evaluation of numerical methods for turbomachines based on
Transcription
evaluation of numerical methods for turbomachines based on
UNIVERSIDAD PONTIFICIA COMILLAS ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA (ICAI) INGENIERO SUPERIOR INDUSTRIAL PROYECTO FIN DE CARRERA EVALUATION OF NUMERICAL METHODS FOR TURBOMACHINES BASED ON EXPERIMENTAL DATA FROM A FRANCIS PUMP-TURBINE IN PUMP MODE. AUTOR: ANA FABA TORTOSA MADRID, junio de 2008 EVALUACIÓN DE MÉTODOS NUMÉRICOS PARA TURBOMÁQUINAS BASADO EN DATOS EXPERIMENTALES DE UNA BOMBA-TURBINA FRANCIS EN SENTIDO BOMBA Autor: Faba Tortosa, Ana. Director: Braun, Olivier. Entidad Colaboradora: École Polytecnique Fédérale de Lausanne. RESUMEN DEL PROYECTO El objetivo de este proyecto es la validación, a partir de su comparación con datos reales experimentales, de la capacidad de los métodos CFD para modelar el comportamiento del flujo dinámico para fenómenos transitorios en turbomáquinas hidráulicas. El caso estudiado concierne a una bomba-turbina Francis en sentido bomba. Un modelo a escala reducida de dicha máquina con una velocidad específica ν = 0.19 es el utilizado en las medidas de laboratorio. La turbomáquina está compuesta de 9 álabes en la rueda y 20 canales en el difusor. Dos puntos de operación distintos, uno de carga parcial y otro de elevada tasa de descarga, son descritos y analizados por métodos de simulación, en términos de fluctuaciones de presión a lo largo de los canales del estator. Estas fluctuaciones generadas por la interacción entre el rotor y el estator (RSI) y los fenómenos que afectan al comportamiento del flujo a causa de ello forman parte del estudio realizado en este proyecto. La simulación numérica de los flujos transitorios está representada gracias al paquete de software de ingeniería ANSYS CFX.11. para cuatro dominios de computación diferentes: 3 posibles dominios parciales y también la máquina completa. Aunque inicialmente se realizan análisis generales para los cuatro casos, tras la evaluación de la calidad de los resultados, se detalla la investigación para el último caso exclusivamente. Solamente un dominio de computación que incluya la máquina entera se considera que minimice los errores de CFD y que consiga así una simulación fiable. Esos resultados numéricos precisos y detallados son comparados con las medidas obtenidas en los test del laboratorio, con la intención de validar el método y determinar los principales fallos de los métodos CFD. Las medidas de presión en el marco estacionario, cuyos puntos suponen básicamente todo el núcleo del análisis, están tomadas mediante diminutos sensores de presión piezoresistivos, situados en diferentes puntos de dos de los veinte canales del distribuidor: el primero y el último. En general, se descubre muy buena concordancia entre la simulación y los resultados del laboratorio para el punto de operación con elevado caudal de descarga pero existen algunas diferencias para el de carga parcial. Las posibles causas y la descripción del comportamiento del flujo están desarrolladas a lo largo de este proyecto. Por ejemplo, torbellinos y pequeños remolinos pueden nacer debido a la falta de adaptación entre la geometría de la máquina y los triángulos de velocidad del agua en cargas parciales. Este tipo de fenómenos aleatorios e instantáneos crean un régimen no permanente y un grado de inestabilidad en el comportamiento del flujo que no llega a predecirse ni a simularse completamente bien con los actuales métodos de CFD. Este proyecto puede ofrecer una primera referencia de análisis para más detallados estudios en el futuro. De acuerdo con esto, las diferencias más relevantes entre los resultados de CFD y los datos experimentales pueden ayudar a advertir dónde y cómo los métodos numéricos pueden ser mejorados y corregidos. Por otra parte, como los la simulación en CFD ofrece resultados de la presión y todas las variables y propiedades del flujo en cualquier lugar y posición de toda la máquina, para cualquier canal y cualquier punto en la bomba-turbina, de la observación de los resultados que parezcan insospechados o más sorprendentes pueden ayudar a sugerir dónde colocar un nuevo sensor de presión para el próximo test de laboratorio. Para concluir, la comparación de ambos resultados puede ayudar a mejorar el desarrollo de las máquinas hidráulicas, tanto encontrando los puntos débiles de las máquinas actuales, sus riesgos y problemas de operación así como corrigiendo los métodos de CFD con la intención de mejorarlos y aplicarlos al diseño y rediseño de dichas máquinas EVALUATION OF NUMERICAL METHODS FOR TURBOMACHINES BASED ON EXPERIMENTAL DATA FROM A FRANCIS PUMP-TURBINE IN PUMP MODE The aim of this project is to validate, by comparison with real experimental data, the ability of CFD methods to model dynamic flow behaviour for unsteady phenomena in hydraulic turbomachines. The study case concerns a Francis pump-turbine in pump mode. A reduced-scale model ν = 0.19 is used for the laboratory measurements. It is composed of 9 runner blades and 20 diffuser channels. Two different operating points, one of partial load and another of high discharge, are described and analysed by simulation methods, in terms of pressure fluctuactions along the stator channels. These fluctuacions generated by Rotor Stator Interaction (RSI) and the phenomena that affect the flow behaviour caused by them are part of the study of this paper. The numerical simulation of the unsteady flow is performed with ANSYS CFX.11 for four computing domains: 3 possible partial domains and also the entire machine. Although previous general analysis are done for the four cases, after the evaluation of the quality of their results, detailed investigations are finally made for the last one exclusively. Only a computing domain of the entire machine is considered to be the best to minimize the CFD errors and to get a reliable simulation. These computer results are compared with laboratory measurements in order to to validate the method and to determine main CFD simulation failures. The pressure measurements in the stationary frame, whose points are the mainly analysis issue, are performed with piezoresitive miniature pressure sensors located in several locations at two of the 20 distributor channels: the first and the last one. Very good agreement between simulation and laboratory results is found in general for the high discharge operating point but some discrepancies are discovered for the partial load one. Posible causes and the description of the flow behaviour are included in this paper. For example, eddies and small whirls can be born due to the lack of fit between the machine geometry and water velocity triangles at partial load. This type of phenomena creates instability and unsteadiness that it is not completely well simulated by CFD methods. This project can provide a first reference analysis for more detailed studies in the future. According to this, most relevant differences between CFD results and experimental data can help to advice where and how the numerical method should be improved and checked. On the other hand, as CFD results are obtained for the whole machine, all around the channels and at all locations in the pump-turbine, unexpected or surprising results discovered by the observation of CFD analysis for any point would suggest to include a sensor at that location during the next laboratory experimental test. To sum up, the comparison of both results can help to improve the hydraulic machines development, by finding current machines weaknesses, risks and operating problems as well as by checking the CFD methods errors, in order to improve them for machine design and redesign. Table of Content TABLE OF CONTENTS 1 Introduction. ..................................................................................2 1.1 THE HYDRODYNA PROJECT.....................................................................3 1.2 Brief overview of hydro power sector situation. ......................................3 1.3 Introduction to Hydraulic Turbomachines. ................................... 7 1.3.1 Variables definition. Classification of Turbomachines ...... 7 1.3.2 Nomenclature............................................................................ 12 1.3.3 1.3.4 Power conversion and Balance............................................... 14 Euler equations ......................................................................... 18 1.3.5 1.3.6 Rotor-stator Interaction in Francis Pump-Turbines. .......... 22 Frequency Analysis. ................................................................. 26 2 Numerical Simulation and CFD Methods ............................ 28 2.1 Introduction to Computational Methods...................................... 28 2.2 CFD (Computational Fluid Dynamics) ......................................... 30 2.2.1 The Choice of the Physical Approach. .................................. 31 2.2.2 The mathematical model ......................................................... 32 2.2.3 2.2.4 2.3 2.4 2.5 3 4 The Discretization Method..................................................... 33 The implementation of numerical algorithms and Solution. 36 CFD for Turbomachinery. ............................................................... 37 CFD historic evolution.................................................................... 38 CFD Error Estimation ...................................................................... 39 Introduction to the study case ................................................. 43 3.1 3.2 Description of the Hydromachine.................................................. 44 Partial simulations and Studied points ......................................... 46 3.3 Expected rotor-stator behaviour in Hydrodyna .......................... 47 Numerical Simulation. Procedure and Results.................... 50 4.1 Introduction. The Procedure followed ........................................ 50 4.2 Previous study: Turbine mode results .......................................... 51 4.2.1 Parametric studies of the Model............................................. 51 4.2.1.1 Mesh Sensitivity ...................................................................... 51 4.2.1.2 Time Step Sensitivity .............................................................. 52 4.2.1.3 Convergence Sensitivity.......................................................... 52 4.2.2 Conclusions for the turbine mode.......................................... 53 4.3 Pump mode ....................................................................................... 54 Table of Content 4.3.1 Complete machine simulation................................................ 54 4.3.1.1 Generation of Velocity triangles ............................................ 57 4.3.1.2 Results . Rotor-Stator Interaction (RSI) ................................. 62 4.3.1.2.1 STUDY FOR ϕ = @0.043 .............................................. 65 4.3.1.2.2 STUDY FOR ϕ = @0.028 ............................................... 70 4.3.1.2.2.2. Conclusions for simulation with ϕ = @0.028 ............ 77 4.3.2 5 Partial simulations.................................................................... 78 Experimental Data Results ....................................................... 85 5.1 5.2 Measurement techniques................................................................. 85 Experimental Results for ϕ = @0.043 ........................................... 87 5.2.1 Frequency Analysis .................................................................. 90 5.2.2 Conclusions for ϕ = @0.043 ..................................................... 92 5.3 Experimental Results for ϕ = @0.028 ............................................ 93 5.3.1 5.3.2 6 Comparaison of results. Conclusions .................................. 101 6.1 6.2 7 8 9 frequency Analysis................................................................... 95 Conclusions for ϕ = @0.028 ..................................................... 99 Detailed Comparaison of Results................................................. 101 Final Conclusions ........................................................................... 107 Attached documents ................................................................ 109 7.1 Sensors locations............................................................................. 109 7.2 7.3 7.4 7.5 Hydrodyna’s Nomenclature ........................................................ 111 Hydrodyna’s hill chart................................................................... 112 Example of a List of set-up expressions ...................................... 113 Code for creating Velocity triangles session............................. 119 7.6 7.7 Matlab file: frequency analysis and multiple comparisons..... 129 Comparison CFD and Experimental Results Plots.................... 135 REFERENCES ........................................................................... 140 ACKNOWLEDGEMENTS ..................................................... 144 1 Introduction. The Hydrodyna Project Introduction 1 INTRODUCTION. Flow in a turbomachine is three dimensional, turbulent and unsteady. Furthermore, fluid dynamic interactions appear between the flow fields in both rotationary and stationary parts, resulting in a more unsteady and secondary flow fields. In particular, the interaction between impeller blades and guide vane, known as Rotor-Stator Interaction (RSI), can generate pressure fluctuations of high amplitude in pump-turbines, especially in the points situated next to the rotor-stator gap, becoming one of the most important sources of vibration or blade cracking in this kind of turbomachines. As a consequence of these fluctuations not only runner blades can be serious damaged but also other machine components. The importance of the consequences of the noise, the erosion and the vibrations, depends on the turbomachine use. For example, it can cause hydraulic and electric instability in machines installated in power production plants, or disturb the confortability in ship passengers in propulsion engines. Anyway, erosion provokes a mass loss in runner blades so that it must be repaired and periodically inspected, with the following costs. In order to lessen this damage and reduce these problems it is important to find the main causes and try to combat them. Extensive numerical and experimental studies are required to understand and predict the flow phenomena. In general, not enough experimental data referred to this problem exist nowadays and computational methods for unsteady flow have to be still improved, calibrated and validated. There is still much research to do and more experimental data to analyse enabling to validate the unsteady numerical calculations and helping to find solutions to the problems linked to this phenomena. 2 Introduction 1.1 3 THE HYDRODYNA PROJECT Hydrodyna Project objectives[EURE99] are to improve the availability and the reliability of hydraulic turbomachinery, to make posible the optimization of the nowadays technical and scientific challenge in electricity market: The need of extending the operating range and the reliability of modernized hydro units according to variable demand. Development of advanced fluid and mechanical instrumentation, computation of fluid mechanics for a deeper knowledge are necessary in order to achieve these goals. They are to be obtained by developing a base of knowledge from experimental and computational results got from measures and tests in a Francis pump-turbine. The Hydrodyna Project intends to study dynamic behaviour of hydraulic machines to bring under control damage and risks derived from it and to upgrading efficiency and design technologies especially for pump-turbines. The aim of this paper is to help to achieve some of this objectives from the study of numerical and experimental results. 1.2 BRIEF OVERVIEW OF HYDRO POWER SECTOR SITUATION. Changes in the electricity market over the last decade create a demand for widing and becoming more flexible the operating range in hydroelectric plants. Nowadays power demand varies considerably and continuously, so a better energetic ressources gestion will allow to adapt electric production to the demand. Changes in demand oblige to start and stop electricity production plants in operating conditions far from the one they were designed to work in. Hydraulic power becomes the most convenient generation form to respond to this challenges at the moment. The great combination of its characteristics Introduction and properties (renewable green energy, moderated price, variable production rate and especially storage capability ) makes hydrounits full of advantages. When possible, modern plants tend to use reversible pump-turbines that can be run in one direction as turbines and in the other direction as pumps. The system is joined to reversible electric motor/generators. During the storage part of the cycle the motor drives the pump, while the generator produces the electricity during the hydraulic discharge from an upper reservoir. The refurbishment and the upgrading of the hydraulic power plants is the major concern in the energetic domain for the next years. Some real factors influenciating this fact are : • The capability of hydraulic storage installations to answer to quickly electricity demand variations ( the only ones together with gas turbines). Besides, they represent the only possibility in these moments to store large energy quantities with high global efficiency (around 80%). • The difficulty and the drawbacks of solar and wind energy accumulation. • The obligation of greenhouse gas emission reduction –Kyoto commitment [ERKM04] • The polemic and problematic issues of the nuclear energy source. Though the development reached for the hydroelectric resources in Europe during last century, it remains economical interest in the renovation of the existing power plants, as investors know it is quite profitable. Hydroelectric production coming from pump-storage installations functioning with power grid electricity dominates renewable electric sector in Europe, as shown in the figure (fig. 1a). [ENER07] 4 Introduction 5 figure (1.a) Distribution of electric renewable generation sources in Europe in 2005 In Switzerland, the hydropower production represents 56% (from which 32% of storage installations) of the total electricity production, compared to 40% from nuclear and the rest from other energy sources such as solar, wind, fuel and gas [ASEU04]. (fig. 1.b) Electricity Production Plants in Switzerland in 2003 5 4.3 24 Hydraulic Plants 1 Storage Hydraulic Plants 2 Nuclear Plants 3 4 40 Thermal Plants 32 5 Others figure (1.b) Distribution of electric generation sources in Switzerland in 2004 by installation type. Introduction 6 There are someones who consider hydro-power the most reliable way to obtain the green energy increase, but the main advantage of this electric ressource is that it is the only one capable to adapt to the peaks in the demand or to profit the economic periods to store. However, most European hydro power plants were designed many years ago and are not well adapted to the current operating requirements.Innovations and new design technologies for future plants and for the modernization of old power stations can increase considerably their efficiency as well as their power. The liberalisation of Energy Markets makes it profitable and reasonable to leverage the hydraulic potential of Southern Central Europe to cover peak consumption instead of continuous load. A special challenge in the refurbishment of existing power plants is to fit a machine with a wider operating range in the space available in the civil engineering installation, which has to be unchanged for cost and infeasibility reasons. These are the basis which motivated the Hydrodyna Project. A first part of this Project has already been done and it was focused in the turbine mode of the Francis machine. During the initial experimental study case pressure measures were also taken for some operating points in the pump mode. These measures will be analysed in this paper and they will be compared with numerical results from CFD methods in order to calibrate and validate them, in an attempt to derive conclusions which could help even to predict, prevent, control and reduce unsteady flows from causing damage and lowing efficiency. Introduction 1.3 7 INTRODUCTION TO HYDRAULIC TURBOMACHINES. In this paragraph a general overview about hydraulic turbomachines will be tackled. It is not the aim of this paper to explain in detail fluids mechanical theory or turbomachines lessons; however, some basic variable definitions, nomenclature and conclusions dealing with the subject must be introduced. Turbomachines are rotating machines that transfer energy between a rotating part (called rotor) and fluid. Depending on the energy transfer direction we distinguish turbines (transferring energy from a fluid to the rotor and normally the rotor to a transmission shaft) or a compressor (it exchanges energy from the rotor to the fluid). For hydraulic turbomachines the fluid is liquid water, and compressors are known as pumps. Energy transfer is obtained from a simultaneous, “reaction machines”, exchange of pressure potential energy and kinetic energy. In some machines (the most famous ones are Pelton turbines) only a conversion of kinetic energy takes place. They are called “action machines”. 