage flow. The concave hub, Experimental and Numerical Study of an Axial Flow Pump 4.23
Chapter 4: Experimental Results of Axial Flow Pump upstream of the midchord, provides a boundary layer that is quite thin and will grow when leaving the impeller, because of the convex curvature of the hub at the end of the impeller. At the inlet of the impeller, the pitch-averaged tangential velocity component is quite constant over the full span, except near the bulb of the axial flow pump, where the flow has a swirl. From the leading edge of the impeller, the tangential velocity component increases towards the trailing edge. The tangential velocity profile in the cross section of the impeller is of the free vortex type; at the hub the tangential component has a higher value than at full span. Downstream the midchord of the impeller, the tangential component near the tip increases because the tip leakage flow increases. The experimental blade-to-blade velocity profiles in the impeller are shown in figures 6.23 to 6.44, together with the numerical results, and in figures 4.15 to 4.17 for the tip region. For the first and the second survey (figures 6.23 and 6.24), located at 40% and 20 % upstream of the leading edge of the impeller, the potential blockage of the rotor blade causes a depression in the axial velocity profile for every spanwise location. The velocity gradient is lighter at the suction side and the blockage due to the leading edge of the blade can be identified as velocity minima. This blockage of the axial velocity is also perceptible at the span of the tip. The passage vortex is recognizable in the plots of the tangential velocity (see figure 6.32 and 6.33). The amplitude of the swirl of the passage vortex near the hub is higher than at midspan and the casing. As shown in the blade-to-blade plots of the cross section at the leading edge (see figures 6.25), the axial velocity near the pressure side of the blade will first decrease and later increase very strong at any distance blade surface. At the suction side, the axial velocity will not increase near surface of the blade. This is because the high incidence angle of the flow creates a recirculation region near the hub. In the pictures more downstream the leading edge, the axial velocity near the blade boundary increases after the region of the recirculation region. The influence of the flow with a high incident angle is not visible anymore in these pictures. For the planes downstream the leading edge of the rotor, the axial velocity distributions (see figures 6.26 to 6.29) increase linearly from the pressure side to the suction side, except in the boundary layer of the blades where on both sides of the blades the boundary layer has been developed during the passage. Some velocity profiles staggers because of the lower acquisition data rates. The flow in this region is very complex as there is interaction of the rotation with the boundary layer formed on the bullet shaped hub and the large curvature of the blades close to the leading edges. The axial velocity at the suction side increases strong with regard to the leading edge. The large velocity gradients near suction surface indicate that the wall shear stresses on the suction surface are larger that those acting on the pressure side. The local maximum of the axial velocity near the pressure side of the blade decreases downstream and disappears at midchord. The recirculation region disappears before midchord. A boundary-layer flow analysis could be carried out using the momentum integral technique, but he evolution of the blade wakes in the span and chord direction cannot be made due to the missing data of the points in the blade shadow. The measurements of the boundary-layer characteristics (three components of velocity, turbulence intensities, Reynolds stresses, shin-friction coefficient, and limiting streamline angle) could carried out utilizing the tri-axial hot-wire probes or five-hole pitot tubes rotating with the rotor inside the passage.
4.24
Chapter 4: Experimental Results of Axial Flow Pump At the outlet, the velocity distributions of the axial (see figure 6.30 and 6.31) and tangential velocity (see figure 6.39 and 6.40) are flat out of the blade wake, which is noticeable as a velocity depression, because the surface of the cross section increases due to the convex hub and the thickness of the blades that becomes smaller. Figure 4.20 shows the axial velocity deficits associated with the rotor blade wakes. Notice how the wake structures have much deeper and broader profiles when the flow goes more downstream. The blade-to-blade tangential velocity profiles are shown in the figures 6.32 to 6.40 and in 6.43 to 6.44 for the distribution in the tip. The passage vortex is recognized in all plots upstream the midchord of the impeller. The tangential value has a positive value at the suction side of the impeller, and a negative value at the pressure side, which indicates the swirl of the passage vortex. The tangential velocity has higher values at the hub than near the tip, because the pitch-wise distance between the blades is shorter at the hub than at full span. The plots of the tangential velocity component before the leading edge of the impeller indicate that the passage vortex is already absent before entering the impeller. The potential blockage of the blades influence the flow upstream the impeller. In the cross section of the leading edge, the leading edge vortices in the suction side of the blades and the tip leading vortices are recognized. The sign of the tangential velocity changes locally. Near the hub (at 10% and 20% span), the swirls of the leading edge vortices are much smaller because of the strong radial velocity components near the hub that are due to the strong concave surface. At the pressure side of the blade, the leading edge vortices are sucked by the recirculation region that exists on the pressure side of the blade near the leading edge. The leading edge vortices at the suction side disappear in the cross section downstream the midchord of the impeller. In each cross section inside the impeller, the tangential velocity profile distribution decreases with increasing spanwise, following the free vortex type. For every constant spanwise position, the pitch-averaged tangential velocity increases in flow direction, due to the rotation of the pump. At the trailing edge, the tangential velocity has a larger value at the suction side of the trailing edge because the flow is sucked over the trailing edge from the pressure side to the suction side. This effect decreases further from the trailing edge and the tangential velocity profile becomes broader. A trailing edge vortex is formed over the complete trailing edge of the impeller. These trailing edge vortices are moved to the suction side of the impeller In the tangential profiles of the blade-to-blade plots at the tip near the suction side of the blade, the tip leakage vortex is clearly visible (see plots 4.15 to 4.17). At the pressure side, a separation vortex (vena contracta) exits on the blade gap surface just upstream of the pressure side corner (see plot of 25% axial downstream the leading edge of the tip). This is because the flow along the pressure surface has a large radial velocity component, which turns around sharply at the pressure side corner to from the leakage flow. This tip leakage vortex is now bigger than at the tip leading edge. In the plot of the axial velocity near the casing, a negative axial velocity in the cross section of the leading edge indicates a reverse flow due to the tip leakage flow. The detail plot of the tip in the cross section of the midchord (see figure 4.16) show that the axial velocity near the suction side of the blade is increased and decreased near the pressure side due to the narrower tip gap. At that moment, the tip leakage Experimental and Numerical Study of an Axial Flow Pump 4.25
Chapter 4: Experimental Results of Axial Flow Pump vortex is weaker and merges with the blade wake flow and will disappear in the downstream plots. Even the vena contracta has disappeared. When the gap downstream the midchord increases, the maximum axial velocity near the suction side has a broader profile and becomes more uniform. The turbulent kinetic energy has a maximum near the tip of the leading edge where the tip leakage vortex is formed (see figure 4.18). Downstream the leading edge, the profile becomes broader and the position of the maximum turbulent kinetic energy goes further from the suction side to the pressure side (see figure 4.18 and 4.19). The non-deterministic unsteadiness increases because of the increased turbulence levels in the vortex. This means that during the passage through the impeller, the tip leading vortex moves from the tip gap to the pressure side over the casing. At the trailing edge, a new peak in the profile of the turbulent kinetic energy appears due to the trailing edge vortex. Even this peak becomes broader downstream. After the trailing edge, the turbulent kinetic energy profile is very broad, not only because of the increased turbulence levels in the vortex, but also because of the vortex that is wandering and kinking. Freestream turbulence, vortex structure instabilities and the influence of the casing endwall influence the vortex wandering and kinking. This type of unsteady motion, and similar unsteadiness associated with other vortices and separation regions, increase the turbulent kinetic energy over the whole area downstream the trailing edge.
Figure 4.20: Comparing axial (left) and tangential (right) velocity of the blade wake at 20% and 40 % downstream of tip rotor trailing edge at 82 % span
4.7 Conclusion
This chapter discussed the measurement procedure and the performance of the axial pump. Measurement errors were analysed and fulfilled the ISO 3555. The performance test indicated that the measured design flow rate is 274 l/s at a rotation speed of 755 rpm instead of the theoretical design point of 302 l/s at the same rotating speed. The measurement of the flow field within the impeller is reported here. The experimental measurements and subsequent data analysis have improved our physical understanding of this complex flow field One might observe that the impeller has produced a rather two-dimension flow field. However, a further analysis of the data shows that three-dimensional effects are significant. The considerable boundary layer growth is observed from the hub to mid radius, while the flow from mid radius to tip
4.26
Chapter 4: Experimental Results of Axial Flow Pump is found to be highly complex, due to the interaction of pressure and suction surface boundary layers and the resulting radial inflow in the first part of the impeller passage. The passage vortex, tip leakage flow, trailing edge vortices and the vena contracta are recognized in the flow field. Flow field measurements fail to exhibit other vortices (described in the literature review: horseshoe vortices and vortices emanating from the three-dimensional separation regions on the blades) since there are either to weak or are filtered out in the measurement locations. The two dominant secondary flow structures within the impeller are passage vortex and the rotor blade tip-leading vortex. This tip leakage vortex also has a strong impact on the flow unsteadiness downstream of the rotor blades. The nondeterministic unsteadiness increases because of the increased turbulence levels in the vortex and because of the vortex wandering and kinking. Freestream turbulence, vortex structures instabilities, and the influence of the casing endwall all influence vortex wandering and kinking. This type of unsteady motion, along with a similar unsteadiness associated with other vortices and separation regions, will certainly add to the difficulties associated with turbulence modelling during numerical computations of the flow field in the axial flow pump. However the laser Doppler anemometry is a good instrument to measure the velocity components of the particle in the flow, it has handicaps in our application. The velocity field cannot be measured completely due to the blade shadow, the high reflections of the laser light by the blade and the hub, and/or the low data rate near the hub. It means that the velocity profiles of the blade and the hub boundary layers can never been measured. Due to the unsteadiness characters of some vortices, it is difficult to create a database that represents the complete flow field in the impeller. Also due to the curved casing and the different refraction indexes of window and water, it is difficult to measure the two (or three velocity) components in a measurement point at one time. The Reynolds stress component will be limited to the averaged axial and tangential square fluctuations. To solve these problems, other measurement instruments as five holes or hotwire probe (CTA), that rotate together with impeller will make attempts.
4.27
Chapter 5: Numerical Simulations of the Axial Pump Flow
Chapter 5: Numerical simulations of the Axial Pump Flow

5.1 Introduction
In order to investigate the pump performance and to make a flow field analysis numerically in the impeller, a numerical model has been created and the time marching finite-volume based Navier-Stokes solver EURANUS of FINETM Turbo by NUMECA International was applied. The three dimensional compressible Reynolds averaged continuity, Navier-Stokes and energy equations are solved using structured multiblock meshes that modelled the propeller pump completely (impeller, gap, stator and the inlet duct). The meshing of the grid was created using IGGTM/Autogrid by NUMECA International which is a mesh generator for optimising multiblock structured meshes for orthogonality and mesh point clustering for an accurate description of viscous effects in the boundary layer. Two different models, one for the validation of the performance of the propeller pump and another for the validation of the flow field in the impeller have been approached.
5.2 Prediction Method

The steady incompressible Reynolds-averaged Navier-Stokes equations are solved for the flow past the current propeller pump, i.e.,
U i =0 xi
U U j 2 U k 1 p eff i + + U iU j + 2 ijk jU k = ij ui u j + Fi x j x j xi 3 xk xi x j
5.1
5.2
where Ui and ui are mean and fluctuating velocities, j is the angular velocity and
eff = laminar + turbulent
5.3
To close the system of equations, a closure model for the turbulence stress should be applied. Currently many turbulence models are available for large-scale threedimensional flow calculations. Although none of the existing closure models represent all of the salient flow features accurately, major differences between the flow fields of different hydrodynamics designs can be calculated with the existing models. It is well known that neither of the conventional mixing length type turbulence model nor any standard two-equations type turbulence models describe the turbulence stresses properly in the separated flow regions. For the current study, two different turbulence models are modified to include the low Reynolds number effects following the studies of Baldwin-Lomax (Baldwin-Lomax, 1978) and Spalart-Allmaras (Spalart-Allmaras, 1992).
5.1
5.3 Turbulence Models

5.3.1 Turbulence Model of Baldwin-Lomax
The Baldwin-Lomax algebraic turbulence model (Baldwin-Lomax, 1978) is a two layer model where the turbulent viscosity in the inner layer is determined using Prandtls mixing length model, and the turbulent viscosity in the outer layer is determined from the mean flow and a length scale. The strain-rate parameter in Prandtls mixing length model is taken to be the magnitude of the vorticity. The influence on the mean flow equations through the turbulent kinetic energy is neglected. The turbulent viscosity is given by
(t )i ' n nc (t )0 ' n > nc
t =
5.4
where n is the normal distance to the wall, and nc is the smallest value of n at which the inner and outer viscosity are equal. The inner viscosity is
(t )i = l 2
where
l = kn 1 e y
5.5
/ A+
5.6 5.7
w w y+ = w
and
i = ijk
with the ijk Kronecker symbol. The outer viscosity is
u j xk
5.8
(t )0 = KCcp Fwake FKleb (n )

where Fwake is the smaller of
5.9
nmax Fmax
Cwk nmax
5.10
2
[ (u
+ v 2 + w2
max
(u
+ v 2 + w2
min
] /F
2
max
5.11
The term nmax is the value of n corresponding to the maximum value of F, Fmax where
F (n ) = n 1 e n and
/ A+
5.12
5.2
Chapter 5: Numerical Simulations of the Axial Pump Flow FKleb = 1 + 5.5(nC Kleb / nmax )
6 1
5.13
The constants used are hard coded and equal A+=26, Cwk=1., Ccp=1.6, k=0.41, Ckleb=0.3, K = 0.0168. No attempt was made to optimise the constants of the turbulence model in this study.
5.3.2 Turbulence Model of Spalart-Allmaras

The Baldwin-Lomax algebraic turbulence model has been widely implemented for complex flow problems. However, it has severe limitations when applied to vortical flows because of difficulties in the definition of an unambiguous wall-algebraic length scale. Furthermore, models of this type do not take into account the transport and diffusion of turbulence, and so turbulence history effects are not captured. These effects are important for blade tip flows, recirculation flows and separated flows. The Spalart-Allmaras turbulence model (Spalart and Allmaras, 1992) is a oneequation turbulence model, which can be considered as a bridge between the algebraic model of Baldwin-Lomax and the two equation models. The Spalart-Allmaras model has become quite popular in the last years because of its robustness and its ability to treat complex flows. The main advantage of the Spalart-Allmaras model (SA model) when compared to the one of Baldwin-Lomax is that the turbulent eddy viscosity field is always continuous. Its advantage over the k- model is mainly its robustness and the lower additional CPU and memory usage. On the other hand, the SA model is more difficult to use. This is not because of the extra storage, but because they require finer grids near a wall, involve strong source terms that degrade the convergence, and demand non-trivial upstream and free stream conditions for the turbulence variables. The principle of this turbulence model is based on the resolution of an additional transport equation for the eddy viscosity. The equation contains an advective, a diffusive and a source term and is implemented in a non-conservative manner. The implementation is based on the papers of Spalart and Allmaras (1992) with the improvements described in Ashford and Powell (1996) in order to avoid negative ~ values for the production term ( S in equation 5.24). While solving the Reynoldsaveraged Navier Stokes equations and a transport equation for the turbulence model, the Reynolds stresses are given by equation 5.17: u i u j = 2. t S ij where u u Sij = i + j x j xi The turbulent viscosity t is given by ~f v =v
t
5.14
5.15
v1
5.16
~ is the turbulent working variable and f a function defined by where v1
5.3
f v1 =
3 3 + cv1
5.17
~ and the molecular viscosity , with is the ratio between the working variable ~ v = 5.18 v ~ obeys the transport equation The turbulence-working variable
~ r v ~ = 1 { [(v + (1 + c )v ~ )v ~] c v ~ ~ + V v b2 b 2 v } + Q t
r where V is the velocity vector, Q the source term and , cb2 constants.
5.19
The source term includes a production term and a destruction term:

~P(v ~) v ~D (v ~) Q=v
5.20
where
~~ ~P(v ~) = c S v b1 v
5.21
5.22
~ 2 v ~ ~ v D(v ) = cw1 f w d The production term P is constructed with the following functions: ~ v ~ S = Sf v 3 + 2 2 f v 2 d
5.23
fv2 =
1 (1 + / cv 2 )
f v3 =
(1 + f v1 )(1 f v 2 )
5.24
where d is the distance to the closest wall and S the magnitude of vorticity. In the destruction term (equation 5.25), the function fw is
6 6 1 + cw 3 fw = g 6 6 g +c w3 1
5.25
With g = r + cw 2 r r
6
)
cw2=0.3 cb2=0.622 cw3=2.
=0.41
r= ~ 2 2 S d
~ v
5.26
The constants arising in the model are c w1 = cb1 2 + (1 + cb 2 )
cv1=7.1
=2/3
cv2=5
cb1=0.1355
The equation 5.22 is solved with the appropriate boundary conditions: on solid wall ~ = 0 , along the inflow boundaries the value of is specified ( ~ is obtained by using t
Chapter 5: Numerical Simulations of the Axial Pump Flow a Newton-Raphson procedure to solve equation 5.22) and along the outflow boundaries it is extrapolated from the interior values. Again, no attempt was made to optimise the constants of the turbulence models in this study.
5.4 Solving Algorithms

All governing equations are integrated in time using an explicit four stage RungaKutta scheme and convergence towards steady state can be accelerated using multigrid combined with residual smoothing and local time stepping. A pseudocompressibility method and a second-order dual-time stepping procedure that are implemented by Hakimi (1997), accelerate the convergence of the basic explicit timestepping scheme. The convective fluxes are treated through a second-order Jameson scheme with second- and fourth-order scalar dissipation. Using quasi-steady mixing planes approach, which pitching averaging of the flow solution, is performed at the rotor/stator interface and the exchange of information at the interface depends on the local direction of the flow simulated the rotor/stator interaction. For more theoretical details and background information, I refer to the manual of FINETM turbo
5.5 Numerical Model of Axial Pump

In the next paragraph, every step of the simulation procedure will be discussed in detail. The current computational method has been validated against many flow fields to examine the accuracy and grid-dependency of numerical results.
5.5.1 Computational Domain

Ideally all 3 rotor and 8 stator blades should be modelled in order to simulate the true flow field and to detect any asymmetry of the flow. However, most simulations of the flow field in turbomachines take advantage of the geometrical symmetry of the axial pump and simulate only one rotor and stator passage. The subject of placing external boundaries as inlet and outlet is of mayor importance. The position of the inlet boundary has an impact on the calculated flow field in the pump. In order not to affect the flow field inside the impeller, the inlet section was located four times the diameter of the inlet pipe upstream of the leading edge of the rotor tip for a simulation of a fully developed turbulent channel flow and the outlet boundary was located at 90%Dtip downstream the trailing edge of the stator tip, where the pressure taps of the experiment set-up are mounted. Two different grids were made: one without the adjustment of the shroud (see mesh 1, 2,3 of table 5.1, 5.2 and 5.3), and one with the adjustment of the shroud. The integration of the glass window in the shroud makes the curvature of the shroud smooth, so that the impeller doesnt rotate in a groove of the shroud as the original case (see figure 3.4). The distance from the tip of the impeller until the shroud becomes longer than in the original situation. The first model was used to calculate the numerical pump performance and the second was used to validate the flow field in the impeller. The geometrical outline of the computational domain is shown in figure 5.1 and 5.2. Figure 5.1 outlines the computational domain without adjustment of the shroud by the window and figure 5.2 outlines the computational domain with the adjustment of the Experimental and Numerical Study of an Axial Flow Pump 5.5
Chapter 5: Numerical Simulations of the Axial Pump Flow window. The pictures show the different in the casing because of the smoothing of the window.
Figure 5.1: Geometrical outline of the computational domain of the model without adjustment of the shroud (complete length of inlet pipe and outlet are not show).
Figure 5.2: Geometrical outline of the computational domain of the model with adjustment of the shroud for the window (complete length of inlet pipe and outlet are not show).
5.5.2 Boundary Conditions

Having defined the geometrical layout of the numerical model of the propeller pump the next subject is the application of boundary conditions. As inflow condition, mass flow rates associated with the targets for the simulations of off-design and design flow are specified. The direction of the absolute velocity vector is axially imposed at the inlet pipe. The pipe at the inlet of the impeller is long enough to develop a fully velocity profile before the flow enters the impeller. This is necessary because the radial velocity profile at the inlet of the impeller could not be measured with the 2D LDA. At the front of the impeller there are important radial Experimental and Numerical Study of an Axial Flow Pump 5.6
Chapter 5: Numerical Simulations of the Axial Pump Flow velocities that are important for the flow field in the impeller (see figure 5.25). Due to difficulties of setting up realistic turbulent inflow conditions only the mass flow and the direction of the velocity are imposed, hoping that the turbulence will develop due to interaction with the geometry. Therefore, in this context, the development of the turbulent structures requires an excessive long inlet section. Lund (1998) provides an overview of existing inflow generation techniques. The influence of the length of the inlet pipe is studied in figures 5.3 and 5.4 by comparing the axial and tangential pitchaveraged velocity profiles at 50% pipe diameter upstream of the tip rotor leading edge. The study is done for 3 different pipe lengths of 3, 4 and 6 times the pipe diameter, upstream the tip leading edge of the rotor. Before to increase the length of the inlet pipe, grid sensibility of the pipe was first studied with a separated grid that doesnt include a model of the stator to reduce the calculation time and to have more freedom with number of grid points in the inlet and impeller, without the need to care about the computer capacity. The comparisons of the velocity profiles of the different lengths were done for both turbulence models.
Figure 5.3: Numerical study of the length of the inlet pipe by comparing the axial and tangential pitchaveraged velocity profiles at 50% pipe diameter upstream of rotor leading edge (calculated with BaldwinLomax model).
Figure 5.4: Numerical study of the length of the inlet pipe by comparing the axial and tangential pitchaveraged velocity profiles at 50% pipe diameter upstream of rotor leading edge (calculated with SpalartAllmaras model).
Figures 5.3 and 5.4 show that the pitch-averaged velocity profiles in the cross section of 50% pipe diameter upstream the tip rotor leading edge dont change more when the
5.7
Chapter 5: Numerical Simulations of the Axial Pump Flow pipe inlet condition is at 4 x pipe diameter upstream the tip rotor leading edge. Only minor variations are observed between the tangential velocities of both turbulence models. This means that the inlet pipe is long enough to develop a fully velocity profile before the flow enters the impeller. Comparisons with the experimental data in chapter 6 show that it is a fact. Noteworthy is that the model of Baldwin-Lomax has difficulties to estimate the velocity profiles near the singular line of the inlet pipe (the boundary condition of the symmetry axis of the inlet pipe is a singular line). Eventually, this singular line could be changed in an Euler walls condition, but this approach is not checked in this work. It will be better to impose a measured 3D velocity field of the suction pipe at any distance from the rotor tip leading edge. The number of grid points, that will be reduce in the inlet pipe, will than be recuperate in region with high density of turbulence in the impeller and stator. The 3D velocity profile could be measured with a five-hole pitot tube. For simulations with the turbulence model of Spalart-Allmaras, the turbulent kinetic energy k has to be quantified at the inlet of the model, and it is derived from the turbulent intensity Tu defined by:
Tu = u'2 U
freestream
5.27
For internal flows such as turbomachinery flows, typical values are Tu = 5%. With these considerations, k can be calculated in the considering anisotrope turbulence:
k = 1.5(Tu .U ref
5.28
with Uref being the streamwise velocity. At the outlet of the model, a uniform pressure is described, thereby not posing any restrictions on the velocity. A no-slip condition is utilized on the solid walls, which are the shroud, the hub and the pressure and suction surfaces of the blades. These blades of the impeller are rotating with the same angular velocity together with the hub, while the shroud and stator blade are stationary. Periodic boundaries are used upstream and downstream of the blade leading and trailing edges, respectively. Rotation boundaries are utilized at the rotor-stator interface.
5.5.3 Discretisation and Grid Resolution

