THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS

345 E 47th St, New York, N.Y. 10017 96-ZT738

The Society shall not be responsible for statements or opinions advanced in papers or discussion at meetings of the Society or of its Divisions or
Sections. or printed in its publications. Discussion is printed only il the paper is published in an ASME Journal. Authorization to photocopy
material for internal or personal use under circumstance not falling within the fair use provisions of the Copyright Act is granted by ASME to
libraries and other users registered with the Copyright Clearance Center (CCC)Transactional Reporting Service provided that the base fee of $OSO
per page is paid directly to the CCC, 27 Congress Street Salem MA 01970. Requests for special permission or bulk reproduction should be addressed
to the ASME Technical Publishing Department .

CopynghtC 1996 by ASME All Rights Reserved Printed in U.SA

QUASI-THREE-DIMENSIONAL AND THREE-DIMENSIONAL FLOW CALCULATION IN A

FRANCIS TURBINE

Manfred J. Sallaberger 1
Institute of Hydraulic Turbo-Machinery IllIllI ltI II 11111 II till!!
BREAK
University of Technology Graz
Graz
Austria

ABSTRACT V volume
The complex three-dimensional flow in the wicket gate and the vector of absolute velocity
runner of a Francis turbine is investigated by applying both a quasi- vector of relative velocity
three-dimensional and a three-dimensional computational method. The a = arctan ( v„, / v e )
computations were conducted on a double grid containing the = arctan ( w e / we )
stationary wicket gate and the rotating runner. The equations for Pb blade angle
inviscid and incompressible flow are solved, assuming that the relative 8 blade angle
flow field in the runner is stationary. In the quasi-three-dimensional blade angle
method the governing equations are solved on stream surfaces using efficiency
a Finite-Element-Method. In the three-dimensional method, the density
equations of continuity and motion are solved bY a Finite-Volume angular velocity
technique using Denton's code for incompressible flow. Both methods stream function
are used in order to compute the flow in a Francis-runner of high
specific speed at the operating point of optimum efficiency. The results Subscripts
of the calculations are compared with measurements taken at the draft- LE leading edge
tube inlet. Differences between results of computations and meridional direction
measurements are presented. TB trailing edge

NOMENCLATURE
A Area INTRODUCTION
a velocity of sound The increasing capabilities of Computational Fluid Dynamics
normal or circumferential thickness of stream filament during the last years have brought great advance in hydro turbine
vector of blade force design. These calculations began with two-dimensional potential
pressure head methods for the blade-to-blade surface and two-dimensional
enthalpy per unit mass of fluid circumferential averaged methods for the flow in the meridional plane.
total energy Quasi-three-dimensional methods give a solution of the flow by
relative total energy iteratively solving the flow on two kinds of stream surfaces. They can
static pressure not yield three-dimensional effects but are able to treat rotational
cylindrical coordinates effects. Three-dimensional potential flow methods take account of
time three-dimensional effects but can not cope with rotational effects.
Alternatively the three-dimensional Euler methods represent a
significant progress as three-dimensional and rotational effects are
'Current address: Sulzer Hydro Ltd, 01-8023 Zurich, Switzerland treated correctly. By solving the momentum equations for inviscid

Presented at the International Gas turbine and Aeroengine Congress & Exhibition
Birmingham, UK — June 10-13,1996
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 flow for the primitive variables pressure and velocity, inviscid vortices caused by turbulent mixing or effects due to different boundary-layer
and inviscid secondary flows can be shown correctly and only viscous thickness on suction side and pressure side of a profile. Nevertheless,
effects are neglected. momentum produced three-dimensional effects can be shown clearly.
By solving the Navier-Stokes equations and applying a turbulence
model like the well known k-E model or the Baldwin Lomax mixing-
length model also viscous effects and losses can be calculated. GOVERNING EQUATIONS
Application of these codes, however, is rather time consuming, The three-dimensional flow in stationary and rotating parts of
expensive and occassionaly implies some degree of uncertainty, turbomachinery is governed by the following set of thermodynamic
especially with the validity of the turbulence models. equations written in an invariant vector form.
Besides these numerical methods, experimental investigations are The principle of conservation of mass leads to the equation of
essential to the design of hydraulic machinery. Model tests give an continuity.
insight to the real flow phenomena and hence show possibilities for
optimization of a turbine. Furthermore, these experimental results are ag (1)
V*(0) = 0
a base for the improvement of numerical methods. at
With help of these highly developed calculation methods, a number of
design steps of a runner can be performed by numerical design Newton's second law of motion gives
methods. Variations of parameters can be tested quickly saving time
aV 1 (2)
and money. The fine tuning again has to be conducted in the model (V-V)V = Vp
test, but the number of modifications to be tested is reduced as the at
design is close to the optimum. Also efficiency and cavitation
guaranties have to be verified by model testing. Limiting the following considerations to hydraulic turbomachinely
In the following, the inviscid three-dimensional flow in wicket gate allows some simplifications. The equations are solved for
and runner of a Francis turbine will be calculated. The application of incompressible flow. Furthermore, the flow is assumed to be
Wu's quasi-three-dimensional method (Wu 1952, Wu et al. 1984, Wu stationary, in the rotating parts of the machinery the relative flow field
et al. 1992) and a fully three-dimensional method (Denton 1975, shall then be stationary. Treating the flow as inviscid eliminates the
Denton 1976. Denton 1982) will be investigated. Viscous effects and energy equation.
boundary-layer effects are not considered.
At first, the flow is calculated by Wu's quasi-three-dimensional
method (Wu 1952). Integration of the differential equations is
performed with a Finite Element method. QUASI-THREE-DIMENSIONAL FLOW CALCULATION
Then, the flow is calculated by a fully three-dimensional method using In his general theory of three-dimensional flow in turbomachinery
a suitable Euler 3D solver. The program used is based on Denton's C. H. Wu (1952) presented a method for the solution of both the direct
well known Euler 3D code (Denton 1982), solving the three-dimensio- and the inverse problem. In this theory, the three-dimensional flow is
nal inviscid and incompressible flow. With slight modifications, it was calculated iteratively by solving the flow on two different kinds of
applied to the solution of the flow in a Francis turbine. intersecting relative stream-surfaces SI and S2. On these surfaces, the
The complex geometry of a Francis turbine and the highly three- flow is treated mathematically in a two-dimensional manner. Thus the
dimensional flow in this kind of turbomachinery make high demands three-dimensional flow is calculated iteratively by a two-dimensional
on every computational method. In this turbine a swirling flow is method. The equation of continuity is combined with the equation of
entering the wicket gate radially, where swirl is increased usually. motion by use of a stream function defined on these surfaces and as
Leaving the wicket gate the flow enters the runner radially and by a result only one governing equation has to be solved.
changing moment of momentum and energy it is turned to axial The first kind of stream surfaces SI extends from blade to blade.
direction, leaving the runner and entering the draft tube axially. In At the hub and shroud (or casing wall) these SI surfaces coincide with
order to get a more accurate approach of this three-dimensional flow, the bounding wall and are surfaces of revolution, the SI surfaces in
the equations are solved in wicket gate and runner simultaneously. The between may be twisted. The second kind of stream surfaces 52
computations were conducted on a double grid containing the extends from hub to shroud. The S2 surfaces near the suction side and
stationary wicket gate and the rotating runner. In each of the grids the the pressure side roughly coincide with the blade contour. The flow
flow is assumed to be stationary. The values of the flow properties are channel is devided into a number of SI surfaces and a number of S2
transmitted from one blade row to another with help of a mixing plane surfaces. The three-dimensional flow is approximately calculated by
between the grids. iteratively solving the flow on a number of SI surfaces and on a
The applicability of the calculation methods to a Francis turbine of number of S2 surfaces. The shape of one kind of these stream surfaces
high specific speed is considered. The quality of the numeric solutions is defined by streamlines calculated on the other kind of surfaces. Thus
is investigated and the results of the calculations are compared with the SI surfaces are created by joining the corresponding streamlines
results of measurements at a cross section at draft tube inlet, i.e. on the S2 surfaces from the previous calculation. The shape of the 52
runner outlet. surfaces is determined by combining the corresponding streamlines on
When comparing computations and measurement, inviscid momentum the SI surfaces, which devide the flow channel into sections of equal
caused three-dimensional effects and viscous effects have to be mass flow
differentiated. So the inviscid calculation is not able to show effects In this way, the flow channel is described by a number of SI and S2