1.3.1 VARIABLES DEFINITION. CLASSIFICATION OF TURBOMACHINES Definitions for main physical magnitudes will be treated in the following paragraphs. - Hydraulic power Ph , also known as flow power, is the power between high and low pressure machine sections. It is calculated as follows: b c Ph = ρQB gH I @gH @I = ρQ B E [1.3.1.a] Introduction 8 where ρ is density, Q is the discharge (see formula 1.3.1.d) , g is gravity acceleration and H is installation head (see formula 1.3.1.e), with subscript I for high pressure region and @ I for low pressure (figure 2). Specific hydraulic energy (E) definition is included in this formula too. figure 2. Section showing high and low pressure in a turbine installation. As has just been discussed before, depending on hydraulic energy transfer direction, we find turbines or pumps. - Turbines are work-generating, so fluid hydraulic power is decreased. Power is transferred from the water to the rotor (by the runner blades) and then to the transmission shaft to the generator (typically electric generators). - Pumps are work-absorbing, so the rotor transfers power (taken from a external motor) to the fluid, increasing the hydraulic power of the flow. Pumps are receptors. According to this, hydraulic power is defined as positive for turbines and negative for pumps. Ph > 0 Turbines Introduction Ph < 0 9 Pumps - Mechanic Power P is defined for turbomachines as : @Q @ @Q @ T P= w where [1.3.1.b] @ @Q @ @ wQ is the rotation speed and T the total moment in the turbomachine. - Transferred Power P t is defined as the real power converted inside the rotor, the runner blades. It does not include friction couples or leakage. The formula to calculate it from T t (the moment resulting from the flow action over the runner blades) is: @ @Q @Q Pt = T t @ w [1.3.1.c] @ where @ wQ is the rotation speed of the runner. - Flow rate, or discharge, is measured in [m3/s] . It is calculated as: @ @ Q =Z Z @ cQ @ nQ dA [1.3.1.d] A - Installation Head multiplied by acceleration due to gravity is called Mean Specific Hydraulic Energy. It is defined as : h i @ @ Q2 @ @ @ p c cQ @ nQ ffff fffffffffffffffffffff k fffffffffffffffffffffffffffffffffffff gH =Z Zj + gZ + dA ρ 2 Q A [1.3.1.e] Introduction 10 - Specific velocity is an essential non-dimensional parameter used in turbomachines. It owes its importance to its definition variable, since it combines flow rate, rotation speed, and hydraulic energy. Moreover it gives an idea of the type and geometry of the runner blades. These properties make specific velocity a reference parameter, serving to classificate machines and to choose a particular machine for an installation. There are several specific velocity definitions, depending on the units in which the variables are expressed and the relation among them. The most important one is ν =ω w w w w w w w w w w w w w w w w w w w w dw e Q fffffff s π fffffffffffffffffffffff ` a34ffff [1.1.3.f] 2E but also are commonly used: - For turbines: w w w w w w w ns = p P N fffffffffffffffffffff H 5ffff 4 [1.1.3.g] where N is in [rev/min] and P is in [KW] or [CV]. - For pumps and sometimes also for turbines w w w w w w w w nq = qQ N fffffffffffffffffffff H 3ffff 4 [1.1.3.h] Introduction 11 A classification for the most typical examples of hydraulic turbomachines is made in table 1. Extra characteristics information is given in the same table. In figure 3 a chart with these examples distributed by its specific velocity and its installation height is provided. Name TURBINES Type Flow Specific geometry Velocity Pelton Action Tangential Low Francis Reaction Radial and Medium radial-axial Deriaz Reaction Diagonal Medium Kaplan, Reaction Axial High Reaction Radial and Low Bulb PUMPS Radial radial-axial Diagonal Reaction Diagonal Medium Axial Reaction Axial High Table 1. Hydraulic turbomachines classification Introduction Figure 3. Turbomachines classification by specific velocity and Installation Head 1.3.2 NOMENCLATURE Now that turbomachines have been introduced, it is the time to meet their main components and their positions. As the study-case in this paper deals with a Francis pump-turbine, a picture containing its parts is given in picture 4. 12 Introduction 13 Figure 4.Francis turbine section. [IMAG07] For the rest of types of machines, nomenclature is the same, with the exception of pumps where, due to the role they play, some parts receive differents names. Table 2 shows the correspondances. TURBINE PUMP Runner Impeller Diffuser/Draft tube Inlet Pipe/Bend/Casing Distributor/ Guide Vane Vaned /Diffuser Spiral Casing Volute Table 2. Differents names for the machine components in pumps and turbines. Introduction 1.3.3 14 POWER CONVERSION AND BALANCE. All power definitions above suggest a global energetic balance. Hydraulic power results from the total dissipated power by the fluid in differents parts of the machine and the power exchanged in the runner. Ph = Prsc + Prv + Pro + Pt + Prq + Prb + Prd [1.3.3.a] where : Ph is the hydraulic power Prsc is power losses in the spiral casing Prv is power losses in the state vanes Pro is power losses in guide vanes Pt is power transferred by the runner Prb is power losses in the runner blades by friction and others Prq is power losses by leakage Prd is power losses in diffuser Power is defined from energy. Tranferred energy in the runner is associated to the homologous power and the flow rate going through the runner Qt: Pt = ρQt Et [1.3.3.b] In turbines, the total energy in the runner is E = Et + Erb [1.3.3.c] E = Et @Erb [1.3.3.d] While in pumps is Introduction 15 In the following two pages there are four pictures (pictures 5a. 5b. 5c. 5d.) representing the power conversion for both, turbines and pumps. While 5a. and 5c. Show only the mechanical exchange 5b and 5d provide a complete and detailed power balance. The legend for the variables in these pictures is: - Pm Runner mechanical power - P Machine power - PLm External mechanical Power Losses - Pt Extracted power (turbine) or Supplied Power (pump) - Prm Internal mechanical power losses - X Pre Flow power dissipation in turbine - X Prq Leakage flow power in turbines - X Pq Leakage flow power in pumps - X Pr Flow power dissipation in pumps It is essential to highlight that since power direction is different between pumps and turbines, efficiency gets different definitions too. In turbines η= P ffffff Ph [1.3.3.e] η= Pffffff h P [1.3.3.f] While in pumps Introduction 16 Figure 5a. Turbine Mechanical Balance Figure 5b. Turbine Power Balance Introduction 17 Figure 5c. Pump mechanical balance Figure 5d. Pump Power Balance Introduction 1.3.4 18 EULER EQUATIONS Euler equations describe in fluid dynamics how pressure, density and momentum are related in a moving fluid. They are a coupled system of differential equations, since they have to be solved simultaneously because all the dependent variables appear in all the equations. Although they appear to be very complex, they are actually simplifications of the Navier-Stokes equations. The Euler equations are only valable for inviscid flow, since they consider zero viscosity. The solution gives therefore only an approximation of the reality. [NASA06] They represent the conservation of mass, flow and momentum. ` a ∂fffffff ρ + 5 Apu = 0 ∂t [1.3.4.a] b` a c ∂fffffffffff ρu + 5 ρu N u + 5 p = 0 ∂t d [1.3.4.b] e b c ∂ffffffff E +5 u E + p = 0 ∂t [1.3.4.c] where E is the total energy per unit volume b E = ρe + c ρfffffffffffffffffffffffffffffffffffffff u 2 +v 2 + w 2 2 [1.3.4.d] and u, v,w are the velocity components, e is the total internal energy per unit mass flow, p is pressure, ρ is density and u is the fluid velocity. [WIKI07] There are only three equations and four unknowns, so it is required to add one more equation to solve the system. Normally, an “equation of state” linking pressure and density is used. Introduction 19 In practice more simplifications are considered. In hydraulic turbomachines, from long mathematical developments, simplification assumptions and concrete power balance for this kind of machines, the called “Euler’s Global Equation” is derived for the runner. b jjjjkjjj jjjjjjkjjj Et = Kcu c 1e u 1e @Kcu 1E d c ffff 1E jjjjkjjfffjj jjjjjjjkjjfffj e c 1 e u 1 e @Erb [1.3.4.e] where K are flow distribution coefficients depending on the machine design, kjj jkjjj c and u are respectively the fluid velocity and rotation linear velocity and fff the subscripts 1 and 1 mean high and low pressure section for the external (e) streamline. It could be perfectly applied to the internal streamline (i) too. Machine and flow characteristics, such as energy transferred or discharge rate, are inferred from this equation and its geometrical meaning (velocity components: sizes and projections) and representated by “velocity triangles”. These triangles provide a graphic implementation of this equation, from which Figure 6 gives an example. Introduction 20 Figure 6. Example of Velocity Tirangles [WIKI07] In partial loads velocity triangles are modified in the following way. The β angle between rotational and relative velocity (called ) is theorically considered to be imposed by the geometry of the runner and to be independent of the flow, so that for partial flow as the cu component is decreased all the rest of parameters from the velocity triangle vary as shown in figure 7. From this hypothesis, “Euler Characteristic” correspondance between energy transfer ψ (fig. and the flow ϕ 8) shows (for a fixed the β , so for a concrete machine) as a straight line. But this approach is just ideal, due to the fact that in reality β is not constant, because the runner geometry can not dominate completely the flow behaviour. In pumps, the flow against pressure gradient is complicated and recirculation and detachments appear degenerating the ideal “Euler Characteristic”. Later studies in this paper will take this idea up again and check it results. (see chapter 4, paragraph 4.4.1) Introduction 21 Cu1 Cm1 W1 C1 α1 β1 U1 Figure 7. Velocity triangles at the outlet for differents flow rates ψb 2 Et 1 Q1 = 2 1 − 2 U1b tg β1b A1U1 Q1 ϕ1 = πωR13 ψ 1b = 2 ψ 1b = 2 1 − 1 π R12 tg β1b A1 tg β1b ϕ1 φ Figure 8. Euler’s Characteristic The also better known Bernoulli equation is derived by integrating Euler's equations along a streamline. This one, under the assumption of constant density, the equation of state and neglecting the losses, is usually easily applied to solve fluid circuits in pipes and for simple geometric problems. When a relative observer is considered for the rotor study in turbomachines the same equation is called “Rothalpy”. For further information on these topics see [WHIT03]. Introduction 22 Finally, an important conclusion to draw related to this paragraph is the fact that to upgrade the power of a machine at a given radius, the increase of its speed is needed. At the same time, as a result, pressure fluctuations will be increased too , and consequently all the problems in terms of the structure damage and instability. 1.3.5 ROTOR-STATOR INTERACTION IN FRANCIS PUMP-TURBINES. The heart of this paper will be focused on the RSI analysis. Here an introduction to the phenomena behaviour sets the basis for the later studies in detail. The phenomenology of rotor-stator interaction is a consequence of the combination of inviscid and viscous flow and it is believed to be influenced by the design of the machine (for example, too large wicket gate thickness, high rotation speed, etc.) With respect to viscous effects, flow incidence angle in the distributor channels, blockage and flow detachments related to partial load and wakes, play major roles in the RSI behaviour. Effects of inviscid flow , also called potential effects, are associated to relative blades motion: the flow in distributor channels is periodically perturbed by the rotating impeller blades, and this effect is propagated in both upstream and downstream directions of the flow, generating pressure fluctuations and disturbances troughtout the entire machine flow fields. Once transient effects have finished, an expression for stationary and periodic pressure fields can be stablished for the rotor ( pr) and stator ( ps) flows. It must be highlighted that they are only right expressions if no unsteady effects appear for the concrete operating point. Thanks to the help of Fourier Series [GARC98] complex pressure fields in the following way: we can express these Introduction 23 b c 1 b p s θs , t =X Bn cos nzo θs + φn c [1.3.5.a] n=1 b c 1 b p r θr , t = X Bm cos nzb θr + φm c [1.3.5.b] m=1 where m and n are the armonics orders, Bx and φx the amplitude and the phase shift for the appropiate harmonic, θx the angle coordinates for the rotative (r) or stationary (s) system and zb and zo the number of blades and of guide vanes respectively. In the limit regions between stator and rotor the pressure field is a combination of the two ones above [ZOBE07], [NICO06], [TANA90], [OHUR90] whose expression is got from the product of both of them for each value of m and n from 1 to 1 . With the help of the trigonometrics functions properties for the product ` a ` a cos a cos b = a ` aC 1fffB ` cos a + b + cos a @b and considering the stator reference 2 system, those products result in the following expression : b c p mn θs , t = b c A b c ` a ` a Affffffffffffff ffffffffffffff mn cos mz b ωt @ mz b @nz o θs + φn @φm + mn cos mz b ωt @ mz b + nz o θs + φn @φm 2 2 [1.3.5.c] The connection between the two reference systems is given by the rotation velocity, since θr = θs @ω t and this equation is a function of two variables: time and space. Introduction 24 Flow field distorsion caused by Flow field distorsion caused the runner pressure field Combination of the guide vane effects both figure 9. Modulation process between impeller blade and guide vane flow fields Numerous studies made by Ohura et al. [OHUR90], Tanaka et al. [TANA90] and Chen [CHEN61] describe how this interaction induces pressure fluctuactions that are propagated in the entire machine. From their conclusions two different types of pressure waves are distinguished: - Diametrical mode rotating in the region between the guide vane and the impeller blades. - Standing waves in the spiral case. The first type is more relevant to our concrete study case. In fact, the modulated pressure form p mn is the combination of two diametrical pressure modes, with high and low order numbers, k1 and k2: k 1 = mzb @nzo [1.3.5.d] k 2 = mzb +nzo [1.3.5.e] Positive values for mode number indicate that the wave, the diametrical mode, is rotating in the same direction as the rotor, while negative values indicate it rotates in the opposite direction. The lowest absolute values of these numbers represent the highest energy for the diametrical modes. It means that the pressure fluctuations are stronger, have more energy and propagate further along the entire machine for the Introduction 25 lowest k absolute values. As a result, it is usually more important the impact of k1 than k2 because of the definition as a subtraction. Finally, to illustrate it figure 10a gives an example representing the diametrical modes shape in 3D. Figure 10a. 3D pictures of diametrical modes. [OHAS94] Depending on the value and the sign of k, in figure 10b, areas marked with – mean they are dominated by a low pressure wave and + denote high pressure wave domain. The direction of the arrow is related to the rotation speed direction: the same as the diametrical mode for positive k and the opposite on the contrary. k= -1 k= 2 Figure 10b. Diametrical modes shape patterns according to k values Introduction 1.3.6 26 FREQUENCY ANALYSIS. As a conclusion of the paragraph above, periodical waves are expected in the analysis of pressure fluctuactions inside the hydraulic machine. As has been explained, several frequency will take part of these pressure signals so that to understand and better analyse them a descomposition will be necessary. Based on the Fourier Series Theory [GARC98], [STOR02] the Fourier Transformation enables the decomposition of periodic wave or signals into the sum of several sinusoidal functions whose frequency are a multiple of the wave main frequency. The general mathematical expression to express this transformation is : ` a 1 p t = a0 + X n=1 d b c b ce a n B sin 2πfB t + b n B cos 2πfB t [1.3.6.a] where p(t) is the signal in the time domain, an and bn are unknown coefficients of the series. From them, the amplitude and the phase of the signal for each frequency can be calculated. The parameter f, that has units of Hertz[Hz], corresponds to the fundamental frequency of the wave. From the values obtained for the unknown coefficients for this decomposition the signal is “translated” to the “frequency domain” which enables to infer which are the most important frequencies defining the wave and its propagation throught the entire machine. A Matlab code has been developped for the frequency analysis of the pressure signals obtained in the simulation and in the experimental data too. It will be used for the comparisons and conclusions, as well. It is attached in chapter 7. 