The meshing of the domain was created using AutogridTM which is a mesh generator optimising the mesh of orthogonallity and mesh point clustering for an accurate description of viscous effects in the boundary layer. The geometry of the propeller pump was given by AutoCAD coordinates, delivered by the factory. The H-topology of the mesh is used. A H-mesh is such that the number and the distribution of the grid along two periodic extensions are the same. The main quality of the H-topology is that it ensures a very good mesh quality, in terms of cells orthogonality.
5.8
Chapter 5: Numerical Simulations of the Axial Pump Flow Figure 5.5 illustrates a generic blade passage with the appropriate upstream and downstream extensions. This becomes the geometry that is modelled in the rotation reference frame for the impeller and in the stationary frame for the stator.
Figure 5.5: Solution domain (coarse grid)
Before to start with the final calculation for the comparison of CFD results with empirical data, the sensitivity of the grid is studied. Three different meshes were considered, one, the finest mesh, has 1.065.332 nodes (see mesh 1, table 5.1), the second has 578.724 nodes (see mesh 2, table 5.2) and the coarse mesh has 206.616 nodes (see mesh 3, table 5.3). A multiblock approach has been applied and the total of cells distributed in four blocks (inlet, rotor, gap and stator) are represented in table 5.1, 5.2 and 5.3. As the spacing between the rotor and the stator is relatively large, a mixing plane is established approximately from the physical transition rotating-static hub to the shroud centred between the two blade rows.
Mesh 1:
Block number 1 2
Domain
Streamwise direction 57 33
Spanwise direction 73 33
Azimuthal direction 113 49
Total number of points 470.193 53.361
Number of blades 3 3
Rotor Gap
5.9
Chapter 5: Numerical Simulations of the Axial Pump Flow 3 4 Bulb Stator 57 49 73 49 65 113 Total: 470.193 271.313 1.065.332 3 8
Table 5.1: Characteristics of the mesh 1 in the computation domain (used for the pump performance).
Mesh 2:
Block number 1 2 3 4
Domain
Streamwise direction 39 13 39 49
Spanwise direction 39 13 39 49
Azimuthal direction 137 73 57 113 Total:
Total number of points 208.377 12.337 86.697 271.313 578.724
Number of blades 3 3 3 8
Rotor Gap Bulb Stator
Table 5.2: Characteristics of the mesh 2 in the computation domain.
Mesh 3:
Block number 1 2 3 4
Domain
Streamwise direction 41 13 41 25
Spanwise direction 41 13 41 25
Azimuthal direction 69 73 57 57 Total:
Total number of points 115.989 6.253 48.749 35.625 206.616
Number of blades 3 3 3 8
Rotor Gap Bulb Stator
Table 5.3: Characteristics of the mesh 3 in the computation domain.
While creating the mesh particular care has been taken in order to ensure a sufficient resolution along the blades, hub and shroud as well as across the blade leading and trailing edges. Due to strong variations in velocities near the solid wall areas, it is important to have sufficient amount of points inside the boundary layer to properly capture those high gradients in the numerical simulations. The wall variable y+ varies between 1 and 10 for the low Reynolds turbulence models of Baldwin-Lomax and Spalart-Allmaras. To estimate ywall as a function of a desired y+ value, a truncated series solution of the Blasius equation (Hinze, 1959) is used:
5.10

-7 8 18
U y wall =6 ref
L ref 2
y+
5.29
where Vref can be taken from an average at the inlet, and the reference length, Lref, is the distance between hub and shroud curves that exist upstream of the first row of blade. This is an approximation, of course, as the thickness of boundary layers will vary widely within the computational domain. It is only necessary to place y+ within the range and not at a specific value. The y+ indicates that the near-wall nodes are within the laminar sublayer. From the boundary, the grid is stretched, which is a good compromise between the mesh points and the computational time. Figures 5.6 to figure 5.8 compare the axial and tangential velocity profiles from the three meshes in three different cross sections at 50% axial chord downstream the tip leading edge, at the trailing edge and at 40% axial chord downstream the trailing edge of the impeller. The velocity profiles were chosen in these sections because of the high turbulence and the last section include also the mixing plane of the rotor and the stator. Only minor variations are observed between the solutions obtained with the second (mesh 2) and the finer grids (mesh 1). However, it was determined that the second mesh (with the number of points as table 5.2) provided sufficient resolution, all the following results of the performance of the propeller pump are presented using the finer grid solutions The impeller has 470.193 nodes with 57 nodes blade to blade, 73 nodes hub to shroud, and 113 nodes inlet to outlet of the block. The stator model has a total of 271.313 nodes with 49 nodes blade to blade, 49 nodes hub to shroud, and 113 nodes inlet to outlet of the block. The number of grid levels is (3/3/4). The pitch-averaged grid used for the study of the pump performance is shown in figure 5.9.
Figure 5.6: Study of grid sensibility of axial and tangential pitch-averaged velocity at midchord of impeller.)
5.11
Figure 5.7: Study of grid sensibility of axial and tangential pitch-averaged velocity at trailing edge of impeller.
Figure 5.8: Study of grid sensibility of axial and tangential pitch-averaged velocity at 20% axially chord downstream of trailing edge impeller.)
Figure 5.9: Pitch-averaged grid used for the study of pump performance.
5.12
Chapter 5: Numerical Simulations of the Axial Pump Flow Another grid was made with the adjustment of the shroud (with the integration of the glass, see figure 3.4). In the study of the flow field, the numbers of points in the impeller and gap were increased. In order to have a better quality regarding the cells orthogonality in the tip, the butterfly mesh topology in the tip gap was applied. Table 5.4 shows the main characteristics of the mesh used for the flow yield comparison. The multigrid level is also (3/3/3).
Mesh 4:
Block number 1 2 3 4 5
domain
Streamwise direction 89 25 89 17 49
Spanwise direction 89 25 89 25 49
Azimuthal direction 121 161 65 65 113 Total:
Total number of points 958.441 100.625 515.865 27.625 271.313 1.873.869
Number of blades 3 3 3 3 8
Rotor Gap Bulb Gap2 Stator
Table 5.4: Characteristics of the mesh used for the velocity field validation (mesh 4).
The pitch-averaged grid used for the study of the flow field in the impeller is shown in figure 5.10.
Figure 5.10: Pitch-averaged grid used for the study of flow field in impeller
5.5.4 Physical Modelling

The turbulence models of Baldwin-Lomax and Spalart-Allmaras are used for all simulations. A detailed explanation of these turbulence models was given in the
5.13
Chapter 5: Numerical Simulations of the Axial Pump Flow previous paragraph. The Reynolds number, based on the tip speed of impeller at the leading edge and its diameter, is 2,5.106.
5.5.5 Initial Solution

The initial solution is defined by constant values. The physical values are uniformly used as initial solution over all the blocks, except on the boundaries. The static pressure is the atmospheric pressure, the static temperature is 273 K and only the averaged-axial velocity at the inlet of the propeller pump is specified. For the simulation with the Spalart-Allmaras model, the turbulence kinetic energy is specified using equation 5.31 (with Tu = 5% and Uref = 3,30 m/sec).
5.6 Discussion of Results and Numerical Flow Field

In this paragraph, the numerical flow field, calculated with both turbulence model of Baldwin-Lomax and Spalart-Allmaras will be discussed. All numerical velocities are normalized with the tip speed (Vtip = 13.9 m/s). Each solution for all grids show iterative convergence by displaying at least 3 orders of magnitude drops in residuals. The global residuals in the code (FINE) are defined as a flux balance (the sum of the fluxes on all the faces of each cell). For the mass flows, the difference between the inlet flow and the outlet flow is less than 0.5 % at an average of about 4000 time steps. As described in the boundary conditions the inflow condition is axially imposed. When the flow rate change with a constant rotating speed of the impeller, than the incidence of the flow at the leading edge of the impeller will change too. This incidence has influence for the performance of the pump. The selection of the optimal incidence is a function of blade camber, blade shape, space-chord ratio, stagger angle, Reynolds number, solidity and it is described, for examples, by the correlation coefficients of Howell and Lieblein (Dixon (1975) and Japikse, (1976)). The stagger angle of the impeller blades is high and it is not constant over the span of the leading edge, so that the incidence at the leading edge of the impeller is not the same over the full span. Figures 5.11, 5.12 and 5.13 plot the relative vector field at the leading edge of the impeller near the hub, midspan and tip for 4 different flow rates (75 %Qd, 90 %Qd, Qd and 110 %Qd). Comparing the incidences near the shroud (see figure 5.11) and midspan (see figure 5.12), there are no different between them, but near the hub (see figure 5.13), where the blade angle and the thickness of the blade are different, there is a complete different image. The incidence is very high and a recirculation region at the pressure side exits near the leading edge of the impeller. This region increase with decreasing flow rate. The stagnation point near the leading edge of the impeller moves more downstream with increasing the flow rate. Picture 5.14 and picture 5.15 show the streamlines of the pressure side of the impeller for different flow rates. The surface flow visualization patterns over the pressure surface show a very two-dimensional flow. However, one very significant pattern is show up on the pressure surface. A radial outward flow exists over the top 8 percent of the span. This radial flow initiates the flow from the pressure side to the suction side through the tip leakage. Figure 5.15 shows that the stagnation line evolutes more downstream with decreasing the flow rate (and thus increasing incidence). The stagnation line is the line where the relative velocity is zero. Looking more in detail
5.14
Chapter 5: Numerical Simulations of the Axial Pump Flow on the picture 5.15, which presents the streamlines at the pressure side for a flow rate of 75% Qd and 90% Qd, at the feet of the leading edge a small reverse flow is absent, while this is not visible in the pictures of the higher flow rates.
Figure 5.11: Flow direction at the leading edge at tip (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
The numerical calculations were done for a small working regime of the pump. If the incidence will increase, stall will happen. Impeller blade stall is directly equivalent to simple airfoil stall. If the flow rate is reduced while maintaining the machine at constant speed, then the inlet angle will increase in magnitude, thus giving an increase in incidence angle. As the incidence continues increase, the flow is required to accelerate locally more and more around the leading edge. After a critical incidence, the flow starts separating from the surface and thus creating a stall condition. Stall is a manifestation of boundary-layer separation, due to the adverse pressure gradient and viscous shear stresses. When the viscous shear effect and adverse pressure gradients are sufficient to reduce the fluid velocity to zero, it is forced to deviate from the surface and the flow is said to have stall.
5.15
Figure 5.12: Flow direction at the leading edge at midspan (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
Figure 5.13: Flow direction at the leading edge at zero span (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
5.16
Figure 5.14: Streamlines of the pressure side of the impeller (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
a) Flow rate = 75%Qd
b) Flow rate = 90%Qd
Figure 5.15: Detail plot of the streamlines of the pressure side of the impeller
Flow deviation is also of paramount importance. Under normal conditions, the primary flow leaving a bladed passage will always deviate from the pressure towards the suction side as the forces on each side of the blade come into balance. An instantaneous balancing of pressure at a trailing edge occurs only in an idealized potential flow, but does not happen with real blades in real fluids. In fact the details of a trailing edge flow permit some pressure imbalance as the streamlines bend around a (slightly) blunt trailing edge. Missing the flow angle by a few degrees can imply missing the head requirement and the flow requirement by 2%, 5% or perhaps even more. The deviation is a function of blade camber, blade shape, space-chord ratio and stagger angle. It will be observed by our experiment that the fluid receives its maximum guidance on the pressure side and that this diminishes almost linearly
5.17
Chapter 5: Numerical Simulations of the Axial Pump Flow towards the suction side of the channel. It will be interesting to look to all the deviation of all flow rates. Figures 5.16, 5.17 and 5.18 show the flow direction at the trailing edge at zero span, midspan and tip.
Figure 5.16: Flow direction at the leading edge at tip (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
The pictures show that the flow will follow the blade surface for all flow rates. But the flow angle will be different at the trailing edge of the blade, because the adverse pressure gradient of the trailing edge disappears suddenly. At the hub near the trailing edge of the impeller, another recirculation region at the suction side is shown (see figure 5.18). The structure of the recirculation region is less depending of the flow rate than the circulation region at the leading edge. The two-dimensional flow field on the suction side of the impeller is more complicated and will be discussed in chapter 8 (3D flow structure) with the aid of topological roles. Figure 5.19 plots the pitchwise-averaged relative flow angle at the leading and trailing edge in function of the flow rate.
5.18
Figure 5.17: Flow direction at the leading edge at midspan (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
Figure 5.18: Flow direction at the leading edge at zero span (left up: Flow rate = 75% Qd, up right: Flow rate = 90% Qd, left down: Flow rate = Qd, right down: Flow rate = 110% Qd)
5.19
Figure 5.19: Influence of the relative velocity angle at leading and trailing edge in function of flow rate.
The incidence of the flow increases with decreasing the flow rate. The blade angles at the hub and the tip of the leading edge of the impeller are 55.24 and 76.83 respectively. The relative velocity angles are equal at the blade angles at design flow. The blade angles at the hub and the tip of the trailing edge of the impeller are 40.79 and 70.57 respectively. The different of the flow angles at the trailing edge are small for all flow rate. This means that the flow is guided well by the blade row. A big different at the low flow rate happened in the boundary near the hub. This difference is due to the recirculation region near the hub at the suction side of the trailing edge. As explained before in the literature review of the axial pumps, the energy from the impeller is converted into pressure by applying the angular momentum to the constant mass flow of liquid going through the impeller blade. The increasing of the static pressure and the total pressure in the rotor passage is very clear in the meriodinal view of the pitch-averaged pictures 5.21 and 5.22. These figures describe the energetic transmission that happens in the passage row. Upstream the tip rotor leading edge, the static and total pressure is constant until the bulb of the pump. Through the concave curvature of the casing and the strong convex curvature of the hub, the cross section narrows and the axial velocity (see figure 5.23) increase with a decreasing the static pressure on the same streamline, because no energy is added so that the total pressure is constant. The axial velocity, that is uniform and axial imposed at the inlet of the pump, will first accelerate near the casing and later on near the hub. Thereby the static pressure will decrease with the radial direction in the same cross section. This radial distribution of the pressure to the outlet of the impeller can be explained by the radial equilibrium equation (equation 1.8). This basic equation of motion in the radial direction is derived from the balance of pressure and inertial forces on a fluid element for an axisymmetric flow ( = 0 ). In an axial pump, the radial component of the velocity is small with regards to vt2 , so that the two first elements can be neglected. This means that the static pressure near the tip will be always higher near the hub because the third element of the left side is always positive. Remarkably is that the turbulence model of Baldwin-Lomax estimates a higher static and total pressure at the tip trailing edge of the impeller than the model of Spalart-Allmaras at low flow rates. This means that Baldwin-Lomax estimates fewer losses in the tip region than the model of Spalart-Allmaras. Looking more in detail to the evolution of the pitch-averaged axial velocity of figure 5.23, at the inlet of the domain, the velocity field is uniform and axial imposed. When Experimental and Numerical Study of an Axial Flow Pump 5.20
Chapter 5: Numerical Simulations of the Axial Pump Flow the radius of the casing decreases, the axial velocity will increase too because of the constant mass flow. This effect, together with the curvature of the hub and casing explains the distribution of the axial velocity before entering the impeller. In the impeller, the flow continues accelerating because the blade thickness of the impeller increases before midchord and will decrease downstream. At the beginning of the bulb of the axial flow pump, there is a very low axial velocity component. Hereby, a separation due to the strong curvature of the hub will take place and the static pressure is very high, to compensate these low speed. While the static pressure increases in the radial direction to the casing due to the concave curvature of the hub, the velocity will increase. The same reasoning is applicable to the stator. The axial velocity in the stator will first increase and it will decrease downstream because the thickness of the blade is higher at the leading that at the trailing edge. A reverse flow is visible in the tip region for all flow rates. Figure 5.26 shows detailed meriodinal plots of the pitch-averaged axial flow in the tip region of picture 5.23. The streamlines in the tip region indicate the reverse flow in the gap of the impeller for all flow rates. Reducing the flow rate will increasing the reverse flow and moving upstream to the tip leading edge. The turbulence model of Baldwin-Lomax predicts a bigger region of reverse flow than the turbulent model of Spalart-Allmaras. For higher flow rate (110 Qd), the turbulence model of Spalart-Allmaras doesnt predict reverse flow more. The numerical results revealed that downstream of the discharge plane of the impeller blades, the flow (on the casing) moved downstream along with a tangential component is the direction of rotation. Analyse of the numerical vector plots notices that on the suction side of the blade near the tip, there exists a chordwise location where the flow reversed direction and started moving upstream near the tip. The reversed flow remains attached to the blade and shroud until it interacts with the fluid from upstream, when it leaves the blade and flow upstream along the interior of the shroud. Upon existing the shroud, the fluid mixes with the leakage flow interior to the shroud and forms part of the upstream swirling backflow as soon in figure 7.16. The reversal point on the suction side of the blade depends on the extend to which backflow has developed and moves closer to the leading edge with decreasing rate and increasing backflow. The flow downstream (chordwise) of the reversal point continues to flow downstream. There is no other separation of flow on the blade, except near the hub. The leading edge flow remains attached to the blade surface under all conditions and has a large radial component. The flow on the pressure side also exhibits a radial component of velocity that becomes small close to the shroud and is smaller than that on the suction side. Close to the leading edge the flow has a strong radial component of velocity. The upstream backflow consists of two components; (a) the discharge-to-suction leakage flow and (b) separation of flow on the suction surface of the blade. While the leakage flow may be attributed to the adverse pressure gradient exterior to the shroud, the separation at the blade is not only due to the angle of attack at off-design flows. The cause for the reversal flow that occurs in a region close to the rotating shroud and is influenced by the shroud boundary layer. Shear forces due to the rotating shroud boundary layer act in a direction in opposite to that of the flow in the presence of an adverse pressure gradient. When these shear forces combined with the adverse pressure gradient become larger than the momentum of the fluid (from the shear pumping effect), the flow reverses direction. The flow on the shroud boundary layer within the blade passage area also reverses direction and finally moves upstream.
5.21
Chapter 5: Numerical Simulations of the Axial Pump Flow Observations on off-design flows in non-cavitating axial flow inducers found that the leakage flows were the cause of the upstream swirling backflow in unschrouded inducers (Abhijit et all, 1993). As the flow rate is reduced, the discharge pressure over the tip rises, until at a certain flow rate the adverse pressure gradient causes the low energy radial flow from blade passage to reverse direction in the tip clearance region. The backflow in the impeller leads to a drop in efficiency and a high power requirement. In the meriodinal view of the pitch-averaged contour plots of the axial velocity (figure 5.23), the boundary layer on the hub at the leading edge of the impeller is small because the concave surface, but it grow quick at midchord because the convex surface at the trailing edge of the impeller. The boundary layer grows fast because the negative pressure gradient at the hub. For high flow rate, the turbulence model of Spalart-Allmaras predicts back flow in that area. Picture 5.20 is a zoom in the figure of the pitch-averaged streamline at the trailing edge of the rotor near the hub for a flow rate of 75%Qd. The flow reverses downstream on the hub and it re-enters the blade passage region as seen by the streamlines. The recirculation is very clear by two vortices near the trailing edge of the rotor, and before the leading edge of the stator, due to the negative pressure gradient. This re-entrant flow continues on the hub until that trailing edge of the impeller where it meets the fluid upstream. This interaction causes the hub boundary layer flow to separate in the blade passage area. The reduction of the axial velocity on the hub at the blade trailing edge together with the strong radial component could create an adverse pressure gradient at the discharge end causing the flow to reverse on the hub and re-enter the blade passage area. The reentrant flow causes the flow coming from upstream along the hub to separate. This interaction cause the vortices on the hub. The reason of the low velocity area after the trailing edge can be due to the bad design of the hub, which has a lot of curvature deformations in the axial direction. Such deformations have to be prevented because of the losses.
Figure 5.20: Zoom in figure of the pitch-averaged streamline at the trailing edge of the rotor near the hub for a flow rate of 75%Qd, calculated with the turbulence model of Spalart-Allmaras.
5.22
Chapter 5: Numerical Simulations of the Axial Pump Flow The meridional views of the pitch-averaged tangential velocities show in figure 5.24 that the tangential velocity in the impeller increases during the passage in the impeller. For low flow rate, higher tangential velocities at the tip are reached than at high flow rates. The rotor presses a pre-swirl flow in the inlet pipe of the propeller pump, which has the same direction of the rotation of the impeller. At the symmetrical axis of the axial flow pump, the tangential velocity has a contra-rotation direction. Towards the trailing edge of the stator the velocity profile become uniform over the spanwise direction. Figure 5.25 shows that the flow at the entrance of the impeller has a strong radial inflow because the convex hub and the concave casing. The radial velocities at the shroud can reach values of around 50 percent of the axial velocity. At the hub near the leading edge of the impeller, the value of the radial velocity component is also very high. Because of the thickness of the leading edge blade near the hub, the flow following the hub will move more inwards the rotor. This effect will be stronger with increasing the flow rate. At the trailing edge of the impeller, the flow follows the concave hub and it explains the negative value of the radial flow component. The larges values of radial velocity confirm the highly three-dimensional characteristics of the impeller and emphasize the necessity for a three-dimensional theory for accurate flow analysis. Figure 5.27 and 5.28 are plots of the vector field of different cross sections in the inlet pipe, 25 % and 75 % chord axially downstream of the tip leading edge of the rotor and at midchord of the stator. The pre-swirl flow coming from the inlet pipe continues turning at the entrance of the impeller and disappears during the passage in the impeller towards the stator. The tangential velocity profile is smoothed out and will have the same rotation as the impeller towards the stator.
5.23
a) Q = 75% Qd (Baldwin-Lomax)
b) Q = 75% Qd (Spalart-Allmaras)
c) Q = Qd (Baldwin-Lomax)
d) Q = Qd (Spalart-Allmaras)
e) Q = 110% Qd (Baldwin-Lomax)
f) Q = 110% Qd (Spalart-Allmaras)
Figure 5.21: Pitch-averaged static pressure, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.24
Q c) c) Q= Qd (Baldwin-Lomax) d) Q = Qd (Spalart-Allmaras)
Figure 5.22: Pitch-averaged total pressure, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.25
a)
Q = 75% Qd (Baldwin-Lomax)
b) Q = Qd (Baldwin-Lomax)
Figure 5.23: Pitch-averaged axial velocity field, calculated with the turbulence models of BaldwinLomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.26
a)
c)
Q = Qd (Baldwin-Lomax)
d) Q = 110% Qd (Baldwin-Lomax)
Figure 5.24: Pitch-averaged tangential velocity field, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.27
Figure 5.25: Pitch-averaged radial velocity field, calculated with the turbulence models of BaldwinLomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.28
a)
e)
Figure 5.26: Pitch-averaged axial velocity field, detailed in the tip region, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.29
a) 100 % pipe diameter upstream of the rotor leading edge
b) 25 % chord axially downstream of rotor tip leading edge Figure 5.27: Vector field of different cross sections, calculated with the turbulence models of SpalartAllmaras for flow rate = Qd
5.30
c) 75 % chord axially downstream of rotor tip leading edge
d) midchord of the stator Figure 5.28: Vector field of different cross sections, calculated with the turbulence models of SpalartAllmaras for flow rate = Qd (continue)
5.31
Figure 5.29: Static pressure field at midspan, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.32
a) Baldwin-Lomax (75%Qd)
b) Spalart-Allmaras (75%Qd)
c) Baldwin-Lomax (Qd)
d) Spalart-Allmaras (Qd)
e) Baldwin-Lomax (110%Qd)
f) Spalart-Allmaras (110%Qd)
Figure 5.30: Static pressure field at midspan, 90 % span and 95 % span of the impeller, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.33
a)
Figure 5.31: Total pressure field at midspan, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.34
a)
Figure 5.32: Axial velocity field at midspan, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.35
a)
Figure 5.33: Tangential velocity field at midspan, calculated with the turbulence models of BaldwinLomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.36
a)
Figure 5.34: Radial velocity field at midspan, calculated with the turbulence models of Baldwin-Lomax (left) and Spalart-Allmaras (right) for 75%Qd, Qd and 110%Qd
5.37
Chapter 5: Numerical Simulations of the Axial Pump Flow Figure 5.29 represents the static pressure at midspan of the impeller and the stator. The impeller rotates from up to down. Figure 5.30 plots the static pressure distribution on the blades of the impeller as a function of the flow rate for several radii (midspan, 90 % span and 95 % span). At flow rates below the expected design point, it can be seen that there are large pressure differences at the leading edges of the blades corresponding to the positive angle of attack. Above the expected design flow rate the angle of attack is negative and the resultant pressure distributions are crossed over. The static pressure on the suction side increases from the leading edge to the trailing edge. The difference of pressure at the suction side is higher with decreasing the flow rate. At the pressure side of the impeller, the pressure is constant for all flow rates and span positions, except near the leading edge, where the pressure is very high because the high flow incidence. At the suction side of the leading edge, it creates a very low pressure. At the trailing edge, the static pressure at the pressure side and the suction side decrease due to the acceleration of the flow. No big different between the different turbulence models are notified. Figure 5.31 represents the total pressure at midspan of the impeller and the stator for 75% Qd, Qd and 110% Qd, calculated with both turbulence models. The total pressure increases from the leading edge to the trailing edge of the impeller, and also increases from the suction side to the pressure side at the leading edge in a cross section of the impeller. However the velocity is higher at the suction side of the impeller (see figure 5.32), the total pressure is lower due to the lower static pressure at the suction side of the blade. Towards the trailing edge, the total pressure becomes more uniform in the cross section. Looking to the figures of the axial velocity at midspan (see figure 5.32), The axial velocity is higher at the suction side than at the pressure side. A minimum value of the axial velocity is observed at the suction side near the leading edge, which will increase towards the trailing edge. At the pressure side, the axial velocity has a higher value near the leading edge of the impeller, which will decrease towards the trailing edge. The high value at the leading edge is due to the high flow incidence. Analogue remarks can be made for the stator. For low flow rate, different velocity profiles are observed for the different turbulence models in the rotor as in the stator. The turbulence model of Baldwin-Lomax predicts higher velocity in the suction side of the impeller and in the pressure side of the stator. Even in the suction side of the stator predicts the turbulence model of Baldwin-Lomax lower velocity profiles than the Spalart-Allmaras model. For the higher flow rates, no noticeable different in the axial velocity profiles are observed in the plots. Figure 5.33 represents the normalized tangential velocity at midspan in the impeller and the stator. A secondary flow (passage vortex) enters the impeller, and the flow leaves the impeller with a constant tangential velocity. The passage vortex moves form the suction side to the pressure side at the tip and return at the hub to the suction side. At the leading edge of the suction side, a local high tangential velocity is due to the high incident of the flow. These velocity component increases with decreasing the flow rate. The tangential velocity at the leading edge of the stator is constant and will decrease when leaving the stator. Figure 5.34 shows the normalized radial velocity at midspan calculated with the turbulence models of Spalart-Allmaras and Baldwin-Lomax. At the leading edge of the impeller, the flow is radial inwards to the hub at the suction side and upwards at
5.38
Chapter 5: Numerical Simulations of the Axial Pump Flow the pressure side. The radial velocity increases towards the trailing edge. High values are reach at the trailing edge of the impeller. The evolutions of the pitch-averaged meridional velocities are shown in figure 5.33. These plots represent the evolution of the pitch-averaged meridional velocities in 4 different cross sections (at tip leading edge, 50 % chord downstream of leading edge, at trailing edge and 50 % downstream the trailing edge). The flow enters the blade passage and accelerates due to the narrower cross section in the impeller blade passage. The flow accelerates with a velocity peak at the hub, because the blade near the hub is thicker than in the tip, and also because the convex curvature of the hub consolidates these effects
a) flow rate = 75% Qd
b) flow rate = 90% Qd
c) flow rate = Qd
d) flow rate = 110% Qd
Figure 5.35: Evolution of the pitch-averaged meridional component of velocity in different cross section for different flow rates
The flow distribution before the tip leading edge is the same for all the flow rates: the distribution is uniform until the midspan to accelerate to the tip due to the convex curvature of the casing, while the velocity is first reduced in the concave part and later to accelerate in the convex part. The difference in velocity outside the boundary near the hub and tip is small. In the next station (at 50 % chord downstream tip leading edge rotor), the effects are opposite: the velocities are higher but with the peak at the side of the hub due to the curvature of the hub and due to the smaller blade passage near the hub than near the tip, which increase the speed. This effect consolidates with increasing flow rate. The distribution is quite uniform and the difference in velocity at
5.39
Chapter 5: Numerical Simulations of the Axial Pump Flow the hub and tip increases with the flow rate. The plot of the cross section at the trailing edge of the impeller (see figure 5.32) clears very well the descent of the meridional velocity component near the hub. This descent is faster for lower flow rates.
a) flow rate = 75% Qd
b) flow rate = 90% Qd
c) flow rate = Qd
d) flow rate = 110% Qd
Figure 5.36: Evolution of the pitch-averaged tangential component of velocity in different cross section for different flow rates
Figure 5.36 plots the pitch-averaged tangential velocities in 4 different cross sections (at tip leading edge, 50 % chord downstream of leading edge, at trailing edge and 50 % downstream the trailing edge). In every plot, the distribution in every cross section is uniform in the radial direction with an increasing of the tangential velocity component Vt at the entrance of the impeller and a light increasing downstream the trailing edge of the rotor. The tangential component increases when passing the impeller (and thus also the angular moment). This moment is necessary for converting the rotating energy changes in the axial pump and it is flow rate dependent. Comparing all the plots, the tangential velocity is higher for higher flow rates. The numerical results of the performance of the propeller pump and the flow field will be compared with the experimental data in chapter 7 and further discussed in chapter 8.
5.40
5.7 Discussion of the different flow structure caused by the adoption of window in the casing
In this paragraph, the flow structure in the axial flow pump with the original casing will be compared with the flow structure of the pump that includes the optical window in the shroud. As discussed before in paragraph 5.5.1, the integration of the glass window in the shroud makes the curvature of the shroud smooth, so that the impeller doesnt rotate in a groove of the casing. Another numerical model had to be made for the flow field. Figure 5.37 compares numerical results of the normalized axial and tangential velocity at midchord of the impeller, at the trailing edge and at 20 % chord axially downstream the trailing edge of the impeller. The velocity profiles are calculated with the turbulence model of Spalart-Allmaras. The pictures show that the structure of the flow is changed by the integration of the window. The axial velocity at the tip is lower and the boundary is smaller than in the case with the original casing due to the bigger tip gap and the smooth surface of the casing. The axial velocity magnitude of the reverse flow near the casing is also less high. Rightly, the structure of the axial velocity from the hub to the midspan is not changed. The tangential velocity profile is more smoothed over the complete span. The tip leakage flow is higher because the tip gap is increased.
Figure 5.37: Comparing the normalized pitch-averaged axial and tangential velocity profiles in the axial flow pump with the original casing and the integration of the window.
5.41
Figure 5.38: Comparing the normalized pitch-averaged axial and tangential velocity profiles in the axial flow pump with the original casing and the integration of the window (continue).
5.8 Summary and Conclusions