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 stream surfaces. Each of these SI and S2 surfaces can be thoutht to Now a stream function Ns is defined satisfying the continuity equation
be the mean surface of thin layers of flow. These flow layers are identically.
bounded by the adjacent stream surfaces. So e.g. the SI stream surface
number 2 is the mean surface of a flow layer bounded by the St
am
= b0we (5)
surfaces number I and number 3. The flow on the mean stream-surface
is equivalent to the flow in a thin layer of flow bounded by these = -berm.
stream surfaces. The thickness variation of this flow layer is ae
determined by the normal distance between surface number I and
number 3 when the distance is sufficiently small. In this way, the flow With help of this stream function iv the equation of continuity and the
in layers of arbitrary thickness variation can be described. For the 52 equation of motion are combined to a principal equation (WuJ952)
stream surfaces the thickness variation is determined in the same way
using the normal distance of the bounding 52 surfaces. a Van ±V-Pin
am to am ae br = - 2werET
(6)

In a simplified calculation, the number of S2 surfaces is replaced by

one mean surface 52„ which is assumed to be the representative S2
surface. This mean surface 52„ devides the flow channel into equal This equation is solved using a Finite Element method which is
parts of massflow. In this case, the SI surfaces are created by rotating described in a following section (Fig. I).
the calculated streamlines of the S2 surface around the axis. Thus the
SI surfaces are surfaces of revolution. But the calculation of the S2 111111111.a*
surface differs from usual circumferential averaged solutions. In the
quasi-three-dimensional calculation, the thickness variation of the
manb itio
stream layer is determined by the normal distance of the neighbouring

malitICS
streamlines like described above, whereas in usual two-dimensional
calculations, only the blade thickness is considered. The three-
dimensional flow field is calculated by solving the flow on a number
of SI surfaces and on one S2, mean surface. This approximation
greatly simplifies the calculation of the flow field. In the following
flow analysis of a Francis turbine first a simplified calculation and 4At#4
- arOwinsn
then the extended quasi-three-dimensional method using a number of
S2 stream-surfaces is applied.
isat
-rcti
ens
Calculation of flow on Si stream layers
ltvan
%S auna
As the SI surfaces are considered as surfaces of revolution, the
equation of continuity and the equation of motion are treated in a two-
dimensional way on these surfaces. The equation of continuity is then Fig. 1 Finite Element grid of a S1 surface
written as

(3)
Vlbg0) = 0 In turbomachinery, the flow through one channel of the blading
is calculated and is thought tope representative for the entire Wading.
So in the SI calculation, the flow in an finite region stretching one
where b is the normal thickness of the stream-layer. In the calculation blade-spacing has to be solved. The calculation of the cascade flow on
of the S I flow, only the dimensionless variation of the layer thickness boundaries roughly parallel to the blade mean-line is not a classical
is needed, so the flow layer can theoretically be infinitely thin. boundary value problem, thus periodic boundary conditions have to be
On each SI stream-surface, an orthogonal coordinate system (me) is introduced. At points situated on the same radius one blade spacing
defined, where m is the length along the meridional projection of a apart, the flow conditions have to be equal i.e the static pressure and
streamline and 0 is the circumferential direction. the velocity have to be equal. This is performed with the use of link
elements, which connect the corresponding points on the boundaries
Then the equation of motion for two-dimensional flow on surfaces of which are positioned roughly parallel to the blade camber line. With
revolution written in this coordinate system leads to help of these link elements, only the difference in the value of the
stream function its is defined and hence a definite mass flow forted.
(4)
awn - a(rwe) 2wr The effect of these link elements is compareable to spacers consisting
ae am am of springs with very high stiffness in structural problems.
At the entrance of the computational region the mass flow and the
circumferential velocity are defined. Both quantities are taken from the
results of the previous S2 computations.
There is still an additional unknown boundary condition, the value
of the stream function along the surface of the profile and therefore