2 Numerical Simulation Methods. CFD Methods. 2 NUMERICAL SIMULATION AND CFD METHODS 2.1 INTRODUCTION TO COMPUTATIONAL METHODS With the development of computer technology, numerical simulation, also called computational methods, has turned out to be more and more widely used in fields of our society. Simulation techniques not only play very important roles in engineering studies , taking part on what it is called Computer Aided Engineering (CAE), but also many other sciences take advantage of them for their studies. Simulation is applied in biologic and physic sciences, for example for the study of atmospheric, oceanic or internal earth flows or also for chemical reactions as combustion. Even social sciences can use these methods for their studies. For instance, in finances they can be used for banking investigations or for the Stock Market predictions. Numerical simulation is the kind of simulation that uses numerical methods to quantitatively represent the evolution or the state of a concrete system. Numerical simulation can deal with many processes at the same time. Even nonlinear processes can be solved by these methods. The evolution of the system must obey some rules that govern the real processes in the simulated situation. From the outcomes of such simulation we will be able to draw proper conclusions and to get a deeper knowledge of the system. It is important to emphasize that knowledge of background processes and understanding of the simulated subject can be obtained from the results. In this paper these methods will be applied for the concrete study case of hydraulic turbomachines. Computational Fluid Dynamics (CFD), the specific computational simulations for this scientific area, has become an essential part of the engineering design and analysis environment of many industries who require the ability to predict the performance of new designs or Numerical Simulation Methods. CFD 29 processes before they are ever manufactured or implemented or who want to redesign and improve old versions. Fluid dynamics are used in different technologies including aerospatial, locomotive, power generation, air conditioning, heating and ventilation devices, chemistry and biomedical, oil and gas conductions, etc. Fluid dynamics are applied in all measureable technology scales, for a wide range of cases, from ventilation installations in huge structures to the smallest micro-pumps and the most accurate nanotechnology. Physical models are translated into mathematical models and performed by complicated computer algorithms are used to solve these engineering problems. All these techniques have experimented a marvellous developement recently, stimulated by the increasing computer power, so that now they can provide their advantages to analysis, investigation and design tasks, removing and replacing traditional expensive experiments. These simulation methods bring some advantages in relation to the old experimental practices. For example, money invested in creating the test reduced models for the experiences is saved, as well as the time and personal employed in them; it also avoids accidents ocurred during them and eases the sometimes complicated acquisition of measuress which, moreover, are not always reliable. Another important role for numerical simulation is that it enables extrapolation of the observations from a region in which measuress have already been made to unknown regions. If running a numerical simulation it accurately predicts the observed values of parameters in the regions where the values were known, it gives confidence to the simulation outside this region. Furthermore, it is important to point out that simulation makes possible to find new phenomena and processes that might be not found in experimental or theoretical study. Numerical Simulation Methods. CFD However, limitations of numerical simulation still exist. The development of numerical simulations started after the computer technology, so it is a fairly young method. As with theoretical studies, numerical simulation results need to be confirmed by experimental data, which will be one of the most important aims of this paper. Though numerical simulations can deal with much more complex problems than analytical study, sometimes they can not include all the physical processes that exist because their equations are not known or the numerical simulation is not feasible. In these cases, models appear and with them approximations. In fact, in the majority of cases there are many simplifications made in numerical simulations, which leads to errors in the results. The methodology follows a sequence of steps where approaches and asumptions are required and whose consequences in accuracy will be mentioned below. Focusing on the CFD methods we will now present these steps and some of the approximations they introduce. [FERZ99] [TUTO05] 2.2 CFD (COMPUTATIONAL FLUID DYNAMICS) The methodology that follows CFD concerns these four main steps: 1. The choice of the physical approach for the real problem. 2. The choice of the mathematical model for the simulation. 3. The discretization method. 4. The implementation of numerical algorithms to solve the equation systems. The following pages will introduce the theorical aspects of these steps in order to help to understand the procedure followed to the studied case in 30 Numerical Simulation Methods. CFD 31 this paper and because the possibility of manipulating and adjusting them could produce interesting improvements in the CFD methods. 2.2.1 THE CHOICE OF THE PHYSICAL APPROACH. The first step of the procedure is the determine the governing equations for the fluid dynamics problem. When these equations are know accurately, in theory exact solutions can be achieved with any accuracy desired. Normally the governing equations of physical real problems are groups of Ordinary Differential Equations (ODE) or Partial Differential Equations(PDE). In our case, for fluid mechanics, these equations are the Navier-Stokes conservation laws for the mass, the momentum and the energy. They are a complex system of partial differential, in time and in space , equations. Numerous conditions and fluid properties influence the flow behaviour. The conservation laws for the mass, energy and the momentum apply to all flows but they are non-linear and coupled, which make them difficult to solve. In the domain of hydraulic turbomachines, flow is assumed to be isothermal, incompressible and Newtonian. Such incompressible CFD simulations can not reproduce the propagation of the hydroacoustics waves generated by RSI, although they can provide boundary conditions for acoustic modeling of these phenomena [NICO06]. These simplifications from the general conservation laws provides the wellknown incompressible Navier-Stokes equations [2.2.1.a]. Considered a real description of the Newtonian flows, they constitute the fundamental basis of our hydraulic problem. [2.2.1.a] Numerical Simulation Methods. CFD 32 However, the real problem might not follow some of the initial physics hypothesis. For instance, we cannot assure that the flow rate is constant all the time or that there are no leakage, etc. These assumptions made in the physical approach are the first introduction of approximations in simulation methods. 2.2.2 THE MATHEMATICAL MODEL A mathematical model is an abstract representation of aspects of an existing/designed system using mathematical language to describe its behaviour. A mathematical model usually describes a system by a set of variables and a set of equations that establish relationships between the variables. The objectives and constraints of the system can be represented as functions of the output variables or state variables. In conclusion, the model presents knowledge of that system in usable form. Mathematical model and physical approaches are strongly linked. Physical approaches are chosen and simplified in order to make possible a mathematical modelisation. Sometimes it translates to a easier definition of the geometry, as for example 2-D problems, or considering constant or neglecting a variable.(density for incompressible flows or inviscid flows, respectively). These kind of simplifications of the physical truth transforms the NavierStokes into other easier. It is the case of the Euler equations for inviscid flow or potential equation for irrotational velocity flows. All these simplifications can result appropiate in certain and special flow conditions, whithout meaning a loss of accuracy. If not, these approximations introduce more inaccuracy in the method. Numerical Simulation Methods. CFD One of the most important mathematical model in CFD and especially for our study case is the “Turbulence model”. Approaches to predict turbulent fluid flows exist but they have still to be improved and calibrated. The most important ones (and the ones we use for our calculations) are ReynoldsAveraged Navier-Stokes equations (RANS), which consist in the averaging of motion equations over time or over a coordinate in which the mean flow does not vary. This approach leads to a set of partial differential equations which is not closed and, as a consequence, requires a mathematical model to approximate it, known as “turbulence models”. They introduce an artificial viscosity to model the turbulent effects of the flow. The most famous models are the κ @ε and the κ @ω . Main consequences of this approximation, deduced from comparaison to experimental data, are: excessive production of shear stress, suppression of separation along curved walls, excessive level of turbulence in regions of strong normal stress and wrong response to swirl. We could also mention other modeling approaches as Large Eddy Simulation. There are also Direct Numerical Simulation (DNS) which don’t include any approximation to Navier-Stokes equations, so they are very accurate but extremely expensive and still out of reach for turbomachinery flows for the next few decades. Besides, for meshfree techniques (which will be mentioned in the next section), the Vortex method turbulent model is used. 2.2.3 THE DISCRETIZATION METHOD. After the mathematical mode is set up, it is necessary to transform the continuous nature of the elements (the space, the time...) to a finite quantity of elements (grid or mesh generation) upon which the model equations will be placed. Discretization method approximate these equations by a system of 33 Numerical Simulation Methods. CFD algebraic equations of variables. This discrete representation is necessary to set a problem computationally solvable. Regrettably, both of these two steps, force to build in some tolerances, so more errors are acummulated. In applications dealing with PDE on a finite volumetric domain, the most usual approach is to genere a grid or a mesh that discretize the domain volume, composed by a certain (usually very large) number of small elements, which are typically called cells. Variables are computed at nodes located on cell center and there the governing equations will be calculated. They can be structured (regular) grids, block-structured grids or unstructured grids, and each one is more appropiated to a kind of problem and suggest a different discretization method. Depending on what is represented in a cell, two main types are distinguished: the Lagrangian-grid and the Eulerian-grid. Lagrangian grid is associated to a material description of the model frame, to the substance particles (figure 11) [IMAG07], while Eulerian grid is fixed to the time and the space. figure11 . The lagrangian-grid can be deformated with the particles. Both are applicable to CFD problems. However, as in fluids deformations are frequent, sometimes the Lagrangian grid becomes more complicated and is more associated to solid elements in CSM (Computational Solid Mechanics). Eulerian-grid are more tipically used for CFD. It was the one used to 34 Numerical Simulation Methods. CFD generate the mesh of the Francis pump-turbine that will be studied in this paper. (figure 12) figure12. A partial view of the mesh used for the Hydrodyna Pump-Turbine There are other special methods to proceed without a mesh-based method. These alternatives, called meshfree or meshless methods, are based on the idea of setting randomly distributed nodes with don’t follow any kind of grid and but which have been well defined to make sure that the equations of the model for all the domain, including the boundary and limit conditions, are satisfied. The best example for this kind of methods in fluid dynamics is SPH (Smoothed Particle Hydrodynamics). As has already been said, for the computational procedure time will also have to be discretized. In unsteady problems, the time step choice is another parameter of inaccuracy for the results. Once the grid and time steps are defined, it is the time for the discretization methods. Three main discretization approaches are used nowadays for CFD. In this case, discretization refers to the differential equations, which are converted into several algebraic evaluations. They are: 35 Numerical Simulation Methods. CFD • The Finite Difference Method (FD), which replaces the partial derivatives by approximations of the nodal values of the function for each grid cell. • The Finite Volume Method (FV), which first subdivide the domain in many control volumes (CV) and then applied the interpolated conservation laws to each one. This method is the most stable because every each of its cells must verify the conservative laws. • The Finite Element Method (FE), similar to FV but multipling the equations by a “weight function”. For further references see [FERZ99]. 2.2.4 THE IMPLEMENTATION OF NUMERICAL ALGORITHMS AND SOLUTION. Once the discretization is made, it is the time to translate the equations into a computational code that allows to solve the mesh conditions for the model. The algorithms can be written in several differents programming languages, it depends on the engineer or the programmer preferences, who will try to make them the simplest but the most accurate too. This kind of numerical algorithms are considerably complex and the enormous quantity of equations and variables to be taken into account obliges to utilize modern sophisticated and powerful processors and clusters to implement and solve the millions of calculations they consist on. However, to solve the equation systems iterative methods are used, as they are too complicated and not convenient to use direct methods. introducing once more an error, related to the convergence residual value. Iterative methods need a stop criteria, i.e. a admissible value of a variable to stop the iterations. This value can be fixed to any desired level of accuracy. 36 Numerical Simulation Methods. CFD Finally, once the algorithms are implemented and solved, one can use the solution to obtain conclusions for the real initial problem. Outcomes are usually analysed with proper post-processing programs to get the maximum benefit of them. 2.3 CFD FOR TURBOMACHINERY. In the domain of turbomachinery, CFD is today’s essential tool. The major challenge for CFD in solving problems of this kind of machines is related to the relative motion between rotor and stator. Once the physical and mathematical models are set up the problem arise when defining the relation between the rotationary and the stationary meshes, as well as the rotor-stator interface. While the rotor is calculated by a reference system in movement, the stator uses a stationary frame. The procedure followed, called “Multiple Frame of Reference”, connects them. It consists in linking at each cell of the rotor interface mesh the conditions for the cells in the corresponding stator interface mesh at that precise moment. Moreover, difficulty is increased for partial simulations approaches. Although this procedure can save time to the simulation process, it usually provides less accurate solutions, due to the fact than only the full domain takes into account all the aspects that influence the flow behaviour. Boundary conditions, turbulence sources, periodical surfaces and interfaces are seriously perturbed by this kind of approaches. Flow throughout the entire machine is connected, so that interactions make that what happens in no matter what region has a considerable impact on the rest. The skip of part of the complete machine volume provokes disarranges on the results, because it is not true to solve the flow in one part and then reproduce translated periodic conditions for the rest. The number of 37 Numerical Simulation Methods. CFD channels of stator and rotor is normally unequal to avoid resonance phenomena, and to connect the meshes in partial simulations, the rotor-stator region has to be critically modified. To represent a portion of the machine that includes both, rotating and fixed parts, it is necessary to fit a relation between the number of impeller blades and diffuser vanes, usually with no common division. The decision of this relation, known as pitch ratio, plays a decisive role in the triumph of the outcomes in partial simulations. The more the pitch ratio approaches to the real ratio between the total number of impeller blades and the number of guide vanes, the better. On the contrary, if a large different number is chosen, periodicity effects will be severely perturbed and badly representated by the model. Mistakes and fake results will possible appear in the outcomes. The approach in these cases used in CFD programs for approximating to have periodic conditions in partial simulations used to be called “sliding mesh” . 