Centrifugal Pump Design Tuzson 2000

******Created by ebook converter - www.ebook-converter.
com******
******ebook converter DEMO - www.ebook-converter.com*******

******Created by ebook converter - www.ebook-converter.com******
CENTRIFUGAL
PUMP DESIGN

CENTRIFUGAL
PUMP DESIGN
John Tuzson
Evanston, Illinois


CONTENTS
Preface
1. INTRODUCTION
Scope of book
Information sources
References
2. FLUID MECHANIC FUNDAMENTALS
Practical numerical calculations
Physical principles
Continuity
Balance of forces
Conservation of energy
Potential flow, rotational flow and shear
Curved and vortex flow
Coordinates relative to rotating axes
References
3. INSTABILITY, TURBULENCE AND SEPARATION
Reynolds number
Instability and turbulence

Vortex flow
Separation
Lift
Drag
References
4. PIPES, BENDS AND DIFFUSERS
Pipes and bends
Diffusers
Obstructions in pipes
References
i. PUMPING SYSTEMS
Pump specifications and selection
System modeling
References
6. GENERAL PUMP THEORY
Pump performance
Off-design operation
Pump types and specific speed
Pump configurations
References
7. PUMP FLUID MECHANICS

Impeller work input
Flow at impeller inlet and exit
Slip
Relative velocity in the impeller and rothalpy
Flow separation in the impeller
Jet-wake flow pattern
Blade-wake mixing
Vaned diffuser
Volute
Incidence
Cavitation
Disk friction on the impeller
Axial thrust on the impeller
Leakage
References
8. BOUNDARY LAYERS IN PUMPS
Basic concept
Turbulence
Boundary layer thickness
Boundary layers in closed flow passages
Skin friction

Boundary layers on rotating disks
Boundary layers in vaneless diffusers
Boundary layers inside rotating impellers
Practical pump design calculations
References
9. CAVITATION IN PUMPS
Description
Net positive suction head
Cavitation at design conditions
Cavitation at off-design conditions
Example
Cavitation damage correlation
Inducers
Vibration
References
10. PERFORMANCE CALCULATION
Calculation procedure
Impeller work input
Flow losses
Calibration on test data
Results

References
11. PUMP DESIGN PROCEDURE
Published design procedures
Choice of impeller diameter, exit width, and exit blade
angle
Impeller inlet diameter
Inlet blade angle
Vaned diffuser and crossover inlet
Volute design
Mechanical design
Impeller blade design
Manufacturing requirements
Blade coordinates
Computer calculation of blade coordinates
References
12. DESIGN EXAMPLES
Preliminary design and performance estimate
Inlet velocity calculation
Detailed three-dimensional blade design
Two-dimensional blade design
Return passage design

Volute design example
Concluding remarks
13. ROTOR DYNAMICS OF PUMPS
Fundamentals of shaft vibration
Shaft and impeller supported on two bearings
References
14. INLET AND EXIT RECIRCULATION
Recirculation phenomena
Inlet recirculation
Exit recirculation
References
5. TWO-PHASE FLOW IN PUMPS
Two-phase fluid bulk properties
Experimental pump performance data
Calculation models
Fully separated flow model
Comparison with test data
Fill-spill fluid coupling
References
16. HIGH-VISCOSITY PUMPS
Performance estimates at high viscosity

Design for high viscosity
References
17. SAND AND SLURRY EROSION IN PUMPS
Applications and requirements
Slurry erosion mechanisms
Erosion rate estimates
Typical erosion pattern in slurry pumps
Slurry erosion testing
References
18. EJECTOR PUMP SYSTEMS
System description
System characteristics
Ejector
Ejector specifications and selection
Ejector system performance map
Design example
References
19. EVALUATION ON NEW FLUID MACHINERY CONCEPTS
Development criteria
Energy transfer
Specific speed

Fluid machinery types
Fluid machinery components and systems
Seals
Bearings
Drives
Cost
References
20. COMPUTER PROGRAMS FOR PUMPS
Computer calculations
References
APPENDIX
INDEX

PREFACE
The publication of this book was prompted by my desire to make advanced fluid
mechanic fundamentals available to practicing industrial pump designers.
Unfortunately, a profound rift exists between the academic or research community and
those on the factory floor. The unintended consequence of peer review is that
researchers are motivated by the approval of their colleagues instead of by the users of
advanced technology, the product designers. The academic rule of "publish or perish"
generates a flood of publications that are incomprehensible, and therefore useless, to
those who could make profitable use of the technology. This lack of technology transfer
represents a great loss to the economy, a waste of resources, and ultimately, a loss to
all humankind.
Recent books on centrifugal pump fluid mechanics and design could be listed in two
groups. One recalls valuable empirical industrial experience but lacks a comprehensive
description of fundamental fluid mechanical phenomena. The other describes advanced
laboratory experiments and analytical studies using a formidable mathematical
apparatus. The pay scale and job description of most industrial engineers, employed in
pump fabrication and application do not allow the hiring of master's-level engineers
having a background in advanced fluid mechanics and partial differential equations.
Product designers must be generalists, not specialists. The challenge before us is to
close this gap by insisting on fundamental understanding, but using a mode of
presentation and communication that is comprehensible to those with a minimal
engineering education.
Here is the place to thank all those who had a part in the ultimate appearance of this
book. Room does not suffice to thank individually my teachers in Budapest, Innsbruck,
Paris, and at MIT in Cambridge, as well as my many colleagues in industrial R&D and
in ASME who made me think and challenged me to make sense. However, I must make
an exception by expressing my special gratitude to Paul Hermann and Steve Zakem for
their many helpful comments and suggestions.
John Tuzson
(Tougeon, as in treasure, or tou(t) jaune in French)
Evanston, Illinois


INTRODUCTION
SCOPE OF BOOK
The intent of this book is to offer to those interested in pump design, a clear, detailed
picture of the flow pattern in centrifugal pumps and a simple pump design procedure
based on fundamental understanding, not on empirical formulas. There is the hope and
expectation that readers will learn the skill to design pumps rationally without falling
back on unexplained formulas. Teachers might find a practical outline for their program
of instruction. Since most elements of computer programs can be calculated by hand,
such hand calculations can serve as exercise problems, the results being known from
the computer program printout.
In today's competitive world, pump designers are tempted to just quickly copy
existing pumps, or fall back on empirical design formulas established by statistical
surveys of existing pumps. Such guidelines can be programmed into computers and can
produce pump designs with the push of a button. Great problems arise, however, when
the designs do not perform as expected. Without a deeper understanding of the flow
pattern in pumps, the reasons for malfunctioning are incomprehensible. A competitive
market also demands pump designs outperforming the competition. Statistically average
designs may be adequate but will not provide a competitive edge and will leave the
manufacturer in a "me too," catch-up posture. Superior designs can come only from a
detailed understanding of the flow pattern in pumps.
Today's pump designers need not one but a portfolio of calculation procedures and
computer programs. Even at the high speed of present computers and anticipated even
higher speeds of future computers, the execution of complex computer calculations
requires time, effort, and corresponding expense. Usually, the major portion of the time
is not consumed by the running of the program but by the time needed to collect and
input the necessary data. Reduction, plotting, and interpretation of the calculation results
takes additional time. The amount of time and expense spent on the design of a certain
pump must be in proportion to the value of the pump. Simple, fast, approximate
calculations may suffice for smaller, inexpensive pumps. Large, expensive custom-
engineered pumps warrant complex calculations. It is estimated that a satisfactory three-
dimensional flowvelocity calculation in a centrifugal impeller may take anywhere from
200,000 to 1 million computational mesh points (Gopalakrishnan et al. 1995, Kaupert et
a l . 1996). The design procedure presented here describes, at each design step, the
calculation that needs to be performed and defines the results that are expected from the
calculations. Simple personal computer (PC) programs and their program listings are
offered here to execute the calculation steps. These will give fast, approximate results
which may be adequate for simple pumps. However, if more precise results are desired
and are warranted, more detailed, commercially available computer programs can be
substituted for the simple ones presented here. Regardless of the type of computer
program being used, the designer must know and understand what needs to be
calculated, how the results will be used to improve pump design, and how far the
accuracy of the computer program can be trusted. Computer programs are only tools.
The creative talent and skill of the designer are still needed. The designer must judge
the amount of effort warranted by the design of the particular pump under consideration,
and therefore must select those design tools that are appropriate from the available
portfolio of computational procedures.
Modern hydraulic design procedures of axial flow pumps, which are not treated here,
are based on very extensive airfoil theory. These concepts have been developed during
the early years of aviation and gas turbine development. Because of strategic
importance, vast amounts of test data and empirical information were accumulated from
airfoils and rows of airfoils, cascades, in wind-tunnel experiments. Excellent design
procedures were developed using analytical approaches combined with empirical
correlations. Specialized concepts, variables, and a specialized vocabulary (span,
chord, camber, stagger, pitch, cascade, solidity, deviation angle, stall) describe these
design procedures. Attempts have been made to graft these concepts onto centrifugal
turbomachinery design procedures. Up to a first order of approximation, a similarity
exists between axial flow patterns in axial flow machinery and radial flow in
centrifugal machines. Mathematically, flow patterns around airfoils in axial flow can be
mapped onto a radial, circular field. But because of the effect of the Coriolis
acceleration, which is negligible in axial flow but very important in radial flow, or the
phenomenon of the jet-wake flow pattern in centrifugal turbomachinery, for example, the
similarity remains defective. Wislicenus pointed out this fact in 1941 (Wislicenus
1941). Pump design procedures, some of which predate most aircraft development,
generated over the years distinct concepts, and a different vocabulary, which demand
separate treatment.

INFORMATION SOURCES
An unprecedented wealth of useful information is stored and is freely available in the
technical libraries (Cooper 1999, Engeda 1999). However, despite computer searches
in the Engineering Index and Applied Mechanics Review, the specific information
needed remains difficult to find, especially on practical design issues. In addition to the
references quoted in appropriate chapters of this book, collections on pump technology
have been published, first of all, by the great engineering societies: the Fluids
Engineering Division of the American Society of Mechanical Engineers (ASME), the
Institution of Mechanical Engineers in England, the Vereins Deutscher Ingenieure
(VDI) in Germany, the Societe Hydraulique de France, the Japanese Society of
Mechanical Engineers (JSME), and the International Association for Hydraulic
Research (IAHR). Each has an archival journal containing technical articles, available
from technical libraries, but special symposia volumes often need to be ordered
directly from the associations. Research institutions such as the Electric Power
Research Institute (EPRI), the Von Karman Institute (VKI), and the Pfleiderer Institute,
have their own series of publications. The Societe Hydraulique de France reports on
French hydraulic machinery research in special issues (La Houille Blanche). Testing
a n d rating standards have been established and published by the International
Standards Organization (ISO), the Hydraulic Institute, ASME, IAHR, the American
Petroleum Institute (API), and the VDI. Several universities have centers specializing
in turbomachinery: Michigan State University, Pennsylvania State University, $cole
Polytechnique Federal de Lausanne, and Braunschweig and Karlsruhe Universities.
The Turbomachinery Laboratory of Texas A & M University has hosted 15 pump user
conferences. Other conferences and refresher courses are offered by universities, trade
associations, and private consultants. Manufacturers' trade associations of several
European countries are represented by EUROPUMP, which supports the trade journal
World Pumps, published by Elsevier in Oxford, a successor of the publication of the
British Hydromechanic Research Association (BHRA), Cranfield, Bedford, England.
Other trade journals, including Pumps & Systems and Pumping Technology, and trade
journals in the chemical industry in the United States, contain practical information.
REFERENCES
Cooper, P. (1999): Perspective: The New Face of R&D-A Case Study of the Pump
Industry, ASME Journal of Fluids Engineering, December, pp. 654664. Also
Chemical Engineering, February, pp. 84-88.

Engeda, A. (1999): From the Crystal Palace to the Pump Room, Mechanical
Engineering, February, pp. 50-53. Also ASME Paper 98-GT-22.
Gopalakrishnan, S., Cugal, M., Ferman, R. (1995): Experimental and Theoretical

Flow Field Analysis of Mixed Flow Pumps, 2nd International Conference on Pumps
and Fans, Tsinghua University, Beijing, October 17-20.
Kaupert, K. A., Holbein, P., Staubli, T. (1996): A First Analysis of Flow Field
Hysteresis in a Pump Impeller, ASME Journal of Fluids Engineering, Vol. 118, pp.
685-691.
La Houille Blanche (1968): No. 2/3.
La Houille Blanche (1985): No. 5.
VDI Berichte, (1980): Nos. 371, (1981): No. 424.
VDI Forschungshefte (1968): Nos. 527 and 528, (1969): No. 535, (1970): No. 548
Wislicenus, G. F. (1941): Discussion, in C. A. Gongwer, A Theory of Cavitation

Flow in Centrifugal-Pump Impeller, ASME Transactions, January, pp. 38-40.