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 also along the stagnation streamline. ibis value of the stream function equation of continuity and the equation of motion are combined to a
determines the circulation and thus the lift of the profile. The value of principal equation, Wu (1952).
the blade circulation is determined b y the Kutta-Joulcowslci trailing
edge condition. For profiles with sharp trailing edges the rear 1 allr a( as) (10)
stagnation point must coincide with the sharp trailing edge, because or( br 8r) äz) 7(r'z)
the fluid cannot be forced to flow around the trailing edge with infinite
velocity. This assumption seems to be satisfied for profiles with weak In this equation y(r.z) represents the vorticity and is defined by the
rear loading and where full y attached boundary layers can be expected. gradient of the specific moment of momentum (rv e) and the gradient
For profiles with blunt trailing edges another formulation of the Kuria- of the total energy km or the relative total ener gy h, and the radial
loukowski condition is used. At those points in the rear part of the component of the blade force F r For stationary parts y takes the form
profile, where in real flows the transition of the two boundar y layers
into the wake would occur, the pressure must be equal at both sides of ait707. (11)
the blade adjacent to the trailing ed ge. This condition is realized with v(r,z) = reill_weafrve)
r ar r
or a, - F
help of the following formulation. The direction of the velocity vector
at the rear sta gnation point must coincide with the direction of the
blade mean-line at this point. This is performed by superposition of and for rotatin g parts it gives
two different solutions of the flow field. The resulting solution then
satisfies the condition above. As the considered partial differential ah ) (12)
equation is linear, this superposition is justified and once more is a Y(r,Z) = rbrat (_wea(rve) ar, Ft
ar r er
solution of the differential equation itself.
According to the procedure described above, for two base solutions the
value of the circumferential velocit y at the outlet of the computational The absolute total energy is defined as
region is defined in the way, that the trailin g edge condition is
satisfied approximatel y. By superposition of these two solutions the (13)
4 720
correct value of the circumferential velocit y is calculated, satisfying
the trailing edge condition exactl y.
In this equation the internal ener gy u is neglected as it keeps constant
for this type of flow.
Calculation of flow on S2 stream layers The relative total energy h, is defined by
The $2 surfaces are described in a c ylindrical coordinate s ystem
hit = It - rev s (14)
(r,z,(3). The calculation of the flow on the $2 surface is performed in
the meridional plane (r,z), (Fig2). Like before the continuit y equation
is The equation above is solved usin g a Finite Element method. With
given distributions of y(r.z) and b(r.z), the usual Dirichlet and
(7) Neumann boundary conditions are applied. Both y(r,z) and b(r.z) are
V.(bp0) = 0
determined b y the preceeding solution on Si surfaces, iteratin g to the
final solution. Because of the nonlinear terms in the expression for
where b is the circumferential thickness of the stream la yer. This y(r,z), an appropriate underrelaxation techni que has to be used.
thickness is defined by the circumferential distance of the neighbourin g
streamlines on the SI surfaces and a gain the onl y dimensionless
thickness variation is needed. F w. a(rw e +r2o) w a(rwer2o)
= ' t
The equation of motion for two dimensional flow in the meridional .r az r Or
plane is written as
sin e
= Foca p (15)
awz aw, Fwe a(ree) aoR F 1 cos(e -A)
cose
ar az w: r är är Fz = Fe cotI3
cos(e -A)

h p w2 (rtu)2
Again a stream function tir is defined as R g 2 2

at z, brwz The distribution of the blade force inside a blade row is calculated
ar (9) from the gradients of the specific an gular momentum. The stream
ael _ -brw : angle ft and thus the circumferential component of the velocit y ve is
az obtained from the previous solution of the flow on SI surfaces. In this
way the circumferential component F e of the blade force F is
satisfying the continuity equation. With this stream function Itt, the calculated from the momentum equation in circumferential direction.

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 The additional condition Fb.r/ L. 0 , i.e. the blade force is normal to velocity gradients are needed as it was shown in the preceeding
the blade surface, gives a relation between F e and Fr sections. accurate numerical methods for the determination of these
GU DE- values are are of great importance in order to get good results. For the
VANES Finite Element solution, the program system Mises3 ( Beer, Haas
1978) is used. The program uses isoparametric elements of the
serendipity type with eight nodes in two-dimensional calculations. For
the numerical integration the Gauss-Legendre quadrature with two
Gauss-points in each local direction is used. For stress analysis, it has
been shown that in such elements the gradients of the primary
variables are much more accurately calculated in the Gauss-points than
in any other point in the element including boundary nodes. (Hinton
Campbell 1974). If the distribution of gradients is needed these values
should be determined by means of Gauss-point extrapolation with local
least square smoothing. The final value in a node is found as the
average value of the contributions of the elements adjacent to the
node. In Keck, Haas (1982) the application of Gauss-point integration
to flow problems and particularely to the solution of the Laplace
equation is shown. Accuracy and convergence are checked for the
solution of the Laplace equation with curved-sided eight-node
isoparametric elements and the numerical results are compared with an
exact analytic solution. It is shown that using Gauss-point integration
and linear extrapolation reduces the error drastically compared to
integration in the nodes of the elements.