2.4 CFD HISTORIC EVOLUTION As has already been said, CFD method are quite young. However, they have experimented a fast evolution and development. At the beginning, these techniques could only resolve linearized potential equations in two dimensions, later Euler equations were solved and huge progresses have been done so that nowadays these methods manage to solve the Navier-Stockes equations. Firstly, only 2-D codes were developed and recently 3-D codes are available in numerous software packages. Figure 13 shows the fast development in computer technology since 1993. 38 Numerical Simulation Methods. CFD Figure 13. Computers performance evolution from 1993-2007 [TOPL07] From the observation of the first picture one can see that for example, in only three years, between 2001 and 2004, supercomputers increased 10 times its power. Just a last remark to emphasize the revolution in computers world: todays laptops have approximately the same computing power of a 1995’s super-computer. 2.5 CFD ERROR ESTIMATION It has already been explained above that even in the most accurate numerical simulation a few approximations are introduced. The differences between reality and numerical simulation results can be due to them: - The differential equations governing the problem may contain approximations or idealizations. - Discretization approaches are approximations, by definition. - Iterative methods to solve the equation systems introduce the “truncation error”. (maximal residual value of convergence) 39 Numerical Simulation Methods. CFD - Round-off error, related to the machines precision. It means that computer precission has a finite number of decimals. To close this chapter, we can summarize the CFD procedure as normally it is followed in practice by software packages: - The preprocessing, where the physical model equations, the geometry and and the boundary conditions are described and fixed. In case of a transient problems the initial conditions must be here defined. The discretization of the volume and generation of the mesh is part of the preprocessing. - The computational simulation. In many cases the simulation is repeated and many iterations are done till a good result is obtained. Different causes, coming from the preprocessing phase or not, can provoke this need to repeat the calculations; However, in most of the cases these repetitions are made to control the quality of the results, as in parametric sentivity studies. - The post-processing of the results. It is during this part when the real goal of computational methods is attained. Many developped tools are employed to draw out the maximum information and conclusions from the results. 40 Numerical Simulation Methods. CFD Pre-processing Processing: Computational simulation Post-processing 41 • • • • Choice of the models Mesh generation Physical properties and constraints Initial and boundary conditions Iterations to solve equations system Analysis of the results Figure 14. CFD sofware procedure 3 Study Case: Introduction and description. Study Case: Introduction and description 3 43 INTRODUCTION TO THE STUDY CASE In the first chapter the Hydrodyna Project was introduced. To get deeper knowledge in fluid turbulent flows in pump-turbines several studies and experiments have been made. As it was said in the first chapter, turbine mode study has already been investigated, so in this paper we will focus fundamentally in the pump mode. The design of pumps is mainly based on steady flow assumptions throughout all the machine components (runner, diffuser...). This kind of approach suits reasonably for design operating conditions, but to understand and to take into account wider ranges in pump designs it is necessary to increase the knowledge of unsteadyness and RSI. Two main tasks should be covered according to the project objectives: the RSI phenomenon and the fluid/structure coupling. Undesirable events like flow detachments, ungovernable turbulence and cavitation phenomena, structure vibrations, wakes transport and dissipation will be studied in the context of experiments in hydraulic turbomachines. Experimental data is currently very required for a better optimisation of the pump design. The best option to improve pump design is to develop powerful numerical simulation methods of internal flow. Small improvements to rotating machinery design can translate into large operating savings and other important advantages. However, only validating these design methods from their comparison to the observations obtained in laboratory tests, a real production increase, longevity improvements and waste decrease will be possible to come true. As an objective of this paper, all the analysis and the results obtained from experiments in pump-mode will be explained for a better perception of the phenomena ocurred in the machine. Study Case: Introduction and description 3.1 44 DESCRIPTION OF THE HYDROMACHINE The turbomachine used for the experiments is a reduced scale model of a Francis pump-turbine (figure 13) composed by 20 stay vanes, 20 guide vanes and 9 runner blades. The machine casing is the spiral type and its specific speed is ν = 0.19 (or what is the same nq= 29.98) so it belongs to Francis low speed range. See picture 3 in chapter 1 to locate it. Its Efficiency and Energy/Discharge coefficient in pump-mode are given in figures 16 and 17 only for the 20º vane opening angle position this study will work with. Plots containing more other positions are attached in chapter 7. See “Hydrodyna’s Hill charts”. figure 15. Cross section of Hydrodyna reduced scale-model. Runner Type Francis Number of Number of Number of Nominal Specific blades guide vanes stay vanes flow rate velocity channels channels 20 20 0.228 [m3/s] ν=0.19 9 nq=29.98 Table 3. Machine definition characteristics Study Case: Introduction and description 45 η [-] HYDRODYNA 0.9 0.89 0.88 0.87 0.86 0.85 0.84 0.83 0.82 0.81 0.8 0.79 0.78 0.77 0.76 0.75 0.74 0.73 0.72 0.71 0.7 -0.05 20 deg -0.045 -0.04 -0.035 -0.03 -0.025 -0.02 -0.015 ϕ [-] figure 16. Hydrodyna’s efficiency chart HYDRODYNA 1.2 1.15 1.1 1.05 ψ [-] 1 0.95 0.9 0.85 0.8 20 deg 0.75 0.7 -0.05 -0.045 -0.04 -0.035 -0.03 -0.025 -0.02 -0.015 ϕ [-] Figure 17 Hydrodyna ψ. ϕ coefficients chart To have detailed views of the scale-reduced Francis model see “Attached documents” in chapter 7. Study Case: Introduction and description 3.2 46 PARTIAL SIMULATIONS AND STUDIED POINTS The choice of points was decided from experimental data measures. During previous studies made in generating mode, some measures were also taken for the pump-mode. Among the existing measures, there was a special interest in the comparison of two of them, due to the different behaviours observed during the rig experiments. These points were corresponding to ϕ = @0.028 and ϕ = @0.043 . All the studies in this paper for both results, computational methods results and experimental data measures, will be only about them. The characteristics of these two Hydrodyna’s pump mode studied points are summarized in table 4. ψ EXP ψ CFX ηEXP ηCFX ϕ = @0.028 1.07 1.028 0.855 0.879 ϕ = @0.043 0.83 0.729 0.863 0.795 Table 4. Hydrodyna’s characteristic values for the studied points The explanation for the differences between these values between experimental measures and computational simulations has its basis on the power balance. As it has introduced in the first chapter (see figures 5), there are many different types of losses throught all the energy transfer between the machine components. Some of these losses are not considered for the model in CFD, so the real values calculated from the laboratory measures can not be exactly equal to the ones calculated from the simulations results. With respect to the computer domain of study, three partial computational domains studies were made. Study Case: Introduction and description 47 It has already been commented that partial simulations allow to solve a smaller problem by assuming periodicity of the flow, this means identical flow in all the channels. Another issue of such partial simulation is that the portions of the rotorstator interface do not cover the same angle αB on the impeller side and αo on the diffuser side. For the study case of this paper with a real relation of 9 blades/20 guide vanes, the following combinations are considered: Pitch ratio ZB αB Zo αo 1 40º 2 36º 1.111 2 80º 4 72º 1.111 3 120º 7 126 0.952 αfffffff B αo Table 5. Parameter values of the three partial simulation studied cases. The legend for this table is: ZB: number of blades. α B :Angle between blades. Zo :Number of guide vanes channels. α o : Angle between guide vanes channels. Conclusions based on the quality of their results compared to the full simulation will be commented too. 3.3 EXPECTED ROTOR-STATOR BEHAVIOUR IN HYDRODYNA Recently, RSI has been investigated in depth. The conclusions of all those studies and investigations have been introduced above. Particularly, the Study Case: Introduction and description 48 diametricals mode behaviour explained in the first chapter in the rotor-stator zone and the pressure propagation up and downstream is expected to be applicable to our case. Hydrodyna, with 9 impeller blades and 20 guide vanes, will present the following order values for the diametrical modes, calculated in table 6: n m k1 k2 fs/ fb= mzb 1 2 -2 38 18 1 3 7 47 27 2 4 -4 76 36 2 5 5 85 45 Table 6. Expected rotor-stator diametrical modes for Hydrodyna. The fifth column includes the frequency expected to dominate the pressure fluctuactions created by this phenomenon. It is calculated from the number of blades in the machine and the harmonic order corresponding to the rotor (m) for the minimum k number. Since k1 is negative and m value is 2, the wave is expected to rotate in the opposite direction to the rotor and to have the fundamental frequency for a value twice the blade passage frequency (BPF). In this case 18 times the rotating frequency. 4 Numerical Simulation. Procedure and Results Experimental Data Results 50 4 NUMERICAL SIMULATION. PROCEDURE AND RESULTS 4.1 INTRODUCTION. THE PROCEDURE FOLLOWED As it was discussed in previous chapters, modern software packages and powerful processors are vital tools to resolve CFD problems of the type we are dealing with in this paper. The program used for the calculations was the version ANSYS CFX-11.0 of ANSYS CFX, containing three subproducts (Pre, Solver-Manager and Post). The three of them were applied to the problem. [LMH07] The program enables the analyst to choose among multiple possibilities for the conditions and the models that will be applied and considered in the resolution of the problem (angular average velocity profile, constant average pressure at the outlet, Log law for the solid surface, k-epsilon turbulent models, stage, frozen rotor or transient rotor-stator simulations...). The set-up plays a major role in the success of the problem. An input file created for one of the first simulations of this particular problem is attached in chapter 7. See “Example of a List of set-up expressions”. Once the set-up is completely and correctly defined, it is sent to the solver to be solved and to generate a solution. For this study, numerous simulations were calculated. Partial domains, different points of work, correction of errors in the initialization... all of them thought to get the most of the information possible. These calculations take several days to be finished even for the high-speed supercomputers, so it is convenient to make sure the set-up is carefully made in order not to waste time, money and energy. Before considering the outcomes as ultimate it is always recommended to check them and judge them, to assure that the set-up was correct and that no mistakes will be committed in studying these results. Then continue with the next and final analysis step. Experimental Data Results At the end of the procedure, if everything was properly done, the objectives of the method can be reached. The post simulation tools are used for analysing in depth the results. 4.2 PREVIOUS STUDY: TURBINE MODE RESULTS 4.2.1 PARAMETRIC STUDIES OF THE MODEL As the pump mode will make use of many conclusions and tools already used during the turbine study, parametric study conclusions from the model in this generating direction are enclosed here. These parametric simulation results should not disagree with the pump mode ones as they are general for the global model and they should not vary within the machine mode. To take subjective approximations for decisions about the constraints imposed to the problem conditions without analysing before the influences of some variables, can lead to overkill the process with calculs and larger time computing or , on the contrary, have not strong or not reliable results.[KUEN99] The most significant parameters to study their sensitivity on the problem are the mesh, the time step and the convergence. 4.2.1.1 MESH SENSITIVITY Mesh sensivity was investigated on a partial computational domain because of computational memory limitations. Structured meshes were alwasys used. They included two stay vanes and guide vanes channel, one impeller blade to blade channel and the corresponding part of the cone. The three sensivity 51 Experimental Data Results 52 propositions had the same shape but different grid refinement. The possibilities were: Mesh Stator/channel Rotor/channel Coarse 30,000 70,000 Medium 60,000 170,000 Fine 130,000 400,000 Table 7. Mesh sizes for size sensitivity analysis The conclusion was that, weighting up the accuracy of results and the computational effort, a medium mesh was enough and satisfactory. 4.2.1.2 TIME STEP SENSITIVITY For the time step study, a medium mesh was applied to three possible time steps and at three different points of the mesh. Time steps corresponding to 0.5º, 1º and 2º of the impeller rotation and at two points from the guide vane and one at the blade leading edge. Results suggested to use the time step corresponding to 0.5º. 4.2.1.3 CONVERGENCE SENSITIVITY Convergence sensitivity refers to the maximum residual values for the numerical methods. Results for Cp obtained using three different residual values ( 10-2 , 10-3 and 10-4 ) were compared. The mesh used was medium and the simulation time step was the corresponding to 0.5º according to what had already been set. The evaluation of the influence of this parameter for the three cases concluded to establish 10-3 as maximum residual value. Experimental Data Results 4.2.2 53 CONCLUSIONS FOR THE TURBINE MODE The turbine mode investigation for the maximum discharge operating condition present the following major conclusions: • Pressure fluctuations for the distributor channel obtained from the numerical simulation are in very good agreement with experimental data. However, fluctuactions amplitudes in the rotor frame are around 25% higher for the simulations results than the experimental value. • A computing domain modelling the entire machine geometry enables to minimize the errors. It overcomes the influence of boundary conditions, pitch-ratio of rotor-stator interface and non-uniformities in spiral-case flow. • The numerical result analysis shows the variation of the pressure fluctuations at blade passage frequency (BPF) and its harmonics along a distributor channel of the Francis pump-turbine. • The maximum pressure amplitude of BPF takes place in the rotorstator region. However it decreases quickly, backward to the stay vanes. • The diametrical mode resulting from the modulation of RSI flow field corresponding to the highest energy, has a k number value of -2, which means that the pressure amplitude dominating the fluctuactions generated in the rotor-stator zone spreads to the spiral casing with a dominating first harmonic of 2 BPF. Third and fourth frequency modes are still visible in the guide vane but dissapear in the stay vane channels. Experimental Data Results 4.3 PUMP MODE From this point, the crux of this chapter will be handled : The pump mode simulation and its results. The same structured mesh that had been generated for the machine in turbine mode was applied for the inverse mode. Only the vanes angle position of the diffuser had to be adapted to the position of the operating point that was going to be studied. It was simple to do, just consisting in a rotation of the mesh. Two adaptations were made: one for an opening guide vane angle of 18º and a second one for an opening guide vane angle of 20º. At the end, only this last one was studied in depth, because there were no experimental pressure measures for 18º. 4.3.1 COMPLETE MACHINE SIMULATION As it has been mentioned ANSYS CFX-11.0 was used for the simulations. The physical model was fixed as incompressible flow, unsteady RANS (Reynolds Averaged Navier-Stokes) equations and “sliding mesh” for RotorStator interfaces. Between the available turbulence models, in this case, the SST was chosen, which combinates κ @ε and κ @ω characteristics. Boundary conditions imposed in the simulation software set up are given in table 8. 54 Experimental Data Results 55 Rotating Frame Outlet Inlet Stationary Frame Figure 17. The complete machine computational domain. Boundary conditions Location Characteristic Inlet Draft tube inlet Constant mass flow rate Uniform velocity Outlet Spiral casing Constant average static pressure Walls Solid surfaces Log Law Table 8. Boundary conditions For the space discretization, the Finite Volume (FV) method is applied. It uses a blended discretization scheme for hybrid meshes. These meshes (figure 18), which were taken from the turbine study, are mostly structured, except for a small region of the spiral casing. The characteristics for each component of the machine are summarized in table 9. In total, around 4,500,000 nodes were simulated. Experimental Data Results 56 Nodes per Channel Minimum Angle Mean y+ Spiral casing 500,000 23° 120 Stay vanes 100,000 35° 100 Components Mesh Software ICEMCFD Guide vanes Impeller 185,000 39° 60 Cone 135,000 41° 270 Mesh topology Structured 5 Table 9. Mesh characteristics Figure 18. Complete machine runner meshes To obtain the complete grid from only the basic mesh containing one blade for the runner and for the stator two vaned diffuser channels, it was neccesary to make nine copies of the impeller blades and displace them and analogously for the diffuser channels. Besides, the rest of the regions composing the complete machine (the cone and the spiral casing) were added and defined. The time discretization used a second order backward euler, which is implicit and quite stable, allowing CFL >1 . The second order approach takes Experimental Data Results 57 into account tn-1, tn, tn+1 and the time step was set to 1º (although for parametrics studies 0.5º was recommended). The convergence criteria for coefficient loop presented a RMS residual value of 510-5 and a maximum value of 5 10-3. 