The RANS-CFD simulations are presented for the propeller pump. A symmetrical one rotor and stator passage of the 3 impeller and 8 stator blades was meshed. A part of the inlet pipe is modelled to insure the full turbulence flow at the inlet of the propeller pump. At the inlet, the mass flow is imposed, and a uniform pressure is prescribed at the outlet. At the solid walls a no-slip condition is utilized. Periodicity is specified in the circumferential direction. The turbulence models of Baldwin-Lomax and SpalartAllmaras were used to simulate the flow in the pump. Four different grids were meshed. After the study of the grid sensibility, mesh 1 is retained to simulate the pump performance and it has a total of 1.000.000 node points with approximately 520.000 nodes in the impeller passage and 270.000 nodes in the stator passage. Another mesh (mesh 4) is used to simulate the detailed flow field in the impeller and it has a total of 1.900.000 node points with approximately 1.086.700 nodes in the impeller passage and 270.000 nodes in the rotor passage. Table 5.5 resumes for every simulation case, the kind of mesh that is used with their configuration. Grid Mesh 1 Mesh 2 Mesh 3 Mesh 4 Configuration Original Original Original Modified Total nr. of points 827.028 578.724 206.616 1.917.165 Simulation Pump performance Flow field
Table 5.5: Summary of the meshes used for the simulations
Before the numerical results will be validated with the measurements, a grid sensibility study was done and the length of the inlet pipe was studied. The length of the inlet pipe has to be at least 4 x the inlet pipe diameter to ensure that a fully velocity profile is develop before the flow enters the impeller. Two effects are responsible for the redistribution of the flow entering the impeller: The change of the cross section (due to the effect of the convex hub and concave casing) and the pre-swirl caused by the rotation of the impeller. Experimental and Numerical Study of an Axial Flow Pump 5.42
Chapter 5: Numerical Simulations of the Axial Pump Flow The pre-swirl disappears during the passage in the impeller. For all flow rates, recirculation regions exit near the hub at the leading of the impeller, due to the high flow incidence, and near the hub at the trailing edge of the impeller due to the high pressure gradient between the suction and pressure side of the impeller. Marked difference between both turbulence models are the reverse flow in the tip region of the impeller. The turbulence model of Baldwin-Lomax predicts more reverse flow than the model of Spalart-Allmaras. These reverse flow increases with decreasing flow. For low flow rate, the Spalart-Allmaras model predicts also a recirculation region between the trailing edge of the impeller and the leading edge of the stator due to the negative pressure gradient near the hub.
5.43

6.1 Introduction
In this section, the numerical results of the turbulence models of Baldwin-Lomax and Spalart-Allmaras of the described code will be compared with the experimental results. The intent is to evaluate the capability of the numerical method to predict the performance of the propeller pump and the detailed wake structure of the impeller.
6.2 Comparing Results
Experimental
Performance
versus
Numerical
In this paragraph, the numerical pump performances of the propeller pump calculated with the turbulence models of Baldwin-Lomax and Spalart-Allmaras, are compared with the experimental measurements. Different flow rates are imposed at the inlet of the domain, varying from 230 l/s (surging point) until 330 l/s with intervals of 20 l/s, including the design point (302 l/s). The rotation speed for all flow rates is 755 rpm. The results are plotted in non-dimensional parameters as defined in equations 4.4 to 4.7. The next figures represent the numerical versus the experimental performance. Figure 6.1 compares the numerical head coefficient with the experimental data in function of the flow coefficient. Figure 6.2 represents the numerical and experimental power coefficient in function of the flow coefficient. In figure 6.3, the estimated efficiency is compared with the experimental efficiency. The calculation of the shaft power is given by equation 2.4 and the efficiency by equation 2.5. In the graphics, the uncertainties of the quantities, as defined in table 2.7, are added to the measurement points. In the global performance of the axial flow pump, the numerical working point of mesh 4 is also plotted in the graphics as Baldwin Lomax Redesign and Spalart Allmaras Redesign. The head and power coefficients are lower than in the original casing because the tip gaps at the leading and the trailing edge of the impeller are longer than in the original one. Hereby the numerical efficiency of the propeller pump is also lower because more losses are produced in the tip gap.
6.1
Figure 6.1: Comparison performance data: Flow-Head Coefficient
6.2
The measured performance of the propeller pump shows very good agreement with the design data, which is Q= 302 l/s, H = 1,24 m, P = 4,4 kW for a rotating speed of 755 rpm. Comparing both turbulence models with the experimental data, the models estimate the experimental performance well. The more the mass flow differs from the design point, the more the estimated performance parameters differ from the experimental data. The turbulence model of Spalart-Allmaras estimates quite well the experimental data. Table 6.1 compares the proportional difference in the numerical averaged total pressure (Ptot) increase over the impeller and decrease of the stator for the two turbulence models and for different flow rates. Flow rates 90 % Qd Qd 105 % Qd 110 % Qd % Diff. Ptot Impeller 4.0 4.6 5.2 6.3 % Diff. Ptot Stator 8.5 32.7 33.4 33.7
Table 6.1: Proportional Difference of the total pressure of both turbulence models for different flow rates.
6.3
Chapter 6: Evaluation-Comparison of test data and CFD This table shows that the proportional differences in the numerical averaged total pressure increase over the impeller are minimal compared to the total pressure drop over the stator. The differences of both turbulence models are also shown in the bladeto-blade total pressure profile at 20 % chord axially downstream the trailing edge of the impeller and 60 % and 90 % span (see figure 6.4).
Figure 6.4: Absolute total pressure at 20 % chord axially downstream rotor (90% span and midspan) for design flow (Qd) rate and Q = 110% Qd.
The difference of the numerical pressure drop over the stator is much higher than at the rotor and it is also visible in the blade-to-blade profiles at 5 % chord axially downstream the stator at 60% and 90 % span (see figure 6.5).
6.4
Figure 6.5: Absolute total pressure at 5 % chord axially downstream stator (90% span and midspan) for design flow (Q = Qd) rate and Q = 110% Qd.
Figures 6.6, 6.7, 6.8 and 6.9 represent the velocity components and the static pressure in the same axial and radial positions as in plot 6.5.
Figure 6.6: Axial velocity profile at 5 % chord axially downstream stator (90% span and midspan) for design flow (Q = Qd) rate and Q = 110% Qd.
6.5
Figure 6.7: Tangential velocity profile at 5 % chord axially downstream stator (90% span and midspan) for design flow (Q = Qd) rate and Q = 110% Qd.
Figure 6.8: Radial velocity profile at 5 % chord axially downstream stator (90% span and midspan) for design flow (Q = Qd) rate and Q = 110% Qd.
6.6
Figure 6.9: The static pressure profile at 5 % chord axially downstream stator (90% span and midspan) for design flow (Q = Qd) rate and Q = 110% Qd.
The differences of the predicted velocity components are small, except for the axial velocity component. Baldwin-Lomax turbulence model predicts higher values of the axial velocity at the suction side of the stator, especially near midspan, where the difference of the flow angle and the blade angle at the leading edge of the stator is higher than at the tip and the hub (see figure 6.10). At midspan, the flow structure deviates more from the inviscid flow structure (potential flow) so that viscous effects are more important. The total pressure predicted with the Baldwin-Lomax turbulence model is higher than the one predicted with Spalart-Allmaras. Comparisons of the numerical efficiencies of both turbulence models with the experimental one show that the turbulence model of Baldwin-Lomax predicts fewer losses than Spalart-Allmaras. The higher predicted axial velocity with the turbulence model of Baldwin-Lomax is due to the lack of the viscous damping.
Figure 6.10: Vector field at the leading edge of the stator (left at full span, right at midspan)
6.7
Chapter 6: Evaluation-Comparison of test data and CFD The Baldwin-Lomax turbulence model predicts higher head and efficiency due to the lack of viscous damping effect. The discrepancy of the predicted and measured pressure ratio and efficiency can be explained by the empirical inputs, which are still necessary to represent the turbulent nature of the flow. Amongst the most significant source of inaccuracy in these is the criterion for laminar-turbulent transition. Transition in turbomachinery is still a subject of very active research and discussion, for example, Mayle (1991) and Walker (1992). There are special problems associated with transition in turbomachinery: the high levels of freestream turbulence and disturbance, and the strong non-dimensional pressure gradients and Reynolds number based on chord. The basic structure is easily described. A strong adverse gradient can cause a laminar boundary to separate, but once separated the shear layer is highly unstable and the flow can rapidly undergo transition. The turbulence entrains more fluid than the laminar flow and as a result the turbulent shear layer bends towards the solid wall causing the flow to reattach as a turbulent boundary layer. The Baldwin-Lomax however, predicts much larger vortices due to the lack of viscous damping effect. The present viscous solution is slightly underestimating the size of the vortexes. Nevertheless, the general performance was simulated well by the present prediction method.
6.3 Comparing Experimental versus Numerical Flow Field

The measurements of the velocity field reported here are for the flow rate = 0.0886 and head = 0.015970, as indicated in the H-Q curve of figure 6.1. In this section, the velocity measurements are compared with the numerical results of the two turbulence models for the nine selected sections axially distributed inside and outside the rotor and for five chosen hub to tip spanwise radii (10%, 20%, 40%, 60%, 80% and 90% span). The velocities were non-dimensionalized by dividing them with the blade tip speed and the pitchwise coordinates are normalized with local blade spacing (the lower angle is the suction surface and the higher angle is the pressure side). The time averaging is performed in the relative frame of reference, rotating with the rotor blades. These averages are carried out to remove any turbulent part in the velocity signal. In the contour plots, the rotor turns counter clock-wise. The right side is the pressure side and the left side is the suction side of the blade. Due to the blade shadow, the pitch-averaged of the measured velocities is not always available for the complete blade-to-blade profile, because of the missing information of the quantities in the blade shadow. Only in surveys A and B, 40% and 20 % upstream of the tip leading edge of the rotor, it was possible to pitch-averaged the quantities of the complete revolution. In the other surveys, the pitch-averaging of the experimental data is also done, but keep in mind that the points of the blade shadow are not accounted (with the numerical pitch averaging, all points are accounted). In the pictures of the distribution over a blade-to-blade azimuth of the averaged axial and tangential velocities in a rotating frame reference, upstream, inside and immediately downstream of the rotor row are compared with the numerical models for different blade spanwises H. The blade spanwise distance is defined between the hub and the casing: Experimental and Numerical Study of an Axial Flow Pump 6.8

r rh rt rh
H=
6.1
where rh is the radius of the hub, rt is the radius of the casing and r is the radius of the blade-to-blade sections. The pitchwise positions of each graph indicate the distance from the suction to the pressure side i.e. 10 % pitch is near the suction side, while 90 % pitch is near the pressure side. Five different radial positions are compared with the numerical turbulence models and plotted in the next pictures.
6.9
6.3.1 Contours Plot of Normalized Velocities at 40% Chord Axially Upstream of Rotor Tip Leading Edge
a) Experimental
b) With turbulent model of Baldwin-Lomax
c) With turbulent model of Spalart-Allmaras Figure 6.11: Contour plots of the normalized axial (left) and tangential (right) velocities at 40% chord axially upstream of rotor tip leading edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.10
6.3.2 Contours Plot of Normalized Velocities at 20% Chord Axially Upstream of Rotor Tip Leading Edge
a) Experimental
c) With turbulent model of Spalart-Allmaras Figure 6.12: Contour plots of the normalized axial (left) and tangential (right) velocities at 20% chord axially upstream of rotor tip leading edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.11
6.3.3 Contours Plot of Normalized Velocities at Rotor Tip Leading Edge
a) Experimental
c) With turbulent model of Spalart-Allmaras Figure 6.13: Contour plots of the normalized axial (left) and tangential (right) velocities at rotor tip leading edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.12
6.3.4 Contours Plot of Normalized Velocities at 25% Chord Axially Downstream of Rotor Tip Leading Edge
a) Experimental
c) With turbulent model of Spalart-Allmaras Figure 6.14: Contour plots of the normalized axial (left) and tangential (right) velocities at 25% chord axially downstream of rotor tip leading edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.13
a) Experimental
6.14
a) Experimental
6.15
6.3.7 Contours Plot of Normalized Velocities at Rotor Trailing Edge
a) Experimental
c) With turbulent model of Spalart-Allmaras Figure 6.17: Contour plots of the normalized axial (left) and tangential (right) velocities at rotor tip trailing edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.16
6.3.8 Contours Plot of Normalized Velocities at 20% Chord Axially Downstream of Rotor Tip Trailing Edge
a) Experimental
c) With turbulent model of Spalart-Allmaras Figure 6.18: Contour plots of the normalized axial (left) and tangential (right) velocities at 20% chord axially downstream of tip trailing edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.17
6.3.9 Contours Plot of Normalized Velocities at 40% Chord Axially Downstream of Rotor Tip Trailing Edge
a) Experimental
c) With turbulent model of Spalart-Allmaras Figure 6.19: Contour plots of the normalized axial (left) and tangential (right) velocities at 40% chord axially downstream of tip trailing edge (Experimental, turbulence model of Baldwin-Lomax and Spalart-Allmaras).
6.18
6.3.10 Pitch-averaged Velocity Profiles of the Measured Locations
Figure 6.20 Pitch-averaged velocity profiles of the measured locations (left normalized axial pitch averaged velocity, right: normalized tangential pitch averaged velocity)
6.19
Figure 6.21: Pitch-averaged velocity profiles of the measured locations (left normalized axial pitch averaged velocity, right: normalized tangential pitch averaged velocity)
6.20
Figure 6.22 Pitch-averaged velocity profiles of the measured locations (left normalized axial pitch averaged velocity, right: normalized tangential pitch averaged velocity)
6.21
6.3.11 Blade-to-blade Axial Velocity at 40 % Chord Axially Upstream of the Tip Rotor Leading Edge
Figure 6.23: Blade-to-blade normalized axial velocity at 40 % chord axially upstream of the tip rotor leading edge
6.22
6.3.12 Blade-to-blade Axial Velocity at 20 % Chord Axially Upstream of the Tip Rotor Leading Edge
6.23
6.3.13 Blade-to-blade Axial Velocity at Rotor Leading Edge
Figure 6.25: Blade-to-blade normalized axial velocity at rotor leading edge
6.24
6.3.14 Blade-to-blade Axial Velocity at 25 % Chord Axially Downstream of Rotor Leading Edge
Figure 6.26: Blade-to-blade normalized axial velocity at 25 % chord axially downstream of the tip rotor leading edge
6.25
6.26
6.27
6.3.17 Blade-to-blade Axial Velocity at Rotor Trailing Edge
Figure 6.29: Blade-to-blade normalized axial velocity at tip rotor trailing edge
6.28
6.3.18 Blade-to-blade Axial Velocity at 20 % Chord Axially Downstream of Rotor Trailing Edge
Figure 6.30: Blade-to-blade normalized axial velocity at 20 % chord axially downstream of the tip rotor trailing edge
6.29
6.3.19 Blade-to-blade Axial Velocity at 40 % Chord Axially Downstream of Rotor Trailing Edge
6.30
6.3.20 Blade-to-blade Tangential Velocity at 40 % Chord Axially Upstream of the Tip Rotor Leading Edge
Figure 6.32: Blade-to-blade normalized tangential velocity at 40 % Chord Axially Upstream of the Tip Rotor Leading Edge
6.31
6.3.21 Blade-to-blade Tangential Velocity at 20 % Chord Axially Upstream of the Tip Rotor Leading Edge
Figure 6.33: Blade-to-blade normalized tangential velocity at 20 % Chord Axially Upstream of the Tip Rotor Leading Edge
6.32
6.3.22 Blade-to-blade Tangential Velocity at Rotor Leading Edge
Figure 6.34: Blade-to-blade normalized tangential velocity at rotor leading edge
6.33
6.3.23 Blade-to-blade Tangential Velocity at 25 % Chord Axially Downstream of Rotor Leading Edge
Figure 6.35: Blade-to-blade normalized tangential velocity at 25 % chord axially downstream of the tip rotor leading edge
6.34
6.3.24 Blade-to-blade Tangential Velocity at 50 % Chord Axially Downstream of Rotor Leading Edge
6.35
6.3.25 Blade-to-blade Tangential velocity at 75 % Chord Axially Downstream of Rotor Leading Edge
6.36
6.3.26 Blade-to-blade Tangential Velocity at Rotor Trailing Edge
Figure 6.38: Blade-to-blade normalized tangential velocity at tip rotor trailing edge
6.37
6.3.27 Blade-to-blade Tangential Velocity at 20 % Chord Axially Downstream of Rotor Trailing Edge
Figure 6.39: Blade-to-blade normalized tangential velocity at 20 % chord axially downstream of the tip rotor trailing edge
6.38
6.3.28 Blade-to-blade Tangential Velocity at 40 % Chord Axially Downstream of Rotor Trailing Edge
6.39
6.3.29 Blade-to-blade Velocity in tip (95 % Span)
Figure 6.41: Blade-to-blade normalized axial velocity at tip (95% span)
6.40
Figure 6.42: Blade-to-blade normalized axial velocity at tip (95% span) (continue)
Figure 6.43: Blade-to-blade normalized tangential velocity at tip (95% span)
6.41
Figure 6.44: Blade-to-blade normalized tangential velocity at tip (95% span) (continue)
6.4 Discussion: Numerical versus Experimental Results