I
MEASURING PLANE t--1 THREE-DIMENSIONAL SOLUTION
With quasi-three-dimensional methods the complex three-
dimensional flow through the runner of a Francis-turbine can only be
calculated approximately. Now a full three-dimensional method is
OUTLET
applied to the flow, the equations of continuity and motion for inviscid
,z flow are solved directly using a suitable numerical scheme.
The method used is an explicit, time marching Finite Volume
method. The program used was developed by J.D. Denton of the
Fig. 2 Finite Element mesh of a 52 surface Whittle Laboratory at Cambridge University, United Kingdom (Denton
1975, Denton 1976, Denton 1982 Denton 1983, Denton 1990).
Frimarely. the code has been developed for compressible flow, a
In the quasi-three-dimensional method the set of equations is switched version for incompressible flow has been derived using the method of
from statoinary to rotating system when the streamline is entering the artificial compressibility. With little modifications it was used for
runner. There is no distinct mixing plane and the computational grid calculations of the flow in the Francis-turbine. The basic version of the
for the 52 flow is continous from guide vanes to runner-blades. The program can cope with any kind of single blade rows with axial.
flow between the guide vanes is treated congruent to the blade mean- mixed or radial floW of compressors or turbines. A special version can
surface, hence the flow angles for the 52 computations are defined by treat two or more blade rows in relative rotation, thus the flow through
the blade geometry. The flow on SI surfaces is only solved for the the guide vanes and the rurmer of a Francis turbine can be treated in
runner-blades and the iteration between SI and S2 solutions only takes one calculation.
place for the runner. The solution procedure is an explicit time marching method of the
Convergence of the iteration between SI and S2 surfaces is taken Finite Volume type. The integral form of the equations of conservation
to occur, when the changes in radius, in z-direction and in stream-layer of mass and momentum are applied to a number of finite control
thickness at the trailing edge of the nmner blade are smaller than one volumes. These control volumes share common faces, whereby the
percent. It takes four to six iterations to satisfy these criteria. global conservation of flow properties is obtained better than with a
finite difference scheme and is independent of the distortion of the
grid. The stationary solution is achieved by solving the equations for
Finite Element Method the irtstationary flow and marching on in time. Convergence is taken
• the governing equations for the flow on SI and S2 stream to occur when the change of the properties in time is smaller than a
surfaces are formulated with help of a stream function and therefore preset limit.
only one principal equation has to be solved for each surface. These The equations are solved in a cylindrical coordinate system (43, r, z)
equations are solved by a Finite Element Method. As the velocities and a simple sheared grid is used. A cartesian system would simplify
and the pressure are the variables of primary interest and also the the calculation of the interior mesh points, but would complicate the

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 application of the boundary conditions. The complication using a cuboid. In the eight corners of the cuboids, the flow properties are
cylindrical coordinate system is that the direction of the radial vector stored.
changes around the surfaces of an element. This must be regarded The equations solved are the equation of continuity and motion
when evaluating the radial components of pressure force and for three-dimensional, inviscid, incompressible flow in integral form.
momentum flux. The grid is set up and the equations are solved in the These equations form a system of partial differential equations
physical plane (Fig. 3). connecting the physical variables describing the flow field. The
equations of this system are integrated by Gauss' theorem and applied
to a number of control volumes of finite size.

conservation of mass

6 6
(16)
—
A pAV+ E g v, AA. + ve A449 + EQVr ISAr =O
At ••I s•I 6.1

conservation of x-momentum
( 17)
Av.,
(p v.v.AA.+pv.ve AA0 + pv. vrAk+ptiA.)” AV
• At

conservation of r-momentum
2
6
e V Av
r(pv,44,+pv,v eafie +pv,v,AA,+p&A)-p— A V - g - --&A V
At
(18)

conservation of e-moment of momentum

6
E(erve ysAz + neve 4,40 + rvev,A44 pr AA e) = (19)
nal
Owe )
AV
At

The equations are solved for the absolute velocities. For calculation of
Fig. 3 Computational mesh for Euler 3D calculation rotating blade rows, it seems easier to work in a rotating coordinate
system and to solve for relative velocities. Therefore coriolis and
centrifugal terms would have to be introduced into the equation of
In hydraulic turbomachinery very complex geometries have to be motion bringing the risk of additional instabiltiy in the solution
treated and so the choice of a suitable -grid is of great importance. procedure. Denton (1982) and other authors, however, found it
Denton (1982) poses that use of a grid as simple as possible helps preferable to solve for absolute flow variables in a rotating system.
treating such geometries and keeps the scheme easily understandable. The absolute flow properties calculated from the equations of
In his method, he uses cuboids as finite control volumes with a node conservation are transformed from a fixed grid to a rotating grid which
at each corner. These cuboids are formed by intersection of three kinds coincides with the fixed grid at this instant.
of surfaces and have boundaries of straight lines. The bounding (20)
a( ) = 80 a( ). ae
surfaces are the streamwise surface, the bladewise surface and the
at R. ät Ths ae a:
quasi-orthogonal surface. The streamwise surfaces are similar to the
blade-to-blade surfaces in the quasi-three-dimensional calculation and
are chosen as surfaces of revolution. The bladewise surfaces are The difference scheme used is Denton's opposed difference scheme
arranged in a way that the first and the last contain the suction surface which is described in detail in his papers (Denton 1975, Denton 1982).
and the pressure surface of the blade and so form a blade channel. The In order to ensure stability in the strearnwise direction, upwind
quasi-orthogonal surfaces are arranged roughly perpendicular to the differencing is used for the fluxes of mass and momentum and
meridional flow direction and are surfaces of revolution. At the inter- downwind differencing is used for the pressure. In the pitchwise
section of three kinds of surfaces, there are grid nodes forming a direction, central differencing is used for all flow properties.