12 runner revolutions were simulated to arrive to statistically steady state solutions. This complete domain simulation was only done for an opening vane angle of 20º but at the couple of different discharge points: ϕ = @0.028 and ϕ = @0.043 . As it has been already explained, this points were selected because of existing experimental data measures at them. The different behaviours observed during the experiments created a special interest in their comparison and their simulation, in an attempt to find the causes. 4.3.1.1 GENERATION OF VELOCITY TRIANGLES Velocity triangles are a specific graphic representations for turbomachines. It has been mentioned in the beggining of this paper, as a important tool commonly used, since it contains much information in a simple graph. It should be added that it can also be used fot detecting users error, just checking it at the beggining to verify the operating points and main variables had been well defined in the set-up. They are derived by the Euler’s Global Equation and its geometrical meaning. Just having a quickly look to a velocity triangle (its angles, the length of its sides, the comparison between two triangles of the same machine or different machines) is possible to approximate and get an idea of the current operating point or the energy transferred. Therefore, velocity Experimental Data Results triangles are really useful and illustrative analysis tools and worth to be taken into account for any turbomachine investigation. A procedure to get a representation of velocity triangles for these two points (and capable to be used for represent any other case) was designed and implemented using the simulation software. The algorithm, not so evident at the beginning, enables to see the triangle in 3-D situated between the rotorstator and sorrounded but all the rest of the components. It was something new, created for the first time for this study. Pictures 19 and 20 show the outcomes for the particular operating points: Figure 19. Velocity triangle for ϕ = @0 .028 58 Experimental Data Results 59 Figure 20. Velocity triangle for ϕ = @0 .043 α β cfffff u u ϕ = @0 .028 9.8º ϕ = @0 .043 18.82º 13.67º 15.65º 0.5782 0.4513 Table 10. Summarized velocity triangles characteristics Angles between the velocities have been included and summarized in table 10 . Alpha α is the angle between rotation velocity and flow absolute velocity, while beta β is the angle between relative velocity of the flow seen by a rotative observater situated in the rotor and the rotation velocity. Velocity triangle for ϕ = @0.028 shows its correspondent α value of around 9.8º and its β value of 13.67º . It is a point of partial load. Velocity triangle for ϕ = @0.043 has important differences. For this point, while β is very similar with a value of 15.65º, α has bigger value, around double from before: 18.82º. It is very important to highlight that it is quite well aligned in relation with the guide vanes opening angle of 20º; it is a high discharge point. Experimental Data Results 60 On the other hand, first disagreements with the ideal “Euler characteristic” theory, introduced in the first chapter, appear here. One can see that β is not constant, so that it means the runner blade does not always transfer all the energy it was designed to, because of the slip derived from flow going against pressure gradient. It behaves different in partial loads and non-linear losses appear. The “Euler’s characteristic” is perturbed as shown in figure 21. ψ E = Et − ∑Er ∂p >0 ∂l ϕBEP φ Figure 21. Euler’s Characteristic Losses Perturbation Back to our two concrete cases, their differences and resemblaces, having in mind that rotation speed is the same for both, and the relation to the energy of the flow and the discharge rate in each operating conditions provide the following physic interpretations. In what respects to energy, as it was explained in the first chapter for the Euler Global Equation, the energy transformed in the runner can be calculated as: Experimental Data Results b 61 jjjjkjjj jjjjjjkjjj d c Et = Kcu c 1e u 1e @Kcu 1E ffff 1E jjjjkjjfffjj jjjjjjjkjjfffj e c 1 e u 1 e @Erb [4.3.1.1.a] Neglecting again losses term and knowing that inlet is designed to present axial flow (so that energy transferred is optimized), only the first term in the right part of the equation is considered. Hence, energy is directly proportional to the circumferential component of the flow absolute velocity. According to the distribution of the flow velocity components mentioned above depending on α values, the point ϕ = @0.028 presents smaller its circumferential component α , so for the absolute velocity of the flow ( cu ) is bigger than for ϕ = @0.043 . In conclusion, ϕ = @0.028 energy transformation must be greater. A bigger α (for ϕ = @0.043 ) with similar β and the same rotation velocity means that radial components of both, the absolute and the relative velocity of the flow, are increased; However, only the circumferential component of the relative velocity grows, while that component decreases for the absolute velocity as α increases. As absolute radial velocity component is associated to mass flow rate, if it is bigger at the point corresponding to ϕ = @0.043 the corresponding discharge must be bigger than at ϕ = @0.028 . Indeed, this is truth: ϕ = @0.043 presents a flow rate of 0.231 [m3/s] and ϕ = @0.028 only 0.152[m3/s]. The code written to generate the session in CFX-11 is attached in chapter 7. See “ Code for creating velocity triangles session”. Experimental Data Results 4.3.1.2 RESULTS . ROTOR-STATOR INTERACTION (RSI) “Monitors Points” were set in varied locations of the mesh surface. They function as sensors, allowing to study the pressure, the vorticity, the velocity, etc. in the points where they are located. The analysis of these variables using the tools of CFX and exporting the data files to other external programs provided a large amount of pictures, results and plots. However only some of them, considered the most conclusive and/or interesting ones, will be shown in this paper. There were 10 monitors situated on a blade wall, reproducing the same number of the sensors in both faces of a blade, and 8 monitors in each of two stator channels (so 16 monitors in the distributor). For each of these channels, 4 monitors are in the external part of the stator (next to the shroud) and other 4 at identical positons but in the interior part (next to the hub). At the beginning all these points were simulated and analysed, but realising that pressure fluctuations at the blade did not present any interesting remark, the analysis were focused on the stator points, especially on the ones at the rotor-stator interaction zone. To show the “ideal” behaviour plots coming from five monitors points belonging to a streamline at the blade (figure 22) are exposed in figure 23 and figure 24, for the ϕ = @0.028 and the ϕ = @0.043 operating points, respectively. Figure 22. Points from a streamline at the blade 62 Experimental Data Results 63 Cp 0.1 Cp [-] 0.05 0 0 40 80 120 160 200 240 280 320 360 -0.05 -0.1 Bbb10 -0.15 Bbb11 Bbb7 Bbb8 Bbb9 figure 23. Cp fluctuations during a runner revolution for points along a streamline from the blade at ϕ = @0.028 operating point Cp 0.025 0.02 0.015 Cp [-] 0.01 0.005 0 -0.005 0 40 80 120 160 200 240 280 320 360 -0.01 -0.015 -0.02 -0.025 Bbb10 Bbb11 Bbb7 Bbb8 Bbb9 figure 24. Cp fluctuations during a runner revolution for points along a streamline from the blade at ϕ = @0.043 operating point On the following, only the points on the guide and stay vane will be studied carefully, so these are the only ones appearing and named in figure 25. To understand the place where the rest of monitoring points were located, see the pictures in chapter 7: “Attached documents”. Not all of the pressure sensors installed in the real machine for the experiments had been monitorized for the simulation. The table with the correspondency between them is also attached in that final chapter. Experimental Data Results 64 Figure 25. Monitoring points location in the distributor As the aim of this paper is the comparison of experimental and numerical results and as the region that presents the greater unsteadyness and major difficulties for the measurements is the rotor-stator gap, the analysis will pay specific attention to the monitor points located closer to the gap, especially to A and B, in both external (e) or internal (i) walls. Furthermore, the study will be focused on the pressure, the velocity fields on this area, as to investigate what are the causes of vibration and flow detachments and how they can be prevented. The non-dimensional parameter Cp defined as pffffffffffffffffffffffffffffff @p mean Cp =d e 1fffff ρ Au 2 2 [4.3.1.2.a] Experimental Data Results were p is the pressure in the current point in the precise moment, pmean is the average pressure for the point, ρ is the density of the water and u is the linear velocity for the rotation speed. Cp will be used all the time in our analysis. It is a coefficient that gives a normalized idea of the pressure fields behaviour in a point of in relation to the flow kinetic energy. Results are completely different for the two operating points of study. The first example, with a ϕ = @0.028 is much more instable than ϕ = @0.043 , whose Cp fluctuactions are cleaner and almost perfectly periodic. Compared to the rest of the stator points, points next to the rotor-stator interaction region experience more important pressure fluctuations. Their amplitude peak to peak of the Cp periodic signal is bigger than for the points far from this region. However, the absolute values it fluctuates between are smaller. This is due to the fact that a pump-diffuser converts kinetic energy into pressure energy, so, in pump mode, it must be increased along the stator channel. Certains of these conclusions and more detailed ones will be exposed on the next paragraph. 4.3.1.2.1 STUDY FOR ϕ = @0.043 For this operating point, corresponding as shown in velocity triangles to an α angle of 18.82º, the pressure fluctuactions during a runner revolution are representated by the coefficient Cp and shown in figure 26. 65 Experimental Data Results 66 Cp Cp [-] 0.04 -0.01 0 -0.06 40 80 120 160 Point A 200 240 280 320 Point B figure 26. Cp in the stator side for points A and B near the rotor-stator gap for ϕ = @0.043 From the graph, a first glance gives already the phase-shift between the points A and B. They are very similar waves, fluctuating between the same values with similar periods and amplitudes, only displaced by a phase component. As the X axis represents rotation degrees, one can see that peaks are separated by around 20º, agreeing to the real 18º between point A and point B. For each of the points individually periodicity is found every 40º, exactly the phase shift between two blades (there are 9 runner blades for 360º). There is the first RSI consequence. Fluctuations at this couple of points in the stator side are completely determinated by the blade passage frequency (BPF). It will be studied and proved by the following frequency analysis of the signals. More points were studied and compared between them, between the external and the internal walls and also for the whole streamlines in the channels, but no sign of unexpected phenomena was found to be shown as 360 Experimental Data Results 67 interesting. See some pictures showing the “perfect” flow behaviour in figures 27 and 28. Cp [-] More details and comparison will be exposed later. Cp 0.025 0.02 0.015 0.01 0.005 0 -0.005 0 -0.01 -0.015 -0.02 -0.025 -0.03 -0.035 -0.04 -0.045 40 80 120 160 D 200 B F 240 280 320 360 H Figure 27. Cp curves for a stator streamline during one runner revolution from the simulation results for ϕ = @0.043 Cp 0.09 A_(e) Cp [-] 0.06 A_(i) 0.03 0 0 40 80 120 160 200 240 280 320 360 -0.03 -0.06 Figure 28. Comparison of pressure fluctuactions for one runner revolution between the internal and external points located in section A for ϕ = @0.043 . Experimental Data Results 4.3.1.2.1.1 68 FREQUENCY ANALYSIS. As it has been introduced, the expression for the Fourier Transform is : ` a 1 p t = a0 + X n=1 d b c b ce a n B sin 2πfB t + b n B cos 2πfB t [4.3.1.2.1.1..a] The parameter f, in Hertz [Hz], is the fundamental frequency of the wave, and in this study, is calculated corresponding to one runner blade passage for the current machine rotation speed: 1/9 w The developped matlab code (see attached documents) for these frequency analysis allows multiple comparisons at the same time. From all the possibilities only some concrete results will be put into show here. First of all, figure 29 compares frequency spectra for all points in the section next to the rotor stator interface. So, it includes both external and internal points A and B, located in the two stator channels. Figure 29. Pressure Spectra for section next to the RS gap for ϕ = @0.043 Experimental Data Results 69 Conclusions are clearly visible: all the points have very similar frequency spectra and no relevant differences are found between external and internal points at this location. There are small amplitude variations between the channels, with bigger amplitude at the first BPF for channel 20, corresponding to A section. (to remind nomenclature, see Attached Documents chapter 7). Also it is possible to appreciate that while for the first BPF is higher the amplitude of the external points for the second BPF (18) is the contrary, so that internal points have higher amplitude. The following figure shows the propagation along the channel. Lower frequencies tend to “dissapear” and also their amplitudes are lower as one goes far from the rotor-stator interface. For point B which is in this limit region, the amplitude is still relevant till the fourth harmonic. Figure 30. Pressure Spectra for points from the stator channel for ϕ = @0.043 Experimental Data Results 70 4.3.1.2.1.2. CONCLUSIONS FOR SIMULATIONS IN ϕ = @0.043 For this high load operating point , ϕ = @0.043 , the simulation results satisfy really well the expected behaviour. According to the frequency analysis, the fluctuations wave is not propagated very far upstream the stator diffuser. Only the limiting region next to the the rotor-stator gap presents relevant Cp amplitudes fo the BPF harmonics, till the third and forth one. Just a remark can be done according to flow rate distribution between the channels. It has been observed that the Cp amplitude is slightly stronger for the channel 20 than in channel one. It can be due to the influence of the tongue. At channel 20 is the last one, the pressure waves are blocked while in the first one where they can be easier propagated to the volute. The expected behaviour of the pump-turbine is verified from the CFD results. No sign of detachment flow are shown by the simulation for this operating point. Explanations to this fact could be the closeness of this point to the BEP (Best Efficiency Point. In the hill chart corresponding to 20º, for a ϕ = @0.043 the efficiency takes a value around 0.863, while the best efficiency point arrives to 0.88). In these cases, near to the BEP, for which the design was optimalized, velocity triangles are quite well adapted to the machine geometry: it was shown that α angle was of 18.82º , almost the same as the guide vane opening angle of 20º. In these cases, unsteady phenomena does not develop and RSI behaves as expected. 4.3.1.2.2 STUDY FOR ϕ = @0.028 Now it is the time to analyse the ϕ = @0.028 operating point. Corresponding to α angle of 9.8º, much smaller than the previous one, its pressure Experimental Data Results 71 fluctuactions during a runner revolution and for the same points as before (Points A and B) are shown in figure 31. Cp 0.14 Point A Point B Cp [-] 0.09 0.04 -0.01 0 40 80 120 160 200 240 280 320 360 -0.06 figure 31. Cp in the stator side for the points A and B near the Rotor-Stator gap for ϕ = @0.028 Waves are not so similar now, neither the phase-shift is so evident between the two points. At least, a kind of in-stable periodic tendency is inferable for each one, still taking a value of approximately 40º, according to the BPF. To make easier the analysis for these waves a frequency study will be neccesary. Experimental Data Results 72 Cp 0.08 0.06 Cp [-] 0.04 0.02 0 0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 360 -0.02 -0.04 D -0.06 B F H Figure 32. Cp curves for a stator streamline during one runner revolution from the simulation results for ϕ = @0.028 In this case, as the results are more disturbing, more analysis and comparisons are really necessary in order to find the reasons. Cp 0.08 Cp [-] 0.04 0 0 40 80 120 160 200 240 280 320 -0.04 -0.08 -0.12 A_(e) A_(i) Figure 33.. Comparison of pressure fluctuactions for one runner revolution between the internal and external points located in section A for ϕ = @0.028 360 Experimental Data Results Looking for a detachment, pressure fluctuactions from the same points but belonging to the external (e) and internal(e) region of the pump are compared. Results in figure 33 show clear differences compared with the perfect superposition of cp fluctuactions observed for ϕ = @0.043 . From this graph, one can see that pressure fields in the hub and the shroud for this partial load point present a very different behaviour in various moments during one revolution. The several peaks in the graph show these strong differences in pressure tendency and its reasons worth a deeper analysis. 4.3.1.2.2.1. FREQUENCY ANALYSIS As has been discussed and explained in previous paragraphs, it is possible to translate the Cp signal into “frequency domain” to extract the most relevant frequencies appearing on it. For this particular operating point , it will be really useful to help to find what can be the main causes of the mess in the pressure signals and to understand why it happens. The results are shown and commented in the following pictures: 73 Experimental Data Results Figure 34. Pressure Spectra for section next to the RS gap for ϕ = @0.028 There is already here an interesting remark: one can see that in CFD simulation for ϕ = @0.028 at section B are quite different than in section A, and which is more important, they do not present the same amplitude, especially between external and internal faces in A. While for BPF the external A has higher amplitude for 2 BPF is higher the internal. The most important detail to be remarked is the fact that the internal A has higher amplitude for the 2nd frequency than for the 1st. It corresponds to the diammetrical modes. To continue the study for these RSI effects in A, a comparison including the following point C in its same channel is shown in figure 35. From this picture we can observe that this behaviour occurs only for the internal point of section A, while the external present Cp amplitudes decreasing for lower amplitudes. The same observations can be done for points at C section, both internal and external. 