The structure of the flow in the impeller, plotted in the above pictures, is already discussed in paragraph 4.6 and 5.6. This paragraph handles only the difference between the numerical and the experimental results. Figures 6.11 to 6.19 compare the experimental contours plot of the axial and tangential velocity components with the numerical plots, which are calculated with the turbulence models of Spalart-Allmaras and Baldwin-Lomax. The numerical contour plots of both turbulence models represent the global experimental flow structure in the impeller of the axial flow pump. Upstream the leading edge of the impeller, the turbulence model of Spalart-Allmaras (SA) estimates the flow structure better than the turbulence model of Baldwin-Lomax (BL). BL shows a difference in flow structure, because it predicts a bigger reversal region near the casing at the inlet of the impeller than the turbulence model of SA, and the tangential component of the swirl of the passage vortex near the casing is higher estimated by BL than the experimental and the velocity component estimated with the turbulence model of SA.
6.42
Chapter 6: Evaluation-Comparison of test data and CFD As discussed in a previous chapter (see paragraph 5.6: Discussion of Results and Numerical Flow Field), a region of the reverse back flow exists upstream the leading edge of the blade and extends towards the midpitch, where the flow goes more downstream. From the midchord of the impeller, the reverse flow region will decrease more downstream. First the reverse flow region will disappear from the pressure side of the blade and later from the suction side. At the trailing edge of the impeller, the reverse flow was completely disappeared. Comparing the thickness of the reversal flow region of both models (the one with the original casing and the one with the smoothed casing), we indicate that the thickness of the reverse flow region increases with the tip gap. This is also the reason why the numerical efficiency of the turbulence models with the smoothed casing is lower than the original casing. In the impeller, the axial velocity is higher at the suction side than at the pressure side. Due to the concave curvature of the hub, the axial velocity is also higher near the hub than at midspan. Upstream the cross section of 75 % chord axially downstream the leading edge, the boundary layer of the blades and the hub are very thin, because the static pressure increases due to the narrow passage in the impeller and the concave curvature of the hub and the blades. Downstream this cross section, the averaged axial velocity will decrease because of the convex curvature of the hub and the casing. Because of the strong pressure gradient at the hub and the suction side of the blade, the boundary layer on the hub and the suction side of the blade will increase further downstream. The boundary layer on the suction side develops faster than at the pressure side. This pressure gradient creates at the hub near the trailing edge of the impeller a reverse flow region (see figure 5.20). This reversal region could not be measured because of the high reflection of the laser light by the hub (high noise in the scattered light) and/or due to the problems in focussing the laser beams far from the tunnel window. In general, the turbulence model of Baldwin-Lomax predicts bigger boundary layers at the hub than the Spalart-Allmaras. The contour plots of the tangential components show that the flow enters the impeller with the vortex passage moving from the suction side to the pressure side along the casing. The vortex disappears during the passage in the impeller, and the flow will have the same rotation as the impeller. At the entrance of the impeller, the tangential velocity distribution is quite constant in the radial direction, except in the boundary layer. All the turbulence models predict very well the tangential velocity structure in the impeller, except near the casing where the BL model estimates a higher tangential component in the boundary layers. In the contours plots of the experimental axial velocity component of the cross sections at 75 % chord axially downstream of the rotor tip leading edge and at the trailing edge, a region with high axial velocity is detected in the tip region at the suction side of the impeller. This region of high axial velocity is not visible in the numerical contour plots. This local acceleration can be created by the discontinuity of the smooth optical glass in the curvature casing. The pitch-averaged axial and tangential velocity profiles are shown in figures 6.20 to 6.22. The equal numerical and measured velocity profiles at the inlet of the impeller indicate that the uniform flow rate at the numerical inlet of the model is fully developed to an inlet pipe flow with the correct boundary layers. Strange is the tangential velocity profile at the entrance of the impeller estimated with the turbulence model of Baldwin-Lomax. Further research of the local grid configuration has to be done in the future. Keeping in mind the average measurement error of 5.5 %, both Experimental and Numerical Study of an Axial Flow Pump 6.43
Chapter 6: Evaluation-Comparison of test data and CFD turbulence models estimated the averaged axial and tangential velocities of the flow very well, however in the tip region, the tangential velocity components of both models are overestimated and the axial velocity component underestimated. Remember that the uncertainties of the averaged experimental quantities increase going to the hub because the quantities of the blade shadow are not included in the averaging process, while with the numerical results the quantities for all points included. Overall the predicted blade-to-blade profiles (see plots 6.23 to 6.44) of the two turbulence models may be compared with the experiments. The interaction between the pressure surface and suction surface boundary layers would result in an extremely complex flow, especially near the outer half of the blade span. In 7 difference spanwise radii, the axial and tangential velocity profiles are compared with the numerical results. The predictions generally agree well with the measurements. Upstream the leading edge of the impeller, both turbulence models underestimate slightly the axial velocity components near the hub and overestimate it near the casing. The experiment profile near the casing is also broader than the numerical one (see figure 6.23). As shown in the contour plots of the axial velocity component (6.11) and the blade-to-blade profile at 95 % span (see figure 6.41), the Baldwin-Lomax model underestimates the axial component and predicts a bigger region of reverse flow upstream the leading edge than the turbulence model of Spalart-Allmaras and the experimental data. The tangential velocity is overestimated in the hub and underestimated at the shroud. The difference between both turbulence models is very small. At the leading edge, the local acceleration on the pressure side of the blade due to the high incidence of the flow is well predicted. While the experimental plots show that the increased axial velocity due to the high incidence is visible until 60 % span, both turbulence models estimated the influence until 20% span. This can be due to the position uncertainty of the measurement point. A small displacement of the measurement point in the upstream direction can change the presented imagine. In general, Baldwin Lomax predicts higher tangential gradient velocity in boundary layer of the suction side of the impeller blade, both at the leading edge and in all cross section downstream in the impeller. The two models predict the experimental tip leakage vortices and the leading edge vortices. The experimental swirls of the vortices are higher than the numerical ones. Both turbulence models underestimate the tangential velocity near the pressure side of the impeller. The tangential velocity has lower values than the predicted ones. In the next section (25 % chord downstream the leading edge), a recirculation region on the pressure side is formed. At the tip leading edge, the flow incidence angle is so high that a flow will accelerate very fast near the pressure side of the blade and decreases towards the midspan. The experimental recirculation region on the pressure side of the blade is found until 40 % span. The numerical results dont represent this recirculation region more. The difference between the axial and the tangential velocity predicted with both turbulence models is small at midspan of the impeller but increases towards the hub and the tip. At the hub, the Baldwin-Lomax model predicts a higher axial and tangential velocity profile than the Spalart-Allmaras model, in the concave as well as in the convex curvature of the hub. A lack of agreement appears clearly in comparing Experimental and Numerical Study of an Axial Flow Pump 6.44
Chapter 6: Evaluation-Comparison of test data and CFD the casing flow. In the tip region, the difference between both models can change quickly from section to section. Generally, the Spalart-Allmaras model predicts higher axial velocity in the tip region and the Baldwin-Lomax model predicts lower tangential components. The estimated boundary layers on pressure side of the blade are predicted smaller than the measured boundary layers. Large differences occur near the tip behind the midchord. It can be seen that at midspan, the calculated data are very comparable to the experimental data, but a large difference occurs close to the tip. A window was inserted into the duct to allow optical access to the rotor flow, as explained in chapter 3. The installation of the window produced a discontinuity in the casing of the impeller because it smoothes the curved casing. This can alter the character of the flow in the tip region, downstream the midchord of the impeller. Downstream the trailing edge, while the numerical tangential velocity distribution of both models show an increasing curve in the wake of the blade at the suction side, the experimental curve increases very smoothly. The tangential velocity profiles of both models are overestimated in the wakes of the impeller. At the suction side of the blade, the real flow will follow better the blade boundary and will have less overturning over the trailing edge than the predicted ones. Even a difference between both numerical axial velocity distributions downstream the trailing edge of the impeller is detected. Quantitatively the Baldwin-Lomax has a lower axial flow in the tip region because of the reverse flow near the casing, as it was already shown in the contour plots, but the distribution profile is similar to the Spalart-Allmaras ones. Analogue for the tangential velocity profiles, the BaldwinLomax estimates higher values in the tip than the Spalart-Allmaras model, especially near the pressure side of the blade. The Baldwin-Lomax estimates a higher tip leakage flow than the Spalart-Allmaras model. Globally the results of the Spalart-Allmaras estimate better the complex three dimensional flow fields in the impeller. The difference of both models may be related to the limitation of the simple algebraic turbulence model of Baldwin-Lomax. This model would underpredict the flow approaching separation. It may than produce a considerable discrepancy in a separation region, especially in a large separation like that of the tip leakage flow and the leakage vortex. It is clear that the algebraic eddyviscosity model is strictly valid only for two dimensional flows with a mild pressure gradient or is suitable for three-dimensional boundary layers with small cross flow. The model is not valid for curvature, rotation or separation and pressure or turbulence driven secondary flows. The algebraic eddy viscosity model is adequate for the prediction of the mean velocity field. The discrepancy between the computed and measured characteristics may be also attributed to the grid resolution around the blades surfaces being insufficient to resolve the turbulent boundary layer developing on them.
6.5 Conclusions
During the data analysis, the experimental data of the pump performance, contours of the axial and tangential velocity field, pitch-averaged velocity profiles and some blade-to-blade velocity profiles are compared with the numerical solutions obtaining by solving the Reynolds-averaged Navier Stokes equations, with the two different turbulence models (Baldwin-Lomax and Spalart-Allmaras). Experimental and Numerical Study of an Axial Flow Pump 6.45
Chapter 6: Evaluation-Comparison of test data and CFD Comparing the numerical pump performance with the experimental data, both models estimate the performance well. The more the mass flow differs from the design point, the more the estimated performance parameters differ from the experimental data. The turbulence model of Spalart-Allmaras estimates quite well the experimental data. The Baldwin-Lomax however, predicts higher head and efficiency due to the lack of viscous damping effect. Overall the predicted velocity profiles may be compared with the experiments. The predictions generally agree well with the measurements. The estimated boundary layers on pressure side of the blade are predicted smaller than the measured boundary layers. Large differences occur near the tip behind the midchord. It can be seen that at midspan, the calculated data are very comparable to the experimental data, but a large difference occurs close to the tip. In general, the turbulence model of SpalartAllmaras estimates more accurate the complex three-dimensional flow field than the turbulence model of Baldwin-Lomax. The difference may be related to the limitation of the simple algebraic turbulence model of Baldwin-Lomax. This model would underpredict the flow approaching separation. It may than produce a considerable discrepancy in a separation region, especially in a large separation like that of the tip leakage flow and the leakage vortex. It is clear that the algebraic eddy-viscosity model is strictly valid only for two dimensional flows with a mild pressure gradient or is suitable for three-dimensional boundary layers with small cross flow. The model is not valid for curvature, rotation or separation and pressure or turbulence driven secondary flows. The algebraic eddy viscosity model is adequate for the prediction of the mean velocity field.
6.46

7.1 Introduction
In the previous chapter, the numerical simulations of the flow field in the impeller and the performance of the propeller were validated with the experimental data. The performance curve and the velocities profiles, calculated with the algebraic model and the one-equation model, reproduce well the experimental data. Quantitatively, the turbulence model of Spalart-Allmaras gives better numerical results that are representative for the experimental data. Since the comparison of the measured and the calculated results provides sufficient evidence of that the numerical methods predicts the separated flow reasonably well, an attempts was made to use the numerical results to shed new light into the complex flow structure inside the impeller. In this chapter, the three-dimensional flow in the propeller pump will be further studied with the numerical results of the turbulence model of Spalart-Allmaras, to clarify many of the flow phenomena by flow visualization. It is widely accepted that the flow visualization is a very important supplement for outlining a complex threedimensional flow. The visualizations were carried out on both rotor blade surfaces and the endwalls of the impeller for 4 different flow rates (75%, 90%, 100% and 110% Qd). The numerical model for the simulation of the performance will be used in this chapter.
7.2 Evolution of the Flow Structure in the Impeller

The flow visualisation observations are done with the aids of the numerical limited streamlines, immediately adjacent to a solid wall and are analysed by the topological concepts mentioned by Murray and Peake (1982), Kang (1993), Dlery (2001), Surana et al (2005) and Gbadebo et al (2005). The streamlines in the mesh surface closest to a solid wall are considered to be limiting streamlines in this study. Since separated flows are usually characterized by free shear layers as well as strong viscous effects, a highly refined mesh right from the adjoining surfaces of the blade surfaces and/or endwalls toward the free stream was found necessary to achieve a realistic prediction of the size of the separated region. The level of grid refinement have to be insure the adequacy of the thin shear layer approximation for viscosity in the code, which assumes the first grid point from the surface to be at less at a y+ of 10. Pictures 7.1 show a general view of the streamlines near the hub and the blades for the design and an off-design (75%Qd) flow rates. Different separation lines (converging shear stress lines) and (re)attachment lines (diverging shear stress lines) as well as any singular points such as nodes and saddle points are shown in these pictures. The local velocity vectors represent the differences between the separation lines and the reattachment lines. When the velocity vectors are joining, than the line is a separation, and when the velocity vectors separate, the line is a reattachment. A nodal point is called a separation node, if all vector lines are directing towards it. In other words, from there the fluid moves away from the wall. A node is called an attachment node, if all the vector lines are directing away from it. This implies that the fluid sinks to it from the external flow field. Spiral node or focus is a point around which an infinite
7.1
Chapter 7: 3D Flow Structure number of limiting streamlines spiral, for example, can denote a vortex lifting off the surface. With these features, a first idea of the flow field can be developed.
Figure 7.1: Streamlines on the hub and the blades for different flow rates (left: flow rate = 75%Qd and right: flow rate = Qd)
Lets study the critical lines and singular points in more detail. At the inlet of the propeller pump near the hub, there is a stagnation point S1 in front of the leading edge of the impeller and the corresponding nodal point of attachment N1 (see figure 7.2). It is a saddle point (S1) and indicates the formation of the horseshoe vortex. The incoming boundary layer flow separates in front of the blade leading edge and forms a horseshoe vortex (see figure 1.11). This vortex forms two legs (marked by convergence of neighbouring streamlines) evolving along the blade pressure side and suction side, respectively. Under the action of the passage traverse pressure gradient, the suction side leg of the horseshoe vortex, Ls, turns around the leading edge towards the suction side corner and disappears before the midchord. The pressure side leg, Lp, of the horseshoe vortex is enlarged and moves towards mid passage to the suction side corner from where they are rolled into the passage vortex (PV) during its evolution downstream because of the pressure gradient at the hub and the secondary flow (see figure 5.6). On its way downstream, it is incorporated into the passage vortex due to their same sense of rotation. In the present flow conditions, it is difficult to separate one from the other. The horseshoe vortex is straightened with increasing inlet flow angle Table 7.1 gives an indication of the relative position of the saddle point in function of the flow rate. The reference position is the position of the saddle point by design flow rate, calculated with the turbulence model of Spalart-Allmaras. The calculated positions are an indication, because the grid points around the saddle point estimated his coordinates. With increasing the flow rate to the surging point, the position of the horseshoe vortex near the leading edge of the impeller moves more upstream. Due to the higher incident angle at lower flow rate, the pressure side separation line of the horseshoe vortex bents more towards the suction side than for a lower incidence.
7.2
Flow Rate 75%Qd 90%Qd Qd 110%Qd
Pitch wise 0.87 0.93 1.00 1.02
Axial chord wise 1.05 1.01 1.00 0.99
Table 7.1: Relative position of the saddle point S1 near the leading edge
Figure 7.2: Zoom of the streamlines on the hub near the leading edge of the impeller. The colour indicates the static pressure on the hub. The position of the horseshoe vortex is depends the flow rate.
Figure 7.3 gives the streamlines of the pressure side of the impeller. The surface flow visualization patterns over the pressure surface show a very two-dimensional flow. However, one very significant pattern does show up on the pressure surface. A radial outward flow exists over the top 8 percent of the span. This radial flow initiates the flow from the pressure surface through the tip clearance. Figure 7.9 shows the skinfriction lines (LL) associated with the leakage flow across the rotor blade tip section. The pressure side corner of the rotor blade tip section of the propeller pump was rounded to prevent local separation of the leakage flow. This rounding is not implemented in the numerical models, so that the separation of the leakage flow in the numerical models are overestimated.
Figure 7.3: Streamlines of the pressure side of the impeller (left 75% Qd , right Qd)
7.3
Chapter 7: 3D Flow Structure There is a small reverse flow present at the feet of the leading edge, while this is not visible in the pictures of the higher flow rates. The close-up of these streamlines near the leading edge for the flow rates of 75%Qd and 90%Qd are zoomed in picture 7.4, where the reattachment line is very clear. This reattachment line is due to the incidence of the flow at the leading edge. As described in the boundary conditions the inflow condition is axially imposed. When the flow rate changes with a constant rotating speed of the impeller, than the incidence of the flow at the leading edge of the impeller will change too. This incidence has influence on the performance of the pump. The selection of the optimal incidence is a function of blade camber, blade shape, space-chord ratio, stagger angle, Reynolds number and solidity that are described by the correlation coefficient of Howell and Lieblein (Dixon, 1975).
Figure 7.4: Detail plot of the streamlines of the pressure side of the impeller (left 75% Qd , right 90% Qd)
The high blade angle and the thickness of the blade near the hub of the impeller results in the high incidence angle of the flow on the leading edge of the impeller, which gives the reattachment line on the suction side of the impeller. This reattachment line moves more upstream with increasing flow rate. Figure 7.5 shows the streamlines on the suction side of the impeller blade. This figure shows that a rather complex, three-dimensional corner separation exists there where the suction surface trailing edge meets the hub. Also referred to as corner separation, these separations contribute greatly to passage blockage, which effectively places a limit on the loading and static pressure rise achievable by the propeller. In addition, the subsequent mixing of the flow in the separated region with the main passage flow may lead to a considerable total pressure loss and a consequent reduction in efficiency. This figure shows that a line of local separation exits on the suction surface, originating at about 35 % chord from the leading edge at the hub. It stretches diagonally and winds into a focus (vortex F5) near the tip. The same line passes through a saddle point (S5) adjacent to the focus and extends to a distance along the span equal to about 60% chord length from the hub at the trailing edge. The oil-paint surface flow visualization from the HIREP experiment of Zierke et al (1995) showed similar patterns on the suction surface. The downstream zone of separation varies from the three-dimensional corner separation near the hub, including the spiral node of separation, to a more two-
7.4
Chapter 7: 3D Flow Structure dimensional type of separation over most of the span. Just upstream of this local separation line, centrifugal effects cause the skin-friction lines to migrate radially outward. Also, skin-friction lines emanating from the pressure surface have moved around the trailing edge within the separation zone, and also migrate radially outward. Because of the radially outward flow near the trailing edge, the surface of separation that lifts off the blade at the local separation line rolls up into a spanwise vortex. Near the casing, this vortex moves in the circumferential direction, away from the suction surface as the blade rotates in the other direction. The suction side branch, labelled with Ls in figure 7.2, turns round the blade leading edge and interacts with the suction surface before the midchord in both flow cases. This can be seen to the common point from which the suction surface and the endwall separation lines originate. Close inspection of the endwall/surface corner of figure 7.5 suggests that this point takes the form of a multiple (node-saddle) critical point (a node (N2) on suction side and a saddle point (S3) on the endwall). A multiple nodesaddle critical point like this is topologically unstable unless it remains in the corner between surfaces. On the endwall, the nodal point arm of the critical point links up with the trailing edge saddle point to form a pair (N2,S2). On the suction surface, the saddle point arm pairs up with the nodal point of attachment at the trailing edge corner, i.e. (S3, N3). On the hub, at the trailing-edge corner, is a pair of saddle and focal points of attachment (S4, F4). The nodal point of attachment N3 is the origin of the attachment line, which acts as one of the critical lines of saddle point S5. Following the observation of the formation of critical points in figure 7.5, it is clear that the leading edge horseshoe vortex and its associated dividing streamlines that emanate from the leading-edge stagnation (saddle) point, and which form the base of the vortex system, play a mayor role in the mechanism of 3D separation. The suctionside leg of the horseshoe vortex experiences the combined effect of the streamwise and circumferential pressure gradients. The node-saddle critical point (N2 and S3), which occurs as a node on the hub and as a saddle point on the suction surface, marks the origin of 3D separation and is characterized by the emergence of separation lines on both the suction surface and the hub. Due to the evolution of the horseshoe vortex and the generation of the separation bubble, the limiting streamlines in front of the saddle point S3, coming from the midpassage, does not climb along the suction surface, but move along the corner towards downstream. The suction side leg of the leading edge horseshoe vortex stretches downstream along the suction side corner and cannot exist downstream of the midchord, due to the action of the strong adverse pressure gradient and the threedimensional separation bubble.
7.5
Figure 7.5: Detail plot of the streamlines of the saddle point at the suction side of the impeller (at design)
Just downstream of the nodal point arm (N2) on the hub, the limiting streamlines along the corner can have a reversing direction as they approach the suction surface, presumably because of high static pressure that is building along the corner. The limiting streamlines then converge along the separation line, which appears as an envelope of the limiting streamlines. Because this node-saddle critical point occurs as a saddle point on the suction-surface part of the corner, the suction-surface separation line appears to form an asymptote of the neighbouring streamlines, limiting streamlines near the saddle point. The trailing edge saddle point S2, on the hub, acts as a divide between the limiting streamlines undergoing flow reversal (under greater influence of the adverse pressure gradient) on the suction-surface corner and those proceeding normally downstream. Zierke et al (1994) observed that the flow on the suction side of the impeller moves to the tip due to the centrifugal force. A separated vortex surface (trailing edge attachment line) may shed and extend towards downstream, in which the so-called trailing shed vortex (see figure 2.10) evolves. The separated vortex surface moves away from the midspan, together with fluid from the corner separation region, circuits around the spiral node F5 and forms a spiral vortex. This vortex is named here the concentrated shed vortex, using shed as it is shed from the suction surface and concentrated to distinguish it from the trailing edge shed vortex. (CSV, see figure 7.6). Zierke et al (1994) were able to visualize this vortex as it moved away from the suction surface in the circumferential direction by lowering the tunnel pressure until the vortex began to cavitate. They showed that this trailing edge separation vortex lies closer to the casing than the tip leakage vortex, with both vortices rotating with the
7.6
Chapter 7: 3D Flow Structure same sense. Eventually, the trailing edge separation vortex rolls up into the tip leakage vortex as it propagates downstream. Evidently as discussed in the literature review, the higher loaded rotor blades created a stronger vortex from the trailing edge separation.
Figure 7.6: Schematic structure of the secondary flows (Kang, 1998)
Behind the impeller exit, streamlines are sweeping from the pressure side towards the suction side. Approaching to the suction side, they were facing a streamline issued from immediately behind the blade trailing edge, and form the saddle point S2. Then they turned to the mainstream direction. The line parallel is the reattachment line of the passage vortex, while the line from the saddle punt S2 is the separation line of a corner vortex (CV, see figure 7.6). We still call this vortex a corner vortex, even though it is not formed in the suction side corner within the passage. It is concluded that the corner vortex is not the evolution of the suction side edge of the horseshoe vortex, even though they have the same senses. The horseshoe vortex has been disappeared in front of the bubble leading edge. The corner vortex is generated just behind the trailing edge, due to the interaction of the passage vortex and the wake flow. The vortex street consists of two branches with counter-rotating vortices. The two wake branches contain the low-momentum fluid stemming from the pressure side and from the suction side boundary layer of the blade respectively. The pressure side branch of the wake can be found between leading boundary of the wake and its centreline, while the suction side branch is between the centreline and the trailing boundary of the wake. An asymmetry of the wake around its centreline can occur due to the different boundary layer thickness on pressure side and suction side of the blade. Thus it is depending on the inlet flow angle of the blades, respectively the operating point of the axial pump. Figure 7.7 compares the suction surface and the endwall limiting streamline patterns computed for different flow rates. Tables 7.2 until 7.5 give an indication of the
7.7
Chapter 7: 3D Flow Structure relative positions of the singular points (S2, S4, N4 en S5) in function of the flow rate. The reference position is the hub position of the leading edge of the rotor. These calculated positions are also an indication, because the grids points around the singular points estimate their positions. Their positions are normalized with the dimensions of the reference point. Flow Rate 90%Qd Qd 110%Qd Pitch wise 0.797 1.23 1.29 Axial chord wise 1.92 1.81 1.81
Table 7.2: Relative position of the saddle point S2
Flow Rate 90%Qd Qd 110%Qd
Pitch wise 0.98 1.13 1.24
Axial chord wise 1.73 1.73 1.73
Table 7.3: Relative position of the saddle point S4
Flow Rate 90%Qd Qd 110%Qd
Pitch wise 0.93 1.04 1.28
Axial chord wise 1.79 1.79 1.80
Table 7.4: Relative position of the nodal point N4
Flow Rate 75%Qd 90%Qd Qd 110%Qd
Span wise 0.7075 0.85 0.99 1.28
Table 7. 5: Relative position of the saddle point S5
Close inspection of the predicted surface singularities shows that form the five critical points pairs, similar to pairs (S2,N2) to (S5,F5) enumerated in figure 7.5, no apparent critical points occurs on the suction surface within the blade passage. By just decreasing the flow rate until 75% Qd, the 3D separated region grows slowly. The structure of the region changes complete when further decreasing of the mass flow.
7.8
Chapter 7: 3D Flow Structure For the mass flow of 75%Qd, the singular points (S2, S4, and N4) are so close to the endwall separation that no difference could be made. As might be expected, an increase in the chordwise extend of the separated region from the leading edge and the spanwise extend from the hub is evident with decreased flow rate (increased incidence). When the flow rate increases, the saddle point S4 moves more to the trailing edge and joins with the node point F4. The suction streamline pattern at the design condition is almost the same as that at the off-design condition, except for the variation of the separation region. The spanwise position of the spiral node F5 decreases with increasing flow rate.
Flow rate = 75% Qd
Flow rate = 90% Qd
Flow rate = Qd
Flow rate = 110%Qd
Figure 7.7: visualization of the singular points after the trailing edge of the impeller.
Certainly the wake structure develops during the wake propagation. Due to the induced velocities by those vortices an increasing respectively decreasing of velocity is induced at the wake boundaries within the blade passage as well as in the blade surfaces. As a result of the suction side of the impeller blade, a decrease of velocity can be observed at the leading boundary of the wake, while it increases at the trailing boundary of the wake. Within the blade passage effectively an additional secondary flow system appear, which is directed from suction side to pressure side within the passage. The incoming wakes strongly influence the unsteady boundary layer. Within the boundary layer the wake structure can be observed with a time lag to the propagation of the wakes in the passage.
7.9
Chapter 7: 3D Flow Structure Downstream, near the trailing edge of the impeller, the structure of the flow is more complex. Due to the strong curvature of the hub, the adverse pressure gradient creates extra separations. Two saddle points and an attachment node in the midpassage near the trailing edge are recognized for high flow rates. At low flow rates, the saddle points are less clear due to the separation lines and the attachment node by the separation line.
7.3 Flow Structure of the Tip Clearance Flow