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 In the momentum equations, an effective pressure instead of the real circumferential direction information is smeared, whereas the quantities
pressure is used. The effective pressure is determined by the value of may vary in quasi-orthogonal direction. As the mixing plane is very
the pressure in the nett downstream node plus a correction term. This close to guide-vane trailing-edge and runner leading-edge the mixing
correction term is defined by an interpolation procedure and damped should not introduce to much disturbance to the flow.
after every time step. Stability considerations are determined by the analysis of the
region of influence of physical and numerical dependence. For
pdr j • Pj.1 + CF:r (21)
stability, the region of influence of the discredzing scheme expressed
by the discretization equations must contain the physical region of
influence described by the differential equations. During one time step
At, a pressure wave (disturbance) starting at a point may not travel
CF a (1 —RF)-CF RF- CFJart (22) more than one grid spacing. As with all explicite schemes the
Jxt-w JOLD
maximum time step for stability is limited by the CFL condition
(Courant-Friedrichs-Lewi condition). In order to save computation
To solve the conservation equations at the boundaries of the time, instead of the CFL condition for three-dimensional flowthe
computational domain, appropriate boundary conditions have to be condition for one-dimensional flow is considered.
defined. At the inflow boundary the total pressure must be specified
AI (24)
in each node. The values may vary in both circumferential and quasi- At <
orthogonal direction. The inlet flow direction is specified either by the (no)
relative or absolute flow direction in both the streamwise and the
meridional plane, or the angle in the streamwise plane may be replaced For Al the minimum spatial extension of a control volume is taken.
by an absolutes swirl velocity. A further condition is the method used The scheme is first order accurate in time, which is sufficient as only
to extrapolate the static pressure from the interior flow field. This may the stationary solution is of interest. The accuracy in space is of tint
be carried out by defining either the first order derivative or the second order with a correction added to increase the spatial accuracy to
order derivative of the static pressure along a quasi-streamline. In this second order.
calculation the second order derivative is used to calculate the static The incompressible flow is solved with help of the method of
pressure from the interior flow-field. Then the value of the inlet arificial compressibility. In the momentum equations, the value of the
velocity is calculated as the difference of the total pressure and the true density is used, whereas in the equation of continuity an artificial
static pressure at each node. density is used. When the solution is marching on in time, this
artificial density gradient vanishes and the results converge to the
Or a2P o (23) stationary solution. By introduction of an artificial state equation,
ap 2 pressure and artificial compressibility are related and an artificial speed
of sound may be defined.
The three-dimensional calculations are performed for different grid
At the downstream boundary, only the static pressure must be defined. sizes, 29x159r.29, 19x159x19 and 13x159x13. Here the results are
This may be done by defining a spanwise variation or by fixing the shown for the 13x159xI3 grid, the minimum grid size which gives
value of the static pressure only at the hub or only at the shroud. The satisfactory results. Grid stretching in pitchwise and spanwise
mass-flow between inflow and outflow boundaries is effected by a directions and some concentration of the grid lines near leading edges
pressure gradient corresponding to real physics. and trailing edges is used. The computation domain includes guide
At solid walls, only the condition of no flow through the surface vanes and runner of the Francis turbine and is extended from stay vane
is applied. Denton does not resolve the flow velocities parallel to the outlet to draft tube inlet (Fig. 3).
surface. Consequently the final velocity may not be exactly parallel to
the surface. Denton (1982) argues that the errors ocurring by this
scheme would not be worse than those caused by resolving the flow TEST RIG AND MEASUREMENTS
velocities adjacent to the walls. The described calculation methods are applied to a Francis-turbine
At the bladewise surfaces upstream and downstream of the blading, a of high specific speed n, = 297 at the operation point of optimum
periodicity condition is applied. The nodes on these surfaces are efficiency. There the specific speed is defined as
initially treated as interior nodes and then the average value the of two
corresponding nodes is finally used.
As the flow computations are conducted on two grids, which are rts = 3.652 n Q c13 (25)
in relative motion, a special boundary condition at the interface 1035
between the stationary to the rotating grid has to be applied. Denton
(1983. 1990) has defined a mixing-plane in which the flow properties
are transmitted from the staionary to the rotating grid. The values of with H and Q in SI-units and n in rpm. On the combined pump and
the flow-properties in each node are mass weighted averaged in turbine test rig in the laboratory of the institute overall efficiency
circumferential direction and multiplied by the relation of the blade measurements and measurements of the exit flow field are performed.
numbers and then transmitted to the succeeding grid for the next blade The test rig consists of closed loop with a pump forcing a certain
row. This averaging ensures conservation of mass flow. In mass-flow of water in the loop and producing the head for the turbine.

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 The flow rate is determined by measuring the pressure difference in a inviscid calculation this effect of the boundary layer is not considered
ventury tube. The pressure difference is measured by an rotating- and so the circumferential velocities are somewhat different.
piston-manometer. As the ventury tube is calibrated, the accuracy of
determination of the flow rate is 5 %0. For the turbine head the
difference of static pressure between the entrance of the spiral-casing
and the outlet of the draft-tube is measured by a rotating-piston-
manometer with an accuracy of 1 of the end value. The turbine
head is calculated by adding the difference of the pressure-height of
the velocities between turbine inlet and outlet, to the difference of the
static pressure. The accuracy of the determination of the turbine head
should be 2 %.
The measurements of the exit flow field are performed at a plane
located at draft-tube inlet (Fig. 2). A three-hole cylindrical probe with
a diameter of eight millimeter is used to determine the axial
component of the velocity v 7 and the circumferential component v„,. 0 10 20 30 IV 20 20 72 10 110 100 II 20 20 10 150 120 170 180 100 203