FLUID MECHANIC
FUNDAMENTALS
PRACTICAL NUMERICAL CALCULATIONS

The fluid mechanical phenomena in centrifugal pumps are described in terms of
velocity and pressure distributions. Local velocities are measured relative to the flow
passage walls or the objects placed in the flow. The flow pattern is the same whether
an airplane flies through still air or the air flows past a stationary airplane.
The pressure represents the pressure energy of a unit volume of fluid, as can be seen
from its dimensions:
In the case of incompressible fluids such as water, the pressure appears only in the
form of a pressure difference with respect to a reference pressure. In pumps the
pressure level is often given with respect to the pressure prevailing at the pump inlet.
However, the choice of reference pressure is arbitrary as long as it is clearly defined. In
practical applications the reference pressure level is often taken at a free water surface
exposed to the ambient. In pressure-measuring instruments, the pressure is shown either
with respect to ambient pressure, which at standard conditions is taken as 14.7 lb/in2,
or is measured deliberately as a pressure difference, by connecting a manometer or
differential pressure gauge to two locations in the fluid. Barometers measure pressure
level with respect to an absolute vacuum. Consequently, the units of pressure are
designated psig if they are measured from ambient pressure, and psia if they represent
absolute pressure.
Unlike the case of incompressible fluids, absolute pressures must be used when the
flow of compressible fluids is analyzed. Compressible fluids are typically gases, but
two-phase flow, liquid and gas mixtures, also qualify as compressible fluids. In pumps,
compressible flow phenomena appear when the liquid contains gases or vapors. Such a
case is when cavitation appears, when the local pressure is lowered to the point where
the liquid starts boiling and vapor bubbles appear. In compressible fluids, in general,
fluid properties, such as the boiling point at a given temperature, and in particular the
specific volume, depend on the absolute pressure, not on the pressure difference.
Cavitation will appear when the absolute pressure anywhere in a flow dips below the
vapor pressure of the fluid, which is a material property of the fluid depending on the
absolute pressure and temperature. In such cases, absolute pressures should preferably
be used.
Pump fluid mechanic calculation will require algebraic manipulations of the

fundamental equations to arrive at numerical results. Many errors of numerical
calculations can be avoided by careful balancing of the units of physical quantities.
Regardless of the type of units used, SI or English, the three dimensions of mass M,
length L, and time T enter into the units of the physical quantities of interest in pump
design (Table 2.1).
Every term of an equation, consisting of products of variables, must have the same
combination of dimensions. Indeed, they must even have the same units. This
requirement helps to keep track of conversion factors and to observe the relationship
between the units of mass and of force. (Force is mass times acceleration.) Applied, for
example, to Bernoulli's equation, which states that the energy of a fluid element remains
the same as it moves along a streamline, we have

In English units:
In SI units:
The various terms of the Bernoulli equation represent different types of energies. The
ratios of terms yield dimensionless numbers, the magnitude of which expresses the
relative importance of these energies. Such dimensionless numbers are:
where p is the pressure, V the velocity, and p the density.
where V is the velocity, g the gravitational acceleration, and h the typical height, the
density being eliminated.
The Bernoulli equation assumes inviscid flow; therefore, the viscous energy is
neglected and does not appear. Nonetheless, the importance of viscosity can be
expressed as the ratio of the inertia (pV2) and viscous energy (V/D), which becomes
the Reynolds number = VD/v, since the kinematic viscosity v is the ratio of the dynamic

viscosity and the density p. Flow calculations can be simplified considerably if either
the viscous or inertia forces can be neglected. The magnitude of the Reynolds number
serves as a measure to decide which force can be neglected. In pumping machinery
handling water, large Reynolds numbers, typically well above the transition Reynolds
number range of 102 to 103, are usually encountered, which show that viscous forces
are not very important. The prevalence of viscous or inviscid flow conditions in various
pump components is governed by the local Reynolds number based on a typical velocity
V and on the typical geometrical width or size of the component D.
The magnitude of the pressure coefficient indicates, for example, how much pressure
head is recovered in diffusers, and therefore becomes a measure of performance. Since
it is dimensionless, it applies to similar diffusers regardless of their size and of the
actual magnitude of the velocity.
Nondimensionalizing the fundamental equations by dividing through with the typical

variables of one of the terms-for example, the inertia term eliminates all units and
dimensions. The solution can then be obtained in pure numbers or dimensionless
coefficients, which have general validity. Such numbers are, for example, pipe friction
loss coefficients, which specify that fraction of the velocity head, pV2/2, which results
in a pressure loss, p2 -pi.
When exploring the relationship between a number of variables, a dimensionless

correlation can be established by forming dimensionless groups of each of the variables
with the help of selected, representative values of the three basic dimensions-length,
time, and mass-or other, selected variables having a dimension that is a combination of
these basic dimensions. The three basic dimensions suffice for describing mechanical or
fluid-mechanic phenomena. Further fundamental units must be introduced when other
physical phenomena need to be analyzed, such as heat or mass transfer, electricity, or
chemical processes.
Since several possible combinations exist, several possible equivalent sets of

dimensionless groups, including all variables, can exist. The unknown relationship must
involve one complete set of these dimensionless groups, even if the exact nature of their
relationship is not known. Sometimes a relationship consisting of a product of all
dimensionless groups is assumed, and the exponents of the dimensionless groups are
determined by experiment. In particular, heat and mass transfer calculations in chemical
engineering often use such correlations. In fluid mechanics such an empirical correlation
has been established for estimating the pipe friction pressure drop. The dimensionless
groups are the loss or pressure coefficient Cp, the friction coefficient .?, the ratio of
length to diameter LID, the relative roughness k/D, and the Reynolds number Re.
where A becomes a function of k/D, and Re. Indeed, in the laminar and turbulent range,
respectively (Schlichting 1960, p. 525):
Between the laminar and turbulent ranges a transition range exists, which is a more
complex function of k/D and Re and is shown graphically on the empirical plot of the
pipe friction coefficient. Similar but slightly different mathematical expressions have
been derived by various scientists.

PHYSICAL PRINCIPLES
The fundamental equations, which are presented here, are not of the most general form
but have been simplified to be used for designing pumps. Still they derive from the
general physical principles of conservation of mass and energy and the balance of
forces and moments (Batchelor 1967, Comolet 1994, Lamb 1955). Drastic
simplifications are made here by assuming an incompressible fluid, most often water,
and assuming steady flow. The effects of viscosity are neglected and the flow is
assumed to be inviscid. Some special effects of viscosity are discussed in appropriate
chapters.
To derive the mathematical equations corresponding to the principles described

above, a space or volume-a control volume-is defined within the fluid. The closed
boundaries or surface of the volume can be drawn arbitrarily. If the volume is chosen
infinitesimally small-for example, the cube dx, dy, dz in rectangular coordinates the
equations appear in the form of partial differential equations of the three velocity
components and the pressure. The technical literature refers to these partial differential
equations as the Navier-Stokes equations. If the size of the volume is chosen larger than
infinitesimal, the velocities entering or leaving the volume and the pressure acting on the
boundaries must be integrated around the boundaries. In practice, the integration can be
implemented by subdividing the surface into smaller elements, over which the flow
velocity or pressure remains constant, and by adding up flow across these elements and
the pressure forces. In many cases the boundaries can be chosen in such a manner that
the velocity remains uniform and perpendicular or parallel to at least a portion of the
boundary, which simplifies the calculations and leads to simple algebraic equations.
The state of the fluid at any one location is defined by its velocity and pressure. Since
the velocity is a vector quantity and has not only a magnitude but also a direction in
three-dimensional space, components of the velocity in three perpendicular directions
must be known. The total number of unknowns is then four, and therefore four equations
are required to obtain a solution. In many cases the flow can be assumed to take place in
a plane, in two dimensions, or along a line, in which case only three or two unknowns
exist and only three or two equations are required. For example, the flow is often
assumed to be parallel to the flow passage walls.
Continuity
The equation of conservation of mass, also called the equation of continuity, states that

the steady flow into any space must be equal to the outflow. In steady flow the fluid
particles follow certain streamlines. A flow cross section can be defined
perpendicular to the limiting streamlines on the circumference of the cross section (Fig.
2.1). Then the flow through any cross section defined by the same streamlines must be
the same. For example, an obvious choice of streamlines in a pipe corresponds to the
pipe walls. The flow through any pipe cross section must then be the same.

Balance of Forces
The three equations corresponding to the balance of forces must be written separately
for the three perpendicular directions-x, y, z in rectangular coordinates-because they
are independent from each other. The physical principle states that the pressure forces
on the boundaries and the difference between the momentum entering and leaving the
control volume in the chosen direction must be equal. The momentum of the flow is
equal to the mass times velocity. In the simple case of straight, parallel flow, only one
equation is needed. If there is an object in the flow, the force on the object, F, must be
included in the equation (Fig. 2.2):
Conservation of Energy
The principle of conservation of energy states that the energy of a unit fluid element
remains the same along the streamline it follows, provided that there are no losses.
Since energy is a scalar quantity, only one equation is needed, the familiar Bernoulli's
equation (Fig. 2.3):
Figure 2.1 Continuity or conservation of mass equation.

Figure 2.2 Balance of forces equation.
Here po is the total pressure or stagnation pressure, the sum of all the terms. The
variables h, and h2 stand for the height of the two locations, measured from the same
reference level, as shown on Fig. 2.3. When the equation is applied to short piping
components or pumps, the head difference from the level difference is negligible
compared with the head change due to head loss or head rise.
In pump calculations the terms of this equation are often divided through by the
specific weight of the fluid, pg, to convert to the terms of velocity head, pressure head,
and static head all having the dimension of length. The total head or stagnation head, ho,
designates the sum of all these terms, and corresponds to the head that would be reached
if the fluid were brought to rest without any losses:

Figure 2.3 Conservation of energy equation.
The continuity equation and the three balance-of-forces equations are sufficient to
determine the four unknowns: the pressure and the three velocity components. However,
sometimes use of the energy equation is to be preferred to the force balance equation,
especially if the flow is in one direction only. Also in some instances the losses can be
calculated using the energy equation after the velocity and pressure have been
calculated from the other equations. Such an example, that of the sudden expansion loss,
will be given in Chapter 4.

Potential Flow, Rotational Flow, and Shear

If there are no losses and the energy-the total or stagnation pressure as expressed by
Bernoulli's equation-is the same on all streamlines everywhere, then an irrotational,
potential flow exists and calculation is considerably simplified (Flugge 1959, Lamb
1955). The important issue from the point of view of mathematics is whether the flow
velocity vector field satisfies the conditions of irrotation. The important issue from the
point of view of physics is whether the energy on all streamlines is the same. The two
criteria are identical, as can be shown by mathematical deduction. (If the derivative of
Bernoulli's equation is taken in the three coordinate directions, and if the
corresponding Navier-Stokes equation, without viscous terms, is subtracted from each,
it is found that the energy does not change in any direction, provided that the condition
of irrotationality is satisfied.)
The irrotational, potential flow, or equal-energy hypothesis allows that the pressure
be eliminated from the basic equations, which then contain only the velocities and the
geometrical variables. All velocities can be changed by a common factor, positive or
negative, without changing the shape of the streamlines. The flow can even reverse. The
equations are linear and can be superposed. The sum of two or more solutions is also a
solution. At every point, the velocities, calculated from different solutions, can be
added. However, potential flow equations express only purely kinematic relationships.
The flow can be unsteady and the calculated flow pattern can be an instantaneous
snapshot of a time-dependent flow. The dynamics of the flow, the balance of forces,
depends on the pressure distribution, which remains totally undefined until the pressure
distribution on the boundaries has been specified. It can be reasoned that the stability of
a potential flow solution, whether it will persist or change with time, also remains
undefined until the dynamics of the flow is clarified. This consideration would imply
that some potential flow solutions, which are unstable, cannot persist and will not be
found in real life (D'Alambert's paradox, which states that objects in a hypothetical
potential flow can have no drag, an obvious impossibility in real life).
Since the flow entering the pump can usually be assumed to have the same energy on
all streamlines, a potential flow assumption is not unreasonable, provided that there are
no major energy losses on some of the streamlines and provided that the flow follows
the flow passage walls and does not separate. If parallel streamlines have different
energies, a rotational flow exists with shear between neighboring streamlines. Shear
flows can develop when the fluid on one streamline experiences greater energy losses
than on another, or when pumps impart less energy to the fluid on one streamline, than
on the other. In straight, parallel, rotational flow the velocity varies perpendicularly to
the flow direction. The effects of shear are discussed in a subsequent chapter.

Curved Flow and Vortex Flow

If the streamlines are curved, the pressure increases perpendicular to the streamlines in
the direction away from the center of curvature, because of the centrifugal acceleration.
Conversely, if the streamlines are straight and parallel, the pressure must be the same
throughout the cross section perpendicular to the streamlines. These rules help to
visualize flow pattern in a given case and can provide some indication in which
direction pressure and velocity change.
Vortex flow represents a special case of curved streamlines, when the streamlines
are concentric circles around the center of the vortex. The pressure increases with the
radius: dp/dr = pV2/r because of the centrifugal acceleration. A potential vortex exists
if the energy on all streamlines, everywhere, is the same. To arrive at an expression for
the radial variation of the tangential velocity V, the pressure is eliminated from the
energy equation, Bernoulli's equation, by differentiating it with respect to r and
substituting dp/dr from the expression above.

Assuming that the axial velocity in the z direction is constant or zero, the solution of the
resulting differential equation predicts that the tangential velocity V must vary in
inverse proportion to the radius. Consequently, Vr = constant.
The product of the circumferential velocity and the radius Vr corresponds to the
angular momentum of the fluid particle. The finding above states that the angular
momentum of a rotating fluid is conserved if there are no losses or external forces-a
fundamental principle of physics. As will be seen, centrifugal pumps impart energy to
the fluid by increasing its angular momentum. Consequently, this principle plays a very
important role in pump calculations. The corresponding equations in the absolute as
well as in the relative rotating frame of reference are given in Chapter 7 when the flow
in a rotating pump impeller is described.

Coordinates Relative to Rotating Axes

Since flow in pump impellers can be considered axisymmetric, or shows an axial
symmetry, polar coordinates are particularly convenient. Flow patterns relative to the
rotating impeller can be studied by subtracting the rotational speed of the impeller, ter,
from the absolute tangential velocity. However, the additional terms corresponding to
the centrifugal and Coriolis acceleration must be retained. The centrifugal and Coriolis
accelerations distinguish the flow conditions in the impeller from the conditions that
would be present if the impeller would not rotate. Although the flow pattern would be
the same in a rotating and a stationary impeller if potential flow could be assumed, the
pressure distribution would differ greatly. In the case of unequal energy flows, shear
flows, when boundary layers or flow separation were present, secondary flows and
unusual flow patterns can appear, such as a jet-wake flow or blade suction side
separation, which would have no equivalent in an impeller that would not rotate.
Figure 2.4 Axisymmetric flow in cylindrical coordinates: (a) velocities; (b) centrifugal
and Coriolis accelerations.
If the flow is to be defined in terms of the relative velocities Wr, Wo, and W, with
respect to the r, 0, and z coordinate directions rotating with an angular velocity w
around the z axis, the following expressions must be substituted for the absolute
velocities into the equations in cylindrical coordinates: Vr = Wr, Vo = Wo + cur, and V
= W. It is found that in addition to the terms one would see in the equations for a
stationary, cylindrical coordinate system, the following acceleration terms appear in the
radial and the circumferential directions which result from the rotation: -2Woco - co2r
and +2Wrco. One recognizes the centrifugal acceleration term due to the rotational
velocity u02r. The remaining two terms are due to components of the Coriolis
acceleration expressed in the rotating frame of reference. These accelerations would not
exist if the coordinate system would not rotate (Batchelor 1967).

