Sunteți pe pagina 1din 9
A discussion on numerical simulation in Pelton turbines Morten Hana, Norwegian University of Science and Technology, Norway Emait:hanaGtev.ntnu.no Abstract Compared to other turbines types the Pelton turbine has a free surface which compli- cates numerical computation, The two most common methods used for free surface in CFD is briefly presented. One of the methods is tested in two different CFD programs, Additional s multi phase code was tested, The simple test ease was con- structed to investigate some aspects of numerical simulation in Pelton turbines. The purpose of the simulation is to see if the codes can handle the splitting of the jet, the deflection of the high speed flow in the bucket and the flow over the outlet edge. ‘The simulations results is discussed along with how to improve further calculations. The results are also compared to experimental results, Résumé La simulation numérique de l'éconlement dans une turbine Pelton présente des diffi- cultés spécifiques dues & l'existence d'une surface libre. Les deux méthodes les plus couramment appliquées ce type de problémes sont décrites. Des simulations out ‘été effectuées dans un cas simplifié grace & deux codes numériques différents mettant en oeuvre lune de ces méthodes. L'objectif de ce travail était d'appécier s'il est ainsi possible de reproduire les diverses caractéristiques de Iéeoulement dans l'auget. Les résultats sont examinés accompagnés d'une discussion des possibles améliorations et dune comparaison avec des vues obtenues expérimentalement, 1. Introduction ‘The Pelton turbine has some prominent features which does not make numer ieal simulation straightforward. The most important is the free surface. The Pelton buckets assignment is to cut, split and redirect the jet. Together with the relative motion between the jet and buckets this causes the free surface to continuously and rapidly change shape and position, thus making the tracking of the surface important. A graphical method first described by Brekke (1], has been the only tool for theoretical analysis of the flow in Pelton turbines. Recent investigations and improvements of the method, Hana [2], showed the difficulties to develop this method into a fully numerical code, After this a simple particle analysis has been constructed, which analyses a single particle without the influence of other water particles, thus giving some unphysical results like crossed flow paths, However, the particles are under the influence of Coriolis and centrifugal accelerations and gives some indication of the flow patterns seen in high speed visualization. In order to develop this method more a flow solver for the fluid is needed. The purpose of this work is to investigate conventional free surface methods to see if they can be used for numerical analysis of the Pelton turbine. Fither as a substitute for the graphical method or in the further development. of the method. 2. Free surface representation ‘There are two general methods of approaching a free surface problem, namely the Lagrangian and the Eulerian methods. In the former method the liquid regions are subdivided into a finite difference or a finite element grid and each cell is moved and deformed according to the liquid region. The governing flow equations are simple and easy to solve in the Lagrangian formulation, since the momentum equations have no non-linear advection terms. However the movement and deformation of the grid increases the complexity. In an Eulerian formulation the grid remains fixed and liquid regions change shape and location on the mesh, An identification method is needed to carry the information about the fluid on the grid. One way to treat the identification method is to assign a scalar volume fraction function to each cell. This function varies between one and zero, depending whether the cell is full or empty of the fluid, respectively, This leads to the volume of fraction (VOF) methods as first described by Hirt and Nichols [3] In a VOF method initial interface geometry is discarded in favor of the discrete volume fraction data. Given the velocity field interfaces are tracked by evolving fluid volumes in time with the solution of the advection equation oF Bre rao Figure 2.1: Problem setup where f is the scalar volume fraction, u is the velocity, ‘At any time in the solution an exact interface location is not known, i.e. a given distribution of volume fraction data does not guarantee a unique interface topology. Interface is reconstructed from local volume fraction data, The reconstructed interface is then used to compute the volume fluxes necessary for the convective term in the volume evolution equation. A second way to treat the identification between the flnid and the grid is to assign massless markers fluid identities like density and velocity. Interpolation schemes is used to transfer particle information between the particles and the grid. The markers then solely represent the fluid on the grid. On the grid a set of flow equations are solved. ‘This is the marker and cell (MAC) method by Harlow and Welch [4]. Some important features of the method compared to others are compact interface, lack of numerical diffusion and the ability to withstand severe topological changes. But the method suffer from a more CPU demanding calculation and hence is more expensive and time consuming then the VOF method. 