74 Experimental Data Results Figure 35. Pressure Spectra comparison for points external and internal next to the RS gap of channel number 20 for ϕ = @0.028 Checking the other channel, (picture 36), one can find out the similar behaviour: 2nd BPF is more relevant than 1st . However, in this channel for both external and internal points in section B. On the contrary, point D behaves as in the first studied ϕ case. Figure 36. Pressure Spectra comparison for points external and internal next to the RS gap of channel number 1 for ϕ = @0.028 75 Experimental Data Results In order to finish this analyse , to check if it only this effect is found in the points next to the rotor stator gap and to have a global overview of the whole stator streamlines, plots in figures 37 and 38 are attached. Figure 37. Pressure Spectra for points from the stator channel 1 for ϕ = @0.028 Figure 38. Pressure Spectra for points from the stator channel 20 for ϕ = @0.028 76 Experimental Data Results Comparing both channels, conclusions about their differences must be drawn. There are no special comments to do about the points in the stay vane ( E, F, G, H) in either channel 20 or channel 1. On the other hand, point A, B, C and D have differents Cp amplitudes. While channel 20 shows decreasing tendency for lower amplitudes, in channel 1 point B experiments higher amplitude for the 2nd BPF than for the 1st . This effect should be taken into account (first be verified by experimental data), in order no to excite this frequency. Also, we could remark higher values in Cp amplitude for point D than for point B, which is the nearest tothe gap region. 4.3.1.2.2.2. CONCLUSIONS FOR SIMULATION WITH ϕ = @0.028 The results for this partial load case point out that something different and not designed is happening. Some explanations can be based on the analysis of velocity triangles, frequency domain and the theoric knowledge, as well as the comparison with the other studied points. As it has already been said during the frequency analyse, particular and different phenomena on the domain frequency of the signal were found. Differences between channels and between external and internal points amplitudes, have been shown in several of the compared cases, not only in frequency domain but also in time domain. As a first conclusion according to all the CFD observations, it is possible to infer that this disturbance must be due to the fact that in partial loads, far from the designed point (BEP) of the operation, a detachment of the flow takes place in the rotor-stator gap. It is caused by the blockage created by offdesign incidence angle between the flow and the vanes (it can be detected 77 Experimental Data Results 78 from the velocity triangle at this point, which shows a α angle of 9.8º, quite far from the 20º of the guide vane opening). These blockage effects are increased in pump-turbines where the gap is normally bigger and the vanes are thicker. [NICO02], [NICO06]. Wakes, eddies and other unsteady phenomena are developped then, causing vibration, noise and blade cracking. Other unsteady effects influence the irregular distribution of the flow rate among the channels. Further studies could be done on this matter. 4.3.2 PARTIAL SIMULATIONS Three partial computational domain approximations were set and calculated for Hydrodyna Pump. Their analyse can be interesting as they can be used to prove the capacity (or not) of partial approaches performance, in order to evaluate the quality and the power of the numerical simulation methods used. However, it is considered more necessary to evaluate first the full pumpturbine domain simulations and later try to ratify them for the partial approaches. This paper is focused on the first part, so that partial solutions will be only put on show and only for ϕ = @0.028 . Three possible approaches were studied and introduced in table 5 in paragraph 3.2. Keeping in mind that Hydrodyna has 9 runner blades and 20 guide vanes, the first and more basic approach consisted in two guide vane chanels for one blade impeller (see figure 39). Its pitch ratio was admissible and the pre-processing and the processing steps much quicklier. Experimental Data Results 79 Figure 39. View of the first restricted domain simulated (1/2) Initial pressure fluctuactions were already studied for this simulation results at several points. An example, showing the results of Cp calculs (this parameter will be explained in detail later; see the following paragraph about the complete machine simulation) for four points (See figure 25) in the stator channel on the right is shown in the following plot : Cp 0.04 0.03 0.02 Cp [-] 0.01 0 1 21 41 61 81 101 12 1 1 41 1 61 18 1 2 01 2 21 2 41 2 61 28 1 3 01 3 21 3 41 36 1 3 81 401 42 1 44 1 46 1 4 81 501 52 1 54 1 5 61 5 81 601 62 1 6 41 6 61 68 1 7 01 7 21 -0.01 -0.02 -0.03 -0.04 -0.05 D B F H Figure 40. Cp curves for a streamline during a runner revolution from the results of the partial approach 1/2 Experimental Data Results We can easily appreciate lack of any kind of periodicity and no correlation between the curves. This approach should not be taken into account. A second partial approach was made, this time from two blades and four guide vanes (see figure 41). The pitch ratio is the same as before but the rest of global factors influencing partial simulations will change from the first case. Results obtained in this case are shown in the following picture 42. To be compared with the first approach results, the pressure fluctuactions at the same points (see figure 25) are chosen; now corresponding in the figure to the stator upper streamline. Figure 41. View of the second restricted domain simulated (2/4) 80 Experimental Data Results 81 Cp 0.02 0.015 0.01 Cp [-] 0.005 0 1 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321 341 361 381 401 421 441 461 481 501 521 541 561 581 601 621 641 661 681 701 721 -0.005 -0.01 -0.015 -0.02 -0.025 -0.03 D B F H Figure 42. Cp curves for stator streamline during a runner revolution from the results of the partial approach 2/4 Curves are not completely periodic yet , as it had happened for the first approach. On the other hand, some tendencies are already more clearly visible. It is possible to distinguish kind of periods related to the BPF for points B and D, both the nearest to the Rotor-Stator region, which are normally the most influenced by pressure fluctuactions generated by RSI. The last partial simulation was made. The computational domain in this case was composed by three runner blades and seven guide vanes. Pitch ratio was better than in the previous partial cases so that the simulation becomes more reliable and its results expected to be more accurate and in better agreement with reality. The following pictures (43 and 44) show the corresponding computer domain and a example of the analyse of the results. Experimental Data Results 82 Figure 43. View of the third restricted domain simulated, (3/7) Cp 0.02 0.015 0.01 Cp [-] 0.005 0 1 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321 341 361 381 401 421 441 461 481 501 521 541 561 581 601 621 641 661 681 701 -0.005 -0.01 -0.015 -0.02 -0.025 D B F H Figure 44. Cp curves for a stator streamline during a runner revolution from the results of the partial approach 3/7 The same streamline (see figure 25) as before is used for the pressure plot . In this case, some changes arise. The waves seem more periodical and a modulated wave is present. Especially, points B and D, corresponding to the 721 Experimental Data Results ones nearest to the rotor-stator gap on the selected streamline, present a more visible periodic behaviour and BPF can be inferred in them. As a conclusion, we must say that the second and especially the third approximation could allow to have a first impression of the phenomena located in the rotor-stator gap and to distinguish some aspects of the pressure fluctuactions and peaks associated to the RSI . Besides, some parameters as efficiency and ψ were quite well approximated by this partial approaches. However, more investigations for future applications and improvements at this matter are still needed. 83 5 Experimental Data Results Experimental Data Results 85 5 EXPERIMENTAL DATA RESULTS Tests for Hydrodyna Project were done on a reduced scale model of the pump-turbine. This practice is very common for turbomachines, since large sizes of real machinery make difficult their setting on laboratories or their reproduction in 1:1 scales. Hydrodynamic similarity theory proves that results obtained from models are applicable to original devices. These theories include similarity properties not only in a geometric domain but also in other operating variables such as velocity, installation conditions or change of fluid. [TANA90]. For the experiments in this paper, the similarity is applied in geometrical terms. The reduced model keeps geometrical rules and as the same time, the laboratory tests were carried under controlled conditions, stablished to reproduce a concrete operating point. Hence, results obtained for the model will be accurately the same as the expected for the prototype. 5.1 MEASUREMENT TECHNIQUES Pressure measurements were made in the EPFL laboratory (LMH) test rig. LDV(LaserDopplerVelocimeter) [WIKI07] and PIV (Particle Image Velocimeter) [WIKI07] measurements were done for various operating conditions of the Francis model. Not the same techniques could be used for the stator and rotor measurementss. The signal transmission between them was done by wireless photodiodes. The synchronization of the data sampling is performed through a master-slave scheme in the rotating parts. Triggers were led by a tachometer signal. Experimental Data Results 86 In the distributor channels, 48 piezoresistive miniature sensors were located to catch the unsteady pressure in their walls. This kind of sensors are applied for hydro and aerodynamic pressure measurements. In this case of hydraulic pressure measurements, the active zone of the sensor must be waterproof. The sensitive chips are made of silicon and mounted in a Wheatstone bridge. In order to receive the deformation signal, the pressure indicator pieces are located in the limits regions of the membrane where traction and compression effects are expected. The Wheatstone bridge will experiment a disequilibrium in the moment of the pressure application, and it will create a variation of voltage which will be measured and registered. The stator measures used for this paper were taken at a sample frequency of 51.2 kHz. For the rotating parts, an instrumented shaft was developed by the EPFL to acquire the unsteady pressure fluctuactions at the runner blades walls. It consisted in a signal conditioning electronic composed of 32 preamplifiers and filters and installed on board. Main technical characteristics of the applied system are given in table 11. To see all their positions in the model, have a look on chapter 7 pictures: “Sensor locations”. Amplification Factor Range From 1 to 1000 Acquisition Boards Location Pump-turbine Shaft Maximum Sampling Frequency 20 KHz Memory Storage capacity 64000 samples/channel Digitized Data Transfer Rate 1.5 Mbits/s Table11.. Characteristics of measurement techniques for the rotor Simulations for the pump-mode were calculated after experiments, so that we already knew which points to investigate as being the ones measured. Experimental Data Results 87 Although more points than ϕ = @0.043 and ϕ = @0.028 were tested, only these two suggested a deeper analyse after the measures. Laboratory technicians noticed during them changes in the machine behaviour that make them interesting. In the following, results for those two operating conditions will be shown in a similar way as it was done in the last part of the chapter before. 5.2 EXPERIMENTAL RESULTS FOR ϕ = @0.043 As it has already been explained, the interest was finally focused on the points near the rotor-stator gap, so the experimental data results concerning only those points will be shown. To make easier comparisons, the same graphs and parameters as for the simulation are analysed here. It has already been mentioned that the sample frequency for this experience had a value of 51200 Hz. Having in mind that pump speed rotation was of 900 rpm, to have a impeller revolution 3414 pressure measures are needed. Starting by the Cp plot for the points corresponding to the simulated A and B, the pressure fluctuactions for a runner revolution present the following shape (figure 45): Experimental Data Results 88 Cp Cp [-] 0.04 -0.01 1 191 381 571 761 951 1141 1331 1521 1711 1901 2091 2281 2471 2661 2851 3041 3231 Point A -0.06 Point B figure 45. Cp pressure fluctuactions for points A and B during a runner revolution from experimental measures for ϕ = @0.043 Although the noise gives the signal a less regular form, one can see the strong similarities between this plot and the same one corresponding for the simulation results. (see figure 26). A filter is usually applied to this curves to ease the analysis. A priori, the agreement between numerical solution and experimental looks to be admissible and quite satisfactory for this operating point. In these results one can appreciate once again the RSI expected consequences: - Individually, each point experiments a periodic fluctuaction, mostly defined by the blade passage frequency. It has a value corresponding to 40º, which using the relation between the sample frequency and the rotational speed of the machine, corresponds approximately 380 samples. On purpose, the sample-step between two marks used in the graph is 190, so that one can perceive easier the periodicity separated by 2 intervals. -As for the simulation results, phase shift of around 18º (due to equally geometric distribution of the 20 vane guides) is also conserved between pressure measurement at point A and point B. Experimental Data Results 89 Approximately the phase shift correspond to a one interval between two marks Further analysis will be done for this operating point to compare with the “perfect” behaviour results from CFD calculations to verify if in real experiments it behaves as expected. Cp [-] Cp 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 -0.02 -0.025 -0.03 -0.035 -0.04 -0.045 -0.05 1 191 381 571 761 951 1141 1331 1521 1711 1901 2091 2281 2471 2661 2851 3041 3231 D B F Figure 46. Cp curves for a stator streamline during one runner revolution from the experimental results for ϕ = @0.043 Cp 0.09 Cp [-] 0.06 0.03 0 1 191 381 571 761 951 1141 1331 1521 1711 1901 2091 2281 2471 2661 2851 3041 3231 -0.03 -0.06 A_(e) A_(i) Figure 47. Comparison of pressure fluctuactions measured for one runner revolution between the internal and external points located in section A for ϕ = @0.043 Experimental Data Results (Remark: Point H is not included as it used to be for the CFD calculations, because there was no pressure sensor at this point. Anyway, their fluctuations are not really relevant. ) It would be convenient to filter the signals for a global comparaison between them and the CFD ones. Anyway, the agreement with what had been observed for ϕ = @0.043 from the CFD simulations is quite good, leaving the most interesting part of tha analysis to the other operating point ϕ = @0.028 . Figure 46 shows the propagation of the pressure wave is quite normal along the streamline. As one goes further from the Rotor stator gap the pressure fluctuactions dissapear. Figure 47 shows no differences between pressure fluctuaction amplitude in the external point A and the internal. So, according to measures, there is no sign of recirculation or whirls during the revolution for this point. 5.2.1 FREQUENCY ANALYSIS Basically the same procedure that was applied for numerical data will be applied for the experimental measurements in order to analyse the frequency domain of the signals. As it has already been shown in time domain, laboratory measurements have much noise; hence in many cases a phase average over the real pressure measurements has been done to obtain a cleaner signal. First of all, in figure 48 there is a comparison of the frequency spectra for all points in the section next to the rotor stator interface. It includes both external and internal points A and B, located in the two stator channels. 90 Experimental Data Results 91 Figure 48. Pressure Spectra of pressure measurement for section next to the RS gap for ϕ = @0.043 For ϕ = @0.043 no differences between any of these points are found. They all verify the decreasing tendency as expected for the Cp amplitude with frequencies from the first BPF. Their values are precisely similar between external and internal and between one channel and the other. For a global idea about the whole channel behaviour, figure 49 is shown. Everything happens as expected: there is a global tendency to strongly decrease in Cp amplitude for the higher frequencies, which means attenuation. Near the rotor-stator region this amplitude is relevant, dissapearing from the 3rd BPF, but the propagation is not so strong to be important at the stay vanes. Experimental Data Results Figure 49. Pressure Spectra for points from the stator channel for ϕ = @0.043 5.2.2 CONCLUSIONS FOR ϕ = @0.043 As the experimental analysis provides almost the same results as the CFD ones, the conclusions for this operating point are the same. Furthermore, this perfect agreement between simulation methods and experimental measures puts into show that CFD is able to model the real phenomena ocurring in the turbomachines, even for the rotor-stator interaction effects, at least when no unsteady effects are present in the flow. Now, CFD availability has to be still verified for all the operating ranges. The partial load point will be studied from the analysis of the following measurement results at ϕ = -0.028, which according to CFD results should present signs of undesirable phenomena. 