REFERENCES
Batchelor, G. K. (1967): Fluid Dynamics, Cambridge University Press, New York.
Comolet, R. (1994): Mecanique Experimentale des Fluides, Masson, Paris.
Flugge, S. (1959): Encyclopedia of Physics, Vol. VIII/1, Fluid Dynamics I, Springer

Verlag, Berlin, pp. 187.
Lamb, H. (1955): Hydrodynamics, Dover, New York.
Schlichting, H. (1960): Boundary Layer Theory, McGraw-Hill, New York.


INSTABILITY, TURBULENCE,
AND SEPARATION
REYNOLDS NUMBER
Drastic changes take place in the flow pattern when, with increasing velocity, the
inertia forces become greater than the viscous forces. The ratio of these forces, the
Reynolds number, gives a measure of their relative importance. In high Reynolds
number flow, inertia forces predominate.
Here pV2 stands for the inertia forces and pV/L for the viscous forces. The ratio of the
dynamic viscosity p and the density p is generally known as the kinematic viscosity v.
The shear force due to the viscosity of the fluid acts between neighboring streamlines
in the direction of the velocity. Its effect appears most clearly in the case of straight,
parallel flow. In straight, parallel flow a viscous shear force i appears when the
velocity varies perpendicular to the streamlines (Fig. 3.1). It is proportional to the
velocity gradient and perpendicular to the direction of the velocity dV/dy, the
coefficient of proportionality being it the dynamic viscosity of the fluid:

Figure 3.1 Parallel shear flow.
At 68 F or 20 C the numerical values of these quantities for water are p = 21.1 x 10-
6lbf-sec/ft2 = 1.0087cP, v = 10.9x 10-6ft2/sec = 1.008 x 10-6 m2/s, and p = 1.935 lbf-
sect/ft4 = 1.00 x 10-3N s2/m4.
Such shear appears, for example, at the walls of flow passages. The fluid
immediately next to the wall does not move, and the velocity parallel to the wall
increases gradually with the distance from the wall toward the main stream. A small,
low-velocity layer-a boundary layer-exists near the wall, the thickness of which
depends on the Reynolds number, and is typically small in water pumps. Since this
shear force at the wall slows the flow and results in an energy loss, it must be taken into
account in flow calculations, which is done by introducing a wall friction loss and a
dimensionless pressure coefficient, the skin friction coefficient. Shear layers also exist
on the sides of jets, where the high velocity of the jet neighbors a stagnant fluid. Wake
boundaries, behind objects in a free stream, are also shear layers (Batchelor 1967,
Crighton 1985).
It will be noted that in the straight, parallel flow case above, the velocity will vary
from one streamline to the next only if the energy on the streamlines is different. Indeed,
in straight, parallel flow, in the absence of streamline curvature, the pressure cannot
vary perpendicular to the streamlines. Therefore, the sum of the pressure head and
velocity head, the total or stagnation head, cannot remain the same. This state implies
that the flow is rotational and is not a potential constant-energy flow. More generally
stated, shear, and therefore viscous shear forces, can exist only in rotational flows.

INSTABILITY AND TURBULENCE

When viscous forces become small compared with inertia forces, waves and
fluctuations appear on the shear layers between neighboring streamlines with different
energies (Fig. 3.2). At moderately high Reynolds numbers these waves grow and
eventually, farther downstream, roll up in vortices. At high Reynolds numbers an
irregular fluctuation-turbulence-appears. This instability of shear flows is inherent in
the nature of fluid flow and does not depend on viscosity. On the contrary, it appears
when viscosity becomes unimportant. Turbulence is not a viscous phenomenon.
Turbulent fluctuations and excursions of fluid particles transfer energy from one
streamline to the other. They accelerate the slow stream and slow the fast.
Consequently, turbulence acts in a manner similar to viscosity. Therefore, an artificial
eddy viscosity has been defined, which as a first approximation is sometimes used in
calculations, as if it were the real viscosity of the fluid.
Practical flow calculations aim to arrive at the steady value of the flow variables.
Calculation of the instantaneous value of the fluctuating quantities is of little interest.
Therefore, advanced models to calculate turbulent flow operate with time-averaged
values of the instantaneous velocities. These models attempt to represent the average
effect of turbulent fluctuations. Several more advanced and more complex
computational turbulence models have been developed. They still incorporate empirical
data and usually give fully satisfactory results only for those kinds of flow
configurations for which they have been developed.
Above it was stated that in straight, parallel flow, shear layers become unstable,
develop waves, and roll up in vortices. Quite generally, in mechanical systems, the
proof of stability consists of disturbing the equilibrium position and observing whether
the system returns to it-whether the oscillations around the equilibrium position tend to
damp out or amplify. Such calculations become prohibitively complicated in fluid
mechanics. Nonetheless, it was possible to show that shear layers can be stabilized by
an acceleration, and corresponding pressure gradient, oriented perpendicularly to the
shear layer and in the direction in which the energy increases on neighboring
streamlines (Bayly and Orszag 1988; Johnston et al. 1972; Tuzson 1977, 1993).
Acceleration in the opposite direction destabilizes the flow. Knowing the direction of
acceleration in a shear flow provides an estimate of whether rapid mixing will occur or
whether the lowenergy flow region will persist. Applied to impeller flow, for example,
a realistic flow model becomes possible, as will be shown.

Figure 3.2 Unstable shear layer on jet boundary.

Vortex Flow
A vortex offers the most striking example of stabilized flow. In an equal-energy,
potential vortex, the velocity varies in inverse proportion to the radius, Vr = constant
(Fig. 3.3). It will be observed that according to this relationship, when the radius
approaches zero, the velocity approaches infinity. Therefore, such an ideal vortex
cannot exist in real life. Indeed, real vortices possess a stagnant core near the center.
This core rotates as a solid body with constant angular speed w, and the velocity in the
core becomes V = wr. The core boundary forms a shear layer. However, the
centrifugal acceleration acts perpendicular to the streamlines, and points from the core,
where lower energy prevails, radially out. Consequently, the shear layer is stabilized
and can persist. More generally, it is found that if in a vortex with an arbitrary radial
velocity distribution, anywhere the velocity decreases faster than I /r, an unstable shear
layer exists. Such is the case, for example, at the wall of a cylindrical vessel containing
a swirling fluid. The boundary layers are unstable and exceptionally fast mixing
occurs. In the opposite case, when the velocity changes radially less fast than l 1r, the
flow is stable.
The core of a vortex is the only place where in real life stagnant fluid can remain
stable and persist without mixing in a large flow field. In other cases, in wakes behind
obstacles, the shear layers of the wake boundaries become unstable, generate irregular,
unsteady flow fluctuations, and eventually, far downstream, mix with the main flow.
Separation
Further complications occur when the flow cannot follow the passage walls or flow
boundaries and separates. Two typical cases of flow separation can be distinguished.
The first type appears at sharp corners or sudden recesses in the wall, as shown in Fig.
3.4. The flow is unable to turn sharply around the corner, because the streamline
curvature would become very small and the pressure rise perpendicular to the
streamline very large. Since the fluid conserves its energy on the streamline
approaching the corner, according to the Bernoulli equation, the velocity would have to
become extremely high and the pressure at the corner would drop to theoretically
infinitely low values. In real flow the streamline leaves the wall at the corner and
becomes a free streamline or shear layer. The region behind the corner becomes filled
with stagnant or slowly moving, recirculating, low-energy fluid; it is a location where
such low-energy fluid can accumulate and remain without being swept downstream.
The shear layer is unstable and mixes farther downstream, resulting in an energy loss.

The second type of separation appears near gradually curving or even straight but
diverging walls, where the passage cross section increases and the flow would have to
slow down (Crighton 1985). With decreasing velocity, according to Bernoulli's
equation, the pressure increases. Except for very special conditions, generally the flow
cannot tolerate such a pressure increase and the streamline separates from the wall and
becomes a free streamline, a shear layer. As before, low-energy fluid accumulates
beyond the separation point near the wall. The shear layer mixes and some of the energy
of the flow is lost. The downstream pressure cannot rise above its value at separation,
which prevails throughout the separated region. Since the pressure along the free
streamline, bordering the separated region, is constant, the flow velocity on the
separated or free streamline must also remain constant. Such separation occurs, for
example, behind obstacles placed into the fluid stream, or in diffusers, which are
discussed below. Flow separation is responsible for major losses in pumping systems,
and great effort is made to avoid them by streamlining the flow passages.
It is seen that the velocity and pressure distribution beyond the separation point
depend on the location of the separation point, which unfortunately is sometimes
difficult to determine. On gradually curving surfaces the velocity changes slowly ahead
of the separation point, and secondary effects, such as those of viscosity, turbulence, and
upstream conditions, can become important and can influence the exact location of the
separation point. Refined computer calculation methods have been developed for
determining separation points on airplane wings, for example. However, such
calculations are not warranted for pump design, and empirical guidelines have been
developed instead.
Two issues exist in pump design, where the onset and location of separation is
particularly important and for which empirical guidelines have been found: flow in
diffusers and flow through the impeller. These are discussed in subsequent chapters. It
is important to remember that fluid flow can accelerate with very low losses, but
always experiences losses when slowing down.
Figure 3.4 Backward-facing step; shear layer instability.

Lift
All objects placed in a fluid stream experience a pressure force, usually split into two
components, one perpendicular to the main direction of flow, the lift, and a second
component in the direction of the flow, the drag (Fig. 3.5). These forces are calculated
by integrating the pressure on the surface of the object. Most calculation methods use
inviscid potential or equal-energy flow assumptions to determine the pressure
distribution on the surface of the object. However, the potential flow assumption does
not allow for flow separation, which occurs on most objects, except on ideal airfoils at
low incidence. Therefore, potential flow pressure distribution is assumed up to the
location of the separation point, and constant wake pressure beyond. As a first
approximation, separation can be assumed to occur where the velocity near the surface
begins to decrease or slightly beyond.
Lift is particularly important for aircraft wings, but it appears on all objects, except
on those having an axisymmetric (bullet- or droplike) shape, with their axis aligned in
the direction of the flow. The lifting force can be attributed to rotation or circulation
around the lifting body, which can be represented by a vortex with its axis
perpendicular to the flow direction and also perpendicular to the direction of the lifting
force vector. The rotational velocity due to the vortex adds to the main flow on one side
of the body and opposes the main flow on the other. Therefore, the vortex induces higher
velocities, and therefore lower pressures, on one side of the body than on the other,
which results in a lateral resultant of the pressure force, a lift. Since according to fluid
mechanic fundamentals a vortex filament cannot start or end in the free stream, the lifting
vortex has to trail away from the wingtips or sides of the lifting body, leaving a wake in
the form of two opposing vortices at equal distance, with their axes aligned in the
direction of the flow, as shown in Fig. 3.6. Ideally, a vortex filament can only start at a
stagnation point, where the velocity comes to zero. Such trailing vortices can often be
observed on aircraft wingtips when landing in cloudy weather. Trailing vortices also
exist behind road vehicles, for example (Sovran et al. 1978, Weihs 1980), and behind
protrusions and holes on flow passage walls (Tuzson 1999). They are not visible in
water, but are present in pumps, and result in serious erosion when the water carries
sand, for example (Batchelor 1967).

Figure 3.5 Lift and drag forces.

Drag
Drag forces are responsible for most fluid energy losses. As described above, the flow
separates from the surface of the body, and a low-energy wake appears behind the
body. The wake is at approximately uniform pressure, equal to the pressure at the
location of separation, and constant velocity persists on the wake boundaries. The
shear layers at the wake boundaries become unstable. Irregular oscillations and eddies
appear that mix the wake fluid with the main stream. The drag force Fd on various
objects, having a cross section A, have been calculated theoretically and have been
measured experimentally. Typically, the value of the drag coefficient C1 ranges from
0.1 to 1.0:
Assuming a two-dimensional flow, if a control volume is defined around an object in a

uniform stream by drawing boundaries at a great distance from the object, uniform free-
stream conditions prevail on all boundaries except on the downstream boundary, where
the wake trails away, as shown in Fig. 3.7. All forces acting on this volume must
balance. Applying the force balance equation to this volume in the streamwise
direction, the drag force on the object must be balanced by forces on the downstream
boundary. If the low velocity of the fluid in the wake is parallel to the main stream, the
pressure must be uniform and equal to the free-stream pressure on the downstream
boundary. Therefore, the pressure forces cancel on all boundaries of the control
volume. The momentum forces entering the volume, however, are greater than the
momentum forces leaving, because of the velocity defect of the wake. This momentum
defect must therefore balance the drag force of the object. The drag force of an object
in a uniform stream can be measured by adding up or integrating the momentum defect
across the wake (Batchelor 1967):

Figure 3.6 Three-dimensional lifting body and trailing vortices.
Here u is the average velocity in the wake and A is the approximate crosssectional
area of the wake. If the wake contains trailing vortices, the pressure is not uniform on
the downstream boundary of the control volume, and the corresponding pressure force
must be added to the momentum force of the wake.