The clearance between the impeller and the casing leads to a clearance flow and hence loss of efficiency. The leakage flow over the tip occurs as result of the pressure difference between the pressure and suction surfaces and is dominated by a vortex structure around the tip. The flow produced by this vortex reduces the local turning performed by the blade and hence the work extracted from the flow. As a consequence of the viscous effects in the tip clearance gap, entropy is produced. The second major aspect is the subsequent mixing of the flow, which has passed through the tip clearance, with that of the main flow. The leakage issuing from the gap is of high velocity and mixes rapidly with the mainstream. Intense mixing between these two dissimilar and high-speed jets, with widely varying flow angle, produces intense shearing and flow separation. The formation of a vortex depends on various parameters, of which some are configuration parameters (single stage, multistage, cascade, rotor alone), and others are not: inlet flow turbulence and the annulus wall boundary layer thickness, the difference in magnitude and direction of the velocity in the leakage jet and in the mainstream flow, blade speed and stagger angle. These are in addition to the widely accepted parameters such as tip clearance height, Reynolds number, Mach number, thickness of the blade, blade loading, etc. Several investigators have extensively researched the detailed structure of the flow within the tip region, but there are substantial differences in the detailed understanding of the structure of the flow observed in cascades and rotors (Lakshminarayana et al (1995)). Based on the data reported in the literature, a schematic drawing of the structure of the tip clearance in an axial flow propeller can be made, as shown in figure 7.8. The leakage jet emerging from the tip clearance travels towards the suction surface. The mixing of this high-speed leakage jet with the main flow, both of which involve large flow angle changes at the interface, produces a separated region entrains fluid from across the passage as well from the lower radii. This region has very low axial, tangential, and radial velocities, and the losses are very high (in general 2 % until 4 % or more). The radial flow is outward everywhere in the tip clearance region, with the exception of the region very close to the tip of the blade and at inner radii near the suction surface. The multiple tip vortex structure in the tip consists of the well-known tip leakage vortex, a tip separation vortex and a secondary vortex (see figure 7.8).
7.10
Figure 7.8: Schematic of the vortex structure around the tip (Kang, 1993).
The skin-fiction line as shown in figure 7.9, approaching the blade leading edge, splits into two branches at a point in front of the leading edge. This point is a saddle point, labelled as S6 in figure 7.9, resulting in the formation of a leading edge horseshoe vortex (just like at the hub, see figure 2.11)). Its suction side branch turns around the leading edge and joins with the separation line of the tip leakage vortex at about 5% chord downstream. The pressure side branch stretches toward downstream to the midpassage. The most remarkable impression of figure 7.10 is the formation of a reattachment line near the pressure side and a separation line of the tip leakage vortex near the suction side. From a detailed analysis of the visualisations, a separation line close to the separation line of the tip leakage vortex can be identified. The streamlines around the midpitch converge to this line, which is certainly the separation of the passage vortex (see figure 2.10). The flow diverging from the reattachment line of the passage vortex implies a flow towards the endwall. As a result, a node point and a saddle point will be topologically generated at the leading edge region, as indicated with NL and SL in figure 7.11. In the paragraph of the casing wall streamlines near the impeller, it will be shown that the wall static pressure takes a larger value along the reattachment line. Moreover, it is seen in figure 7.9 that the limiting streamlines, issued from the reattachment line Lr towards the pressure side, are all going into the gap at almost right angles to the pressure surface, except near the leading and the trailing edges, where the pressure driven leakage flows are small. The origin of the tip leakage vortex (see figure 7.8) might to be the saddle point SL as indicated in figure 7.11. The apparent rolling-up of the tip leakage vortex starts just downstream of the leading edge in the suction side corner. Inside the gap, the vortex may be mixed with the leakage flow.
7.11
Figure 7.9: Visualization of the saddle point near the leading edge of the tip
Figure 7.10: Visualization of the saddle point near the leading edge of the tip at design flow
7.12
Chapter 7: 3D Flow Structure Comparing figures 7.10 and 7.11 at design and off-design conditions respectively, no significant difference can be identified except for the variation of the limiting streamline turn angle. With an increasing inlet flow angle, the inclination of the streamlines in the middle of the passage and under the blade profiles becomes much stronger towards the tangential direction. In addition, by increasing the inlet flow angle, the separation line of the tip leakage vortex tends to move forward.
Figure 7.11: Visualization of the saddle point near the leading edge of the tip.
From the visualisation of the flow pattern on the blade tip surface (see figure 7.12), one can see that the streamlines, diverging from about the middle of the blade profile, converge towards the tip edges over about the complete chord length. As a result, a saddle S9 and a node point N9 were formed at the leading edge region (see figure 7.13).
Figure 7.12: Visualization of the blade surface flow pattern.
Figure 7.13: Visualization of the saddle point near the leading edge of the tip.
7.13
Chapter 7: 3D Flow Structure The separation line of the tip separation vortex of figure 7.13 is similar to the socalled separation bubble or vena contracta (see figure 7.14), as referred by Rains (1954), Bindon (1986a) and Moore and Tilton (1988). The flow enters the tip gap from the pressure side of the blade, separates from the blade tip and contracts to a jet. However, it is not really a separation bubble, but a vortex aligning along the tip. Both of the separation lines along the blade edges are issued form the saddle point S9 at the leading edge (see figure 7.13) and terminate at node N10 near the trailing edge. Apparently, the tip flow pattern will strongly depend upon the blade thickness, gap size and the pressure difference. The oil pattern of the HIREP experiment (Zierke (1995)) shows that the angle between the relative skin-friction lines and the chordline decreased with increasing the clearance.
Figure 7.14: Tip clearance flow geometry with vena contracta.
Figures 7.15 are the relative velocities in three different radial positions (zero, mid and full span of the tip clearance). The flow from the pressure side to the suction side is weak because the flow in the tip clearance has nearly the same direction of the main flow. The raison of the weak tip clearance is the high blade angle at the tip (76.83 to opposite the axial direction) and it is nearly parallel with the rotation direction. This means that the expected losses due to the interaction of the tip leakage and the main flow will be small.
7.14
Figure 7.15: Relative velocity vector field at tip (up), at midspan (middle) and fullspan (down)
In most open propellers, which generally have an elliptical blade shape, only a single trailing edge vortex is present. The saddle point S6 disappears and no insight of the leading edge horseshoe vortex will be seen. In ducted propellers, typically two welldefined vortices in the tip region are formed. In addition to a trailing vortex formed near the tip trailing edge, a much stronger tip-leakage vortex is generated in the gap region between the shroud wall and the blade tip. These two unequal co-rotating vortices introduce small-scale unsteady motions during vortex merging. These are in addition to upstream turbulent fluctuations and the vortices start wandering (Chen et al. (1999), and Devenport et al. (1999)).
7.4 Casing Wall Stream Lines near the Impeller

The wall streamlines (or wall-shear stress lines) rotor profiles (blade tip contours) may be identified in figure 7.16. The direction of the rotor rotation is from the top to the bottom of the figure and the flow direction is from right to left. The whole picture may be divided into three regions, separated by two characteristic circumferentially directed lines, divide the whole figure into three parallel sections.
Figure 7.16: Casing flow characteristics (left Q/Qd=0.75, middle Q/Qd=1, right Q/Qd=1.10)
In the right region of figure 7.16, the wall streamlines are oriented parallel to the pump axis. In the second region, with an axial length approximately equal to the axial chord of the rotor tip section, the wall streamlines are highly curved, being parallel to Experimental and Numerical Study of an Axial Flow Pump 7.15
Chapter 7: 3D Flow Structure the axis at the beginning of the region and circumferential at the end. Finally, in the third region, the region to the stator, the wall streamlines are again highly curved, this time with a circumferential direction at the beginning and a direction parallel to the axis at the end. The vectors in figure 7.16 indicate the axially forward flow in the first and third section and axially backward flow (i.e., reversed flow) in the second section. This behaviour of the flow defines the first characteristic line to be a separation line and the second characteristic line to be an attachment line. The position of the separation and attachment lines and the axial length of the reverse flow section have been investigated for a wide range of flow rates. The left figure of a flow rate of Q/Qd = 0.75 is similar to the right figure with a flow rate Q/Qd = 1, apart from a somewhat increased reverse flow region and a curve up into the circumferential direction of the wall streamlines near the separation line. More pronounced changes of the axial event of the reverse flow region were observed in the casing flow characteristics for higher and lower values of Q/Qd. At design, i.e. Q/Qd=1, the separation line is just in front of the leading edge (axial position a/lax=0.0) and the attachment line is just ahead of the trailing edge (axial position a/lax=1.0). For higher values of Q/Qd, the reverse flow region contracts more and more without completely vanishing in the flow range investigated. For lower values of Q/Qd, the reverse flow region gradually expands until about Q/Qd=0.75 at which the stall limit is reached. The separation suddenly jumps to some 50 % of the axial rotor chord in front of the rotor while the attachment line stays nearly at the same position. This leads to a substantial increase of the reverse flow region at about Q/Qd=0.75, the first stable operating point of the pomp in stall. When the pump is throttled further, the separation line moves far in front of the leading edge while the attachment line stays nearly fixed at the trailing edge. The sudden jump of the separation line from the rotor leading edge to a position ahead of it is also related to an enormous expansion of the reverse flow region in radial direction and represents the actual reason for the dramatically decreased pump head at Q/Qd=0.75. The reverse flow above the rotor has a large circumferential component, as can be seen in figure 7.16. As long as this reverse flow is a thin layer only and separates from the casing at the rotor leading edge, the main flow is not affected substantially. As soon as the separation line moves ahead from the rotor, the reverse flow loaded with positive angular momentum separates from the casing and recirculates back into the rotor. Thus, the returning axially positive flow at the outer part of the rotor has a large positive circumferential component, which leads to a reduction of pump head. A summary of this development is illustrated in figure 7.16, showing a slowly increasing reverse flow section with different flow rates. The figures, which shows reversed flow near the casing wall between separation and attachment line, is obviously the most critical over tip section since it is here where tip clearance flow, vortex and the main flow interact creating blockage and end wall losses in a highly complicated process. This process may not be fully understood without further detailed information especially concerning the state of the tip clearance vortex. The calculated casing wall pressure distributions of the present pump were used to construct constant static pressure lines for the rotor casing wall region at all flow rates investigated. As figure 7.17 shows the isobars for Q/Qd=0.75. The isobar with Cp=0.0 corresponds to the static pressure of the undisturbed flow in front of the rotor. The difference of Cp between the isobars has a constant value of Cp = 1. A well-defined low-pressure trough with nearly circumferential direction is to seen in figure 7.17. Experimental and Numerical Study of an Axial Flow Pump 7.16
Figure 7.17: Casing wall static pressure isobars (right Q/Qd=0.75, left Q/Qd=1.)
This low-pressure trough indicates the existence of a tip clearance vortex. Figure 7.17 shows also the distribution of the wall pressure coefficient Cp at design flow rate. With decreasing flow rate two important trends of the pressure distributions were observed. The depth of the low pressure troughs increases with decreasing flow rate and the orientation of the low pressure troughs turn more and more onto circumferential direction with decreasing flow rate. A tip clearance vortex in a nearly circumferential position constitutes a considerable blockage for the casing wall flow in front of the rotor. This flow separates along a circumferential separation line and it might well be assumed, that the tip clearance vortex takes part in the separation process.
7.5 Conclusion
In this chapter, a general 3D view of the flow in the propeller pump is reconstructed based on the numerical results of the averaged Navier-Stokes equation and the turbulence model of Spalart-Allmaras, together with the topological rules of the 3D separated flows observed on the endwalls and the rotor blade surfaces. Different singular nodes and lines were recognized on the solid walls of the propeller pump such that a general 3D view of the flow is reconstructed as resumed in figure 7.18. The different vortexes (tip leakage vortex, trailing-edge separation vortex and the horseshoe vortex) are printed on the plot. The vena contracta is absorbed by the tip leakage vortex. The passage vortex is also not plotted.
7.17
Figure 7.18: Schematic of Rotor Blade Secondary Flow [Zierke et al, 1993]
The passage vortex, tip leakage flow, trailing edge vortices and the vena contracta are recognized in the measured flow field. Flow field measurements fail to exhibit other vortices since there are either to weak or are filtered out in the measurement locations. The velocity field can not be measured completely due to the blade shadow, the high reflections of the laser light by the blade and the hub, and/or the low data rate near the hub. It means that the velocity profiles of the blade and the hub boundary layers can never been measured. Due to the unsteadiness characters of some vortices, it is difficult to create a database that represents the complete flow field in the impeller. Also due to the curved casing and the different refraction indexes of window and water, it is difficult to measure the two (or three velocity) components in a measurement point at one time. The two dominant secondary flow structures within the impeller are passage vortex and the rotor blade tip-leading vortex. This tip leakage vortex also has a strong impact on the flow unsteadiness downstream of the rotor blades. The nondeterministic unsteadiness increases because of the increased turbulence levels in the vortex and because of the vortex wandering and kinking. Freestream turbulence, vortex structures instabilities and the influence of the casing endwall all influence vortex wandering and kinking. This type of unsteady motion, along with a similar unsteadiness associated with other vortices and separation regions, will certainly add to the difficulties associated with turbulence modelling during numerical computations of the flow field in the axial flow pump.
7.18
Summary and Conclusions
Summary and Conclusions