92 Experimental Data Results 93 EXPERIMENTAL RESULTS FOR ϕ = @0.028 5.3 In figure 50 Cp fluctuactions in A and B are shown as usual. This time, the plot seems to have more noise than for the case before, and it is harder to determine the periodicity or the phase gap between the signals. Some similar problems were found for the simulation results but in the experimental data analysis the shape are more perturbed and disarranged. Cp Point A Point B Cp [-] 0.04 -0.01 1 191 381 571 761 951 1141 1331 1521 1711 1901 2091 2281 2471 2661 2851 3041 3231 -0.06 Figure 50. Cp fluctuactions for points A and B during a runner revolution from experimental measures for ϕ = @0.028 Pressure fluctuactions for the whole stator streamline are also provided to be compared with the CFD one. More chaotic behaviour than for the high discharge point is found. More particularly for point B, next to the gap, instantaneous and not repeated fluctuation are registered. Experimental Data Results 94 Cp [-] Cp 0.045 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 -0.02 -0.025 -0.03 -0.035 -0.04 -0.045 1 191 381 571 761 951 1141 1331 1521 D 1711 1901 B 2091 2281 2471 2661 2851 3041 3231 F Figure 51. Cp curves for a stator streamline during one runner revolution from the experimental results for ϕ = @0.028 Following the same procedure as for the simulations, pressure comparison between external and internal points of the walls were carried out. The plot obtained was the following (Figure 52) Cp 0.04 0.03 Cp [-] 0.02 0.01 0 -0.01 1 191 381 571 761 951 1141 1331 1521 1711 1901 2091 2281 2471 2661 2851 3041 3231 -0.02 -0.03 -0.04 A_(e) A_(i) Figure 52. Comparison of experimental fluctuations for one impeller revolution between the internal and external points located in section A for ϕ = @0 .028 Experimental Data Results 95 This plot presents many fluctuactions for both points, internal and external. In fact, it does not agree perfectly with the one obtained from the simulation solution, where the period was longer but the differences in peaks amplitude value between external and internal points were larger. To distinguished properly these differences the frequency analysis is followed. 5.3.1 FREQUENCY ANALYSIS Frequency analysis can help to find and to understand the reasons of what happened during the experience at this operating point. The first graph shows the frequency spectra for the measurements in points near to the rotor-stator gap. The most relevant aspect from this graph is the differences between external and internal Cp amplitude as well as the bigger Cp amplitud for the 2nd frequency than for the 1st . Differences in pressure amplitude between external and internal points can be a sign of detachment and recirculation, as it has already been explained for the CFD results. In what respects to the importance of the second frequency, which is observed at the four points, it can be explained for the RSI and the diammetrical modes that where theorically introduced in chapters before. It had already been predicted by the CFD calculations, which suposses a great achievement of this method. It is remarkable to know when operating the machine at this partial load point that the 2 BPF has the highest amplitude, in order to avoid this frequencies. Now, differences in amplitude for external and internal points for both channels are shown in figure 53, giving strength to the possibility of a detachment in the flow. Experimental Data Results 96 Figure 53. Pressure Spectra of pressure measurement for section next to the RS gap for ϕ = @0.028 To closer analysis, one channel is selected and the two points nearest to the rotor-stator gap are studied. (Figure 54) The highest amplitude for the second BPF dissapears at point D, and it has no sign of differences between external or internal position either. Figure 54. Pressure Spectra comparison for points external and internal next to the RS gap of channel number 1 for ϕ = @0.028 Experimental Data Results Just to check if it something different takes place for the other channel, a similar graph is shown in figure 55 for channel 20. Cp amplitudes of points A and B, internal and external, are the same as for the corresponding points in channel number 1. Only a slight difference in Cp amplitude value for external and internal A compared to B could be mentioned as a sign of differences in the flow rate distribution among the channels. Figure 55. Pressure Spectra comparison for points external and internal next to the RS gap of channel number 20 for ϕ = @0.028 Cp amplitudes in frequency domain all along the stator channel 20 and channel 1 are considered in the two following figures and their analysis draw the following conclusion: 97 Experimental Data Results Figure 56. Pressure Spectra for points from the other stator channel 20 for ϕ = @0.028 Figure 57. Pressure Spectra for points from the stator channel 1 for ϕ = @0.028 Both plots show the same characteristic: At this operating point, the highest Cp amplitude is the one of the 2nd BPF frequency for the points at the rotorstator interaction zone. On the other hand, it attenuates for the points belonging to the stay vanes, where the 1st BPF has the higher amplitude. 98 Experimental Data Results 5.3.2 CONCLUSIONS FOR ϕ = @0.028 Conclusions for experimental data are the same as for CFD results. Although the agreement between them must be detailed in the next chapter in general aspects they provide an almost identical final conclusion: For the point of ϕ= -0.028, frequency and time domain analysis show unsteady phenomena in the flow. Pressure differences found between channels and between near points from the same channel or even between symetrical points at different faces of the channel put into show the possible appearance of wakes, flow detachment or swirls and eddies, which are increased by the the pumpturbines design, as it has already been explained for the conclusion in the CFD results. The presence of highest Cp amplitudes for the 2 BPF at the points near the RS interaction region for the laboratory measurements ratifies the diammetrical modes theory, which had already predicted this RSI effect over the flow in the difusser. 99 6 Comparaison of results and Conclusions Comparison of Results. Conclusions 101 6 COMPARAISON OF RESULTS. CONCLUSIONS Finally, it is the time for the evaluation of the methods and the understanding of the results. It has also to be said that once the simulation has been done and having all the measurements, there are a huge amount of possible points and moments to be studied in order to find differences between channels, between Cp amplitudes, between particular moments... Some of the most interesting graphics analysed during the preparation of this paper, are attached in chapter 7 for further detailed examples and information. Though the aim of this paper was basically the evaluation of the methods, it has provided an initial analysis to this study case, in order to make easier future further studies and measures. In fact, conclusions in chapters before and all the theory and explanations to the dynamic flow phenomena included along all the paper gives evidence of it. 6.1 DETAILED COMPARAISON OF RESULTS As the individual characteristics of each point have already been described and analysed in the two chapters before, now only the differences between CFD and Experimental results will be highlighted. It has already been said that the number of possible comparisons has become too large, the selected points to be compared were the following: Comparison of Results. Conclusions - 102 Four points: the two ones in section A and the two ones in B, from different channels but all at the limit region and in both internal and external faces. - A whole streamline line in the stator channel, more precisely, the external one in channel number 1. (points B, D, F and H in the external face) Comparison of Results. Conclusions 103 Firstly, the high discharge case ϕ = -0.043 compared results are shown: figure 58. Comparison for ϕ=- 0.043 for points in the RS region The agreement is quite accurate in both time and frequency domains for all the points. The amplitudes are perfectly estimated, except for the 2nd harmonic which according to measures is lower than CFD prediction. Comparison of Results. Conclusions 104 figure 59. Comparison for channel 1 at ϕ = -0.043 The good agreement between experimental and CFD results is confirmed by figure 59. It is also remarkable the fact that CFD predicts the the highest Cp amplitude for the 2nd BPF at point F, which is ratified by the experimental measures. Comparison of Results. Conclusions 105 Now, the partial load case ϕ = -0.028 results are evaluated: Figure 60. Comparison for ϕ=- 0.028 for points in the RS region The most remarkable event here is the disagreement about the mean pressure level between channels. While CFD predicts a similar level for both channels, measurements result show a lower mean pressure level for channel 20 (points A) than for channel 1 (points B). In what respects to frequencies, although there is agreement about the highest amplitude for the 2nd BPF, in general CFD had predicted higher values than what measures results show, especially for the 1st harmonic. Comparison of Results. Conclusions 106 figure 61. Comparison for channel 1 at ϕ = -0.028 In time domain, the same points present higher pressure recovery than in CFD simulation results. It is especially remarkable between points F and H. In frequency domain, there is a large difference between CFD amplitudes value prediction at the 1st BPF and measurements results. However, CFD has quite well prediction for the propagation from B to D and for the fact that at point D the 1st BPF dominates over the rest. Comparison of Results. Conclusions 6.2 107 FINAL CONCLUSIONS • The capacity of numerical methods for representing operating points where no unsteady behaviour takes place is validated from experimental data. It is proved to be well developped according to the strong agreement between numerical and experimental data for the entire computational domain simulations at high discharge point. • Operating points presenting unsteady behaviour, as usually occurs for partial loads in pumps, are quite well simulated by CFD methods. However, some disagreements are found with experimental data in what respects to amplitude and to attenuation of the flow pressure fluctuations. • The presence of diametrical modes has been validated from these analysis. The RSI is well simulated by the CFD methods, detecting the majority of the cases where the 2 BPF had the highest Cp amplitude. • Partial load affects critically pump-turbines pressure fluctuactions. Flow detachment is developped near the rotor-stator interaction zone, creating a source of disturbance and instability at this region that propagates perturbing the whole machine flow. • Investigations must improve and evaluate partial simulations as it has been introduced in this paper. Admissible results are achieved for the partial simulation 3/7, suggesting the possibility of full development for these approaches. 7 Attached documents Attached documents 7 ATTACHED DOCUMENTS 7.1 SENSORS LOCATIONS 109 Attached documents 110 Attached documents 7.2 111 HYDRODYNA’S NOMENCLATURE Stator Upper (i) Lower (l) EXP(#ch) Sensor CFX XL Col EXP(#chl) Sensor CFX XL Col Ch 20 47 78 UpDist22 AV 21 122 LowDist22 V Section o_2 A B Ch 1 39 59 UpDist7 AN 24 107 LowDist7 Y Ch 20 44 53 UpDist1 AS 20 101 LowDist1 U Section o_3 C D Ch 1 36 54 UpDist2 AK 23 102 LowDist2 X E F Ch 20 43 29 UpSt1 AR 19 41 LowSt1 T Section v_4 Ch 1 35 30 UpSt2 AJ 22 42 LowSt2 W Ch 20 42 35 UpSt7 AQ 18 47 LowSt7 S Section v_5 G H Ch 1 34 36 UpSt8 AI - 48 LowSt8 - Rotor Blade #1 EXP(# ch) 12 13 14 15 16 17 Sensor P12 P13 P14 P15 P16 P17 CFX Bba1 Bba2 Bba3 Bba4 Bba5 Bba6 XL Col M N O P Q R Blade #2 EXP(# ch) 1 2 3 4 5 6 7 8 9 10 11 Sensor P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11 CFX Bbb1 Bbb2 Bbb3 Bbb4 Bbb5 Bbb6 Bbb7 Bbb8 Bbb9 Bbb10 Bbb11 XL Col B C D E F G H I J K L Blade #3 EXP(# ch) 19 20 18 Sensor P19 P20 P18 CFX Bbc10 Bbc11 Bbc7 XL Col T U S Attached documents 7.3 112 HYDRODYNA’S HILL CHART η [-] HYDRODYNA 0.9 0.89 0.88 0.87 0.86 0.85 0.84 0.83 0.82 0.81 0.8 0.79 0.78 0.77 0.76 0.75 0.74 0.73 0.72 0.71 0.7 -0.05 12 deg 14 deg 16 deg 18 deg 20 deg 22 deg 24 deg 26 deg 28 deg -0.045 -0.04 -0.035 -0.03 -0.025 -0.02 -0.015 ϕ [-] HYDRODYNA 1.2 1.15 1.1 1.05 12 deg 14 deg ψ [-] 1 16 deg 0.95 18 deg 0.9 20 deg 22 deg 0.85 24 deg 26 deg 0.8 28 deg 0.75 0.7 -0.05 -0.045 -0.04 -0.035 ϕ [-] -0.03 -0.025 -0.02 -0.015 Attached documents 7.4 EXAMPLE OF A LIST OF SET-UP EXPRESSIONS LIBRARY: CEL: EXPRESSIONS: R1 = 0.261875 [m] Aref = pi*R1^2 omega = 652.9 [rev min^-1] omegaDL = omega/1.0 [rad] DTime = 1.0/abs(omegaDL) DTimeNr = 360 DTimeAngle = 360 [deg]/ DTimeNr DTimeFr = 0.5[rad]/abs(omega) DTimeTrn = DTimeAngle/abs(omega) nb = 9 rho = ave(Density)@Asc I zb = 9 P1 = -1.0 * massFlowInt(ptotstn)@Int b o Side 2 / rho * zb / nb Q1 = -1.0 * massFlow()@Int b o Side 2 / rho * zb / nb E1 = P1/(Q1*rho) nv = 20 zv = 20 P2 = massFlowInt(ptot)@Int b o Side 1 / rho * zv / nv Q2 = massFlow()@Int b o Side 1 / rho * zv / nv E2 = P2/(Q2*rho) P3m = massFlowInt(ptotstn)@Ad 3m/ rho * zb / nb Q3m = massFlow()@Ad 3m/rho * zb / nb E3m = P3m/ (Q3m*rho) P5 = -massFlowInt(ptot)@ Asc I/ rho * zv / nv Q5 = -massFlow()@Asc I/rho * zv / nv E5 = P5/ (Q5*rho) ERef = 0.5*omegaDL^2* R1^2 TB = (torque_z()@Bb + torque_z()@Sb i +torque_z()@Sb e )*zb/nb Pm = -1.0 *TB* omegaDL 113 Attached documents Eta1 = (P2 - P3m)/ Pm Eta5 = (P5 - P3m)/ Pm Phi = 0.02853948 QPhi = pi*omegaDL*R1^3*Phi Phi b 1 = - massFlow()@REGION:RS ROTOR / rho * zb / QPhi Phi b 2 = - massFlow()@REGION:RS ROTOR 2 / rho * zb / QPhi Phi b 3 = - massFlow()@REGION:RS ROTOR 3 / rho * zb / QPhi Phi b 4 = - massFlow()@REGION:RS ROTOR 4 / rho * zb / QPhi Phi b 5 = - massFlow()@REGION:RS ROTOR 5 / rho * zb / QPhi Phi b 6 = - massFlow()@REGION:RS ROTOR 6 / rho * zb / QPhi Phi b 7 = - massFlow()@REGION:RS ROTOR 7 / rho * zb / QPhi Phi b 8 = - massFlow()@REGION:RS ROTOR 8 / rho * zb / QPhi Phi b 9 = - massFlow()@REGION:RS ROTOR 9 / rho * zb / QPhi Phi v 1 = - massFlow()@REGION:AV56 / rho * zv / QPhi Phi v 10 = - massFlow()@REGION:AV56 10 / rho * zv / QPhi Phi v 2 = - massFlow()@REGION:AV56 2 / rho * zv / QPhi Phi v 3 = - massFlow()@REGION:AV56 3 / rho * zv / QPhi Phi v 4 = - massFlow()@REGION:AV56 4 / rho * zv / QPhi Phi v 5 = - massFlow()@REGION:AV56 5 / rho * zv / QPhi Phi v 6 = - massFlow()@REGION:AV56 6 / rho * zv / QPhi Phi v 7 = - massFlow()@REGION:AV56 7 / rho * zv / QPhi Phi v 8 = - massFlow()@REGION:AV56 8 / rho * zv / QPhi Phi v 9 = - massFlow()@REGION:AV56 9 / rho * zv / QPhi Psi1 = (E1-E3m) / ERef Psi2 = (E2- E3m) / ERef Psi5 = (E5- E3m) / ERef Qb = QPhi/ (2.0 * zb) R1carre = 0,0685785 [m^2] URef = abs(omegaDL)*R1 END MATERIAL: Water Material Description = Water (liquid) Material Group = Water Data, Constant Property Liquids Option = Pure Substance 114 Attached documents Thermodynamic State = Liquid PROPERTIES: Option = General Material Thermal Expansivity = 2.57E-04 [K^-1] ABSORPTION COEFFICIENT: Absorption Coefficient = 1.0 [m^-1] Option = Value END DYNAMIC VISCOSITY: Dynamic Viscosity = 8.899E-4 [kg m^-1 s^-1] Option = Value END EQUATION OF STATE: Density = 997.0 [kg m^-3] Molar Mass = 18.02 [kg kmol^-1] Option = Value END REFERENCE STATE: Option = Specified Point Reference Pressure = 1 [atm] Reference Specific Enthalpy = 0.0 [J/kg] Reference Specific Entropy = 0.0 [J/kg/K] Reference Temperature = 25 [C] END REFRACTIVE INDEX: Option = Value Refractive Index = 1.0 [m m^-1] END SCATTERING COEFFICIENT: Option = Value Scattering Coefficient = 0.0 [m^-1] END SPECIFIC HEAT CAPACITY: Option = Value 115 Attached documents Specific Heat Capacity = 4181.7 [J kg^-1 K^-1] Specific Heat Type = Constant Pressure END THERMAL CONDUCTIVITY: Option = Value Thermal Conductivity = 0.6069 [W m^-1 K^-1] END END END END [.................................................] SIMULATION TYPE: Option = Transient EXTERNAL SOLVER COUPLING: Option = None END INITIAL TIME: Option = Automatic with Value Time = 0 [s] END TIME DURATION: Number of Timesteps per Run = 180 Option = Number of Timesteps per Run END TIME STEPS: Option = Timesteps Timesteps = DTimeTrn END END INITIALISATION: Frame Type = Stationary Option = Automatic 116 Attached documents INITIAL CONDITIONS: Velocity Type = Cylindrical CYLINDRICAL VELOCITY COMPONENTS: Option = Automatic with Value Velocity Axial Component = 0 [m s^-1] Velocity Theta Component = -5 [m s^-1] Velocity r Component = -3 [m s^-1] AXIS DEFINITION: Option = Coordinate Axis Rotation Axis = Coord 0.3 END END EPSILON: Eddy Length Scale = 0.001 [m] Option = Automatic with Value END K: Fractional Intensity = 0.05 Option = Automatic with Value END STATIC PRESSURE: Option = Automatic with Value Relative Pressure = 0 [Pa] END [.....................................................] OUTPUT FREQUENCY: Option = Timestep Interval Timestep Interval = 40 END END MONITOR POINT: Bba2 Cartesian Coordinates = 219.1267024 [mm], -111.3269521 [mm], \ 0.006628096 [mm] 117 Attached documents Option = Cartesian Coordinates Output Variables List = Pressure,Total Pressure,Velocity,Velocity in \ Stn Frame,Total Pressure in Rel Frame,Velocity u,Velocity \ v,Velocity w,Turbulence Kinetic Energy END [..............................................] CONVERGENCE CRITERIA: Residual Target = 0.0001 Residual Type = MAX END TRANSIENT SCHEME: Option = Second Order Backward Euler TIMESTEP INITIALISATION: Option = Automatic END END END 118 Attached documents 7.5 CODE 119 FOR CREATING VELOCITY TRIANGLES SESSION turbo more_vars ! $RSInterfaceRFR = "Int b o Side 1"; ! $CScaleMax = 100.0; ! $ArrowScale = 3; ! $LOff = 0.017 * $ArrowScale; ! $RPnt = 0.265; ! $ThetaPnt = 10 * 3.1415927 / 180.0; ! $XPnt = $RPnt * cos($ThetaPnt); ! $YPnt = $RPnt * sin($ThetaPnt); ! $XOff = -1.0 * $LOff * sin($ThetaPnt); ! $YOff = $LOff * cos($ThetaPnt); ! $XBeta = $XPnt + $XOff; ! $YBeta = $YPnt + $YOff; LIBRARY: CEL: EXPRESSIONS: CPlot = sqrt(($CScaleMax [m s^-1])^2 - Cu 1^2 - Cr 1^2) * step(0.04(X^2+Y^2)/\ 1[m^2] ) Cr 1 = Q1 / area()@$RSInterfaceRFR Cu 1 = massFlowAve(Velocity in Stn Frame Circumferential)@$RSInterfaceRFR U 1 = ave(Rotation Velocity)@$RSInterfaceRFR UPlot = sqrt(($CScaleMax [m s^-1])^2 - U 1^2 ) * step(0.04-(X^2+Y^2)/1[m^2] ) URef = abs(omegaDL)*R1 Attached documents 120 WPlot = sqrt(($CScaleMax [m s^-1])^2 - (Cu 1 - U 1)^2 - Cr 1^2) * step(0.04(X^2+\ Y^2)/1[m^2] ) BetaAve1 = atan2(Cr 1, U 1 - Cu 1) * 1 [rad] AlphaAve1 = atan2(Cr 1, Cu 1) * 1 [rad] END END END USER VECTOR VARIABLE:C Ave 1 Boundary Values = Conservative Calculate Global Range = On Recipe = Expression Variable to Copy = Pressure X Expression = Cu 1 * Theta Direction X + Cr 1 * Radial Direction X Y Expression = Cu 1 * Theta Direction Y + Cr 1 * Radial Direction Y Z Expression = CPlot END USER VECTOR VARIABLE:U Ave 1 Boundary Values = Conservative Calculate Global Range = On Recipe = Expression Variable to Copy = Pressure X Expression = U 1* Theta Direction X Y Expression = U 1 * Theta Direction Y Z Expression = UPlot END USER VECTOR VARIABLE:W Ave 1 Boundary Values = Conservative Calculate Global Range = On Recipe = Expression Variable to Copy = Pressure X Expression = (Cu 1 - U 1)* Theta Direction X + Cr 1 * Radial Direction X Attached documents Y Expression = (Cu 1 - U 1) * Theta Direction Y + Cr 1 * Radial Direction Y Z Expression = WPlot END POINT:Point 1 Apply Instancing Transform = On Colour = 1, 1, 0 Colour Map = Rainbow Colour Mode = Constant Colour Scale = Linear Colour Variable = Pressure Colour Variable Boundary Values = Hybrid Culling Mode = No Culling Domain List = All Domains Draw Faces = On