REFERENCES
Batchelor, G. K. (1967): Fluid Dynamics, Cambridge University Press, New York,
pp. 348-353, 580-590.
Bayly, B. J., Orszag, S. A. (1988): Instability Mechanisms in Shear-Flow Transition,

Annual Review of Fluid Mechanics, Vol. 20, pp. 359-391.
Figure 3.7 Drag force and wake behind an object in a uniform stream.
Crighton, D. G. (1985): The Kutta Condition in Unsteady Flow, Annual Review of

Fluid Mechanics, Vol. 17, pp. 411-445.
Johnston, P. J., Halleen, R. M., Lezius, D. K. (1972): Effects of Spanwise Rotation on

the Structure of Two-Dimensional Fully Developed Channel Flow, ASME Journal
of Fluid Mechanics, Vol. 56, No. 3, pp. 533-557.
Sovran, G., Morel, T., Mason, W. T. (1978): Aerodynamic Drag of Bluff Bodies and
Road Vehicles, Symposium at the General Motors Research Laboratory, Plenum
Press, New York.
Tuzson, J. (1977): Stability of a Curved Free Streamline, ASME Journal of Fluid

Mechanics, September, pp. 603-605.
Tuzson, J. (1993): Interpretation of Impeller Flow Calculations, ASME Journal of

Fluids Engineering, September, pp. 463-467.
Tuzson, J. (1999): Slurry Erosion from Trailing Vortices, ASME Fluids Engineering
Conference, San Francisco, FEDSM99-7794.
Weihs, D. (1980): Approximate Calculation of Vortex Trajectories of Slender Bodies

at Incidence, AIAA Journal, Vol. 18, pp. 1402-1403.


PIPES, BENDS, AND

DIFFUSERS
PIPES AND BENDS

For engineering calculations it can usually be assumed that the flow velocity in straight
pipes is uniform across the cross section. A small boundary layer exists near the walls,
for which allowance can be made by assuming a blockage of a few percent and a
corresponding reduction in the cross-sectional area. The equation of continuity
determines the velocity from the flow rate and cross-sectional area. If pipes branch or
merge, the flow will split in a proportion of the cross-sectional areas of the branches.
In reality, upstream disturbances convect far downstream, and the establishment of

uniform flow conditions requires about 10 pipe diameters of straight pipe run.
Disturbances arise from piping components which change the direction or velocity of
the flow: enlargements, bends, branches, tees, enlargements. These produce separation,
trailing vortices, and turbulence, which extract energy from the flow and produce
pressure losses. The lost energy is still present in the fluid in the form of irregular
fluctuations and vortices, which only dissipate far downstream, but such irregular flow
is of no use, and the corresponding energy represents a loss. Such losses in pipes and
piping components have been measured. The loss is proportional to the velocity head of
the flow. Loss coefficients are on the order of 1. More precise values for various piping
components can be found in most engineering handbooks.
The pressure loss in straight pipe runs is correlated in the form of a pressure
coefficient C,,, or loss coefficient A:
The curves shown on the Fig. 4.1 correspond to the correlation of Moody (1994) and
Nicuradse (1933), which are also given by the following analytical expressions
(Schlichting 1960, p. 525), where the Reynolds number corresponds to Re = DV/v:

In this expression k is the average size of the surface roughness elements. As an

example, typical cast iron pipes have a roughness height of k = 0.01 in = 0.25 mm.
Turbulent flow usually exists in those piping components that operate at Reynolds
numbers Re beyond 100,000 (note that "log" stands for the base 10 logarithm)
(Hydraulic Institute 1961).
In the expression above for the pipe friction loss, L is the length of the pipe and D the
diameter. If the pipe cross section differs from the circular shape, D becomes the
hydraulic diameter Dh, which corresponds to four times the crosssectional area A
divided by the circumference C: Dh = 4A/C. Losses of piping components are specified
by an appropriate pressure coefficient C1,. Practical calculation procedures express the
pressure drop from piping components as the pressure drop from an equivalent length of
straight pipe, as shown in Table 4.1. Typical flow velocities in pipes range from 5 to 25
fl/sec (1.5 to 7.5 m/s).
Figure 4.1 Pipe friction coefficient as a function of Reynolds number.

Figure 4.2 Orifice flow and pressure drop.
In case of constrictions or orifices, the pressure drop is often known, and the flow
rate Q is to be calculated, d being the orifice diameter (Fig. 4.2).
Here Cd is a discharge coefficient that accounts for constriction of the streamlines in

the orifice and for the variation of the velocity across the cross section. For sharp-
edged orifices the discharge coefficient is on the order of 0.7. With careful rounding it
can approach 1.
Reaction forces on 90 bends can be calculated by applying the momentum equation

in the direction of the incoming and exiting flows separately, since the forces in the two
directions are independent (Fig. 4.3). The reaction forces must be resisted either by the
piping connections or by external supports. The direction of the resulting external force,
the vectorial sum of the forces in the two directions, is at 45 to the direction of the two
components.
In either direction the reaction force is
Figure 4.3 Reaction force on a pipe bend.
In this expression V is the flow velocity, p the local static pressure at the inlet or exit
flange, and A the cross-sectional area of the pipe. Water hammer pressure pulses can
generate considerably larger, instantaneous reaction forces and should be considered.
Various standards regulate the required strength of piping connections in practical
applications.

DIFFUSERS
A great portion of the energy imparted to the fluid by pumps consists of a velocity
head: pV2/2, whereas the energy is usually desired in the form of pressure.
Consequently, an important component of pumps is a diffuser, which transforms the
velocity head leaving the pump impeller into a pressure head by slowing down the
flow gradually. As explained in Chapter 3, the flow has a strong tendency to separate
when it encounters a pressure rise. It takes a very careful design to avoid separation
and achieve a significant pressure recovery in a diffuser.
Diffusers achieve pressure recovery or diffusion by gradually increasing the flow

passage cross section (Fig. 4.4). According to an empirical rule of thumb, in straight-
wall diffusers the walls should diverge by an angle of about 7. More precise, but still
empirical guidelines have also been developed (Carlson et al. 1967, Johnston 1998,
Muggli et al. 1997, Reneau et al. 1967) and are shown in Fig. 4.5. Experiments show
that the least losses occur when the flow is on the verge of separating. In practice, inlet
disturbances and a nonuniform inlet velocity distribution require a more conservative
design.
Figure 4.4 Diffuser.

Figure 4.5 Preferred diffuser geometry.
Diffuser performance is illustrated by plotting the pressure recovery Cp, which is the
fraction of inlet velocity head converted to pressure head, against the length of the
diffuser or against the increasing cross-sectional area. The performance of actual
diffusers can be bracketed on this plot by two theoretical curves: the perfect loss-free
diffusion and the sudden expansion loss, which can be considered (with some
qualification) the worst case.
The ideal, loss-free case can be calculated from the continuity and energy equations
applied to a straight two-dimensional diffuser.
Continuity:

Energy:
Substituting V2 = VI(AI/A2) and rearranging yields
The sudden expansion case corresponds to the flow configuration when a flow passage
suddenly opens into a much larger one, as shown in Fig. 4.6. The flow separates and a
jet issues into the larger passage. As the direction of velocity is along the straight flow
passage, in this case the continuity equation and the force balance equation in the
direction of the velocity can be used to obtain a solution, provided that the flow is
uniform and completely mixed at the end of the larger passage, at the cross section A2.
The passage configuration and the boundaries of the control volume are shown in Fig.
4.6. Note that the pressure in the separated region equals the pressure at the separation
point of the flow.
Continuity:
Force balance:
Figure 4.6 Sudden expansion.

Substituting V2, dividing by A2, and rearranging yields
The pressure recovery of practical diffusers typically falls between these two cases
(Fig. 4.7). It can be less than the pressure recovery for sudden expansion if the diffuser
is not long enough and the flow is not completely mixed at the exit. Practical diffusers
rarely exceed area ratios greater than 2. At an area ratio of 2, ideally 75% of the inlet
velocity head could be recovered, while a sudden expansion would still recover 50%.
Most smaller pumps use short diffusers to save space, and only 50% or less of the
velocity head is recovered. Also, the flow velocity at the diffuser inlet is nonuniform. It
leads to early separation, beyond which no further kinetic energy can be recovered.
Since both the pressure and velocity at the diffuser inlet and exit are known, the
equation of conservation of energy, Bernoulli's equation, can be used to calculate the
energy lost between the inlet and exit. Experiments with curved diffusers lead to the
conclusion that turning through 30 or less leaves the pressure recovery essentially
unaffected (Fox and Kline 1962).
In diffuser design the greatest attention should be given to the front portion of the
diffuser, where the velocity is still high, and where disproportionately more pressure
can be recovered. For example, in a straight-walled diffuser about half of the energy is
recovered in the first third of the diffuser. In some centrifugal pumps a radial diffuser
section immediately follows the exit of the impeller. The radial diffuser can consist of
passages formed by airfoil-shaped guide vanes arranged in a circle, or of several
straight, diverging diffusers. In Chapter 11 we offer details on the design of diffusers in
pumps. Diffuser loss accounts for the greatest part of pump losses, and therefore its
design offers the greatest challenge in pump design.

Figure 4.7 Pressure recovery coefficient versus area ratio for a loss-free diffuser and
for a sudden expansion.

Obstructions in Pipes
The sudden expansion flow model can also be used to estimate the pressure drop from
objects obstructing the flow in pipes. Only the net flow cross sections enter into
calculation of the sudden expansion loss; their particular configuration remains
unspecified. Therefore, the net free cross-sectional area around the object can be taken
as the inlet cross section Al. In the case of sharp-cornered objects, such as a flat plate
perpendicular to the flow, allowance must be made for a contraction of the flow
moving past the object and the cross-sectional area must be multiplied by a contraction
coefficient. Depending on the rounding of the corners of the object, the contraction
coefficient may vary from 0.7 to 1.0. As before, a condition for the validity of the
sudden expansion flow model is that the flow be fully mixed and the velocity be
uniform at the downstream cross section.
Such an obstruction loss model is sometimes applied to the specific case of the blade
wake mixing loss behind blade rows. Blades and vanes have some definite thickness
and leave a wake in the flow behind them. It can be assumed that the cross section of the
flow passage between the blades of the blade row suddenly increases by the amount of
the blade thickness as the flow leaves the blade row. Since the blade thickness is
considerably less than the passage width, the area increase and the corresponding loss
are relatively small and may or may not be negligible. A sudden expansion loss
calculation can provide an estimate of the magnitude of such a wake mixing loss.
Although sometimes applied to rotating impellers, such a wake mixing loss estimate
should only be applied to stationary blade rows. The flow leaving the impeller is
unsteady when considered by a stationary observer in an absolute frame of reference.
The absolute flow velocity is not parallel but approximately perpendicular to the wake.
The mixing process is therefore entirely different from a steady wake mixing. The fluid
in the wake is not low-energy fluid as in the case of a wake behind a stationary object,
but can move with a higher velocity (approximately with the tip speed) than the main
flow. In addition, trailing vortices are likely to be present in the corners between the
blade and the hub and shroud surfaces. Very fast mixing has been observed in tests,
resulting in surprisingly low losses. The explanation was offered that energy transfer
takes place by normal, unsteady pressure forces-by pressure exchange (Foa 1958) rather
than by viscous or turbulent mixing.
REFERENCES

Carlson, J. J., Johnston, J. E, Sagi, C. J. (1967): Effects of Wall Shape on Flow

Regimes and Performance in Straight, Two-Dimensional Diffusers, ASME Journal
of Basic Engineering, March, pp. 151-160.
Foa, V. J. (1958): Pressure Exchange, Applied Mechanics Review, December, pp.

655-657.
Fox, R. W, Kline, S. J. (1962): Flow Regimes in Curved Subsonic Diffusers, ASME

Journal of Basic Engineering, September, pp. 303-312.
Hydraulic Institute (1961): Pipe Friction Manual, HI, New York.
Johnston, J. P. (1998): Review: Diffuser Design and Performance Analysis by a

Unified Integral Method, ASME Journal of Fluids Engineering, March, pp. 6-12.
Moody, L. E.(1944): Friction Factors for Pipe Flow, Vol. 66, AMSE, New York, p.
671.
Muggli, F. A., Eisele, K., Casey, M. V., Gulich, J., Schachenmann, A. (1997): Flow
Analysis in a Pump Diffuser, Part 2: Validation and Limitations of CFD for Diffuser
Flows, ASME Journal of Fluids Engineering, December, pp. 978985.
Nicuradse, J. (1933): Stromungsgesetze in rauhen Rohren, VDI Forschungsheft No.

361.
Reneau, L. R., Johnston, J. E, Kline, S. J. (1967): Performance and Design of Straight,

Two-Dimensional Diffusers, ASME Journal of Basic Engineering, March, p. 141.
Schlichting, H. (1960): Boundary Layer Theory, McGraw-Hill, New York.