According to the geomerty of the draft-tube cone the maximum angle RADIUS (MAI I
between the axial velocity and the meridional velocity is seven
degrees. Therefore the difference between meridional velocity and the
Fig. 4 Axial velocity at the measuring plane
axial component is very small and errors occuring hereby are
considered neglectable. The direction of the velocity is determined by
rotating the probe until the static pressure in both holes left and right
of the stagnation point is equal. The value of the velocity is evaluated
by measuring the stagnation pressure. The pressure is measured by an _ndintinide
inductive pressure-transducer. The process of measurement is idandifidndirri
Ennindedillir

••
controlled by a microcomputer, two stepping motors are positioning 1 22
the probe and the evaluation of the results is also preformed by the 00
computer. The maximum difference between flow-rate measured by Eddiddlndle Ina
the ventury-tube and the integrated flow-rate of the probe is 2,5 %.

ell
II -CI
0

a EASUREMENT

COMPARISON BETWEEN COMPUTATIONS AND nunn

MEASUREMENTS 10 20 XI 10 03 70 02 CO 102 11 120 110 1.0 IN 100 170 1 20 IN 720

Calculations and measurements are compared for the flow-rate and RADIUS mm j

the head of the operating point of best efficiency. The flow-rate of

both computations are within 1 % of the value of the measured flow- Fig. 5 Circumferential velocity at the measuring plane
rate.
The values of the deviation of the flow angle A13 at the trailing edge
of the runner and the values of v z and ve in the measuring plane at In Fig. 6 the distribution of the deviation of the flow angle Ap at
draft tube inlet are compared. There, the results of the calculations are the trailing edge of the runner is shown. The flow angle 13 is the angle
mass weighted and circumferentially averaged values of the variuos between the direction of flow and the circumferential direction. The
flow properties. As the measurements are only performed at one radial blade angle 0„ is the angle between the direction of blade-congruent
line their, results may not be exactly circumferentially averaged and flow and the circumferential direction. The difference between 13 and
thus some uncertainty is implied. 13„ ,i.e. the angular deviation is shown in Fig. 6 , calculation and
In discussing the results of measurement and calculation it should measurement are compared. Attention has to be paid to the fact, that
be mentioned, that the equalisation of the flow by turbulent mixing in the values of the flow angle deviation of the measurement were
the real viscous flow cannot be shown by the inviscid calculations. calculated back from measurements at a plane in some distance of the
Considering this fact, the agreement of the results of the quasi-three- trailing egde. This calculation is performed under the assumption of
dimensional calculation with the the measurement seems to be constant specific moment of momentum rv e along the meridional
satisfactory. In the distribution of the axial velocity v z (Fig. 4) streamlines of the quasi-three-dimensional calculation. Furthermore, it
calculation and measurement agree very well from shroud till 30% of has to be mentioned that the inviscid calculation does not consider the
the radius. At radii closer to the hub, the turbulent mixing of the flow different boundary layer thickness on pressure side and suction side of
on the long distance between trailing edge and measuring plane may the blading. Considering these facts, the agreement between calculation
cause the difference between inviscid and viscid flow. No explaination and measurement looks satisfactory.
is found for the unsteadiness of the computed results at a radius of 155
mm. The distribution of the circumferential velocity v e (Fig. 5) also
shows satisfactory agreement, here the difference of the value may be
explained by the effect of the boundary layer in real flow. In the

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 The differences between the real viscous flow and the inviscid
computation also have to be mentioned when discussing the results of
the three-dimensional solution. In Fig. 7 the distribution of the axial
• MOSUPIDGENT
velocity v, is shown. On radii greater than 30 % of the radius of the
o eon shroud, the values agree well. On the other band, at radii closer to the
II 0

• 31DELLEF1
hub, the turbulent mixing of the real flow causes differences between
inviscid calculation and the real viscous flow like explained above.
ICC Furthermore it has to be noticed that in the diagram, the zero value is
suppressed and therefore the difference appears exaggerated. For the
distribution of the circumferential velocity v o (Fig.8) also the missing
C
'
boundary layer thickness and the long distance between trailing edge
and measuring plane may cause the large differences at the radii close
to the hub.
o a a so 40 SI 10 70 00 10 1C0 110 In 110 14 193 140 170 Ile GO 7,3 The distribution of the deviation angle. AB (Fig. 6) is shown
GROAN ONO% OP TRADING EDGE Owl BOO together with the results of the quasi-three-dimensional calculation. For
interpretation of the results, the same explaination is before is valid.
The agreement is satisfactory, but at the radii close to the shroud the
discrepancy between measurement and computation is rather large.
Fig. 6 Deviation of flow angle at runner trailing edge
Evidently the three-dimensional computation shows better results than
the quasi-three-dimensional computation, especially at smaller radii.
This may indicate that the complex three-dimensional flow in a Francis
turbine can be better simulated by the three-dimensional method.
In Fig. 9, the distribution of the total pressure near suction side
(SS) and in Fig. 10 near pressure side (PS) of an extended quasi-three-
dimensional calculation is shown. Contrary to the theorie of Wu in this
extended Q3D solution the calculations on SI surfaces are done on
8 surfaces of revolution and in circumferential direction several surfaces
of revolution are connected. So the general S I surfaces are
approximated in circumferential direction by a stepwise surface. The
total pressure is shown on the S2 surfaces near SS and near PS.
The distribution of the total pressure head of the three-dimensional
IDA 47.0 no OA OD IWO 17.0 no DOA MA IDA PAO MA OA 170.0 110.0 170.0 AO
calculation is shown near SS (Fig. II) and near PS (Fig. 12) . Both
RADIUS fmmj solution methods show the steady reduction of total pressure along the
meridional flow direction. The three-dimensional method, however,
shows some pressure gradients inside the blade, which are not shown
Fig. 7 Axial velocity at measuring plane by the quasi-three-dimensional method. This may be due to the fact,
that in the quai-three-dimensional method the flow is fixed to the
stream-surfaces and thereby will be smoothed.
In Fig. 13, the meridional projection of the streamline distribution

I 11-- 1. 1 1 1 1 of the extended quasi-three-dimensional (Q3D) solution is shown.