Draw Lines = Off Instancing Transform = Default Transform Lighting = On Line Colour = 0, 0, 0 Line Width = 2 Max = 0 [Pa] Min = 0 [Pa] Node Number = 1 Normalized = Off Option = XYZ Point = $XPnt [m], $YPnt [m], 0 [m] Point Symbol = Crosshair Range = Global Specular Lighting = On Surface Drawing = Smooth Shading Symbol Size = 1.0 Transparency = 0.0 Variable = Pressure Variable Boundary Values = Hybrid 121 Attached documents Visibility = Off OBJECT VIEW TRANSFORM: Apply Reflection = Off Apply Rotation = Off Apply Scale = Off Apply Translation = Off Principal Axis = Z Reflection Plane Option = XY Plane Rotation Angle = 0 [degree] Rotation Axis From = 0 [m], 0 [m], 0 [m] Rotation Axis To = 0 [m], 0 [m], 0 [m] Rotation Axis Type = Principal Axis Scale Vector = 1 , 1 , 1 Translation Vector = 0 [m], 0 [m], 0 [m] X = 0 [m] Y = 0 [m] Z = 0 [m] END END VECTOR:Vector c Add Sample Vertex Normals = On Apply Instancing Transform = On Colour = 0.75, 0.75, 0.75 Colour Map = Rainbow Colour Mode = Use Plot Variable Colour Scale = Linear Colour Variable = C Ave 1 Colour Variable Boundary Values = Hybrid Coord Frame = Global Culling Mode = No Culling Direction = X Domain List = Vbd Draw Faces = On Draw Lines = Off 122 Attached documents Instancing Transform = Default Transform Lighting = On Line Width = 3 Location List = Point 1 Locator Sampling Method = Vertex Max = 30 [m s^-1] Maximum Number of Items = 100 Min = 0 [m s^-1] Normalized = Off Number of Samples = 100 Projection Type = None Random Seed = 1 Range = User Specified Reduction Factor = 1.0 Reduction or Max Number = Reduction Sample Spacing = 0.1 Sampling Accuracy = High Sampling Aspect Ratio = 1 Sampling Grid Angle = 0 [degree] Specular Lighting = On Surface Drawing = Smooth Shading Surface Sampling = Off Symbol = Line Arrow Symbol Size = $ArrowScale Transparency = 0.0 Variable = C Ave 1 Variable Boundary Values = Hybrid Visibility = On OBJECT VIEW TRANSFORM: Apply Reflection = Off Apply Rotation = Off Apply Scale = Off Apply Translation = Off Principal Axis = Z 123 Attached documents Reflection Plane Option = XY Plane Rotation Angle = 0 [degree] Rotation Axis From = 0 [m], 0 [m], 0 [m] Rotation Axis To = 0 [m], 0 [m], 0 [m] Rotation Axis Type = Principal Axis Scale Vector = 1 , 1 , 1 Translation Vector = 0 [m], 0 [m], 0 [m] X = 0 [m] Y = 0 [m] Z = 0 [m] END END VECTOR:Vector u Add Sample Vertex Normals = On Apply Instancing Transform = On Colour = 0.75, 0.75, 0.75 Colour Map = Rainbow Colour Mode = Variable Colour Scale = Linear Colour Variable = Rotation Velocity Colour Variable Boundary Values = Hybrid Coord Frame = Global Culling Mode = No Culling Direction = X Domain List = All Domains Draw Faces = On Draw Lines = On Instancing Transform = Default Transform Lighting = On Line Width = 3 Location List = Point 1 Locator Sampling Method = Vertex Max = 30 [m s^-1] Maximum Number of Items = 100 124 Attached documents Min = 0 [m s^-1] Normalized = Off Number of Samples = 100 Projection Type = None Random Seed = 1 Range = Global Reduction Factor = 1.0 Reduction or Max Number = Reduction Sample Spacing = 0.1 Sampling Accuracy = High Sampling Aspect Ratio = 1 Sampling Grid Angle = 0 [degree] Specular Lighting = On Surface Drawing = Smooth Shading Surface Sampling = Off Symbol = Line Arrow Symbol Size = $ArrowScale Transparency = 0.0 Variable = U Ave 1 Variable Boundary Values = Hybrid Visibility = On OBJECT VIEW TRANSFORM: Apply Reflection = Off Apply Rotation = Off Apply Scale = Off Apply Translation = Off Principal Axis = Z Reflection Plane Option = XY Plane Rotation Angle = 0 [degree] Rotation Axis From = 0 [m], 0 [m], 0 [m] Rotation Axis To = 0 [m], 0 [m], 0 [m] Rotation Axis Type = Principal Axis Scale Vector = 1 , 1 , 1 Translation Vector = 0 [m], 0 [m], 0 [m] 125 Attached documents X = 0 [m] Y = 0 [m] Z = 0 [m] END END VECTOR:Vector w Add Sample Vertex Normals = On Apply Instancing Transform = On Colour = 0.75, 0.75, 0.75 Colour Map = Rainbow Colour Mode = Use Plot Variable Colour Scale = Linear Colour Variable = W Ave 1 Colour Variable Boundary Values = Hybrid Coord Frame = Global Culling Mode = No Culling Direction = X Domain List = Vbd Draw Faces = On Draw Lines = Off Instancing Transform = Default Transform Lighting = On Line Width = 3 Location List = Point 1 Locator Sampling Method = Vertex Max = 30 [m s^-1] Maximum Number of Items = 100 Min = 0 [m s^-1] Normalized = Off Number of Samples = 100 Projection Type = None Random Seed = 1 Range = Global Reduction Factor = 1.0 126 Attached documents Reduction or Max Number = Reduction Sample Spacing = 0.1 Sampling Accuracy = High Sampling Aspect Ratio = 1 Sampling Grid Angle = 0 [degree] Specular Lighting = On Surface Drawing = Smooth Shading Surface Sampling = Off Symbol = Line Arrow Symbol Size = $ArrowScale Transparency = 0.0 Variable = W Ave 1 Variable Boundary Values = Hybrid Visibility = On OBJECT VIEW TRANSFORM: Apply Reflection = Off Apply Rotation = Off Apply Scale = Off Apply Translation = On Principal Axis = Z Reflection Plane Option = XY Plane Rotation Angle = 0 [degree] Rotation Axis From = 0 [m], 0 [m], 0 [m] Rotation Axis To = 0 [m], 0 [m], 0 [m] Rotation Axis Type = Principal Axis Scale Vector = 1 , 1 , 1 Translation Vector = $XOff [m], $YOff [m], 0 [m] X = 0 [m] Y = 0 [m] Z = 0 [m] END END TEXT: Text Alpha Colour = 0, 0, 0 127 Attached documents Font = Sans Serif Position Mode = Three Coords Text Colour Mode = Default Text Height = 0.018 Text Position = $XPnt [m], $YPnt [m], 0 [m] Text Rotation = 0 [degree] Visibility = On X Justification = Center Y Justification = None TEXT: Text Beta Colour = 0, 0, 0 Font = Sans Serif Position Mode = Three Coords Text Colour Mode = Default Text Height = 0.018 Text Position = $XBeta [m], $YBeta [m], 0 [m] Text Rotation = 0 [degree] Visibility = On X Justification = Center Y Justification = None 128 Attached documents 7.6 MATLAB FILE: 129 FREQUENCY ANALYSIS AND MULTIPLE COMPARISONS %% Initialisation and data aquisitation clear ; % Definition of operating points and corresponding data sources (all XL at % the moment OPS = ['028'; '043']; Phis = [0.028, 0.043]; Types = {'CFD','Raw Data','Phase Average'}; PSources = struct('File', {'MonP_vvo_3640_Phi028.xls','MonP_vvo_1440_Phi043.xls'}, ... 'Desc', 'CFD',... 'Type', 'XLColumn',... 'Start', {2+18, 2+18},... % -6 Times steps phase shift (40 - 6 = 34) 'Length', {3600, 1080},... 'NsPerRev', {360, 360},... 'XL2Pa', 1.000,... 'WLLow', 1,... 'CircShift', 0); PSources = [PSources; struct('File', {'538_08_22S_0_phi028.xls','538_08_16S_0_phi043.xls'}, ... 'Desc', 'Exp',... 'Type', 'XLColumn',... 'Start', {10, 10},... 'Length', {3*3414, 3*3414},... 'NsPerRev', {3414, 3414},... 'XL2Pa', 101300.0,... 'WLLow', 10,... 'CircShift', 0)]; PSources = [PSources; struct('File', {'538_08_22_PhaseAve_Stator.xls','538_08_16_PhaseAve_Stator.xls'}, ... 'Desc', 'Exp',... 'Type', 'XLColumn',... 'Start', {12, 12},... 'Length', {720, 720},... 'NsPerRev', {720, 720},... 'XL2Pa', 101300.0,... 'WLLow', 1,... 'CircShift', 200)]; % Definition of Sensor Locations in structures with all necessary info Secs = ['o2'; 'o3'; 'v4'; 'v5']; SecCell = {'o2'; 'o3'; 'v4'; 'v5'}; MnC = {'Dist', 'Dist', 'St', 'St'}; Attached documents 130 Sfcs = ['i'; 'e'; 'b']; MnS = {'Up';'Low'}; Chls = [20 1]; SLog = strcmp(cellstr(Secs),'o2'); % Declaration and dummy definition of Structure for a Monitor/pressure tap % point conctaining info about: % names in different namig schemes, channel numbers and so on % Structured info about its location etc. % the first index represents the section in the machine % the second: if it is exterior (Lower) or Interior (Upper) % the third: which diffuser channel we're in (20 or 1 at the time being) MPts = repmat(struct('Sec', 'o2', 'Sfc', 'i', 'Chl', 20, ... 'Cnl', 0, 'XlE', 'A', 'Pnt', 0, ... 'MnP', 'LowDist' , 'XlC', 'A', 'Mnn', 1 ),... [4 2 2]); % Selon la structure des indexes choisis les données sont placées %section [ ... o_2 ... ¦ ...o_3... ¦ ... v_4 ... ¦ ... v_5 ...] %HubShr [ i ¦ e ¦ i ¦ e ¦ i ¦ e ¦ i ¦ e ] %channel [20 01 20 01 20 01 20 01 20 01 20 01 20 01 20 01] Cans = [47 39 21 24 44 36 20 23 43 35 19 22 42 34 18 34] Pnts = [78 59 122 107 53 54 101 102 29 30 41 42 35 36 47 48] MnCX = [22 7 22 7 1 2 1 2 1 2 1 2 7 8 7 8] ColCFD = ['K' 'M' 'C' 'E' 'J' 'L' 'B' 'D' 'N' 'O' 'F' 'G' 'P' 'Q' 'H' 'I'] RepName = ['A' 'B' 'A' 'B' 'C' 'D' 'C' 'D' 'E' 'F' 'E' 'F' 'G' 'H' 'G' 'H'] for i = 1:4 for j = 1:2 for k = 1:2 MPts(i,j,k).Sec = Secs(i,:); MPts(i,j,k).Sfc = Sfcs(j); MPts(i,j,k).Chl = Chls(k); MPts(i,j,k).Cnl = Cans(4*(i-1)+2*(j-1)+k); MPts(i,j,k).XlE = Can2Col(Cans(4*(i-1)+2*(j-1)+k)); MPts(i,j,k).Pnt = Pnts(4*(i-1)+2*(j-1)+k); MPts(i,j,k).XlC = ColCFD(4*(i-1)+2*(j-1)+k); MPts(i,j,k).RpN = RepName(4*(i-1)+2*(j-1)+k); if MnCX(4*(i-1)+2*(j-1)+k)==0 MPts(i,j,k).MnP = 'None'; else MPts(i,j,k).MnP = strcat(MnS(j),MnC(i),sprintf('%d',MnCX(4*(i-1)+2*(j-1)+k))); end Attached documents 131 end end end for i = 1:16 MPts(i) end %% rho= 997; u= 25.062; %% % Choices of Pressure Data to display in Raw Data Plot OPs = [1 1 1 1]; % 1:028, 2:043 Tps = [1 2 1 2]; % '1:CFD, 2:Raw Exp, 3:Phase Average' nplots = numel(OPs); SubMean = 1:nplots; %Which mean to substract, default 1 2 3 4... Use 0 for none, do not use higher SubMean = [1 2 1 2]; %Example substract own mean from two first, and same two from 3rd and 4th data set CFD-EXP ...CFD-EXP %SubMean = [1 1 1 1] %PAve = [360 360 360 360]; %Do additional phase ave on data, 0 do nothing, for averageing of CFD Data % PAveM = [1 1 1 1]; % Reappend Phase Averaged data n times for graphics representation PAve = [0 0 0 0]; %Do additional phase ave on data, 0 do nothing, for averageing of CFD Data PAveM = [1 1 1 1]; % Reappend Phase Averaged data n times for graphics representation iPs=[]; iPs (1,1)= iPs (1,2)= iPs (1,3)= iPs (1,4)= PInd('o2','e', PInd('o2','e', PInd('o2','i', PInd('o2','i', 20, 20, 20, 20, Secs, Secs, Secs, Secs, Sfcs, Sfcs, Sfcs, Sfcs, Chls, Chls, Chls, Chls, size(MPts)); size(MPts)); size(MPts)); size(MPts)); figtype = '-dpng'; Title = sprintf('Comparison of Results CFD vs EXP in Rotor-Stator gap for Phi=%5.3f',... Phis(OPs(1))); PlotName = 'Cp_o2_i_e_Ch20_EXP_PhAve_028'; RawPlotName = [PlotName '_Time']; %% for i = 1:nplots MPts(iPs(i)) LStart = PSources(Tps(i), OPs(i)).Start; Length(i,1) = PSources(Tps(i), OPs(i)).Length; Attached documents NsPerRev(i,1) = PSources(Tps(i), OPs(i)).NsPerRev; switch Tps(i) case 1 XLCol = MPts(iPs(i)).XlC; case 2 XLCol = MPts(iPs(i)).XlE; case 3 XLCol = MPts(iPs(i)).XlE; end fscale(i,1) = {linspace(0,NsPerRev(i,1)/2,Length(i,1)/2+1)}; XLR = sprintf('%s%d:%s%d',XLCol, LStart, XLCol, LStart + Length(i,1)-1); Tmp = PSources(Tps(i), OPs(i)).XL2Pa * xlsread(PSources(Tps(i), OPs(i)).File, XLR ); if PSources(Tps(i), OPs(i)).WLLow > 1 wSize = PSources(Tps(i), OPs(i)).WLLow; Tmp=filtfilt(ones(1,wSize)/wSize,1,Tmp) end if PSources(Tps(i), OPs(i)).CircShift ~= 0 Tmp=circshift(Tmp,PSources(Tps(i), OPs(i)).CircShift); end if PAve(i) ~= 0 q = PAve(i); nph = floor(numel(Tmp)/q); for j = 1:(nph-1) Tmp(1:q) = Tmp(1:q) + Tmp(j*q+1:j*q+q); end Tmp(1:q)=Tmp(1:q)/nph; Tmp(q+1:end,:)=[]; Length(i,1) = q; for j = 1:PAveM(i)-1 Tmp(j*q+1:j*q+q) = Tmp(1:q) end Length(i,1) = PAveM(i)*q; end if i==1 PMean1 = mean(Tmp); end PMean(i) = mean(Tmp); Times(i,1) = {(0:Length(i,1)-1)/NsPerRev(i,1)}; if SubMean(i) ~= 0 cP(i,1) = {(Tmp-PMean(SubMean(i)))/(0.5*rho*u^2)}; else cP(i,1) = {Tmp/(0.5*rho*u^2)}; end end %% Raw Data Plot figraw = figure('Name','Cp fluctuation (time)','NumberTitle','off',... 'Position',[50 50 800 600], ... 'PaperPosition',[0 0 8 6], ... 'PaperUnits','inch', ... 'PaperPositionMode', 'manual'); tmp = Times{1,1}; for i = 1:nplots plotdata(2*i-1)= {Times{i,1};}; plotdata(2*i)= {cP{i,1}}; 132 Attached documents end h_cp = plot(plotdata{1,:}, 'LineWidth', 1.0); for i = 1:nplots % LegString = sprintf('P_{%s-%s-%d} - %s', MPts(iPs(i)).Sec, MPts(iPs(i)).Sfc, MPts(iPs(i)).Chl, char(Types(Tps(i)))); LegString = sprintf('{%s-%s} - %s', MPts(iPs(i)).RpN, MPts(iPs(i)).Sfc, char(Types(Tps(i)))); set(h_cp(i),'DisplayName',LegString) end legend(gca,'show'); xlabel('t f_n','fontsize',12,'fontweight','b'); ylabel('c_p','fontsize',12,'fontweight','b'); axis([0 1 -inf inf]); %axis([0 1 -inf 0.5]); title(Title); print (figtype, RawPlotName); saveas(figraw, RawPlotName, 'fig'); %% Choices of Pressure Data amongst the former to display for frequency analysis nfft = [1 2 3 4]; % Referrring to the Data formerly loaded, choice of as opposed to the 1..nplots loop FFTPlotName = [PlotName '_FFT']; % Fix Sample Rate per revolution to downsample higher sampled signals p = 360 % Fix number of periods to phase average longer time series nrev = 1 % only use 1 at the moment q=nrev*p fscale = linspace(0,p/2,q/2+1); fmax = 90 + 1; for i = nfft if NsPerRev(i,1) ~= p FftTmp = resample(cP{i,1},p,NsPerRev(i,1)); else FftTmp = cP{i,1}; end rslength = numel(FftTmp); if rslength > q nph = floor(rslength/q); for j = 1:(nph-1) FftTmp(1:q) = FftTmp(1:q) + FftTmp(j*q+1:j*q+q); end FftTmp(1:q)=FftTmp(1:q)/nph; FftTmp(q+1:end,:)=[]; end fft_tmp = fft(FftTmp) / q; fft_tmp(1) = 0.0; 133 Attached documents cP_Red(:,i) = FftTmp; cP_Cmplx(:,i) = fft_tmp(1:fmax); cP_Amps(:,i) = abs(fft_tmp(1:fmax)); cP_Phase(:,i) = angle(fft_tmp(1:fmax)); end %% FFT Amplitudes Stem Plots ftick = 9; xticks = 0:ftick:floor(fscale(fmax)); MCList = {'blue'; 'white'; 'red'; 'green'}; MList = {'o';'s';'d';'p'}; MEdgeList = {'b';'k';'r';'g'}; figfft = figure('Name','Pressure Spectra','NumberTitle','off',... 'Position',[100 100 800 600], ... 'PaperPosition',[0 0 8 6], ... 'PaperUnits','inch', ... 'PaperPositionMode', 'manual'); h_ch01 = stem(fscale(1:fmax)',cP_Amps, 'MarkerSize', 5); for i = nfft % LegString = sprintf('P_{%s-%s-%d} - %s', MPts(iPs(i)).Sec, MPts(iPs(i)).Sfc, MPts(iPs(i)).Chl, char(Types(Tps(i)))); LegString = sprintf('{%s-%s} - %s', MPts(iPs(i)).RpN, MPts(iPs(i)).Sfc, char(Types(Tps(i)))); set(h_ch01(:,i),'MarkerFaceColor',char(MCList(i)),... 'MarkerEdgeColor',char(MEdgeList(i)),... 'Marker',char(MList(i)),... 'DisplayName',LegString ) end axis([0 fscale(fmax) 0 0.01]); xlabel('f_n','fontsize',12,'fontweight','b'); ylabel('c_p Amplitude','fontsize',12,'fontweight','b'); set(gca, 'XTick',xticks, 'FontSize',11); legend(gca,'show'); title(Title); print (figtype, FFTPlotName); saveas(figraw, FFTPlotName, 'fig'); %% FFT Transfer Function for BPF's ntp = size(nfft,2)-1; ftick = 9; xticks = 0:ftick:floor(fscale(fmax)); nt = size(xticks, 2)-1; cP_Transfer = zeros(nt,ntp); for i = nfft(2:end) cP_Transfer(:,i-1) = cP_Cmplx(xticks(2:end)+1,i)./cP_Cmplx(xticks(2:end)+1,1); end cP_TAmp = abs(cP_Transfer); cP_TPhase = angle(cP_Transfer); 134 Attached documents 7.7 COMPARISON CFD AND EXPERIMENTAL RESULTS PLOTS 135 Attached documents 136 Attached documents 137 Attached documents 138 8 References References 140 8 REFERENCES [EURE99] www. eureka.be/inaction/ACcshowProjectOultine.do [ERKM04] Erkman, S. , « Vers une ecologie Industrielle « , Mayer Charles Leopold Eds, 2004. [ASEU04] Office federale de l’energie , « Statistique Suisse de l’electricité 2003 »Bulletin ASE/UCS n.8 , Berne, 2004. [WHIT03] White, F. M., “Mecánica de Fluidos “, McGraw Hill Professional, 2003. [GARC98] Garcia, A., A. López de la Rica y A. De la Villa. “Cálculo I: Teoría y problemas de Análisis Matemático en una variable” CLAGSA, Madrid, 1998 and “Calculo II: Teoría y problemas de funciones en varias variables” CLAGSA, Madrid, 2002. [ZOBE07] Zobeiri, A., J.L. Kueny, M. Farhat, F. Avellan, «Unsteady Pressure due to Rotor-Stator Interactions in Generating Mode of a Pump-Turbine: Numerical and Experimental Investigations”, 2007. [NICO06] Nicolet, C., N. Ruchonnet and F. Avellan. “Hydroacoustic Modeling of Rotor Stator Interaction in Francis PumpTurbine”, 2006. [TANA90] Tanaka, H. “Vibration behaviour and dynamic stress of runners of very high head reversible pump-turbines.” Proceedings, 15th IAHR Symposium, Belgrade, Yugoslavia,1990. References [OHUR90] 141 Ohura, Y., M. Fujii, O. Sugimoto, H. Tanaka and I. Yamagata, “Vibration of the powerhouse structure of pumped storage power plant.”, Proceedings, 15th IAHR Symposium, Belgrade, section U2., 1990. [CHEN61] Chen, Y. N., “Water-Pressure Oscillations in the Volute Casings of Storage Pumps.”, Sulzer Technical Review, Research Number,pp. 21-34, 1961 [OHAS94] Ohashi, H. , “Case Study of Pump failure due to Rotor-Stator Interaction”, International Journal ofRotating Machinery, 1994, Vol. I, No. 1, pp. 53-60. [BREN94] Brennen, C.E., “Hydrodynamics of Pumps.”, NREC, 1994. [FERZ99] Ferziger, P., M. Peric, “CFD Computational Methods for Fluids Dynamics”. Springer. 1999. [STOR02] Storey, B.D., “ Computing Fourier Series and Power Spectrum with Matlab” , 2002 [KUEN99] Kueny, J.L. and A. Guedes, “Identification of Rotor-Stator Unsteady Parameters”, September 1999. [JURI05] Juric, M., “Rotor-Stator Interaction”, 2005. [WIKI07] www.Wikipedia.com, 2007 [NASA06] www .grc.nasa.gov/WWW/K-12/airplane/nseqs.html, 2006. [LMH07] www. lmh. ch, 2007. [TUTO05] Tutorials CFX 11.0, 2005. References [ENER07] 142 Picture taken from : http://www. energies-renouvelables.org [IMAG07] Picture taken from: Images .google [WIKI07] Picture taken from: wikimedia.org [IMAG07] Pictures taken from: Images.google [TOPL07] Picture taken from: www.top500.org/lists/2007/06/performance_development [NICO02] Nicolet, C., F. Avellan. P. Allenbachd, A. Sapin and J.-J. Simond, “Proceeding of the XXI st IAHR Symposium and Systems” Vol. II. 2002, p.p. 799, 800,814-818, 823-828, 834, 848856, 881. 9 ACKNOWLEDGEMENTS Acknowledgements 144 9 ACKNOWLEDGEMENTS Finally, I will like to thank all the people who have helped me to do this project. Especially I thank my coordinator Olivier Braun, who was working with me sometimes till very late and who have dedicated many hours to solve my questions and my problems. Thank you also to the rest of the Numerical Methods Team in LMH, Alireza Zobieri and Cecile Muench, whom I have worked with during the first months and who have provided me many information and all the explanations and help when I needed it. I especially thank Alireza Zobieri, who had already studied the turbine mode, whose results I used in many times as a reference. I will like also to thank previous investigators (Christophe Nicolet, Mohamed Farhat, Jean-Louis Kueny..) because I have learnt much from their results and their publications. I would like to thank M. Avellan and my university coordinators José Ignacio Linares too, for offerring and allowing me this opportunity, respectively. I do not forget all the LMH staff , whom I was talking to and seeing everyday during the project and who get to make me an easier and more pleasant work. Last but not least, I thank all my family and friends, because without them, without their support and encouragement I would have not succeed to finish this project.