PUMPING SYSTEMS
PUMP SPECIFICATIONS AND SELECTION

Before pump design or selection can begin, specifications need to be established which
express several requirements. Only requirements pertaining to the hydraulic
performance of the pump are considered here. Other considerations are reliability,
durability, noise and vibration, maintenance, installation, and control. These concerns
affect primarily the mechanical design of the pump: material selection; auxiliary
component specifications, such as bearings, seals, and instrumentation; and design for
ease of assembly, service and maintenance.
Among the specifications affecting hydraulic performance, the most basic are the
nominal efficiency, speed, flow rate, pressure or head, and power. Efficiency is
important from the point of view of energy cost. Because of the maturity of pump design
technology, the nominal efficiency of commercially available, competitive pumps is
within 1 or 2% test data rarely being more accurate than I%. Power losses arising from
piping system design and losses due to control considerations can amount to 30% or
more of the energy supplied by the pump. In process applications pumps are often
followed immediately by control valves, which produce a head loss of 20 or 30% at
nominal conditions. Flow control is achieved by increasing or decreasing the head loss.
Considerable energy savings can be achieved by variable-speed drives, for example.
Consequently, energy conservation efforts should preferably be directed at these losses
rather than at 1 or 2% pump efficiency improvements. The competitive advantage of
high efficiency therefore appears to reside in reflecting high technical competency on
the part of the manufacturer rather than in energy savings.
If the power of the drive motor is limited, nominal efficiency might also be a
consideration. Depending on the application, motors are usually selected assuming a
certain service factor, which multiplies the nominal power requirement of the driven
equipment. A factor greater than l assures a reserve power margin. Actually, maximum
temperature rise limits the ultimate power that an electric motor can deliver. In
instances where the pump and motor are sold as an assembly, as in appliances for
example, pump and motor are closely matched, assuming a service factor of 1. The
motor load can reach the ultimate power, limited only by the permissible temperature
rise. In such a case the pump efficiency determines the maximum pressure or flow rate

that can be obtained from the assembly with a given motor.
Pump speed will depend primarily on the driver. Electric motor speeds follow the
network frequency: 60 Hz in the United States and 50 Hz in Europe, corresponding to
nominal 3600- and 3000 rpm synchronous speeds of single-pole motors. Smaller
motors, up to perhaps 100 hp, usually turn at 3600 rpm. Larger motors turn at an even
fraction of the synchronous speed: 1800, 1200, 900, or 720 rpm, depending on the
number of poles of the motor.
Large pumps have large inlet diameters. The impeller inlet circumferential velocity
increases with the diameter but must stay in reasonable proportion to the inlet flow
velocity to avoid blade angles too much inclined toward the tangential direction. The
danger of cavitation also increases with the velocities at the inlet. These reasons suggest
lower rotational velocities for large pumps. Under full load, motor speed typically
decreases 2% from the synchronous speed. At a price, variable-frequency drives are
available, some of which can reach 5400 rpm, but the synchronous speed limits the
maximum speed of most induction motors. Internal combustion engines drive pumps in
remote, outdoor, often mobile applications, such as irrigation, dewatering, and flood
control. Their rated speed varies but remains relatively low compared with that of
electric motors, on the order from 1000 to 2000 rpm. Their speed is usually governed to
remain constant regardless of the load.
The rated flow rate of pumps can match the required flow directly since leakage
losses are normally insignificant. The nominal flow rate also determines the inlet and
outlet pipe diameter and flange size, which are standardized. The code number of
commercial lines of pumps often includes the inlet and outlet pipe diameters measured
in inches. Pipe and flange diameters are sized to result in water flow velocities of 5 to
15 ft/sec (1.5 to 5 m/s), but velocities can reach 30 ft/sec (10 m/s), in large high-
specific-speed pumps.
Specifying the pressure or head rise produced by the pump requires the most care. It
is defined as the total head, static and dynamic, at the pump exit flange minus the total
head at the inlet flange. The nominal head required by the system must be estimated by
keeping in mind the maximum possible head that might ever be encountered. A reserve
margin may be necessary. However, an excessive margin may make the pump operate
normally at low efficiency or at some adverse condition. Most pumps are capable of
operating over a range of higher or lower flow rates and pressures on either side of the
nominal. Such off-design operation, at exceptional conditions, should be taken into
account when estimating the nominal head, in order to minimize the need for a large
reserve margin.
The head rise required from the pump is estimated by adding up the suction head at
the inlet, including pipe friction losses, valve and pipe friction losses at the exit, and the
net head of the final load, such as the elevation head to an overhead tank (Fig. 5.1). If
the pump inlet is below the water level of a sump or supply tank, the inlet head will be
positive, from which the inlet piping losses, calculated at the rated flow rate, need to be
subtracted. If the pump inlet is above the water supply level, the inlet head is negative
and so are the additional inlet piping losses. The valve and pipe losses at the exit, also
calculated at the rated flow rate, as well as the net head of the load, add to the head
requirement of the pump.
Particular attention must be given to the pump exit velocity head. The rated pump
head rise usually includes the velocity head of the flow leaving the pump. In
performance testing the pressure rise is measured "total to total," implying that the
velocity heads at the inlet and exit flanges are added to the respective static head
measurements to arrive at the total head at each location. This approach assumes that the
kinetic energy of the flow leaving the pump remains useful, which may or may not be the
case. If the pump produces a free jet, like a fire hose, then the velocity head can indeed
be counted as useful energy. If the pump discharges into a stagnant tank, for example, the
kinetic energy is lost and must be added to the piping losses at the pump exit.

Figure 5.1 Pump system installation.
In addition to specifying the rated efficiency, speed, flow rate, and head, several
other requirements need to be established. First, the pump must have sufficient suction
capability without cavitating. Verifying satisfactory suction performance requires two
steps: the suction head the net positive suction head available (NPSHA) from the
system-must be estimated, and a comparison must be made with the net positive suction
head required (NPSHR) by the pump. Pump catalogs list the NPSHR of the pump as a
function of flow rate. The NPSHR values reach a minimum at the design flow rate and
gradually increase at higher flow rates. The absence of cavitation must be assured at the
greatest anticipated flow above the rated flow rate. Catalog data are usually obtained at
conditions conforming to testing standards, typically when the pump head declines by
3 % with decreasing inlet pressure. However, cavitation vapor bubbles may appear
earlier and cause damage to the impeller. Conditions in the actual application may differ
from the standard testing conditions. Excessive aeration of the water ahead of the pump,
supply sump vortexing, or air leaks into the suction piping may cause pump cavitation at
inlet pressures that would appear sufficiently high according to catalog data. Therefore,
the NPSH required should be estimated conservatively, and a higher inlet pressure than
that absolutely necessary should be provided to the pump.
Often, further requirements are imposed, such as the minimum flow rate at which the
pump may have to function, the head at shutoff, and the maximum power demand.
Several undesirable phenomena can appear at reduced flow rate: general instability of
operation, inlet recirculation, exit recirculation, cavitation, and overheating. Details of
such conditions are discussed in subsequent chapters. Some of these can be tolerated in
case of an emergency but would lead to deterioration and failure of the pump if they
were to persist for longer periods. The application will decide how often and how long
the pump may operate at reduced flow and what the cost trade-off might be between a
more expensive pump and damage from an emergency of low probability.
Manufacturers offer guidelines for the safe, minimum flow of their pumps (Cooper
1988).
Instability at reduced flow results from a positive slope of the head-flow curve of the
pump. It depends not only on the pump characteristics but also on the system load curve,
as described further below. It causes severe torque and power fluctuations as well as
vibration and noise, which can damage the bearings or shorten their life. Inlet and exit
recirculation may or may not create problems. It certainly disrupts the normal flow
pattern in the pump, but can often be tolerated for a limited time. The flow rate
corresponding to the onset of inlet and exit recirculation can be calculated with the
procedures and computer programs given in this book. Recirculation can also combine
with cavitation, which might aggravate vibration and power fluctuations, and will result
in progressive cavitation damage. Recirculation is more likely to precipitate cavitation
if the calculated NPSHA does not comprise a sufficient safety margin.
Overheating results when excessive losses, transformed to heat, are retained in the
pump. Pumps handling cold water rarely encounter problems. However, a temperature
rise will affect the vapor pressure of the fluid. When pumping hot fluids near boiling,
cavitation may appear unexpectedly. The head at shutoff is difficult to estimate because
the flow churns completely uncontrolled in the pump casing. It also depends on the
configuration of the piping upstream from the pump. An approximate estimate can be
obtained by extrapolating the performance curve to zero flow. The pressure at shutoff is
specified, for example, in the case of fire pumps. Such a specification may intend to
make certain that the head-flow curve does not have a range of positive slope, which
would lead to instability.
If the power demand of the pump decreases with increasing flow rate, a reserve
power margin will be needed if the pump is likely to operate below the flow rate at the
rated point. The opposite holds for pumps whose power demand generally increases
with the flow rate. Pump characteristic plots in catalogs show the power demand of the
pump as a function of the flow rate. If the plot does not show the power directly, it can
be calculated from the head, flow rate, and efficiency.

SYSTEM MODELING
The desired performance of a pump in a pumping system defines its design. The need
for a pump arises from an overall system design, the purpose of which is to convey
some fluid from one location to another. Proper functioning of the pump and the overall
system must be ascertained before the hardware is procured and the system is built.
Thought must be given not only to steady-state performance, but also to time-dependent
variation and sudden changes in speed, torque, flow rate, and pressure, as for example
during startup, shutdown, or sudden valve closure. The pumping circuit might be part of
a process, which will need to be controlled automatically. Since the pump feeds energy
into the system, it can also power undesirable instabilities, which must be avoided. A
mathematical modeling of the pumping system may be desirable to test its proper
functioning.
System modeling and analysis can also be used as a diagnostic tool. Electric power
variations or disturbances can reveal mechanical defects in deep-well pumps, which
are not accessible to direct scrutiny (Kenull et al. 1997). Highspeed computers and
commercially available computer programs have considerably facilitated the steady and
unsteady mathematical analysis of pumping systems. However, the modeling of
particular systems still demands skill. Excessively simplified models provide only
trivial answers. Excessively complex models become unwieldy, requiring much labor
and cost.
Pumping system components transform and transmit energy which cannot be lost and
must be accounted for. Typically, electric energy, taken from the grid, is transformed to
shaft work by an electric motor; the pump follows, which transforms the shaft work into
hydraulic energy; the valving and piping system transmits the hydraulic energy while
also producing some losses; finally, the energy is stored or absorbed by the load (Fig.
5.2). Since it is the same energy that appears in different forms in these components, the
variables, which characterize them, show certain similarities and suggest a unified
mathematical approach (Blackburn et al. 1960).
The discipline of control theory has developed practical modeling methods to

characterize those systems that were to be controlled (Blackburn et al. 1960). A
similarity exists between electrical, mechanical, hydraulic, and even thermal systems.
Before the widespread use of high-speed digital computers, system analysis was often
performed on analog computers, where electric models of mechanical, hydraulic, or
thermal systems were assembled using the electric equivalents of such components. A
numerical analysis is now possible with modern digital computers.
Each component can be modeled by a black box with two input ports, corresponding
to two input variables, and two output ports, corresponding to the two output variables.
Such four-terminal element models originated in the analysis of electronic systems.
Here, applied by analogy to pumping systems, the dimension of the input and output
variable pair must be such that their product becomes a variable having the dimension
of power. For example, in the case of the electric motor, voltage and current are the
input variables and shaft speed and torque are the output. In pumps, shaft speed and
torque are the input and flow rate and pressure are the output. In the case of a hydraulic
brake shaft, speed and torque are the input, and mass flow rate and the temperature of
the cooling water, or the heat capacity per unit mass of fluid, are the output. Since no
perfect components exist, most have a sidestream of lost energy, usually in the form of
heat. The output power remains less than the input power, their ratio being the efficiency
of the component.
The input variables, xl and yi, and output variables, x2 and y2, are related, by two
equations, for example by f2 and g2:
Since there are four variables and two equations, two variables must be given to
determine the two others. The two equations are derived from governing physical
principles. One of the equations might state that the input power must be equal to the
output power and the losses. This equality expresses the conservation of energy. Other
equations are usually obtained from force balance equations in the case of mechanical
systems.
Figure 5.2 Pump system model.
If the equations are linear, the output variables can be expressed explicitly in terms
of the input variables alone:

In such a linear form the equations are easily manipulated. Any two of the variables
can be calculated in terms of any two others. The four coefficients, some of which
might be zero, fully characterize the particular system element. For example, many
electrical components, consisting of inductances, resistances, and capacitances, and
many mechanical components, consisting of masses, viscous friction elements, and
springs, yield linear equations.
The torque-speed characteristics of electric motors, which correspond to one of the

needed equations, can differ greatly (Fig. 5.3). In general, the torque of commercial
induction motors remains approximately equal to the nominal torque up to a speed of
about 70% of the nominal speed. It rises rapidly to about 200% of nominal torque at
80% speed and drops to zero at the speed, which corresponds exactly to the network
frequency. If the motor starts a loaded pump, instability problems might arise around
70% speed. Large pumps need special starting procedures (Sulzer 1989). Once the
motor has come up to speed, the motor acts very stiff, large torque or power variations
result in little speed change. Often, the speed can be assumed constant and independent
of the torque for modeling purposes.

Figure 5.3 Electric motor torque-speed diagram.
In the case of pumps, the input variables are the shaft speed N and torque T, and the
output variables are the head H and the flow rate Q. The first equation corresponds to
the head-flow curve for a given shaft speed. The flow rate and speed being given, the
head-flow curve determines the output head. The other output variable, the flow rate,
can be calculated from the input power multiplied by the efficiency and divided by the
head. Theoretically, if any two of the variables are given, the other two can be
calculated. As will be seen, the two given variables will be defined by the components
preceding and following the component under consideration.
In terms of the actual variables-speed N (rpm), torque T (ft-lb/sec), head H (ft), flow
rate Q (ft3/sec), efficiency rl, constant geometrical quantities diameter D2, exit width
B2, slip coefficient of the pump a, exit blade angle Q2, and gravitational acceleration g
= 32.2 ft/sec 2 (9.8 m/s2)-the two equations defining the steadystate output of a pump
become
The head-flow equation above corresponds to the form used in the performance
calculation method of this book, which is given in Chapter 10, where all variables are
defined.
Ultimately, this equation derives from application of the principle of conservation of

angular momentum to pump impellers. The following substitutions lead to the equation
above: U = wD2/2, w = 27rN/60, and W, = Q/7rD2B.
The second equation corresponds to the principle of conservation of energy:
These equations are nonlinear and contain implicitly the head H and flow rate Q, as
well as the efficiency q, which depends on the flow rate. Unlike many electrical
components, the practical relationships, characterizing hydraulic components, are
mostly nonlinear, and therefore the variables cannot be calculated explicitly. They
would have to be solved by trial and error, by progressive approximations. However,
powerful computer programs are available today that can directly solve even
nonlinear, implicit equation systems.
A further complication arises when the equations are time dependent: for example,
when the time derivatives or integrals of the variables appear in the characteristic
equations. Such cases occur when inertia or compressibility forces appear (Blackburn
et al. 1960, Vance 1988).
To illustrate calculations, the case of a rotating coupling with inertia I and a torsional
windup spring rate k is considered here (Vance 1988) (Fig. 5.4). The input T1 and
output T2 torques are equal except for an inertia torque, which is proportional to the
time rate of change of the input speed dN1 /dt. The difference between the input speed
Ni and the output speed N2 is equal to the shaft windup, which is proportional to the
time rate of change of the input torque dTi /dt:
dN Tz=Tl - Id 1 dTi + Nz=-k dt Ni
Since the relationships happen to be linear, one can identify the coefficients a using the
symbolic derivation d/dt as
d au=1 a12=-Idt ld azl=kdt a22=1
The inertia of practical couplings used between skid-mounted electric motors and
pumps can become significant, depending on the rate of change of the shaft speed.
Catalogs of commercial couplings often list their weight and inertia. Couplings are
made somewhat compliant to absorb torsional shaft vibrations and misalignment.
However, their total torsional windup, from no load to rated torque capacity, which can
be used to calculate the spring rate, amounts to only a few degrees, and the couplings
can usually be considered rigid.
Figure 5.4 Model of shaft with inertia and torsional spring.
In the case of pumps the inertia of the impeller must be taken into account and must be
added to the torque load when shaft speed changes are expected. The mass of the fluid
contained in the impeller is combined with the mass of the metal to calculate the
rotational inertia. The polar inertia I of a disk of diameter D and thickness L or weight
W has the dimension of lb/ft/sec 2 in English units:

The impeller is fixed rigidly to the shaft, and the compressibility of the liquid
contained in the pump is negligible. Therefore, no shaft windup or compliance needs to
be introduced. The pump delivers to a load, which will typically consist of an exit
pipe, a valve or orifice and a tank at a higher elevation. A long pipe will represent a
steady friction load and an inertia load of the fluid contained in it. A valve or orifice,
generates an energy loss, and an elevated tank will store energy.
Transient, unsteady operation of the pump, such as startup, shutdown, and loss of
power, can drive the pump into unusual operating ranges, such as reversed flow or
reversed rotation. A full, time-dependent analysis of such upset conditions often
requires that the pump characteristics be known in these unusual ranges of operation.
Since the speed may change continually, extensive use is made of the similarity
relationships by normalizing the flow rate, head, and torque with the rotational speed.
Frequent use is also made of a four-quadrant representation of the performance in the
four fields of positive and negative flow and speed. Other representations have been
proposed which are better suited to the computer calculation of pump transients (Martin
1983).
The need to model and calculate the response and transient, dynamic behavior of long
pipelines has generated specialized analytical methods (Wylie and Steeter 1993).
Detailed discussion of the analysis methods would exceed the scope of this book and
cannot be covered here. Pressure wave propagation and water hammer calculations
have become prerequisites of pipeline and power plant construction. The complexity of
most piping systems usually requires that a computer be used for its analysis. If
relatively short piping connections are used, simplified models suffice for a first,
approximate check. Steady pipe friction losses vary with the square of the flow rate.
The inertia load corresponds to the mass of the fluid in the pipe and varies with the time
rate of change of the flow rate. The compressibility of the fluid can usually be
neglected. When the pipe is very long, the pipe walls stretch easily, or when air pockets
exist somewhere, the volume capacity of the piping system increases with increasing
pressure, which has the same effect as compressibility of the fluid. Under such
conditions, which are difficult to quantify, resonant pressure oscillations can be excited
by the pump. The most frequently encountered frequency of excitation is the blade
passing frequency of the pump-the number of shaft revolutions per second times the
number of impeller blades.
The piping and load downstream of the pump are characterized by expressions of the
form

Q2 = QI
In the expression of the output head, Hz, which is the head remaining at the pipe exit,
the static elevation of the tank h, the pipe friction and valve or orifice loss, and the
inertia is deducted from the input head, Hi, which in this instance designates the head at
the pump exit. This equation corresponds to the conservation of energy. The second
equation assumes that there are no leaks and that the input flow is equal to the output
flow, which corresponds to the continuity equation. If the pump delivers into a tank
open to ambient pressure, the output head H2 is a known constant and the expression
relates the input head H, and flow rate Qi, which correspond to the pump output head
and flow rate.
For the sake of illustration we consider steady operation of a pump, which is driven
at constant speed N and delivers the flow Q into an elevated tank at constant head h
throttled by a valve (Fig. 5.5). To calculate the four variables defining the operation of
the pump-N, T, H, and Q-four equations are now available. The shaft speed N is known
to be constant. The head H and flow rate Q can be determined from the head-flow
characteristics of the pump and from the characteristics of the load. Finally, the input
torque T is obtained from the expression equating the input and output power and losses.
A graphic solution can best illustrate the calculation of the head and flow rate of the
operating point. The characteristic curve of the pump on a head versus flow rate plot is
typically convex in the direction of increasing head, as shown in Fig. 5.5. The slope of
the head-flow curve is usually negative. The head decreases with increasing flow rate.
The load curve corresponds to a parabola with its origin on the vertical axis at a value
of h. The slope of the load curve is usually positive. The intersection of the two curves
gives the operating point H and Q of the pump.

Figure 5.5 Head versus flow rate plot showing pump and load characteristics.
Let us now assume that the head-flow curve of the pump has a maximum. The head
decreases in both directions, with increasing and decreasing flow rate, from its value at
the maximum head. Let us further assume that the pump gradually fills a tank in which
the water-level elevation h rises slowly: h = H, and that the head loss in the piping is
negligible. The load curve, which now consists of a horizontal line, intersects the head-
flow curve at two points on either side of the maximum, as shown in Fig. 5.6. Both are
possible operating points, but as it turns out, only the one at high flow rate remains
stable.
The stability of an operating point can be determined by slightly disturbing the

location of it and observing whether the pump tends to come back to its steady operating
point. If the operating point at the lower flow rate were to be displaced to slightly larger
flow rate, the head produced by the pump would also rise slightly. The excess head
would overcome inertia and accelerate the flow rate to even larger values. The
operating point would move away and would not come back to the original operating
point. Therefore, this operating point will be unstable. On the other hand, if the
operating point at the larger flow rate were moved slightly higher, the pressure
produced by the pump would fall and the flow would tend to slow down and come back
to the original operating point. Therefore, the operating point at the lower flow rate is
unstable, and the operating point at the higher flow rate is stable. It will be observed
that a stable operating point exists if the slope of the load curve is greater than the slope
of the headflow curve.
Such a stability analysis can also be performed on a computer. Dynamic system

models are disturbed by imposing a small perturbation, usually a small sinusoidal

oscillation, added to one or the other of the variables, and by calculating whether the
amplitude of the oscillation would grow or die out. Although this analysis method may
appear artificial, it can be reasoned that some unwanted disturbances will always be
present in real life. A system in an unstable equilibrium position will leave it and never
return. Theoretically, an egg can be stood on end, but in practice it will never remain
standing. A system will eventually return to a stable equilibrium position regardless of
the manner in which it is disturbed, and regardless of the path it takes to return.
Figure 5.6 Head versus flow rate plot showing pump characteristics with unstable
range.
As the elevated tank continues being filled, eventually the head will reach the value
corresponding to the maximum of the head-flow curve. The pump cannot produce higher
pressure and the flow will stop. The equivalent operating point will jump to the vertical
axis. It may even surge back and forth between no flow and the maximum operating
point if the tank slowly drains. The pump will never be able to operate at flow rates
below that which corresponds to the maximum head. Because of such unstable surging,
pumps having a head-flow curve with a positive slope and a maximum should be
avoided for general use. In Chapter 14 we discuss the effect of impeller exit
recirculation on the presence of a maximum on the head-flow curve and on the stability
of the pump at a reduced flow rate.

REFERENCES
Blackburn, J. E, Reethof, G., Shearer, J. L. (1960): Fluid Power Control, MIT Press,
Cambridge, Mass., pp. 130-143, 433-497.
Cooper, P. (1988): Panel Session on Specifying Minimum Flow, Proceedings of the

Texas A&M 5th International Pump User Symposium, Houston, Texas, pp. 177-192.
Imaichi, K., Tsujimoto, Y., Yoshida, Y. (1982): An Analysis of Unsteady Torque on a

Two-Dimensional Radial Impeller, ASME Journal of Fluid Mechanics, June, pp.
228-234.
Kenull, T., Kosyna, G., Thamsen, P. U. (1997): Diagnostics of Submersible Motor

Pumps by Non-stationary Signals in Motor Current, ASME Fluids Engineering
Division Summer Meeting, June, FEDSM97, pp. 1-11.
Martin, C. S. (1983): Representation of Pump Characteristics for Transient Analysis,

ASME Symposium on Performance Characteristics of Hydraulic Turbines and
Pumps, Boston, FED Vol. 6.
Ran, L., Yacamini, R., Smith, K. S. (1996): Torsional Vibrations in Electrical

Induction Motor Drives During Start-up, ASME Journal of Vibration and Acoustics,
April, pp. 242-251.
Shadley, J. R., Wilson, B. L., Dorney, M. S. (1992): Unstable Self-Excitation of

Torsional Vibrations in AC Induction Motor Driven Rotational Systems, ASME
Journal of Vibration and Acoustics, April, pp. 226-231.
Sulzer Brothers Ltd. (1989): Sulzer Centrifugal Pump Handbook, Elsevier Applied
Science, New York, pp. 59-64, 208-228.
Tsukamoto, H., Yoneda, H., Sagara, K. (1993): The Response of a Centrifugal Pump
to Fluctuating Rotational Speed, in Pumping Machinery, ASME FED Vol. 154, pp.
293-300.
Vance, J. M. (1988): Rotordynamics of Turbomachinery, Wiley, New York.
Wylie, E. B., Streeter, V. L. (1993): Fluid Transients in Systems, Prentice Hall, Upper
Saddle River, N.J.



GENERAL PUMP THEORY
PUMP PERFORMANCE
Pumps use shaft power to increase the energy, pressure, or head of the fluid. The flow
rate Q, head H, and efficiency q define the overall performance of a pump. The pump
head H corresponds to the increase in total pressure, from inlet flange to exit flange,
divided by the specific weight of the fluid, pg:
In the English system of units (lbf, ft, sec) assuming water (pg = 62.4 lbf/ft3):
The net hydraulic power P delivered by the pump can be calculated from
Shaft power is calculated by dividing the net hydraulic power P above by the
efficiency j.
In the SI system of units (N, m, s, and Pa) assuming water (pg = 9806 N/m3):

since 1 N m/s = IJ/s = I W
The pump head varies as the flow rate changes. Commercial catalogs show pump
performance by means of a plot of the head as a function of the flow rate for a certain
shaft speed. The efficiency-the ratio of the output and input power, also a function of the
flow rate-also appears on this plot (Fig. 6.1).
Catalog plots often show the head-flow curves corresponding to the same pump with
impellers having different diameters. Since the pump head strongly depends on the
impeller diameter, machining or trimming the impeller to a smaller diameter certainly
offers a simple, fast, expedient way to match the pump characteristics to the system
requirement. However, sound design principles show that the flow pattern in such
truncated impellers will certainly not be ideal. Chances are that the blade tips will be
heavily loaded, and relatively high pressure and flow fluctuations would be expected at
the impeller exit. Off-design operation may also be affected, especially in mixed flow
pumps, depending on how the impeller exit is modified, since exit recirculation may
appear and the head-flow curve may change at reduced flow rates. These effects could
be foreseen if a more detailed study of the impeller flow were made.
The nominal rating of a pump corresponds to the head and flow rate at the best
efficiency point. These are the conditions for which the pump was designed. It is
expected that the flow through the pump would be most disturbance-free at these
conditions. Depending on the particular application, pump users desire pumps with
various head, flow rate, and speed combinations and expect these pumps to operate as
efficiently as possible, which calls for commercial lines of different pump sizes,
configurations, and pump types.

Figure 6.1 Typical pump head-flow curve at constant speed.

OFF-DESIGN OPERATION
Although selected or designed for a particular flow rate, in practice, pumps operate
across a certain range of flow rates. At off-design flow rates the flow pattern deviates
from the ideal. The flow separates at the impeller blade tips, in the impeller, and in the
diffuser or volute. Recirculation may appear at the inlet, flow returning from the
impeller eye into the inlet pipe. Recirculation may also occur at the impeller exit, fluid
flowing back into the impeller from the diffuser or volute. Not only does the efficiency
deteriorate at off-design conditions, but cavitation can appear, which, as will be seen,
can destroy the impeller. Excessive vibration appears, which reduces bearing life and
reliability, and even the entire pumping system can enter into resonant oscillations-into
surging. Some of these phenomena are considered in detail below. Catastrophic
failures are unusual, but off-design point operation for a longer period of time seriously
reduces pump life.
In selecting a pump, much care must be given in estimating the margin of excess
capacity that might be required by the system, supplied by the pump. A much too large
margin normally causes the pump to operate far off design, at a bad efficiency value and
at conditions that reduce its reliability and life. It becomes an economic consideration
whether low efficiency at nominal operating conditions and corresponding operating
cost will balance avoiding loss from lack of excess capacity at unusual occasions. Other
approaches, such as the standby capacity from a parallel-connected auxiliary pump, for
example, might offer a more cost-effective installation. Manufacturers usually provide
guidelines for the safe operating range of their pumps (Cooper 1988, Gopalakrishnan
1988).
PUMP TYPES AND SPECIFIC SPEED