La — There the streamlines on.a S2 surface close to the suction-side (SS),
—
-- -- T---T- I-- — "1- on the 52 mean surface and on a S2 surface close to pressure-side (PS)
_ are shown. Fig. 14 shows the streamline distribution of the three-
I • 1 . 1 1 : I dimensional (3D) solution. There in each node at a radius at the inlet
I . of the computational region, the streamlines are started. So the
streamlines have to be imagined circumferentially spaced from SS to
1 i ..,-1 3D-EULER
— PS. Both solutions show qualitatively the same character, inside the
j I ---I I I MEASUREMENT runner the streamlines near PS run closer to the shroud and the
-taa I I I I . —I. •
- t streamline near the SS run closer to the hub. The full-three-
no to sae es to NO ea gas nts a. tas Itt Ls.0 two ina toss BOA dimensional solution shows this effect more distinctly than the Q3D
RADIUS ( mm I solution. The curvature of the streamlines and therefore of the stream-
surfaces is much higher in the three-dimensional solution than in the
quasi-three-dimensional solution. Again the three-dimensional flow is
Fig. 8 Circumferential velocity at measuring plane better resolved by the fully three-dimensional method.

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 9 Total pressure head near SS, extended 03D Fig. 10 Total pressure head near PS, extended 03D
—

CI

' I

ra- 2117.1- '41114."4.1"1 ".1

Fig. 11 Total pressure head near SS, 3D Fig. 12 Total pressure head near PS. 3D

In the following the inviscid runner efficiency of the computations As a result of the three-dimensional method in Fig. 15, the value
is compared to the measured overall efficiency of the turbine. At the of the total pressure head along a streamline on the way from entrance
flow-rate of optimum efficiency the measurements show a turbine head to outlet of the computational region is shown. In Fig. 16, the value
of 10,0 m and an overall efficiency of 92,6 %. In the inviscid quasi- of the specific moment of momentum along the same streamline is
three-dimensional calculation mass weighted averaging yields to a shown. Apart from slight numerical oscillations both values keep
turbine head of 9.67 m compared to 10.0 m of the viscous constant along the region of no bladings, the peaks at the leading
measurement. This corresponds to an inviscid runner efficiency of edges and trailing edges have numerical reasons and might be typical
96.7 % , which looks plausibel compared to an overall efficiency of for such methods. The streamline chosen enters the runner channel at
92.6 % obtained in the measurement The three-dimensional calcula- about mid of the circumferential width. In the guide-vane channel the
tion yields to a mean value of the turbine head of 9.48 m which corre- same streamline runs close to the blade surface and therefore is
sponds to an inviscid runner efficiency of 95 % and is in good following the profile. This explains the deswirling effect of this certain
agreement with the results of measurements and the quasi-three- streamline. The drop of total pressure head is also shown along this
dimensional calculation. certain streamline and therefore it may differ from the mass averaged
mean-value.

10

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 X

Fig. 13 Streamline distribution of extended 030 solution Fig. 14 Streamline distribution of 3D solution

3.0
!

I
SPECI FICMOMENT OFMOMENTUM
TOTALPRESSURE HEAD [m]

A 0 d

15.0 _ . .. 1.01

10.0 w :I 05
Z ,
<
> I

5.0 o I w
z 5 .... ........ 0.01

1 0 I Cr 03

LE TE LE I TE
03 -050
000 0.05 0.10 0.15 0.20 0.25 0.06 05
0.35 0.00 0.15 0.55 0.60 0.05 0.10 020 0.05 0.10 0.15 0.4 0.3 0.50 0.55 MO 0.15 OS 0.55 0.50 0.05 01*
STREAMLINE LENGTH [m] STREAMLINE LENGTH [on)

Fig. 15 Total pressure head along streamline Fig. 16 Specific moment of momentum along streamline

CONCLUSION
The increasing capabilities of Computational Fluid Dynamics have Apart from the radii close to the hub, the flow field at the measuring
brought new possibilities in hydro turbine design. The application of plane was predicted well by both methods. At radii close to the hub,
reliable computational codes enables us to improve our understanding the viscous effects of the real flow dominate and can not be predicted
of the internal flow fields of turbines. The application of a quasi-three- by the inviscid calculation. Of course, three-dimensional effects were
dimensional method and of a three-dimensional method, Denton's code shown more clearly by Demon's program than in the quasi-three-
for incompressible flow, has been verified by comparing computations dimensional scheme. The strong deviation of the streamlines from
with measurements at draft tube inlet. suction side to pressure side within the runner is shown. The full three-
Both codes predict the integral values well, and the conservation dimensional method is better suitable to solve the complex three-
equations are satisfied. The values of total pressure head, massflow dimensional flow in a Francis turbine.
and momentum correspond also to those obtained in the experiment.