3. Numerical examples For the numerical tests the commercial code Flow-3D [5] and CFX-4 was cho- sen. Also an academic code was obtained. This code, named Ripple [6], is a Simulations results from Flow-3D Pressure countours 4) Initial t= 0.05 7 ©)1=0.1008 Figure 3.1: Resultst from Flow-3D showing pressure countours. two dimensional code where the source code is available, Both Flow-3D and Ripple uses VOF for free surface modeling. Since the Ripple code is a two dimensional code, a simple bucket like shape was constructed, as seen in figure 2.1. The problem consists of a 2.cm slit and two circular shaped “buckets” of radius 2 cm. The exit edge was lowered 0,05 em. The purpose of the simulation was to see if the codes could handle the splitting of the jet, the deflection of the high speed flow in the bucket and the flow over the outlet edge. CFX-4 uses a homogenous multi-phase model which assumes that the solution fields are identical for all phases, Different phases are separated by volume fractions which are found by solving separate continuity equations for each phase. Flow-3D and Ripple uses orthogonal grid, in which obstacles are represented by blocking out the appropriate cells. CFX-4 uses body fitted grid. This difference did not represent any problem when creating the 2 dimensional grid. ‘The results from Flow3D are seen in figure 3.1. The strange shape of the nozzle became necessary because droplets otherwise interfered with the jet. Figure 3.2: A result from Ripple showing a large break-up of the surface. ‘The jet front leaving the nozzle in Flow3D was deformed as seen in figure 3.1b. ‘This was not experienced with Ripple and CFX-4. This difference could not be contributed by the different surface tension models in Flow-3D and Ripple, since both were tested with and without surface tension and giving the same results. ‘The Ripple code experienced problems in handling the jet setup profile in figure 2.1. Using a setup similar to the shape in figure 3.1b gave a smoother transition over the splitter and faster calculations. ‘This setup problem demonstrates the problems with commercial codes. There is little information of the physics and numerical methods behind the calcula- tions. And therefore difficult. to decide how to improve the calculations. ‘The Ripple code on the other hand was easy to understand and manipulate, but was very sensitive to stability conditions. Because of the setup profile of the jet a smooth variation of water thickness over the outlet edge was obtained. And the different thickness gave different outlet angles which is of interest when simulating different exit, angels in a Pelton bucket at varying loads. The results from Ripple showed more variations in the surface shape, but not the variations in exit angels as function of water thickness. One result from Ripple is shown in figure 3.2. In figure 3.3.0 result from CFX-4 is shown, The interface is thicker then in the VOF codes, which must be caused by the multi-phase model, In addition to the multi-phase model a surface sharpening algorithm was used to minimize the interface thickness. If this had not been used, the interface would be much more smeared out in time because of numerical diffusion in the volume fraction equations. Volume fractions CFK-4 Figure 3.3: Result from CFX-4 showing volume fractions. 4. Experiment In order to verify the numerical simulations a simple test rig was constructed, It proved difficult to reproduce a two dimensional test case, The most difficult part of the experiment was to align the nozzle with the bucket. One still picture from the experiment is shown in figure 4.1, The photo was taken with an ordinary camera and a strobe light. The picture shows the flow between two plexi glass plates, which is in front and behind the water in the picture. ‘The exit angle near the wall is not correct, because of the boundary layer near the plexi glass. This made the measurements of the angle difficult and crude. 