Similarity relations can be very useful to scale pumps in size or speed. The
performance of very large pumps can be ascertained from testing their geometrically
similar, small-scale model in the laboratory. If the test facility cannot provide
sufficient suction head or NPSHA, cavitation during testing can be avoided by using
low speed and then correcting the data for higher speed. Performance expressed in
terms of dimensionless variables helps in selecting optimal configurations (Balje 1962,
Cartwright 1977, Tuzson 1993).
The flow rate Q increases in proportion to the shaft speed N, and the head H
increases as the square of the speed. A pump rotating twice as fast will deliver twice as

much flow and four times as much head. The following relationships can be written in
terms of the dimensions of length L and time T, provided that the gravitational
acceleration g is also introduced.
The dimension of the first ratio, representing the flow rate, is L3; that of the second
ratio, representing the head, is L2. A dimensionless number, expressing the magnitude
of the flow rate as compared with the head, is obtained if the first group raised to the
second power is divided by the second group raised to the third power:
The symbol n, stands for the dimensionless specific speed which characterizes a pump
configuration that is best suited to deliver a certain head H at a certain flow rate Q
when operated at a certain speed N. The expression remains dimensionless if
consistent units are used, such as o)(1/sec) = N(rpm)(27c/60 see), Q(ft3/sec), g(32.2
ft/sec2), and H(ft), for example. The same dimensionless number will be obtained with
metric units w(l/sec), Q(m3/s), g(9.81 m/s2), and H(m).
In U.S. industrial practice a form of the specific speed is used which is not
dimensionless but serves the same purpose:
It will be observed that low-specific-speed pumps will have high head and low flow,
and high-specific-speed pumps will have low head and high flow rate. Since certain
flow losses in pumps depend on the flow rate and others on the head, pump
configurations are designed to minimize one type of loss rather than the other,
depending on the specific speed. The impellers of low-specific-speed pumps have a
large diameter and narrow flow passages, and those of highspecific-speed pumps have
large flow passages and relatively small diameter. To maximize the available flow
cross section, flow passages in high-specific-speed impellers are inclined from the
radial toward the axial direction. These are called mixed flow pumps, in distinction to
radial flow pumps.
Past pump designers established by experience typical pump efficiencies as a

function of specific speed and flow rate. The plot showing this relationship is shown in
Chapter 11, where as a first approximation, an estimated efficiency is required. Pumps
of a specific speed of approximately N, = 2300 reach the best efficiency. The
corresponding dimensionless specific speed is ns = 0.84. Efficiencies estimated on the
basis of industrywide surveys offer a target for pump designers (Sulzer 1989, Turton
1994). For example, the Hydraulic Institute (1994) offers a publication, Efficiency
Prediction Method for Centrifugal Pumps. Pump technology is old, and excellent
designs have evolved over close to 100 years. Efficiencies lower than these targets
indicate poor designs, which must have evident shortcomings. Efficiencies higher than
these targets testify to outstanding designs, which cannot easily be achieved but require
extensive and lengthy analytical and experimental development.
Geometrical scaling of pumps is a powerful method of development. Often, small-

scale testing verifies the performance of very large pumps or turbines which cannot be
tested in the laboratory. However, small-scale testing raises the question of proper
scaling of the efficiency. Scaling affects primarily wall friction, disk friction, and
leakage losses, which depend on the relevant Reynolds number. Proper scaling of the
efficiency should adjust the individual losses by taking into account the particular
Reynolds number that applies. In the case of wall friction, the hydraulic diameter and
relative flow velocity in the impeller enter into the calculation of the Reynolds number;
in the case of the disk friction, the impeller diameter and tip speed; in the case of
clearances, the clearance width and flow velocity. The corresponding friction or loss
coefficients also depend on the surface roughness. The partly empirical efficiency
scaling formulas of 15 pump experts are reviewed in an article by Osterwalder and
Hippe (1982). A properly adjusted and calibrated performance calculation program can
help to execute such scaling calculations. It should be kept in mind that efficiency also
depends on factors other than the hydraulic design, such as manufacturing methods,
mechanical losses, and leakage.

PUMP CONFIGURATIONS
Whereas the design of pump hydraulics follows the guidelines of specific speed, the
general arrangement and the configuration of the housing conforms to the needs of the
application. In the most common, single-stage, end suction pump configuration (Fig.
6.2), the flow enters horizontally, through the inlet flange, into the impeller eye. The
impeller rotates around a horizontal axis and is either mounted directly on the shaft of
the electric drive motor, or the pump and motor, separately skid mounted, are
connected with a coupling. Closed or open impellers are distinguished depending on
whether the impeller has a shroud or the blades are only fixed to the rear hub. In pumps
with open impellers a close clearance is kept between the front of the rotating blades
and the stationary housing, and some mechanism exists to adjust this clearance from
time to time, since the blade tips may wear. A spiral-shaped volute or collector
surrounds the impeller. In process pumps the fluid leaves through an exit flange, with
its axis vertical, passing through the centerline, the axis of rotation, of the impeller.
These pumps usually carry a designation code consisting of the standard inlet and exit
flange sizes. General water pumps often use a long diffuser, aligned off center from the
impeller, and tangent to the volute, which allows better diffuser configuration.
Double suction pumps are usually larger and are used primarily in water service
(Fig. 6.3). Two back-to-back fused impellers are mounted on a horizontal shaft,
supported by bearings on either side. The flow enters through the inlet flange,
perpendicular to the direction of the shaft, splits in two, and is ducted to the impeller
inlets on either side. A central scroll serves both impellers and leads, through a single
diffuser, to an exit flange. Such an arrangement often results in better efficiency because
it reduces friction on the back side of the impellers, the disk friction loss, and because
by splitting the flow in two, the specific speed of each impeller sometimes becomes
more favorable.

Figure 6.2 Horizontal end suction back pullout (ANSI) pump. (Courtesy of Goulds
Pumps, ITT Industries.)

Figure 6.3 Horizontal split-case double suction pump. (Courtesy of Goulds Pumps, ITT
Industries.)
The third principal arrangement (vertical pumps or column pumps shown in Fig. 6.4)
consists of one or more impellers mounted on a vertical shaft. A vaned diffuser and
return passage, leading to the next stage, follows the impeller. The return passage forms
part of the housing-the bowl. The fluid enters and leaves the pump in the axial direction.
The pump assembly is lowered into a pit. The electric motor, above ground or
submerged in the well, drives the pump shaft through a coupling. This arrangement suits
water or oil well applications particularly well, because they often require several
stages. The well bore diameter limits the impeller size and the pressure rise per stage.
The desired pressure is obtained by stacking an appropriate number of standard,
massproduced stages.
A variety of pumps are designed specifically for particular applications, the

requirements of which put more emphasis on unusual aspects. Sewage pumps need to
pass oversize objects and therefore require exceptionally large flow passages. One-
bladed sewage pumps are standard (Stark 1991). Large, slow running slurry pumps,
made of hard metals or rubber, minimize erosion. Multistage boiler feed pumps (Fig.
6.5), nuclear plant coolant pumps, oil well water injection pumps, and certain large
process pumps-sometimes driven at high speeds-deliver very high pressures and need
high durability and reliability as well as high efficiency. One-of-a-kind large
hydroelectric plant pumpsturbines-which one can walk around in are in a class by
themselves and need custom design. All these designs derive from the same principles,
but require special, advanced development, beyond the scope of this book.

Figure 6.4 Vertical pump. (Courtesy of Goulds Pumps, ITT Industries.)



REFERENCES
Balje, O. E. (1962): A Study of Design Criteria and Matching of Turbomachines, Part
B: Compressor and Pump Performance and Matching of Turbocompo- nents, ASME
Journal of Engineering for Power, January, pp. 103-114.
Cartwright, W. G. (1977): Specific Speed as a Measure of Design Point Efficiency

and Optimum Geometry for a Class of Compressible Flow Turbomachines, Scaling
for Performance Prediction in Rotodynamic Machines, Institution of Mechanical
Engineers, New York, pp. 139-145.

Texas A&M 5th International Pump User Conference, Houston, Texas, pp. 41-47
and 178-192.
Gopalakrishnan, S. (1988): A New Method for Computing Minimum Flow,

Proceedings of the Texas A&M 5th International Pump User Symposium, Houston,
Texas, pp. 41-47.
Hydraulic Institute (1994): Efficiency Prediction Method for Centrifugal Pumps, HI,
Parsippany, N.J.
Osterwalder, J., Hippe, L. (1982): Studies on Efficiency Scaling Process of Series

Pumps, IAHR Journal of Hydraulic Research, Vol. 20, No. 2, pp. 175-199.
Stark, M. (1991): Auslegungskriterien fair radiale Abwasserpumpenlaufrader mit

einer Schaufel and unterschiedlichem Energieverlauf, VDI Forschungsheft, Vol. 57,
No. 664, pp. 1-56.
Sulzer Brothers Ltd. (1989): Sulzer Centrifugal Pump Handbook, Elsevier Applied
Science, New York, p. 19.
Turton, R. K. (1994): Rotodynamic Pump Design, Cambridge University Press, New

York, pp. 10-14.
Tuzson, J. (1993): Evaluation of Novel Fluid Machinery Concepts, ASME Pumping

Machinery Symposium, Fluid Engineering Division Summer Meeting, Washington,
D.C., FED Vol. 154, pp. 383-386.



PUMP FLUID MECHANICS
IMPELLER WORK INPUT

The rotating impeller imparts energy to the fluid. It is the most important, the only
rotating element of the pump. The diffuser, following the impeller, can transform
kinetic energy into pressure energy but cannot increase the total energy of the fluid. The
impeller contains radial flow passages, inclined toward the axial direction in mixed
flow pumps, formed by rotating blades arranged in a circle. A disk, the shroud, covers
the blades on the front. Another disk, the hub, in the back, connects the impeller
assembly to the shaft. The flow enters axially near the center of rotation and turns in the
radial direction inside the impeller (Fig. 7.1).
Flow conditions, velocities and pressures in the impeller, are described conveniently
in terms of cylindrical coordinates: r, 0, and z, in the radial, circumferential, and axial
directions. The corresponding three absolute velocity components of the fluid are
designated Cr, C, and C. The three velocities relative to the impeller are W,., W,, and
W. The impeller rotates with an angular velocity of w, given by w(1 /sec) = N(rpm)
(2it/60).
In mixed flow pumps and in the inlet of radial flow pumps, the streamlines are
inclined from the axial direction toward the radial direction. The velocity along
projections of streamlines in the r-z plane is called the meridional velocity C,,,, which
depends on the axial variable z as well as on the radial r. Note that the meridional
component of the absolute and the relative velocity are the same, Cm=Wm
The shape of the blades and the resulting flow pattern in the impeller determine how
much energy is transferred by a given size impeller and how efficiently it operates (Fig.
7.2). The theoretical energy increase, the theoretical head rise Hth through the impeller,
can be found by applying the principle of conservation of angular momentum prC,.
Considering a control volume of annular shape with an inner radius r, and outer radius
r2, the torque Tapplied to the fluid by the impeller must be equal to the difference
between the angular momentum entering, pC,,r,, and leaving, pCt2r2:

Figure 7.1 Frontal and cross-sectional view of pump impeller.

Figure 7.2 Relative and absolute flow velocity vectors in a rotating impeller.
Applied to pumps, this expression is known as Euler's equation. Pumps are usually
designed for no angular momentum at the inlet, Ct, r = 0. Consequently, the equation
can be simplified:
where U or U2 = core designates the impeller tip speed, the rotational velocity at the
impeller periphery.

FLOW AT IMPELLER INLET AND EXIT

To find the values of the exit angular momentum, the velocities must be examined at the
inlet of the impeller, the leading edge of the blades, and at the impeller exit, the trailing
edge of the blades. The flow follows certain streamlines inside the rotating impeller,
approximately parallel to the blade surfaces. The symbol W designates the
corresponding relative velocity, W, at the inlet, and W2 at the outlet. The velocity
relative to the impeller can be calculated by adding geometrically the absolute velocity
C, measured in the stationary, nonrotating frame of reference, and the rotational
velocity of the impeller, U = wr. Since the velocities are vector quantities, their
components in the radial and circumferential direction-Wr, Cr and W, Ct, U-must be
added separately. Note that the relative radial velocity equals the absolute radial
velocity since rotation affects only the tangential components, W,. = Cr.
The relative velocity at the inlet W, is obtained from the vectorial sum of the absolute
velocity approaching the pump inlet C, and the rotational velocity U, = cor,, as shown in
Fig. 7.3. Similarly, at the impeller exit the relative velocity W2 is obtained by the
vectorial sum of the absolute velocity C2 and the tip speed, U = U2 = core. The
components CO and Cr2 in the circumferential and radial directions make up the
resultant absolute velocity C2 since
Figure 7.3 Velocity triangles at the impeller inlet (a) and exit (b).

From the velocity triangle shown in Fig. 7.3 it is apparent that the circumferential
component of the absolute velocity can be expressed as a function of the radial velocity
component C,.2 and the local flow angle fF2, which is measured here from the radial
direction in the sense opposite to the direction of rotation:
Consequently, the theoretical head, the energy added to a unit mass of fluid by the
pump, can be written
Note that the negative sign appears since the relative velocity W-2 points in the
direction opposite to the direction of rotation. The blades curve backwards.
As will be seen, this expression is very important for calculating pump performance
and for designing new pumps. It assumes that the flow enters the impeller without any
swirl, without any circumferential velocity component, C11, which can usually be
assumed in the case of pumps. Strictly speaking, the relationship is valid only for fluid
flow on one particular streamline. The velocities and flow angle might vary from one
streamline to the next. If the velocity and flow angle are taken as averages on all
streamlines at the pump exit, the theoretical head will also be the average head for all
streamlines.

SLIP
Unfortunately, the flow does not follow the blades exactly. The flow angle t3F2 is not
identical to the blade angle (32, because the relative exit velocity W2 is slightly more
inclined opposite to the direction of rotation, as shown in Fig. 7.4. This deviation
comes about because the fluid retains its orientation in the absolute frame of reference,
and appears to rotate with respect to the rotating impeller in the opposite direction,
which results in a tangential velocity component at the impeller exit opposed to the
direction of rotation (Fig. 7.5). To compensate for this deviation a correction is
applied to the equation above for the theoretical head in the form of a factor to the tip
speed, a, the slip coefficient:
Figure 7.4 Velocity diagram at the impeller exit with slip.

Figure 7.5 Fluid rotation relative to impeller and the resulting slip velocity.

Figure 7.6 Slip coefficient as a function of exit blade angle and number of blades.
Exact values for the slip coefficient have been calculated by Busemann assuming
potential flow past logarithmic spiral blades in a radial impeller (Busemann 1928).
His results have been verified by modern computer calculations (McDonald and
Howard 1973). Wiesner reviewed Busemann's results and the various empirical
expressions proposed by several other experts, in addition to a correlation from 49
impellers (Wiesner 1967), and arrived at the following expression for calculating the
slip coefficient, which is also shown in Fig. 7.6:
In this expression Z stands for the number of blades and i 2 for the exit blade angle
measured from the radial or meridional direction in the sense opposite to the sense of
rotation. In Wiesner's original paper the impeller exit angle is measured from the
tangential direction. This blade angle corresponds to 90 - #2 if the exit blade angle
/#2 is measured from the meridional direction in the direction opposite the sense of
rotation, which is the convention adopted in this book. The form of Wiesner's original
expression is retained in the expression above to facilitate comparison with his
publication. This relationship for the slip coefficient will be used in subsequent
chapters.
Slip does not represent an energy loss. It only affects the magnitude of the head that a
given size impeller can produce. Its effect is equivalent to the effect of a greater blade
exit angle N2.