The purpose of this thesis was to get a better understanding of the complex flow of an axial propeller pump. Therefore an experimental test facility of a one-stage propeller pump was built and provided with equipment to measure the pump performance. After optimisation of the testing conditions and performing a precise calibration of the measuring equipment, the pump performance was measured and compared with the design conditions of the axial flow pump. These design conditions of the pump were found back in the measurements. Uncertainty analysis proved that the instrumentations and the measurement procedure are conforming to ISO 3555. The next step was to prepare the test facility for two-component LDA velocity measurements in the impeller. Therefore optical access of the LDA beams to the impeller was needed. The strong curved casing of the propeller pump and the different refractions indexes of the mediums caused problems when focussing the laser beams in one volume. After studying the different possibilities of having an optical access within the impeller, it was decided to mount a flat polished polycarbonate window in the casing of the impeller. Thereby, the curved casing is locally smoothed over the complete length of the impeller. Refraction of the laser beams, both in transmissions through this window and through the flow, leads to displacement of the measurement volume and changes in the intersection angle of the laser beams. A theoretical analysis of the positional correction and accuracy of the LDA measurement volume using an off-axis alignment of a probe relative to an internal flow has been carried out. We concluded that measurement of the velocity field in an internal flow using LDA through a transparent window with an off-axis alignment could result in quite different measurement parameters compared to the results of measurements using an on-axis probe alignment. Due to the difficulties of the off-axis alignment, only a 2D flow field in the propeller pump is measured by moving the probe in on-axis alignment. Because of the aberration arising from the curved interface of the circular pipe, measurements for axial and tangential velocity components must be carried out separately. Due to the refraction of laser beams at curved interfaces during the LDA measurements in flows, the position of beam intersection and the half angle may undergo non-linear changes in wavelength, compared to the ones without refraction. Therefore, a guideline for the corrections of the measurements point of the two velocity components and the correction for the tangential velocities with respect to the shift and the optical properties of the measurement volumes has been presented. When the test facility was ready for operation, a two-component LDA velocity survey at the design point has been carried out in nine different cross sections, from 40 % axially rotor chord upstream of the leading edge until 40% axially rotor chord downstream of the trailing edge of the rotor. A shaft encoder connected with the propeller pump, collected time stamped encoded data in bin sizes of 1 degree (or 0.5 degrees in the wakes region of the rotor blades) for all spans. The number of spans in one survey varied between 22 to 27 points. In the procedure of quantifying LDV errors, different types of bias sources of uncertainty were discussed. When evaluating these uncertainties, many of these bias errors are negligible. Many of the remaining errors, although not negligible, are
Conclussion.1
Summary and Conclusions relatively small compared to the precision errors. For the turbulence intensity measurements, many of the bias errors become much more important. For the present investigation, however, the individual and the total bias errors are not explicitly calculated for all measurement points. In some measurement points, the average error of the velocity was around 5.6 % for a number of 8000 samples. The experimental measurements and subsequent data analysis have improved our physical understanding of this complex flow field. One might observe that the impeller has produced a rather two-dimensional flow field. However, a further analysis of the data shows that three-dimensional effects are significant. The considerable boundary layer growth is observed from the hub to mid radius, while the flow from mid radius to tip is found to be highly complex, due to the interaction of pressure and suction surface boundary layers and the resulting radial inflow in the first part of the impeller passage. The passage vortex, tip leakage flow and the vena contracta are recognized in the flow field. Flow field measurements fail to exhibit other vortices (described in the literature review: horseshoe vortices and vortices emanating from the three-dimensional separation regions on the blades) since they are either to weak or filtered out by the measurements. The two dominant secondary flow structures within the impeller are the passage vortex and the rotor blade tip-leading vortex. This tip leakage vortex also has a strong impact on the flow unsteadiness downstream of the rotor blades. The non-deterministic unsteadiness increases because of the increased turbulence levels in the vortex and because of the vortex wandering and kinking. Freestream turbulence, vortex structures instabilities, and the influence of the casing endwall all influence vortex wandering and kinking. This type of unsteady motion, along with similar unsteadiness associated with other vortices and separation regions, will certainly contribute to the difficulties associated with the measurement of the axial and tangential velocity components in the impeller. The installation of the window produced a discontinuity in the casing of the impeller, because it smoothed the curved casing. It is difficult to tell how far it will alter the character of the flow in the tip region, downstream the midchord of the impeller. These experimental 2D LDA measurements and the performance of the propeller pump provided an excellent database for comparisons with three-dimensional, viscous (turbulent) flow computations. During this work, a RANS study with the turbulence models of the algebraic equation of Baldwin-Lomax and the one-equation of Spalart-Allmaras model were applied to validate the numerical results with the experimental performance and the 2D velocity field of the impeller. A grid sensibility study was done, and the length of the inlet was first studied before to start with the definitive comparisons. A length of four times the diameter of the pipe, upstream the rotor tip was sufficient to ensure that the constant mass flow at the inlet of the model developed to a full flow in the pipe before arriving in the impeller. Comparing the numerical pump performance with the experimental data, both models estimate the performance well. The more the mass flow differs from the design point, the more the estimated performance parameters differ from the experimental data. The turbulence model of Spalart-Allmaras estimates quite well the experimental data. The Baldwin-Lomax however, predicts higher head and efficiency due to the lack of viscous damping effect.
Conclussion.2
Summary and Conclusions Overall the predicted velocity profiles may be compared with the experiments. The predictions generally agree well with the measurements. The estimated boundary layers on pressure side of the blade are predicted smaller than the measured boundary layers. Large differences occur near the tip behind the midchord. It can be seen that at midspan, the calculated data are very comparable to the experimental data, but a large difference occurs close to the tip. In general, the turbulence model of SpalartAllmaras estimates more accurate the complex three-dimensional flow field than the turbulence model of Baldwin-Lomax. The difference may be related to the limitation of the simple algebraic turbulence model of Baldwin-Lomax. This model would underpredict the flow approaching separation. It may then produce a considerable discrepancy in a separation region, especially in a large separation like that of the tip leakage flow and the leakage vortex. It is clear that the algebraic eddy-viscosity model is strictly valid only for two dimensional flows with a mild pressure gradient or is suitable for three-dimensional boundary layers with small cross flow. The model is not valid for curvature, rotation or separation and pressure or by turbulence driven secondary flows. The algebraic eddy viscosity model is adequate for the prediction of the mean velocity field. Knowing that the turbulence model of Spalart-Allmaras gives good results that are representative for the experimental data, the three-dimensional flow in the propeller is further studied with the topological rules. The topic of the three-dimensional features in the rotor blades of the propeller as described in the literature are found back in the streamlines patterns on the rotor blades and the endwalls of the impeller. Although the laser Doppler anemometry is a good instrument to measure the velocity components of the particle in the flow, it has handicaps in our application. The velocity field cannot be measured completely due to the blade shadow, the high reflections of the laser light by the blade and the hub, and/or the low data rate near the hub. It means that the velocity profiles of the blade and the hub boundary layers can never be measured. Due to the unsteadiness characters of some vortices it is difficult to create a database that represents the complete flow field in the impeller because of its unsteady features. Also due to the curved casing and the different refraction indexes of window and water, it is difficult to measure the two (or three velocity) components in a measurement point at the same time. The Reynolds stress component will be limited to the average axial and tangential square fluctuations. To solve these problems, other measurement instruments such as five hole - or hotwire probe (CTA), which rotate together with impeller, could be used in an attempt to get better results. Although a lot of experimental work is already done on the axial flow test facility, much more experiments should be done to give a more complete and extended view on the (unsteady) flows in the pump. Since the measurements of the velocity components in the rotor wake are not complete, more measurements could be done in the rotor and stator with a finer grid, especially near the tip of the rotor and in the boundary of the stator and rotor blades. Also interesting would be to measure the flow field at different off-design flow rates, with or without the study of the aviation phenomenon of the rotor. The unsteady features of the rotor tip region could also be studied better with static pressure transducers in the casing, and the unsteady interactions of the rotor wake on the stator could be studied with the aid of (five hole) probes to measure the total and static pressure in the rotor wake of the impeller and in the stator. The window in the casing of the pump is perfect to transform it to support the probes without too much man work. Even cavitations phenomena, by lowering the Experimental and Numerical Study of an Axial Flow Pump Conclussion.3
Summary and Conclusions flow rate, could be studied in detail by visualisation of the different vortices. This additional (unsteady) database would give the opportunity to understand better the unsteady flows in this one-stage pump, and could be used to validate unsteady RANS and/or LES simulations. Numerical study of the propeller pump with different k- turbulence models, with or without validation of this database, would give us also the opportunity to understand better the flow through this axial pump.
Conclussion.4
Reference
Bibliography
Abernathy, R.B., Benedict, R.P., and Dowdell, R.B., 1985, "ASME Measurement Uncertainty", ASME J. of Fluids Engineering, Vol. 107, pp. 161-164 Bhattacharyya A., Acosta A.J., Brennen C.E., Caughey T.K. ,Observations on OffDesign Flows in Non-cavitating Axial Flow Inducers, ASME Pumping Machinery, FED-Vol. 154, 1993 Agrawal. Y. Talbot, L. Gong, K., 1978, Laser Anemometer Study of Flow Development in Curved Circular Pipes, ASME J. of Fluid Mech., Vol. 85, pp. 497518 Albrecht, Borys, Damaschke, Tropea, 2003, Laser Doppler and Phase Doppler Measurement Techniques, Springer Verlag Asford, G.A., Powell, K.G., 1996, An Unstructured Grid Generator and Adaptive Solution Technique for High Reynolds Number Compressible Flow, VKI Lectures Series 1996-06 Baldwin, B., Lomax, H., 1978, Thin Layer Approximation and Algebraic Model for Separated Turbulent Flows, AIAA 78-257 Bicen, A.F., 1982, Refraction Correction for LDA Measurements in Flows with Curved Optical Boundaries, TSI Quarterly, Vol. 23, pp.10-12 Binder, A., Forster, W., Mach, K., Rogge, H., 1987, Unsteady Flow Interaction Caused by Stator Secondary Vortices in a Turbine Rotor, ASME J. of Turbomachinery, Vol. 109, pp 251-257 Bindon, J.P., 1986a, Visualisation of Axial Turbine Tip Clearance Flow Using a Linear Cascade, Cued/A-Turbo/Tr 122 Bindon, J.P., 1986b, Pressure and Flow Field Measurements of Axial Turbine Tip Clearance Flow in a Linear Cascade, Cued/A-Turbo/Tr 123 Boadway. J., Karahan, E., 1981, Correction of Laser Doppler Anemometer Readings for Refraction at Cylindrical Interfaces, DISA Info, Vol. 26, pp. 4-6 Boutier, A., 1991, "Accuracy of Laser Velocimetry", VKI Lecture Series on Laser Velocimetry, Rhode St. Gense, Belgium. Bovendeerd, P.H.M., Van Steenhoven, A.A., Van De Vosse, F.N., Vossers, G., Steady Entry Flow in a Curved Pipe, ASME J. of Fluid Mech., Vol.177, pp. 233246 Budwing, R., 1994, Refractive Index Matching Methods for Liquid Flow Investigations, Experiments in Fluids Vol. 17, pp. 350-355 Brennen, C.E., 1994, Hydrodynamics of Pumps, NREC Delery, J.M., 2001, Robert Legendre and Henri Werle: Towards the Elucidation of Three-Dimensional Separation, Ann. Rev. Fluid Mech, pp. 2001.33:129-54 Carter and Cohen, 1946, Preliminary Investigation into the Three-Dimensional Flow Through a Cascade of Airfoils, Aero. Res. Council, R&M 2336
Ref.1
Reference Chen, G.T., Greitzer, E.M.,Tan, C.S., Marble, F.E., 1991, Similarity Analysis of Compressor Tip Clearance Flow Structure, ASME J. of Turbomachinery, Vol. 113, pp. 260-271 Chorin, A.J., 1967, A Numerical Method for Solving Incompressible Viscous Flow Problems, J. of Comput. Phys., Vol. 2, pp. 12-26. Coleman, H.W., Steele, W.G., 1989, Experimentation and Uncertainty Analysis for Engineers, John Wiley and Sons Cumpsty, N.A., 1989, Compressor Aerodynamics, Longman Scientific & Technical De Cecco, S., Yaras, M.I., and Sjolander, S.A., 1995, Measurements of the TipLeakage Flow in a Turbine Cascade with Large Clearances, ASME Paper 95-Gt-77 Denton, J.D., 1985, Solution of the Euler Equations for Turbomachinery Flows, Part I: Basic Principles and Two-Dimensional Applications, Thermodynamics and Fluid Mechanics of Turbomachinery, Vol. 1, A.S. Ucer, P. Stow Denton, J.D., 1993, Loss Mechanisms in Turbo Machines, IGTI Gas Turbine Scholar Lecture, ASME 93-Gt-435. Denton, J.D., and Dawes, W.N. 1999, Computational Fluid Dynamics for Turbomachinery Design, Proceedings of the Institution of Mechanical Engineers, Part C, J. of Mechanical Engineering Science, Vol. 213(2), pp.107-124 Devenport, W.J., Simpson, R.L., 1989, Time Dependent Structure in Wing Body Junction Flows, Turbulent Shear Flows VI, Springer-Verlag, pp. 232-248. Dewhurst, S.J., Martin, S.R., Jayanti, S. and Coatigan, G., 1990, Flow Measurements Using a 3-D LDA System in a Square Section 90 Bend, 5th International Symposium on Application of Laser Techniques to Fluid Mechanics, Lisbon, Portugal, Paper No. pp 19.1 Dixon, S.L., 1975, Fluid Mechanics of Thermodynamics of Turbomachinery, Pergamon Press. Doukelis, A., Founti, M., Mathioudakis, K., Papiliou, K.,1996, Evaluation of Beam Refraction Effects in a 3D Laser Doppler Anemometry System for Turbomachinery Applications, Meas. Sci. Technol., Vol. 7, pp. 922-931. Durret, R.P., Gould, R.D., Stevenson, W.H., Thompson, H.D., 1985, A Correction Lens for Laser Doppler Velocimeter Measurements in Cylindrical Tube. AIAA J. \, Vol 23, pp. 1387-1391. Durst, F., Melling, A., Whitelaw, J.H., Principles and Practice of Laser Doppler Anemometry, Academic Press, 1981 Edwards, R.V., 1987, Report of the Special Panel on Statistical Particle Bias Problems in Laser Anemometry, ASME J. of Fluids Engineering Vol. 109, pp. 8993. Edwards, R.V., Dybbs, A., 1984, Refractive Index Matching for Velocity Measurements in Complex Geometries, TSI Quart, Vol. 10(4), pp.3-11 Eisele, K., Zhang, Z., Muggli, F., 1994, Investigation of the Unsteady Diffuser Flow in a Radial Pump, 7th International Symposium on Applications of Laser Techniques to Fluid Mechanics. Lisbon, Portugal, Paper No. 38.4.1-38.4.7
Ref.2
Reference Emmons, H.W., Pearson, C.E., Grant, H.P., 1955, Compressor Surge and Stall Propagation, Trans ASME, Vol. 79, pp. 455-469. Erdmann, J.C., Gollert, R.I., 1976, Particle Arrivals Statistics in Laser Anemometry of Turbulent Flows, Applied Physics Letters, Vol. 29, pp.408-411 Farokhi, S., 1988, Analysis of Rotor Tip Clearance Loss in Axial Flow Turbines, ASME J. of Propulsion, Vol. 4, No.5, pp.452-457. Figliola, R.S., and Beasley, D.E., 1991, Theory and Design for Mechanical Measurements, John Wiley & Sons, Inc. FineTM/Turbo 6.1, User Manual Foley, A.C. and Ivey, P.C., 1994, Measurement of Tip-Clearance Flow in a MultiStage, Axial Flow Compressor, ASME Paper 94-Gt-431 Francis, T.B., Katz, 1988, Observations on the Development of a Tip Vortex on a Rectangular Hydrofoil, ASME J. of Fluids Engineering, Vol. 110 pp. 208-215. Gbadebo, S.A., Cumpsty, N.A., Hynes, T.P., 2005, Three-Dimensional Separations in Axial Compressors, ASME J. of Turbomachinery, Vol. 127, pp.331-339. Gearhart, W.S., 1966, Tip Clearance Cavitation in Shrouded Underwater Propulsors, AIAA J. of Aircraft, Vol. 3, No. 2. Graham, J.A. H., 1986, Investigation of a Tip Clearance Cascade in a Water Analogy Rig, ASME J. of Engineering for Gas Turbines and Power, Vol. 108, pp.38-46 Greitzer E.M., 1981, The Stability of Pumping Systems, ASME J. of Fluids Engineering, Vol. 103, pp. 193-217 Hakimi, N. 1997, Precondition Methods for Time Dependent Navier-Stokes Equations. Application to Environmental and Low Speed Flows, Phd Thesis, Dept. of Fluids Mechans. Vrije Universiteit Brussel, Belgium. Hawthone, W.R., 1955, Rotational Flow Through Cascade, ASME J. Mech & Appl. Math, Vol. 3, pp. Heyes, F.J.G., Hodson, H.P., Dailey, G.M., 1992, The Effect of Blade Tip Geometry on the Tip Leakage Flow in Axial Turbine Cascades, ASME J. of Turbomachinery, Vol. 114, pp. 643-651. Hinze, J.O., 1959, Turbulence, An introduction to its Mechanism and Theory, Mcgraw-Hill Book Company. Hirsch, Ch. 1994, CFD Methodology and Validation for Turbomachinery Flows, Turbomachinery Design Using CFD, Agard Lectures Series 195. Hirsch, Ch., 1988, Numerical Computational of Internal and External Flows, Volume 1: Fundamentals of Numerical Discretization, Wiley and Sons Hirsch, Ch., 1990, Numerical Computational of Internal and External Flows, Volume 2, Wiley and Sons Hirsch, Ch., Lacor, C., Rizzi, A., Eliasson, P., Lindblad, I., Huser, J., 1991a, A Multiblock /Multigrid Code of the Efficient Solution of Complex 3d Navier-Stokes Flow, First European Symposium on Aerothermodynamics for Space Vehicles, Estec, Noordwijk, The Netherlands.
Ref.3
Reference Hirsch, Ch., Lacor, C., Rizzi, A., Eliasson, P., Lindblad, I., Huser, J., 1991b, A Multiblock/Multigrid Code to Simulate Complex 3D Navier-Stokes Flows on Structured Meshes, Proceedings of the 4th International Symposium on Computational Fluid Dynamics, University of California Davis, Davis, Ca. Inoue, I.H. and Cumpsty, N.A., 1984, Three-Dimensional Structure and Decay of Vortices behind an Axial Flow Rotation Blade Row, ASME J. of Eng. for Gas Turbines and Power, Vol.106, pp.561-569. Inoue, M. and Kuroumaru, M., 1989, Structure of Tip Clearance Flow in An Isolated Axial Compressor Rotor, ASME J. of Turbomachinery, Vol. 111, pp. 250-256 Inoue, M., and Kuroumaru, M., 1989, Structure of Tip Clearance Flow in An Isolated Axial Compressor Rotor, ASME J. of Turbomachinery, Vol. 111, pp. 250256. Inoue, M., Kuroumaru, M., and Fukuhara, M., 1986, Behavior of Tip Leakage Flow Behind An Axial Compressor Rotor, ASME J. of Engineering for Gas Turbines and Power, Vol. 108, pp. 7-14 Iso 3555, 1977, Centrifugaal-, Schroefcentrifugaal- En Schroefpompen, Richtlijnen Voor De Opleveringsproeven, Klasse B, Bin Jameson, A., Schmitt, W. and Turkel, E., 1981, Numerical Simulation of the Euler Equations By Finite Volume Methods Using Runge-Kutta Time Stepping Schemes, AIAA Paper No. 81-1259. Japikse, D. and Baines, N.C., 1994, Introduction to Turbomachinery, Oxford University Press Japikse, D., 1976, Review-Progress in Numerical Turbomachinery Analysis, ASME J. of Fluids Engineering, Vol. 112, pp.592-606. Jensen, K.D., 2004, Flow Measurements, J. of the Braz. Soc. of Mech. Sci. & Eng., No. 4, Vol. 26. Kang, S., Hirsch, C., 1989, Experimental Study on the Three Dimensional Flow Within A Compressor Cascade With Tip Clearance: Part I: the Tip Leakage Vortex, ASME J. of Turbomachinery, Vol. 115, pp. 444-452 Kang, S., 1993, Investigation on the three dimensional flow within a compressor cascade with and without tip clearance, PhD thesis, Vrije Universiteit Brussel Kang, S., Hirsch, C., 1993, Tip Leakage Flow in A Linear Compressor Cascade, ASME Paper Gt-93-303 Kehoe, A.B. and Prateen, V.D., 1987, Compensation for Refractive-Index Variations in Laser Dopple Anemometry, Applied Optics, Vol. 26, No.13, pp. 2582-2591. Kline, S.J., 1985, The Purpose of Uncertainty Analysis, ASME J. of Fluid Engineering, Vol.107, pp.153-160. Kovats, A., 1964, Design and Performance of Centrifugal and Axial Flow Pumps and Compressors, The Macmillan Company Kreid, D.K., 1974, Laser-Doppler Velocimeter Measurements in Nonuniform Flow: Error Estimates, Applied Optics, Vol. 13, pp. 1872-1881. Kuhlman, J.M., Gross R.W., 1993, Three-Component Velocity Measurements in an Axisymmetric Jet Using LDA, Dantec Information, No.12, pp.10-16. Experimental and Numerical Study of an Axial Flow Pump Ref.4
Reference Lakshminarayana, B. and Horlock, J., 1963, Tip-Clearance Flow and Losses for An Isolated Compressor Blade, Arc R&M 3316. Lakshminarayana, B., 1970, Method of Predicting the Tip Clearance Effects in Axial Flow Turbomachinery, ASME J. of Basic Eng. Vol. 92, pp. 467-480. Lakshminarayana, B., 1991, An Assessment of Computational Fluid Dynamics Techniques in the Analysis and Design of Turbomachinery, The 1990 Freeman Scholar Lecture, ASME J. of Fluids Engineering, Vol. 113(3), pp.315-352. Lakshminarayana, B., Sitaram, N. and Zhang, J., 1986, End-Wall and Profile Losses in A Low-Speed Axial Flow Compressor Rotor, ASME J. of Engineering for Gas Turbines and Power, Vol. 108, pp. 22-31 Lakshminarayana, B., Zaccaria, M., Marathe, B., 1995, The Structure of Tip Clearance Flow in Axial Flow Compressors, ASME J. of Turbomachinery, Vol. 117, pp. 336-347 Langston, L.S., Nice, M.L. and Hooper, R.M., 1977, Three-Dimensional Flow Within A Turbine Blade Passage, ASME J. of Eng. for Power, Vol. 99, No. 1, pp. 21-28. Liu, C.H., Vafidis, C.,Whitelaw, J.H., 1990, Flow in the Coolant Passage of an Internal Combustion Engine Cylinder Head, Exp. of Fluids, Vol. 10, pp 50-54 Lund, S.T., 1998, Generation of Turbulent Inflow for Spatially Developing Boundary Layer Simulation, ASME J. of Computational Physics, Vol. 140, pp. 233258 Mayle, R.E., 1991, The Role of Laminar-Turbulent Transition in Gas Turbine Engines, ASME J. of Turbomachinery., Vol. 113, pp. 509-537. Mc Laughlin, D.K., Tiedermann, W.G., 1973, Biasing Correction for Individual Realization of Laser Anemometer Measurements in Turbulent Flow, Physics of Fluids, Vol. 16, pp. 2082-2088 Mcnally, W.D., and Sockol, P.M. 1985, Review-Computational Methods for Internal Flows With Emphasis on Turbomachinery, ASME J. of Fluids Engineering, Vol. 107(1), pp. 6-22. Meyers. J.F., Generation of Particles and Seeding, VKI Lectures Series 1991-08, Nasa Langley Research Center Hampton, Virginia, Usa. Mizushina, T., Usui, H., 1977, Reduction of Eddy Diffusion for Momentum and Heat in Viscoelastic Fluid Flow in A Circular Tube, Phys. Fluids, Vol.20, pp. 100108 Moffat, R.J., 1985, "Using Uncertainty Analysis in the Planning of An Experiment", ASME J. of Fluid Engineering, Vol. 107, pp. 173-178. Moore, J. and Smith, B.L., 1984, Flow in A Turbine Cascade: Part I, Losses and Leading Edge Effects, ASME J. of Eng. for Gas Turbine and Power, Vol. 106, pp. 400-408. Moore, J. and Tilton, J.S., 1988, Tip Leakage in A Linear Turbine Cascade, ASME J. of Turbomachinery, Vol. 110, pp.18-26. Murray T. and Peake, D.J., 1982, Topology of Three-Dimensional Separated Flows, Ann. Rev. Fluid Mech, Vol.14, pp. 61-85.
Ref.5
Reference Parry, A.J., Lalor, M.J.,Tridimas, Y.D.,Woolley, N.H., 1990,Refraction Corrections for Laser Doppler Anemometry in a Pipe Bend, Dantec Information No. 09, pp.4-6 Prasad, C.R.K., 1977, Tip Clearance Effect in Axial Flow Turbomachines, Indian Institute of Science, Report No. Me-Turbo-1-77. Predin, A., Bilus, I., 2003, Influence of Additional Inlet Flow on the Prerotation and Performance of Centrifugal Impellers, J. of Hydraulic Research, Vol. 41, pp.207-216 Rains, D.A., 1954, Tip Clearance Flow in Compressors and Pumps, California Institute of Technology, Mech. Eng. Laboratory, Report 5. Ruck, B., 1991, Distortion of LDA Fringe Pattern By Trace Particles, Experiments in Fluids, Vol.10, pp. 349-354. Salvage, J.W., 1974, Investigation of Secondary Flow Behaviour and Endwall Boundary Layer Development Through Compressor Cascade, VKI Tn-107. Schlichting, H., 1979, Boundary-Layer Theory, McGraw-Hill, New York. Sieverding, C.H. and Van Den Bosch, P. 1983, The Use of Couloured Smoke to Visualize Secondary Flows in a Turbine Blade Cascade, ASME J. of Fluid Mechanics, Vol. 134, pp. 85-89. Spalart, P.R., Allmaras, S.R., 1992, One-Equation Turbulence Model for Aerodynamic Flows, AIAA 92-0439 Squire, H.B.,Winter, K.G., 1951, The Secondary Flow in Cascade of Airfoils in a Nonuniform Stream, J. of Aeronaut. Sc., Vol. 18, pp 271. Stepanoff, A.J., 1948, Centrifugal and Axial Flow Pump, Theory, Design and Application, John Wiley & Sons. Stock, H.W., Haase. W., 1987, The Determination of Turbulence Length Scales in Algebraic Turbulence Models for Attached and Slightly Separated Flows Using Navier-Stokes Methods, AIAA-87-1302 Storer, J.A., Cumpsty, N.A., 1991, Tip Leakage Flow in Axial Compressors, ASME J. of Turbomachinery, Vol. 113, pp. 252-259. Strazisar, A.J. and Powell, J.A., Laser Anemometer Measurements in a Transonic Axial Flow Compressor Rotor, ASME J. of Engineering for Power, Vol. 103, pp.430-437. Suder, K.L., and Celestina, M.L., 1994, Experimental and Computational Investigation of Tip Leakage Flow in A Transonic Axial Compressor Rotor, ASME Paper 94-Gt-365 Van Maanen, H.R.E., Fortuin, J.M.H., 1983, Experimental Determination of the Random Lump-Age Distribution in the Boundary Layer of the Turbulent Pipe Flow Using Laser Doppler Anemometry, Chem. Eng. Science, Vol. 38, pp. 399-423 Vavra, M.H., 1960, Aero-Thermodynamics and Flows in Turbomachines, John Wiley and Sons Wadia, A.R. and Booth, T.C., 1982, Rotor-Tip Leakage: Part I: Design Optimization Through Viscous Analysis and Experiment, ASME J. of Eng. for Power, Vol. 104, pp. 162-169.
Ref.6
Reference Yaras, M., Zhu, Y. and Sjolander, S.A., 1989, Flow Field in the Tip Gap of a Planar Cascade of Turbines Blades, ASME, J. of Turbomachinery, Vol. 111, pp.276-283. Zhang, Z. and Eisele, K., 1995, Off-Axis Alignment of an LDA-Probe and the Effect of Astigmatism on Measurements, Experiments in Fluids, Vol. 19, pp. 89-94. Zhang, Z. and Eisele, K., 1998, On the Overestimation of the Flow Turbulence Due to Fringe Distortion in LDA Measurement Volumes, Experiments in Fluids, Vol. 25, pp. 371-374. Zhang, Z., 1993, Theoretische Untersuchungen Der Einsatzmglichkeit Fr 3Dimensionale Laser-Doppler-Messungen in Kanalstrmungen. Sulzer Innotec Bericht, Nr. Stt.Tb93.006, Winterthur, Schweiz. Zhang, Z., 2004, Optical Guidelines and Signal Quality for LDA Applications in Circular Pipes Experiments in Fluids, Vol. 37, pp. 29-39. Zierke, W.C., Strake, W.A. and Taylor P.D., 1995, The High Reynolds-Number Flow Trough An Axial Flow Pump, Techical Report No. TR 93-12, 1993, Pennsylvania State University Zierke, W.C., Farrell, K.J. and Straka, W.A., 1994, Measurement of the Tip Clearance Flow for A High Number Axial-Flow Rotor: Part 1 - Flow Visualization, ASME Paper 94-Gt-453 Zierke, W.C., Farrell, K.J. and Straka, W.A., 1994, Measurement of the Tip Clearance Flow for A High Number Axial-Flow Rotor: Part 2 Detailed Flow Measurements, ASME Paper 94-Gt-453 Zierke, W.C., Farell, K.J. and Strake, W.A., 1995, Measurement of the Tip Clearance Flow of High Reynolds-Number Axial Flow Rotor, ASME J. of Turbomachinery, Vol. 117, pp. 522-532.
Ref.7
Appendix A: Laser Doppler Anemometry
Appendix A: Laser Doppler Anemometry1