5. Discussion The simulation showed that simple 2D calculations are possible. However this was not straightforward. Setup problems included a unknown effect that deformed the jet front in Flow-3D, However, this deformation gave a setup profile that made calculations smoother over the splitter in the Ripple code. Figure 4.1: Picture from experiment. Ripple had otherwise problems in handling a square front of the jet, but this was improved when the front was modified. Compared to figure 4.1 all codes shows some apparent differences. First the thickness of the water over the inlet edge seems to be thicker in CF'D calcula- tions then in the experiment. Secondly the exit angle seems to be larger for in the CFD simulations. ‘These two differences are connected because more water in the bucket should give a different exit angle. In numerical analysis of the flow in a Pelton nozzle, done by Risberg [7], the 3D limitation of a code like Flow-3D is demonstrated. The circular jet in the nozzle was deformed into a square jet. after the nozzle. This was caused by how the orthogonal grid is designed and how obstacles are representation on the grid. The different grid methods did not give any problem in this 2 dimensional analysis, but in 3 dimensional analysis preliminary tests with Pelton buckets have shown that there are greater differences.. One alternative to the VOF methods is to construct a particle method on the basis of recent successors of the original MAC method, like the particle method described by Rider and Kothe [8]. ‘This particle method shares similarities with the graphical method. In both methods the particles are moved according to a eulerian time step. In a particle method the entire flow domain is represented by particles, each carrying an identity like velocity and density, This obviously demands a lot of particles in a full 3D representation, In a full 3D calculations of the gap, the particle method seems to have an advantage. VOF methods are generally described in literature to have prob- lems with splitting of fluids. VOF also has a tendency, when the fluid becomes thinner than one or two cells, to exhibit strong numerical surface tension as described in Rider et.al. {9}, The main advantage of a MAC method is the ability to adapt to special re- quirements, e.g. the complex geometry of a Pelton turbine, since this is method usually has to be constructed from scratch. However the recent. advancement of Ripple [9] gives the opportunity to produce the surface normal.. This can be used in conjunction with a graphical method to improve calculation. The graphical method needs the surface normal in order to calculate the particle paths on the surface. The different models for surface tracking, the VOF and multi-phase model, did not in this test give a substantially different results, The interface in the CFX-4 result is more smeared out that the corresponding VOF calculation. 6. Conclusion Of the three codes CFX-4 is the most promising code for further simulation. ‘This is because the multi-phase model gives in 2D sufficiently good results, compared to VOF calculations. How this transfers to 3D remains to see. The geometric modelling and interface with CAD geometry in CFX is better then Flow-3D. The geometric accuracy in modeling Pelton buckets in the two codes will be explored in future 3D calculations, It might become a problem to use square Cartesian grid in Pelton turbines. For further development of the graphical method the particle method is a natura] successor. ‘This method easily models the Pelton geometry. But the demand for CPU and storage is a limitation, Also the construction and testing of such a method is time consuming. For future simulations good visualization results from a Pelton turbine is needed. Animated sequences from high speed visualization with a film or CCD camera must be preferred to still photos. However this type of equipment is expensive compared to an ordinary camera and a strobe light. References [1] Brekke, Hermod: "A. general study on the design of vertical Pelton tur- bines", 1984, presented at 25th anniversary Turbuinstitutt, Lublijana [2] Hana, Morten, "Improvements of a graphical method for calculation of flow on a Pelton turbine", Proc. XVII IAHR Symposium, Valencia, Spain, 1996 [3] C.W. Hirt and D.B Nichols: "Volume of Fluid (VOF) method for the dynamics of free boundaries”, J. of Comp, Physics, 39, 210-225, 1981 [4] PH. Harlow and J.B. Welch, "Numerical calculation of time-dependent viscous incompressible fluid flow with free surfaces”, Phys. Fluids, 8, 2182- 2189, 1965" {5} "Flow-3D: Computational Modeling Power for Scientist and Engineers”, FSI-91-00-1, Flow Science Inc. [6] D.B. Kothe and R.C, Mjolsness, "Ripple: A new model for incompressible flows with free surfaces”, AIAA Journal, 30, 2694-2700, 1992 [7| S. Risberg, "Beregning av stromning i en Peltondyse”, Master thesis, NTNU, 1995, in Norwegian [8] W.J. Rider and D.B. Kothe, "A marker particle method for interface track- ing”, 6th international Symposium on Computational Fluid Dynamics, Lake Tahoe, USA [9] W.J. Rider, D.B. Kothe, 8.J. Mosso, J.H. Cerutti, J.1. Hochstein, * Accurate solutions for incompressible multiphase flows”, AIAA 95-0699, ATAA, 1995,

S-ar putea să vă placă și