11

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 ACKNOWLEDGEMENTS Hirsch, C., Lacor, C., 1982, "Rotational Flow Calculations in 3D
The author is indebted to Prof. J.D. Denton of the Whittle labora- Blade Passages", ASME Paper 82-GT-316.
tory at Cambridge University, England, for his advice in use of his Hirsch, C., Lacor, C„, 1985, "Computation of Three-Dimensional,
code and for the discussion of the results. Inviscid, Rotational Flows", AGARD Lecture Series No. 140, Rom
The author also wishes to thank the Austrian Nationalbank for the Keln Paris.
financial support of this work as part of the Jubilaeumsfonds-project Irons, B.M., Zienlciewics, 0.C., 1968, "The isoparametric finite
Nt. 3936. element system - a new concept in finite element analysis", Proc.
Conf. Recent Advances in Suess Analysis. Royal Aero Soc.
Irons, B.M.. 1969, "Economical computer techniques for
REFERENCES numerically integrated finite elements", Int. J. Num. Meth. Eng., 1. pp.
Anderson, D.A., Tannehill, J.C.. Pletcher, R.H., 1984, "Computational 201-203.
Fluid Mechanics and Heat Transfer", Hemisphere-Verlag, New York. Keck, H., Haas, W., 1982, "Finite Element Analysis of Quasi-three
Bathe, 1986, "Finite-Elemente-Methoden", Springer-Verlag, Dimensional Flow in Turbomachines", at R.H. Gallagher, D.H.
Berlin. Nonie, J.T. Oden and 0.C. Zienkiewicz: At Elements in Fluids,
Beer, G., Haas, H. 1978, "MISES3-Users , Manual, Revision 1, WV. Vol. 4, J. Wiley & Sons-Verlag, London
Bosman, C., Highton, J., 1979, "A Calculation Procedure for 3D Time Keck, H., Code, E., Pestalozzi, J., 1990, "Experience with 3-D flow
dependent, Inviscid, Compressible How through Turbomachinery analysis as a practical design tool", IAHR Symposium, Belgrade.
Blades of any Geometry", Journal Mech. Eng. Sci., Vol. 21, No. 1. Koya, M., Kotake, S., 1984, "Numerical Flow Analysis of Fully
Chorin, A.J., 1967, "A Numerical Method for Solving Incompressible Three-Dimensional Turbine Cascades", ASME Paper 84-GT-19.
Viscous Flow,Problems", J. Comp. Phys., Vol. 2, pp. 12-26. Launder. B.E, Spalding, D.B., 1972, "Mathematical models of
Courant, R., Fiedrichs, K.O., Lewy, 1928, "Ober die partiellen turbulence", Academic Press-Verlag, London.
Differenzengleichungen der Mathematischen Physik", Math. Ann., Vol. McDonald, P.W., 1971, "The Computation of Transonic Flow
100. 32. through Two-Dimensional Gas Turbine Cascades", ASME Paper 71-
Dawes, W.N., 1988, "Development of a 3D Navier Stockes Solver for GT-89.
Application to all Types of Turbomachinery", ASME Paper 88-GT-70. Moore, J., Moore, LG., 1982, "Lecture Notes on Viscous Flow
Demon, S.D., 1975, "A Time Marching Method for Two and Three Calculations in Turbomachinery", von Karman Institute for Fluid
Dimensional Blade to Blade Row", ARC R. & M. 3775. Dynamics Lecture Series on Numerical Methods for Flows in
Denton. J.D., 1976, "Extension of the Finite Volume Time Marching Turbomachinery Bladings.
Method to Three Dimensions", VKI Lecture Series on Transonic Flows Rodi. W.. 1984, "Turbulence Models and their Application in
in Turbomachinery. Hydraulics - A State of the Art Review", LAHR State-of-the-Art Paper,
Demon, S.D., Singh, U.K., 1979, "Time Marching Methods for Delft.
Turbomachinery Row Calculation", VKI Lecture Series on Transonic Schilling, R., 1982, "Numerical analysis of the quasi-3D
Flows for Turbomachinery. incompressible inviscid flow in turbomachines", IAHR Symposium,
Demon, S.D., 1982, "An Improved Time Marching Method for Amsterdam.
Turbomachinery Flow Calculation", ASME Paper 82-GT-239. Sallaberger. M.J.,1993,"Quasi-Three-Dimensional and Three-
Demon, S.D., 1983, "3D Flow Calculations on a Hypothetical Steam Dimensional Flow Calculation in a Francis Turbine", Dissertation
Turbine Last Stage", VKI Lecture Series 6. Techn.Univ. Graz.
Demon, ID., 1990. "The Calculation of Three-Dimensional Viscous Wu, C.H., 1952, "A General Theory of Three-Dimensional Flow in
Flow Through Multistage Turbomachines", ASME Paper 90-GT-19. Subsonic and Turbomachines of Axial-. Radial-, and Mixed-Row
Erilcsson, LE, Rizzi, A., 1984, "Numerical Solutions of the Steady Types", NACA Technical Note 2604. Washington,
Incompressible Euler Equations Applied to Water Turbines", AIAA Wu, C.H., Wang, O., Zhu, G., 1984, "Quasi 3D and Full 3D now
Seattle, Washington, Calculations in Turbomachines", ASME Paper 84-GT-185.
Eriksson, L.E., Rizzi, A., 1984, "Computational of flow around Wu, C.H., Wang, Z., Chen, H., 1992, "Three-Dimensional
wings based on the Euler equations", Journal Fluid Mech., Vol. 148, Rotational Flow in Transonic Turbomachines - Pan 1", ASNfE Journal
Eriksson, L.E., Rizzi, A., 1985, "Computation of inviscid incom- of Turbomachinery, Vol 114, pp. 3849.
pressible flow with rotation", Journal Fluid Mech., Vol. 153. Wu, C.H., Zhao, X., Qin, L., 1992. -Three-Dimensional Rotational
Code, E., Ryhming, It.. 1987, "Dreidimensionale Berechnung der Flow in Transonic Turbomachines - Part II", ASME Journal of
Stramung in einem Francislaufrad", Sulzer Technische Rundschau Nr. Turbomachinery, Vol 114, pp. 50-60.
4. Zienkiewics, 0.C., 1971, "The Finite Element Method in
Code, E., Cuenod, R., Bachmann, P., 1988, "Theoretical and Engineering Science", McGraw-Hill-Verlag, London.
experimental investigation of the flow field around a Francis runner",
IAHR Symposium, Trondheim.
Highton, J., 1976, -The Computation of Three-Dimensional Viscous
Flow with Generalized Geometry", Th.Ph.D. University of Manchester.
Hinton, E., Campbell, J., 1974, "Local and global smoothing of
discontinous finite element functions using a least square method", mt.
J. Num. Meth. Eng., 8, pp. 461-4W.

12

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/27/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use