A.1 Introduction
Laser Doppler Anemometry is a non-intrusive technique used to measure the velocity of particles suspended in a flow. If these particles are small, in the order of microns, they can be assumed to be good flow tracers following the flow and thus their velocity corresponds to the fluid velocity. Important characteristics of the LDA technique, listed in the following section, make it an ideal tool for dynamic flow measurements and turbulence characterization.
A.2
Characteristics of LDA
Laser anemometers offer unique advantages in comparison with other fluid flow instrumentation:
A.2.1 Non-Contact Optical Measurement

Laser anemometers probe the flow with focused laser beams and can sense the velocity without disturbing the flow in the measuring volume. The only necessary conditions are a transparent medium with a suitable concentration of tracer particles (or seeding) and optical access to the flow through windows, or via a submerged optical probe. In the latter case the submerged probe will of course to some extent disturb the flow, but since the measurement take place some distance away from the probe itself, this disturbance can normally be ignored.
A.2.2 No Calibration - No Drift

The laser anemometer has a unique intrinsic response to fluid velocity - absolute linearity. The measurement is based on the stability and linearity of optical electromagnetic waves, which for most practical purposes can be considered unaffected by other physical parameters such as temperature and pressure.
A.2.3 Well-Defined Directional Response

The quantity measured by the laser Doppler method is the projection of the velocity vector on the measuring direction defined by the optical system (a true cosine response). The angular response is thus unambiguously defined.
A.2.4 High Spatial and Temporal Resolution

The optics of the laser anemometer are able to define a very small measuring volume and thus provides good spatial resolution and yield a local measurement of velocity. The small measuring volume, in combination with fast signal processing electronics, also permits high bandwidth, time-resolved measurements of fluctuating velocities, providing excellent temporal resolution. Usually the temporal resolution is limited by the concentration of seeding rather than the measuring equipment itself.
This appendix is greatly based on the article from Jensen (2004).
A.1
A.2.5 Multi-Component Bi-Directional Measurements

Combinations of laser anemometer systems with component separation based on colour, polarization or frequency shift allow one-, two- or three-component LDA systems to be put together based on common optical modules. Acoustic-optical frequency shift allows measurement of reversing flow velocities.
A.3
Principles of LDA
A.3.1 Laser Beam

The special properties of the gas laser, making it so well suited for the measurement of many mechanical properties, are the spatial and temporal coherence. At all cross sections along the laser beam, the intensity has a Gaussian distribution, and the width of the beam is usually defined by the edge-intensity being 1/e2=13% of the coreintensity. At one point the cross section attains its smallest value, and the laser beam is uniquely described by the size and position of this so-called beam waist. With a known wavelength of the laser light, the laser beam is uniquely described by the size d0 and position of the beam waist as shown in Figure A.1. With z describing the distance from the beam waist, the following formulas apply:
Figure A.1: Laser beam with Gaussian intensity distribution
4 d0
4 z 1+ 2 d0
2
A. 1
d ( z ) = d0
A. 2
d 2 2 R ( z ) = z 1 + 0 4 z
A. 3
The beam divergence is much smaller than indicated in Figure A.1, and visually the laser beam appears to be straight and of constant thickness. It is important however to understand, that this is not the case, since measurements should take place in the beam waist to get optimal performance from any LDA-equipment. This is due to the wave fronts being straight in the beam waist and curved elsewhere. According to the previous equations, the wave front radius approaches infinity for z approaching zero, meaning that the wave fronts are approximately straight in the immediate vicinity of
A.2
Appendix A: Laser Doppler Anemometry the beam waist, thus letting us apply the theory of plane waves and greatly simplify calculations.
A.3.2 Doppler Effect

Laser Doppler Anemometry utilizes the Doppler effect to measure instantaneous particle velocities. When particles suspended in a flow are illuminated with a laser beam, the frequency of the light scattered (and/or refracted) from the particles is different from that of the incident beam. This difference in frequency, called the Doppler shift, is linearly proportional to the particle velocity.
Figure A.2: Light scattering from a moving seeding particle
The principle is illustrated in Figure A.2 where the vector U represents the particle r r velocity, and the unit vectors ei and es describe the direction of incoming and scattered light respectively. According to the Lorenz-Mie scattering theory, the light is scattered in all directions at once, but we consider only the light reflected in the direction of the LDA receiver. The incoming light has the velocity c and the frequency fi, but due to the particle movement, the seeding particle sees a different frequency, fs, which is scattered towards the receiver. From the receivers point of view, the seeding particle acts as a moving transmitter, and the movement introduces additional Doppler-shift in the frequency of the light reaching the receiver. Using Doppler-theory, the frequency of the light reaching the receiver can be calculated as:
r r 1 ei f s = fi r r 1 es
(Ur c ) (U c )
r
A. 4
Even for supersonic flows the seeding particle velocity U is much lower than the speed of light, meaning that U / c << 1 . Taking advantage of this, the above expression can be linearized to:
r U r r f r r r f s fi 1 + ( es ei ) = fi + i U ( es ei ) = fi + f c c
r
A. 5
With the particle velocity U being the only unknown parameter, then in principle the particle velocity can be determined from measurements of the Doppler shift f. In practice this frequency change can only be measured directly for very high particle velocities (using a Fabry Perot interferometer). This is why in the commonly employed fringe mode, the LDA is implemented by splitting a laser beam to have two Experimental and Numerical Study of an Axial Flow Pump A.3
Appendix A: Laser Doppler Anemometry beams intersect at a common point so that light scattered from two intersecting laser beams is mixed, as illustrated in Figure A.3. In this way both incoming laser beams are scattered towards the receiver, but with slightly different frequencies due to the different angles of the two laser beams. When two wave trains of slightly different frequency are super-imposed we get the well-known phenomenon of a beat frequency due to the two waves intermittently interfering with each other constructively and destructively. The beat frequency corresponds to the difference between the two wave frequencies, and since the two incoming waves originate from the same laser, they also have the same frequency, f1 = f2 = fI, where the subscript I refers to the incident light:
Figure A.3: LDA Set-up left schematic shows beam-splitter (BS) arrangement for creating two separate beams
f D = f s ,2 f s ,1 r r U r r U r r = f 2 1 + ( es e2 ) f1 1 + ( es e1 ) c c r U r r = f I ( e1 e2 ) c r f r r = I e1 e2 . U .cos ( ) c =
A. 6
.2.sin ( / 2 ) .u x =
2sin( / 2
ux
Where is the angle between the incoming laser beams and is the angle between r r the velocity vector U and the direction of measurement. Note that the unit vector es has dropped out of the calculation, meaning that the position of the receiver has no direct influence on the frequency measured. (According to the Lorenz-Mie light scattering theory, the position of the receiver will however have considerable influence on signal strength). The beat-frequency, also called the Doppler-frequency fD, is much lower than the frequency of the light itself, and it can be measured as fluctuations in the intensity of the light reflected from the seeding particle. As shown in the previous equation the Doppler-frequency is directly proportional to the xcomponent of the particle velocity, and thus can be calculated directly from fD:
ux =
fD 2sin( / 2)
A. 7
A.4
Appendix A: Laser Doppler Anemometry Further discussion on LDA theory and different modes of operation may be found in the classic texts of Durst et al (1976) and Albrecht et al (2003).
A.3.3 Calibration
The LDA measurement principle is given by the relation Vx = df .fd where Vx is the component of velocity in the plane of the laser beams, and perpendicular to their bisector, df is the distance between fringes, and fd is the Doppler frequency. The fringe spacing is a function of the distance between the two beams on the front lens and the focal length of the lens, given by the relation df = /[2sin(/2)] where = laser wavelength and = beam crossing angle. Since df is a constant for a given optical system, there is a linear relation between the Doppler frequency and velocity. The calibration factor, i.e. the fringe spacing, is constant, calculable from the optical parameters, and mostly unaffected by other changing variables in the experiment. Hence, the LDA requires no physical calibration prior to use.
A.4
Implementation
A.4.1 The Fringe Model

Although the above description of LDA is accurate, it may be intuitively difficult to quantify. To handle this, the fringe model is commonly used in LDA as a reasonably simple visualization producing the correct results. When two coherent laser beams intersect, they will interfere in the volume of the intersection. If the beams intersect in their respective beam waists, the wave fronts are approximately plane, and consequently the interference will produce parallel planes of light and darkness as shown in Figure A.4. The interference planes are known as fringes, and the distance f between them depends on the wavelength and the angle between the incident beams:
Figure A.4: Fringes at the point of intersection of two coherent beams
f =
2sin( / 2)
A. 8
The fringes are oriented normal to the x-axis, so the intensity of light reflected from a particle moving through the measuring volume will vary with a frequency proportional to the x-component, ux, of the particle velocity:
A.5
Appendix A: Laser Doppler Anemometry 2sin( / 2)
fD =
ux
ux
A. 9
If the two laser beams do not intersect at the beam waists but elsewhere in the beams, the wave fronts will be curved rather than plane, and as a result the fringe spacing will not be constant but depend on the position within the intersection volume. As a consequence, the measured Doppler frequency will also depend on the particle position, and as such it will no longer be directly proportional to the particle velocity, hence resulting in a velocity bias.
A.4.2 Measuring Volume

Measurements take place in the intersection between the two incident focused laser beams, and the measuring volume is defined as the volume within which the modulation depth is higher than e-2 times the peak core value. Due to the Gaussian intensity distribution in the beams the measuring volume is an ellipsoid as indicated in Figure A.5.
Figure A.5: LDA measurement volume
z =
4F EDL sin 2 A. 10
y =
4F EDL
A. 11
x =
4F EDL cos 2 A. 12
where F is the lens focal length, E is the beam expansion (see Figure A.6), and DL is the initial beam thickness (e-2).
A.6
Figure A.6: shows the use of a beam expander to increase beam separation prior to focusing at a common point
As important are the fringe separation and number of fringes in the measurement volume. These are given by:
f =
2sin 2 A. 13
8.F .tan 2 Nf = EDL
A. 14
This number of fringes Nf applies for a seeding particle moving straight through the centre of the measuring volume along the x-axis. If the particle passes through the outskirts of the measuring volume, it will pass fewer fringes, and consequently there will be fewer periods in the recorded signal from which to estimate the Doppler frequency. To get good results from the LDA-equipment, one should ensure a sufficiently high number of fringes in the measuring volume. Typical LDA set-ups produce between 10 and 100 fringes, but in some cases reasonable results may be obtained with less, depending on the electronics or technique used to determine the frequency. The key issue here is the number of periods produced in the oscillating intensity of the reflected light, and while modern processors using FFT technology can estimate article velocity from as little as one period, the accuracy will improve with more periods.
A.4.3 Backscatter Versus Forward Scatter

A typical LDA set-up in the so-called backscatter mode is shown in Figure A.7. The figure also shows the important components of a modern commercial LDA system. The majority of light from commonly used seeding particles is scattered in a direction away from the transmitting laser, and in the early days of LDA, forward scattering was thus commonly used, meaning that the receiving optics was positioned opposite of the transmitting aperture.
A.7
Figure A.7: Mie scatter diagram for different size seeding particles dp is particle size
Figure A.7 shows the mie scatter function for three different particle sizes dp. From the left size < , centre size = and right size > . Most LDA measurements are performed using seeding particles that are larger than the wavelength of the laser light of most lasers. Figure A.7 illustrates that a much smaller amount of light is scattered back towards the transmitter, but advances in technology has made it possible to make reliable measurements even on these faint signals, and today backward scatter is the usual choice in LDA. This so-called backscatter LDA allows for the integration of transmitting and receiving optics in a common housing (as seen in Figure A.8), saving the user a lot of tedious and timeconsuming work aligning separate units.
Figure A.8: Schematic of components for a modern backscatter LDA system
Forward scattering LDA is however, not completely obsolete since in some cases the improved signal-to-noise ratio make forward-scatter the only way to obtain measurements at all. Experiments requiring forward scatter might include: Experimental and Numerical Study of an Axial Flow Pump A.8
Appendix A: Laser Doppler Anemometry High speed flows, requiring very small seeding particles, which stay in the measuring volume for a very short time, and thus receive and scatter a very limited number of photons. Transient phenomena, which require high data-rates in order to collect a reasonable amount of data over a very short period of time. Very low turbulence intensities, where the turbulent fluctuations might drown in noise, if measured with backscatter LDA.
Forward and back scattering is identified by the position of the receiving aperture relative to the transmitting optics. Another option is off-axis scattering, where the receiver is looking at the measuring volume at an angle. Like forward scattering this approach requires a separate receiver, and thus involves careful alignment of the different units, but it helps to mitigate an intrinsic problem present in both forward and backscatter LDA. As indicated in Figure A.5 the measuring volume is an ellipsoid, and usually the major axis z is much bigger than the two minor axes x and y rendering the measuring volume more or less cigar-shaped. This makes forward and backscattering LDA sensitive to velocity gradients within the measuring volume, and in many cases also disturbs measurements near surfaces due to reflection of the laser beams.
Figure A.9: Off-axis scattering
Figure A.9 illustrates how off-axis scattering reduces the effective size of the measuring volume. Seeding particles passing through either end of the measuring volume will be ignored since they are out of focus, and as such contribute to background noise rather than to the actual signal. This reduces the sensitivity to velocity gradients within the measuring volume, and the off-axis position of the receiver automatically reduces problems with reflection. These properties make offaxis scattering LDA very efficient for example in boundary layer or near- surface measurements.
A.4.4 Optics
In modern LDA equipment the light from the beam splitter and the Bragg cell is sent through optical fibres, as is the light scattered back from seeding particles. This reduces the size and the weight of the probe itself, making the equipment flexible and easier to use in practical measurements. A photograph of a pair of commercially available LDA probes is shown in Figure A.10. The laser, beam splitter, Bragg cell and photo detector (receiver) can be installed stationary and out of the way, while the LDA-probe can be traversed between different measuring positions. Experimental and Numerical Study of an Axial Flow Pump A.9
Figure A.10: Photograph of modern commercial fibre optic based LDA probes
It is normally desired to make the measuring volume as small as possible, which according to the formulas governing the measurement volume means that, the beam waist df = 4F EDL A. 15
should be small. The laser wavelength is a fixed parameter, and focal length F is normally limited by the geometry of the model being investigated. Some lasers allow for adjustment of the beam waist position, but the beam waist diameter DL is normally fixed. This leaves beam expansion as the only remaining way to reduce the size of the measuring volume. When no beam expander is installed, then E = 1. A beam expander E is a combination of lenses in front of or replacing the front lens of a conventional LDA system. It converts the beams exiting the optical system to beams of greater width. At the same time the spacing between the two laser beams is increased, since the beam expander also increases the aperture. Provided the focal length F remains unchanged, the larger beam spacing will thus increase the angle between the two beams. According to the formulas in governing the measurement volume this will further reduce the size of the measuring volume. In agreement with the fundamental principles of wave theory, a larger aperture is able to focus a beam to a smaller spot size and hence generate greater light intensity from the scattering particles. At the same time the greater receiver aperture is able to pick up more of the reflected light. As a result the benefits of the beam expander are threefold: Reduce the size of the measuring volume at a given measuring distance. Improve signal-to-noise ratio at a given measuring distance, Or reach greater measuring distances without sacrificing signal-to-noise.
A.4.5 Frequency Shift

A drawback of the LDA-technique described so far is that negative velocities ux < 0 will produce negative frequencies fD < 0. However, the receiver cannot distinguish between positive and negative frequencies, and as such, there will be a directional ambiguity in the measured velocities.
A.10
Appendix A: Laser Doppler Anemometry To handle this problem, a Bragg cell is introduced in the path of one of the laser beams (as shown in Figure A.3). The Bragg cell shown in Figure A.11 is a block of glass. On one side, an electromechanical transducer driven by an oscillator produces an acoustic wave propagating through the block generating a periodic moving pattern of high and low density. The opposite side of the block is shaped to minimize reflection of the acoustic wave and is attached to a material absorbing the acoustic energy.
Figure A.11: Principles of operation of a Bragg cell
The incident light beam hits a series of travelling wave fronts, which act as a thick diffraction grating. Interference of the light scattered by each acoustic wave front causes intensity maximal to be emitted in a series of directions. By adjusting the acoustic signal intensity and the tilt angle B of the Bragg cell, the intensity balance between the direct beam and the first order of diffraction can be adjusted. In modern LDA-equipment this is exploited, using the Bragg cell itself as the beam splitter. Not only does this eliminate the need for a separate beam splitter, but it also improves the overall efficiency of the light transmitting optics, since more than 90% of the lasing energy can be made to reach the measuring volume, effectively increasing the signal strength. The Bragg cell adds a fixed frequency shift f0 to the diffracted beam, which then results in a measured frequency off a moving particle of f D f0 + 2sin ( 2 )
ux
A. 16
and as long as the particle velocity does not introduce a negative frequency shift numerically larger than f0, the Bragg cell with thus ensure a measurable positive Doppler frequency fD. In other words the frequency shift f0 allows measurement of velocities down to
ux >
f0 2sin ( 2 )
A. 17
without directional ambiguity. Typical values might be = 500 nm, f0 = 40 MHz, = 20o, allowing of negative velocity components down to ux > -57.6 m/s. Upwards the maximum measurable velocity is limited only by the response-time of the photomultiplier and the signal-conditioning electronics. In modern commercial LDA equipment, such a maximum is well into the supersonic velocity regime. Experimental and Numerical Study of an Axial Flow Pump A.11
f D = f0 +
2sin ( 2 )
ux
A. 18
Figure A.12: Resolving directional ambiguity using frequency shift
A.4.6 Signal Processing

The primary result of a laser anemometer measurement is a current pulse from the photo detector. This current contains the frequency information relating to the velocity to be measured. The photocurrent also contains noise, with sources for this noise being: Photo detection shot noise Secondary electronic noise Thermal noise from preamplifier circuit Higher order laser modes (optical noise) Light scattered from outside the measurement volume, dirt, scratched windows, ambient light, multiple particles, etc. Unwanted reflections (windows, lenses, mirrors, etc).
The primary source of noise is the photo detection shot noise, which is a fundamental property of the detection process. The interaction between the optical field and the photosensitive material is a quantum process, which unavoidably impresses a certain amount of fluctuation on the mean photocurrent. In addition there is mean photocurrent and shot noise from undesired light reaching the photo detector. Much of the design effort for the optical system is aimed at reducing the amount of unwanted reflected laser light or ambient light reaching the detector. A laser anemometer is most advantageously operated under such circumstances that the shot noise in the signal is the predominant noise source. This shot noise limited performance can be obtained by proper selection of laser power, seeding particle size and optical system parameters. In addition, noise should be minimized by selecting only the minimum bandwidth needed for measuring the desired velocity range by setting low-pass and high-pass filters in the signal processor input. Very important for the quality of the signal, and the performance of the signal processor, is the number of seeding particles present simultaneously in the measuring volume. If on average much less than one particle is present in the volume, we speak of a burst-type Doppler signal. Typical Doppler burst signals are shown in Figure A.13.
A.12
Figure A.13: Doppler burst
Figure A.14: Filtered Doppler burst
Figure A.14 shows the filtered signal, which is actually input to the signal processor. The DC-part, which was removed by the high-pass filter, is known as the Doppler Pedestal. The envelope of the Doppler modulated current reflects the Gaussian intensity distribution in the measuring volume. If more particles are present in the measuring volume simultaneously, we speak of a multi-particle signal. The detector current is the sum of the current bursts from each individual particle within the illuminated region. Since the particles are located randomly in space, the individual current contributions are added with random phases, and the resulting Doppler signal envelope and phase will fluctuate. Most LDAprocessors are designed for single particle bursts, and with a multi-particle signal, they will normally estimate the velocity as a weighted average of the particles within the measuring volume. One should be aware however, that the random phase fluctuations of the multi-particle LDA signal adds a phase noise to the detected Doppler frequency, which is very difficult to remove. To better estimate the Doppler frequency of noisy signals, frequency domain processing techniques are used. With the advent of fast digital electronics, the Fast Fourier Transform of digitised Doppler signals can now be performed at a very high rate (100s of kHz). The power spectrum S of a discretized Doppler signal x is given by
k =
N 1 n =0
exp{ f 2kn / 2 N }
A. 19
where N is the number of discrete samples, and K = -N, -N + 1, , N 1. The peak of the spectrum gives the Doppler frequency.
A.4.7 Data Analysis

In LDA there are two major problems faced when making a statistical analysis of the measurement data, velocity bias and the random arrival of seeding particles to the measuring volume. While velocity bias is the predominant problem for simple statistics, such as mean and rms values, the random sampling is the main problem for statistical quantities that depend on the timing of events, such as spectrum and correlation functions.
A.13
Appendix A: Laser Doppler Anemometry Table A.1 illustrates the calculation of moments, on the basis of measurements received from the processor. The velocity data coming from the processor consists of N validated bursts, collected during the time T, in a flow with the integral time scale I. For each burst the arrival time ai and the transit time ti of the seeding particle is recorded along with the non-Cartesian velocity components (ui, vi, wi). The different topics involved in the analysis are described in more detail in the open literature, and will be touched upon briefly in the following section.
A.5
Making Measurements
A.5.1 Dealing With Multiple Probes (3D Set-up)

The non-Cartesian velocity components (u1, u2, u3) are transformed to Cartesian coordinates (u, v, w) using the transformation matrix C:
u C11 C12 v = C21 C22 w C 31 C32 C13 u1 C23 . u2 C33 u3
A. 20
A typical 3-D LDA set-up requiring coordinate transformation is shown in Figure A.15 where 3-dimensional velocity measurements are performed with a 2-D probe positioned at off-axis angle 1 and a 1- D probe positioned at off-axis angle 2. The transformation for this case is: 1 0 u sin 2 v = 0 w sin(1 2 ) cos 2 0 sin(1 2 ) 0 u1 sin 1 . u2 sin(1 2 ) u3 cos 1 sin(1 2 )
A. 21
Figure A.15: Filtered Doppler burst
A.14
A.5.2 Calculating Moments

Moments are the simplest form of statistics that can be calculated for a set of data. The calculations are based on individual samples, and the possible relations between samples are ignored, as is the timing of events. This leads to moments sometimes being referred to as one-time statistics, since samples are treated one at a time.
Table A.1: Estimation of moments
Table A.1 lists the formulas used to estimate the moments. The table operates with velocity-components xi and yi, but this is just examples, and could of course be any velocity component, Cartesian or not. It could even be samples of an external signal representing pressure, temperature or something else.
A.5.3 Velocity Bias and Weighting Factor

Even for incompressible flows where the seeding particles are statistically uniformly distributed, the sampling process is not independent of the process being sampled (that is the velocity field). Measurements have shown that the particle arrival rate and the flow field are strongly correlated (McLaughlin (1973) and Erdmann (1976)). During periods of higher velocity, a larger volume of fluid is swept through the measuring volume, and consequently a greater number of velocity samples will be recorded. As a direct result, an attempt to evaluate the statistics of the flow field using arithmetic averaging will bias the results in favour of the higher velocities. There are several ways to deal with this issue: Ensure statistically independent samples - the time between bursts must exceed the integral time-scale of the flow field at least by a factor of two. Then the weighting factor corresponds to the arithmetic mean, i = 1/N. Statistically independent samples can be accomplished by using very low concentration of seeding particles in the fluid. Use dead-time mode - The dead time is a specified period of time after each detected Doppler-burst, during which further bursts will be ignored. Setting the dead-time equal to two times the integral time scale will ensure statistically independent Experimental and Numerical Study of an Axial Flow Pump A.15
Appendix A: Laser Doppler Anemometry samples, while the integral time-scale itself can be estimated from a previous series of velocity samples, recorded with the dead-time feature switched off. Use bias correction - If one plans to calculate correlations and spectra on the basis of measurements performed, the resolution achievable will be greatly reduced by the low data-rates required to ensure statistically independent samples. To improve the resolution of the spectra, a higher data rate is needed, which as explained above will bias the estimated average velocity. To correct this velocity-bias, a non-uniform weighting factor is introduced:
i =
ti N 1
j =1
tj
A. 22
The bias-free method of performing the statistical averages on individual realizations uses the transit time, ti , weighting.
A.16

B.1 Introduction
A detailed measurement uncertainty analysis is important when experimental results are used for computational method validation, or when they are compared between various facilities, especially when the data provide insight into vortex formation and evolution of the vortices trailed from the rotor to the stator blade.
B.2
Sources of errors
The experimental uncertainties are well analysed by the methods described by Kline (1985), Abernathy et al (1985) and Mofat (1985). In their study, the biasing errors and other error sources such as the electronic set-up and the aberrations in the probe volume are not included. In fact, it is very difficult to take all these possible error sources into account. An analysis of the various factors, which influence the overall accuracy in LDA measurements, can be found in Boutier (1991). In this present work, the experimental uncertainty levels were estimated by using the methods, which are already mentioned in the literature. We will be interested to estimate the errors of the mean values. These errors can come from several origins, which are considered independently: Errors in the geometry of the propeller pump, leading to a non-axisymmetry Errors of the rotation speed of the axial pump Errors in the estimation of the instantaneous velocities by the LDA system Errors in the estimation of the mean quantities Spatial resolution of the measurement volume
B.2.1 Errors in the geometry of the propeller pump

The first source of errors, coming from the geometry of the pump, is the one that we want to estimate. Both the stator and rotor blades have been machined with CAD/CAM techniques using the Flygt Computer vision computer connected to a 3axis NC-milling machine. The maximum geometrical inaccuracy of the blade shapes is probably less that 0.1 mm. The inaccuracy of the blade general position is less than 0.7 mm. The cycled velocity profiles, measured for a complete resolution, provide a check of the axisymmetry of the propeller pump and this is done in chapter 4.
B.2.2 Errors of the rotation speed of the axial pump

The errors of the rotation speed of propeller pump can be controlled using the incremental shaft encoder recording the rotation speed of the compressor, updated every 0.25 seconds, with a mean speed given with a precision of 0.1 rpm. We chose for the measurements a rotation speed of 750 rpm. Due to fluctuations in the electrical power supply, or to mechanical defaults (belt driving), the compressor rotation speed presents some unavoidable fluctuations. Using a (very sensitive) potentiometer, this Experimental and Numerical Study of an Axial Flow Pump B.1
Appendix B: Experimental Uncertainties rotation speed was maintained inside the interval (745, 755), which corresponds to an error of 0.5%.
B.2.3 Error of the mean velocities by the LDA system

Below we will try to estimate the errors of the mean velocities. At a specific value of the blade azimuth , the phase-resolved velocity V is reported as a mean value V and an uncertainty V at the 95% confidence level, i.e., V ( x, y, z , ) = V V B.1
For clarity, the prime notation of the previous section has been dropped, and it is understood that the velocity in this section is the phase-resolved velocity. The objective of this section is to rigorously determine the measurement uncertainty in the velocity according V to the established industry standards. The velocity, V , is the product of a transformation matrix and a vector of Doppler frequencies, i.e., u V ( x, y, z, ) = = [ A][ F ] v B.2
The transformation matrix is based purely on the geometric configuration of the optics
1 [ A] = 2sin 1 0 .k 2 2sin 2 0
B.3
where is the wavelength, is the half angle of the beams and k is the optical correction factor due to the different mediums. Channels 1 and 2 correspond to green and blue respectively. Details of the optics configuration are given in chapter 3 and appendix A. The Doppler frequency is the difference between the measured burst frequency, , and the shift frequency setting, s, for each channel, i.e.,
[ F ] = f1 = v1 s1
v s
2 2
B.4
Under the assumptions of the fringe model, LDV is based on a direct measurement of , the burst frequency of light scattered from seed particles passing through the fringe pattern of the probe volume. The measurement uncertainty analysis is divided into three steps: 1. Calibration uncertainty: The study of sources of uncertainty in the alignment of the optics [A] 2. Data acquisition uncertainty: The study of the sources of uncertainty in the frequency measurements [F]
B.2
Appendix B: Experimental Uncertainties 3. Data reduction uncertainty: The study of the propagation of the sources of uncertainty [A] and [F] to the final result. For each source, the total uncertainty, , is a function of the source bias limit, B, the source precision limit, P, and a student t-factor, tv,95, i.e.,
= B 2 + ( tv ,95 P )
B.5
The source limits are a combination of elemental types of bias and precision uncertainty (Abernathy, 1985).
B=
Bi2
i =1
N
B.6
P=
Pi2
i =1
B.7
The student t-factor is found easily in statistics books and for a confidence level of 95% tv,95 is 1.96. It is assumed that the sources of uncertainty are random and are independent errors. This allows the sources to be combined in quadrate according to the Kline- McClintock second power law (Figliola et al, 1991). For a generic function g, of N independent variables, xi, we have
g = g ( x1 , x2 ,...., xi ,....., xN )
and the uncertainty is given by
B.8
g 2 g = ( xi ) i =1 xi
N
B.9
B.2.3.1 Calibration Uncertainty

The total uncertainty in the alignment of the optics, [ A] , can be decomposed into two primary source groups. The first group consists of elemental sources contributing to the uncertainty in the fringe spacing, df. The second group consists of elemental sources contributing to the uncertainty in optical correction factor k. The fringe spacing is determined by measuring the half angle of the beam crossing, , using
df =
2sin
B.10
where is the wavelength of the beam. The distance of the focus length and the beam separators determines the transmitting angle. Using equations B.9 and B.10, the uncertainty in the fringe spacing is found to be
B.3
d f
df
2 1 . tg 2 .x 2 + y 2 sin 2 L
B.11
where L is the focus length. The uncertainty of the wavelength is not known and is not included in the calculation of the uncertainty. The error in fringe spacing estimation is due to the difficulty in measuring the crossing angle of the beams accurately and depends on the specific alignment technique. The estimated beam crossing section df/df is 0.026 % for the focus length of 300 mm and a beam separator of 78 mm. The uncertainty of the correction factor k is even difficult to estimate because of the different refraction index of the fluid. The composition of rainwater will vary during the period of measurement. The refraction of water varies between 1.331 and 1.334 so that the uncertainty of the correction factor k is k/k=0.22 %.
B.2.3.2 Data Acquisition Uncertainty

The objective of this section is to estimate the total uncertainty in the Doppler frequency, f. This is approximated using equation B.4 as
f
f
v
v
B.12
The uncertainty in the frequency shift setting is neglected based on the specifications for the Bragg cell. Using equation B.5, the total uncertainty in the frequency measurement is
v = Bv2 + ( tv ,95 Pv )
B.13
which is a combination of a source bias limit, Bv, and a source precision limit, Pv? The source bias limit is based on several elements, namely
2 2 2 2 2 Bv = B 2 f + BV + Bg + Ba + Bs + Bw
B.14
where the terms under the radical are defined as fringe bias, velocity bias, gradient bias, angle bias, seeding bias, and filter window bias, respectively. For a complex flow field such as a rotor wake, it is very important to estimate and minimize the frequency source bias limit, B. For this reason, each LDV bias source warrants its own discussion and will be presented later. The remainder of this section estimates the source precision limit. The source precision limit, P, is based on the phase resolution, res, the standard deviation of the frequency measurement, , and the frequency resolution, res , i.e.,
2 2 v res v resv Pv = + + 2 N v 1 2 2
B.15
B.4
Appendix B: Experimental Uncertainties The first term under the radical is unique to phase-resolved LDV measurements. At a point in space, the velocity varies periodically with the phase () of the rotor blade. The source of this uncertainty is that a finite resolution is used to discretize a continuous velocity signature (see paragraph 3.3). It is important to note that in practice, improving the phase resolution (decreasing res) tends to reduce the number of samples Nv collected during a given sampling period. This is especially true in the case of the rotor wake where flow seeding becomes difficult. To minimize the source precision limit, preliminary measurements are required to understand the seeding issues and to estimate the frequency gradient v . Another reason is that the source precision limit is highly depending on the critical flow features. For example, a vortex core convecting along the rotor wake boundary is characterized by a seed void (low Nv), a substantial turbulence level (high v) and flow reversal (high v ). Each of these contributions increases the source precision limit solely on the basis of the inherent flow physics. Figure B.1 shows examples of evolution of the velocity as a function of the blade azimuth measurement.
Figure B. 1 Tangential velocity profile at 75% chord axially of the tip leading edge of the rotor and 65 % span.
This allows an estimate of the gradient v to be made as a function of the blade azimuthal. The frequency resolution of the autocorrelation based digital burst correlator is resv / v = 0.05%. Figure B.2 shows the final result of the distribution of the source precision term P of equation B.13 across the passage vortex for a single LDV channel.
B.5
Figure B. 2 Frequency source precision
B.2.4 Sources of the LDV Frequency Bias

Laser Doppler anemometers are amongst the most accurate flow measurement devices. However, they are not immune to errors and, as with any other measurement technique, it is important to know the sources of errors when making an LDA measurement. The reported uncertainty in LDV measurements must include a quantitative estimation and discussion of sources of frequency bias found in equation B.14. A complete discussion of filter window bias, Bw and the angle bias is represented in Edwards (1987). The remaining sources of bias will be treated in the following sections.
B.2.4.1 Processor bias
The instantaneous velocities recorded by the LDA system are proportional to the Doppler frequency shift, which is estimated through a Fourier transform. Since this law is linear and the proportionality factor is fixed, the LDA apparatus doesnt need any calibration. The burst spectrum analyser, used to process the Doppler signal, received from the probe volume a high resolution when only valid data are transmitted to the output buffer, meaning that the signal to noise ratio is high. The Doppler frequency is estimated with different resolution, depending on the record length: 14 bits resolution when the record length is 64, 13 bits when the record length is 32, 12 when 16, 11 bits when 8. When introducing a fixed proportionality factor (see Hazarika and Hirsch, 1998 for details), the error is between 0.05 % and 0.006 %, depending on the record length. Since we used a record length of 64 for both beams (axial and radial velocity), we have measurements with errors of 0.006 %. This is thus an infinitesimal source of error compared to the other sources.
B.6
B.2.4.2 Fringe bias

Theoretically, the fringe pattern is always considered as a set of parallel and equidistant fringe, because we assume that two plane waves interfere inside the probe volume. The wave fronts are planes only if the beams are well focused. This is why the beam waists of the crossing beams must be accurately set at the crossing location of the beams, i.e. the probe volume. When the beams are not perfectly focused where they cross, the fringe pattern is formed of a series of hyperbolas and the variations of the fringe spacing inside the probe volume may rapidly reach 1 % or more, which is thus considered as turbulence (which is not true). Even experiments have proved that the LDA signal of a tracer particle passing the centre of the measuring volume can show period length variations, if, at the same time, at least a second particle is present in the measuring volume or in a laser beam near the volume (Ruck, 1991). The fringe pattern distortion is caused by diffraction, provided that the disturbing particle is located between the light source and the signal-producing particle. It could be shown that with increasing number of simultaneously illuminated particles (multiple scattering), the distortion affects larger portions of the measuring volume. Whereas the local fringe distortion seems to have no significant influence on the LDA mean velocity determination, turbulence measurements may be affected. The scatter in fringe spacing due to simultaneously illuminated particles inside or in the vicinity of the measuring volume can induce a broadening of the velocity probability distribution, which leads to a higher measured turbulence level. A theoretical indicator of the uniformity of fringe spacing in the measurement volume has been given by Zhang & Eisele (1998) in terms of a fringe distortion number. Fringe distortion is proportional to the distance between the point of beam crossing and the beam waist. It has been shown by Zhang and Eisele (1998) that the fringe distortion in LDA measurement volume usually results in a negligible overestimation of both the mean velocity and the flow turbulence. Only in flows with very low turbulence intensity dose the error in estimating the turbulence become significant, what is not the case in propeller pumps.
B.2.4.3 Velocity Bias

The source of this form of bias is the dependency of the rate of seed particles travelling through the probe volume on the flow velocity. The result is a non-uniform sampling of the distribution of velocity around a mean value. Higher velocity particles are sampled more often than low velocity particles, and so the mean value of the distribution becomes biased. This source of bias is increased by the level of turbulence and can become severe when the measurement also suffers from poor spatial resolution. For phase resolved LDV a velocity bias correction is applied on a bin-by-bin basis using transit time weighting. The mean velocity at a given point in space and at a fixed blade azimuth is a weighted mean using
B.7
V=
Vi i
i =1 N
i
i =1
B.16
which is based on the transit time of the seed particle? Even though a bias correction is applied, the uncertainty still exists and can be estimated using
1 BV = V N Vi i =1
N 2
B.17
B.2.4.4 Velocity Gradient Bias

The presence of a velocity gradient across a finite length probe volume is the source of this form of bias. In the previous section, velocity bias assumes an infinitely small probe volume and involves the temporal variation of velocity at a fixed point in space. In this section, velocity gradient bias considers the effect of a finite probe volume in the presence of a spatial variation of velocity at a fixed time. This section of the paper is expanded in detail based on the recommendation of Edwards (1987). Historically, gradient bias has been one of the fundamental problems in the measurement of shear layers using a variety of instruments including Pitot tubes, hot wires, and LDV. The error in the measured mean velocity is a function of the spatial resolution of the instrument compared to the scale of the shear layer. Furthermore, the error is proportional to the velocity gradient and can be expressed in terms of
Lm ( u U 0 ) Lm L ( y L) L
B.18
where Lm is the spatial resolution, and L is the length scale of the shear layer. The frequency bias associated with LDV spatial resolution was theoretically investigated by Kreid (1974). For a given shear layer, the uncertainty grows with the square of the spatial resolution, that is
Bg 2
B.19
where is the ratio of the probe length to the vortex core radius. Kreid (1974) compares the estimation error in velocity measurement obtained by LDV with a spatially non-uniform velocity distribution. The errors will be usually negligible except near a bounding surface or in other regions where the flow profile has an extreme curvature. The error becomes increasingly larger when we approach the wall and for increasing values of . The results for turbulent flow, modelled by the universal turbulent velocity profile are represented in figure B.4. The results are represented for a representative range of values for the shear Reynolds number Re*, defined as Experimental and Numerical Study of an Axial Flow Pump B.8
Re * =
u* .b y
B.20
whereby u* is the shear velocity, by is the scattering volume dimension in the y direction (normal to the wall) and is the viscosity.
Figure B.3 Velocity error for turbulent flow (Edwards, 1987)
It is found that the magnitude of the error and the extent of the region where significant errors occur can be minimized by minimizing the size of the scattering volume. The errors would usually be insignificant for locations more than one volume radius from the wall. However very large errors can occur close to the wall, especially at high flow rates. The predicted errors predicted would usually be negligible, except very near to a bounding surface or in other regions where the flow profile has extreme curvature.
B.2.4.5 Seed Particles and Polydispersie-bias

One of the most important elements in laser velocimetry, yet the most neglected, is the small particle embedded in the flow field that scatters the light necessary to make velocity measurements. Since the laser velocimetry measures the velocity of small particles embedded in the flow and not the flow itself, measurement accuracy directly depends on the ability of these particles to faithfully follow the fluid flow. The characteristics of this slowly particle are often ignored in the effort to obtain data. This seems strange since it is the primary cause of measurement errors. If the particle is too large, it will not follow the flow resulting in an inaccurate representation of the Experimental and Numerical Study of an Axial Flow Pump B.9
Appendix B: Experimental Uncertainties fluid velocity. If the particle is too small, it will not scatter sufficient light to provide the signal-to-noise necessary to minimize the measurement uncertainty in the signal processing electronics. Choosing the proper seeding particle for laser velocimetry applications is a classic case of compromise. A smaller particle will more faithfully follow the fluid flow and will increase the measurement accuracy, while a larger particle will scatter more light and increase the signal strength, resulting in greater measurement precision. The chosen particle size is often determined by the ability of the optical system to see that particle, and the accepted inaccuracies. The sensitivity of the optical system is in turn constrained by facility, optical, and financial limitations. Several particle-seeding techniques for laser velocimetry applications have been presented by Meyers (1991). These techniques include atomizers and vaporization/condensation generators using liquids, and fluidised beds and liquid carriers for solid particles. All of these techniques have a major flaw: they generate polydisperse particle distributions. The test results in the 16-foot transonic tunnel clearly show the effects of these distributions on laser velocimetry measurements. Unknown bias errors in the mean velocity due to particle lag and artificial increases in measurement standard deviation may result in unacceptable measurements. If known mono disperse particles could be used, these problems would be avoided. Further, signal-to-noise would be increased since most noise present in the photo multiplier output is due to the small particles that scatter insufficient light to yield a measurable signal. The major categories of particle generation techniques for laser velocimetry applications have been presented by Meyers (1991). Techniques using liquids yield poly disperse particle size distributions that may vary in mean and standard deviation depending on the material, the generating technique, and the external environment. These techniques produce many small particles that do not scatter sufficient light to be measured, and add background light, which decreases the signal-to-noise of particles that can be measured. As illustrated in a companion lecture, low signal-to-noise will result in increased measurement standard deviation and a corresponding increase in statistical uncertainty of the mean velocity. Therefore liquid-seeding particles should be avoided. Solid particles such as aluminium oxide, aluminium silicate, titanium oxide, silicon carbonate, etc. are also poly disperse but typically narrower than liquid distributions. The best seeding particles are polystyrene because they are spherical, light (specific gravity of 1.05) with a high index of refraction (1.59), mono disperse and, when injected with water-ethanol, do not agglomerate. These techniques are interesting when the needed volume of seeding particle is small because of the cost of the particles. In our case, we needed seeding particles for a volume of 25.000 m3, and knowing that the price of the seeding particle is around 43 Euro/gr., the price is too expensive. Rainwater has enough seeding particles that scatter the light. The error is corrected simply by subtracting the extra variance from the measured value.
B.2.4.6 Coincident bias

When performing 2D or 3D laser velocimetry, a coincidence window is used to insure that signals coming from different processors (measuring different components) are created by the same particle, which allows determining the instantaneous velocity vector. A good criterion might be a short duration of this window so that the transit Experimental and Numerical Study of an Axial Flow Pump B.10
Appendix B: Experimental Uncertainties time of a particle through the probe volume is short. But as soon as the flow becomes turbulent, slower particles have a longer transit time and if the coincidence window is too small, signals coming from the same particle may be rejected. Therefore the coincidence window must not be set too small, but should be set in function of the highest turbulence frequency to be measured inside the flow and of the mean flow velocity. To conclude, the choice of a short coincidence window induces a bias towards high velocities, which must be avoided. For instance reducing the coincidence window may reduce the turbulence rate, which is not a good criterion, because the signal comes from a histogram truncation in the low velocity side. It is difficult to estimate the error due to the coincidence bias.
B.2.5 Data Reduction Uncertainty

Having presented the fundamental sources of uncertainty, A and F , this section presents the results of how these sources combine to the final uncertainty in velocity, i.e.,
V = A2 + F 2
The total measurement uncertainty in the two components of velocity is
B.21
Vz
f = + + 2 Vz 1 f1 tg 1
B.22
Vt
f = + + 2 Vt 2 f 2 tg 2
B.23
The non-dimensional velocity is
V =
u' R
B.24
and so the measurement uncertainty is
Vt
u R = + + Vt u R
2 2
B.25
A representative velocity profile with uncertainty bars as shown in figure B.4 illustrates why reporting a single number for uncertainty can be misleading. This is especially true when velocity profiles are used for CFD validation. The uncertainty increases dramatically near the boundary layer of the blade as a result of the small sample size associated with the seeding void and the increased standard deviation of the velocity associated in the turbulence layer boundary. This represents an important step in estimating the magnitude and cause of vortex and boundary layer measurement uncertainty. The uncertainties are done for 10 different measurement points and the average total error is 5.6%. Experimental and Numerical Study of an Axial Flow Pump B.11
Figure B.4 Tangential velocity profile with error bars.
B.3 Conclusion
In the procedure of quantifying LDV errors, different types of bias sources of uncertainty were discussed. When evaluating these uncertainties, many of these bias errors are shown to be negligible. Many of the remaining errors are relatively small compared with the precision errors. The effect of the seeding bias errors on the mean velocity uncertainty was also shown to be small. For the turbulence intensity measurements, many of the bias errors become much more important. For the present investigation, however, the individual and the total bias errors are not explicitly calculated.
B.12

Stephan Geerts

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stephan Geerts

Uploaded by

Copyright:

Available Formats

Department of Fluid Mechanics and Thermodynamics Faculty of Applied Sciences Vrije Universiteit Brussel Pleinlaan 2, B-1050 Brussel Belgium

Experimental and Numerical Study of An Axial Flow Pump

A dissertation submitted to obtain the degree of Doctor of Engineering Sciences

Chapter 2: Experimental Test Facility

Chapter 3: Applications of the Laser Doppler Anemometry Equipment

Chapter 4: Experimental Results of Axial Flow Pump

Chapter 5: Numerical simulations of the Axial Pump Flow

Chapter 6: Evaluation-Comparison of test data and CFD

Chapter 7: 3D Flow Structure

Summary and Conclusions Bibliography Appendix A: Laser Doppler Anemometry

Appendix B: Experimental Uncertainties

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

f fi fs G g Hideal Hstat Htot i,j,k K k L M m r n N Nf Ns Nw n nf ni nw P Pe

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

V V(t) Vn Vt,cor Vx Va Vf vz vt r w W wz wt x,y,x Xi y+

Beam divergence, off-axis angle, angle of absolute velocity Nomenclature.4

Experimental and Numerical Study of an Axial Flow Pump

ijk t t eff lam turb

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

Chapter 1: Literature Review

Chapter 1: Literature Review

1.2 CFD of Hydropumps

1.3 Flow Analysis of Axial Flow Pump

1.3.1 Two-dimensional Analysis

Figure 1.1: Axial flow pump

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

Chapter 1: Literature Review

Figure 1.2: Velocity diagrams for a compressor stage

d r r ( rmv ) =.Q. ( u 2 v2t -u1v1t ) dt

1.3.2 Three-Dimensional Analysis

Figure 1.3: Geometry of a meridional streamtube in a pump impeller (Brennen)

Experimental and Numerical Study of an Axial Flow Pump

Figure 1.4: A fluid element in radial equilibrium (Dixon, 1975)

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

in a plane (at x=xa) contains terms involving both the slope

Experimental and Numerical Study of an Axial Flow Pump

ui' u 'j 1 p D = +U j with x j Dt t x j xi

Experimental and Numerical Study of an Axial Flow Pump

1.4 Global Performance Characteristic

Experimental and Numerical Study of an Axial Flow Pump

Chapter 1: Literature Review

Figure 1.6: Schematic of noncavitating performance of an axial flow pump

Experimental and Numerical Study of an Axial Flow Pump

Experimental and Numerical Study of an Axial Flow Pump

Chapter 1: Literature Review

Figure 1.7: Physical mechanism for inception of rotating stall (Brennen).

1.5 Characteristics of the 3D Flow Field

1.5.1 Forces Active On the Flow

r r r r r r r p W r r + W .W = 2 W ( r ) + 2 W 1.11 t r r r r r Where 2 W is the Coriolis acceleration and ( r ) is the centripetal

Experimental and Numerical Study of an Axial Flow Pump

Chapter 1: Literature Review

1.5.3 Secondary Flow in Axial Flow Pumps

Experimental and Numerical Study of an Axial Flow Pump

Figure 1.9: Secondary low pattern, after Hawthorne (1955)

Experimental and Numerical Study of an Axial Flow Pump

Chapter 1: Literature Review

Figure 1.12: Secondary flow in bend (Vavra, 1960)

Experimental and Numerical Study of an Axial Flow Pump