RELATIVE VELOCITY IN THE IMPELLER AND ROTHALPY

The inlet velocity triangle defines the inlet relative velocity:
Applying Bernoulli's equation to the inlet, we find that
The last term corresponds to the centrifugal head rise. Consequently, a relationship
equivalent to Bernoulli's equation, called the rothalpy equation, applies in the rotating
impeller in terms of the relative velocity W:
Sometimes it is also called relative energy (Wislicenus 1965). This relationship is

illustrated in Fig. 7.7. The rothalpy is conserved and remains constant on a streamline
through the rotating impeller. This relationship allows calculation of the local static
head h, or the pressure, as a function of the radius r anywhere in the impeller, if the
relative velocity Won the streamline is known. The relative velocity W would be
expected to be calculated from the continuity equation, or approximately from the flow
rate, since the flow cross section should be available from the impeller geometry.
However, caution is indicated since the relative velocity varies from the pressure side
to the suction side of the blades in the circumferential direction due to the blade
loading. The pressure side of the blade faces in the direction of rotation. Nonetheless,
if a linear variation of the radial velocity Wr can be assumed between the blades in the
circumferential direction, the flow rate Q divided by the passage cross section 2irrB
will give the correct magnitude of the average radial velocity. The location of the
midstream streamline, however, will be displaced from the center toward the suction
side, where the velocity is greater.
From the exit velocity diagram we have the relationship

Figure 7.7 Conservation of energy equation and rothalpy in the rotating impeller.
Observing the conservation of rothalpy I, and substituting W2 yields
Therefore, the total head of the flow passing through the impeller, the sum of the static
head h and the velocity head C2/2g, increases by the theoretical head, which is added
by the impeller, UCt2/g = Hth, as stated above.
Although the rothalpy equation, strictly speaking, is valid only along a streamline, in
most pumps the total energy at the inlet is the same on all streamlines, and therefore the
rothalpy is also the same for all streamlines. Any pressure drop due to frictional losses
in the impeller must be deducted from the static pressure. At the impeller exit the flow
angle /3F2 and the relative velocity W2-therefore, the tangential velocity component
Crz also-may vary from one streamline to the next. Consequently, the theoretical head,
the energy addition by the pump, may not be the same for all streamlines. The magnitude
of the relative velocity W2 can be affected, for example, if the exit radius r2 varies, as
in mixed flow pumps, with a slanted exit. We consider such a case in Chapter 14.
Friction losses can also make the energy addition uneven, as in the case of a jet-wake
flow pattern, discussed below.
Since one earmark of a good pump is a uniform theoretical head, or equalenergy

addition on all streamlines, the flow variables need to be calculated on several
streamlines. Impeller design calculations consist of determining the angle a along the
trailing edge of the blade-and more generally along the entire blade surface so that
uniform energy be imparted to the fluid on all streamlines. Usually, calculations follow
a streamline along the hub, the shroud, and along the midstream. More detailed
calculations also consider streamlines on the pressure and suction sides of the blades.
The use of computers allows calculations on multiple streamlines covering the entire
flow passage. The choice of computation complexity and time expenditure should
correspond to the value of the pump.
As will be seen, at off-design operation most pumps, mixed flow pumps in particular,
do not impart the same energy on all streamlines, which results in a shear flow at the
impeller exit. This flow pattern leads to flow losses in the form of eddies and flow
fluctuations, and can sometimes result in peculiar behavior of the pump.
Since the velocity and pressure vary from the pressure side-the side facing in the
direction of rotation-to the suction side of the blades at the impeller exit, a stationary
observer looking at the rotating impeller will see a fluctuating, unsteady flow and
pressure field sweeping by. If Z number of blades pass per shaft revolution, an
oscillation frequency f = ZN per minute or f = ZN/60 per second will be observed. This
fundamental pressure oscillation, which is present in all pumps to a greater or lesser
extent, is called the blade passing frequency. It can induce resonant vibrations in the
piping system connected to the pump. The magnitude of the pressure fluctuations varies
in proportion to the blade loading, the pressure difference between the two sides of the
blade, near the impeller exit.


FLOW SEPARATION IN THE IMPELLER

To illustrate flow conditions in the impeller, most designers plot the relative velocity
W, along the pressure and the suction side of the blade, against the streamline length,
measured from the leading edge, as shown in Fig. 7.8. Because of the blade loading-the
pressure difference across the blade-the pressure side velocity is lower than the
suction side compared at the same fraction of the streamline length. The rothalpy
equation expresses this relationship between the local head or pressure and the
velocity. The difference between the velocities on either side of the blade illustrates
the pressure difference or blade loading, but only qualitatively. Actually, the rothalpy
equation relates the pressure not to the velocity but to the square of the velocity. For a
quantitative measure of the pressure difference, the square of the velocities would need
to be plotted. The plot shows that if the blade loading becomes too large, the velocity
on the pressure side approaches zero. The flow separates, which must be avoided.
Figure 7.8 Relative velocity difference in the impeller as an approximate measure of

blade loading.
An approximate estimate for the appearance of a separated flow can be obtained by

applying the principle of the angular momentum change to a very narrow annular
volume, such that r2 = rl + Ar and having width B in the axial direction, as illustrated in
Fig. 7.9. Then the incremental torque AT corresponds to the moment of the pressure
force exerted on the blades at a radius of rl in the circumferential direction. The

pressure on the side of the blade facing the direction of rotation, the pressure side pp
must be greater than the pressure acting on the trailing, suction side ps of the blade. In
the case of an impeller having Z number of blades, and if the incremental increase in the
angular momentum over a small radial distance of Ar is p(ACri), then
It appears from this equality that the energy added to the fluid is proportional to the
pressure difference across the blade, which is usually referred to as the blade loading.
In the flow passage between two blades, the pressure decreases in the circumferential
direction, approximately linearly, from the pressure side of the blade facing in the
direction of rotation to the suction side. As will be seen, there is a corresponding
increase in the velocity in the impeller. If the blade loading, and therefore the pressure
difference, becomes very high, the velocity at the pressure side of the blade can
become very small or vanish, which results in flow separation and losses. Generally
speaking, the blade loading provides a measure of the likelihood of a flow separation.

Figure 7.9 Pressure difference across the impeller blade: blade loading.
The incremental, radial change in the angular momentum corresponds to the radial
derivative of the expression of the angular momentum:
Assuming that the radial velocity component Cr and the blade angle /3 do not change,
or change very slowly, in the radial direction, and that the streamlines are parallel to
the blades in the circumferential direction, the incremental, radial change of the angular
momentum becomes
Substituting into the expression for the pressure difference across the blade, and
remembering that, according to the rothalpy equation, the pressure difference
corresponds to a velocity head difference at the same radius, one obtains:
From this expression those conditions can be determined for which the velocity at the
pressure side of the blade becomes zero (Crp = 0). These conditions signal separation
in the impeller and the onset of a jet-wake flow pattern.
When the average radial velocity Q/(2irrB) becomes smaller than the value of the
radial velocity Cry./2 above, separation appears. For radial blades, when tan J3 = 0,
the radial velocity would have to be on the order of the circumferential speed U to
avoid separation, which would be excessive, and separation would be expected unless
the number of blades Z were large. Indeed, in centrifugal compressors, with blades
nearly radial, 30 blades are not unusual. In centrifugal pumps, with typically about six
blades (Z = 6) and a blade angle of about 68 (13 = 68), the radial velocity would
have to be less than about 15 % of the circumferential speed U for separation to
appear, according to the criterion above, which would rarely be the case in practice at
design conditions.
Good pump design also demands that blade loading, and the corresponding velocity
difference between the pressure and suction sides, gradually grow at the inlet and taper
off at the exit. Consequently, some designers use the plot of Fig. 7.8 to verify a suitable
velocity distribution and satisfactory impeller design (Dallenbach 1961).
The danger of flow separation is also present when the velocity decreases and the
pressure increases too rapidly on any streamline. As discussed above, the velocity
cannot change beyond the separation point, and no additional velocity head can be
converted to pressure head. In centrifugal pumps the inlet relative velocity Wi is usually
larger than the exit velocity W2, and therefore a certain diffusion-slowing of the
velocity must take place in the impeller, which should not be excessive. Empirical rules
exist for a tolerable velocity ratio, usually on the order of W2/W1 = 0.7. Separation and
increasing flow losses would then be present for velocity ratios below this value. The
first flow separation in the impeller usually appears on the shroud and on the suction
side of the blades. The highest inlet velocity occurs at this location, and the greatest
diffusion or deceleration takes place. A sharp curvature of the shroud in the radial plane
contributes to the likelihood of separation, and should be avoided.
Conditions become necessarily critical at some point with decreasing flow rate,
since the inlet velocity W, approaches the inlet circumferencial velocity U1, while the
exit velocity W2 approaches zero as the flow rate is reduced. The ratio of velocities
W2/W1 decreases toward zero. Therefore, flow separation necessarily appears at some
point as the flow rate is reduced below the best efficiency point. Such a separated flow
can trigger, or is replaced by, inlet recirculation, which is discussed in Chapter 14.

JET-WAKE FLOW PATTERN

It was pointed out above that a shear layer, or separation streamline, between two flow
regions of different fluid energy is stabilized if the acceleration and corresponding
pressure rise perpendicular to the shear layer are directed from the low-energy region
toward the high-energy region. When separation appears in the impeller, two flow
regions of different energy can be distinguished: the separated region and the main
stream, as shown in Fig. 7.10. The relative velocity W remains constant, but different,
on either side of the streamline separating the two regions. The acceleration, and
corresponding pressure increase, are directed from the suction side to the pressure
side. A separated region of low energy will be stable and can then persist on the
suction side. A low-energy separated regionfor example, a boundary layer-on the
pressure side would be unstable and would tend to migrate along the hub and shroud
surfaces to the suction side. Such secondary flows do exist and have been observed in
impellers (Bayly and Orszag 1988; Johnston et al. 1972, Johnston and Rothe 1976;
Tuzson 1977). They are discussed in Chapter 8.
In the case of heavy blade loading, potential flow calculations may predict that the
velocity slows to zero at the pressure side and that flow separation will occur on the
pressure side. However, since a separated region cannot persist on the pressure side,
the flow switches to a jet-wake flow pattern. Consequently the ideal potential flow
model gives a false picture of the flow pattern. A flow pattern consisting of a stable
separated region or wake on the suction side and a main stream or jet near the pressure
side of the blades is called a jet-wake flow pattern. Such a flow pattern appears only
under certain conditions (Dean and Senoo 1960, Tuzson 1993).

Figure 7.10 Separated jet-wake flow in impeller.
If /3 is the flow angle of the separation streamline measured from the radial in the
direction opposite to the sense of rotation, the energy addition in each region is given by
The energy addition exactly balances the loss when

Figure 7.11 Jet-wake flow pattern criterion. (From Tuzson 1993.)
Therefore, a stable separation on the suction side and a jet-wake flow pattern can
persist if the ratio of the average velocity on the separating streamline, (W + W)/2, and
the circumferential speed U is larger than sin/3. The flow angle /3 can be taken equal to
the local blade angle. Figure 7.11 illustrates this relationship. It is apparent, for
example, that radial bladed impellers, when /3 = 0, would usually be expected to
operate with a jet-wake flow pattern. The blades in most pumps lean strongly
backward, making sin /j relatively large. Consequently, jet-wake flow pattern would
not be expected in pumps.

BLADE-WAKE MIXING
Although usually thin, impeller blades have a definite thickness and leave a wake in the
flow at the impeller exit. It can be assumed that the cross section of the flow passage
between the blades of the rotating impeller suddenly increases by the amount of the
blade thickness as the flow leaves the impeller. Since the blade thickness is
considerably less than the passage width, the area increase and the corresponding loss
are relatively small and may or may not be negligible. A sudden expansion loss
calculation can provide a rough estimate of the magnitude of such a blade-wake mixing
loss. There is reason to believe, however, that such a flow model is not valid and that
the mixing loss is relatively smaller than would be expected.
In reality the flow leaving the impeller is unsteady when considered by a stationary
observer in an absolute frame of reference. The absolute flow velocity at the impeller
exit is not parallel to the wake. The mixing process is therefore entirely different from a
steady wake mixing. The fluid in the wake is not lowenergy fluid as in the case of a
wake behind a stationary object, but can move with a higher velocity than the main flow,
approximately with the tip speed. In addition, trailing vortices are likely to be present in
the corners between the blade and the hub and shroud surfaces.
Very fast mixing has been observed in tests, causing surprisingly low losses. The
explanation was offered that energy transfer takes place by normal, unsteady pressure
forces, by pressure exchange (Foa 1958, 1973) rather than by viscous or turbulent shear
forces. Pressure exchange is considerably more efficient than shear in transferring
energy. Conceptually, the issue is very important, but since the losses are low and
mixing is fast, it does not affect pump design or pump performance significantly.
However, blade-wake mixing would be expected to leave a high level of turbulence in
the flow, not to mention the trailing vortices.
VANED DIFFUSER
In some pumps a radial, vaned diffuser follows the impeller exit. A short radial
distance, a vaneless space, precedes the diffuser. The diameter of the diffuser inlet is
usually about 5 to 10 % greater than the impeller diameter. As described above, the
pressure and relative velocity vary from the pressure side of the blades to the suction
side at the impeller exit. In the stationary frame of reference a fluctuating flow comes
off the impeller. If the leading edge of the diffuser vanes is too close to the impeller,
the flow fluctuations impinge on the diffuser vanes and can produce undesirable noise

and pressure pulsations at the pump exit. The vaneless space serves to allow the flow
to mix, become more uniform, and adjust to the inlet of the vaned diffuser. The pressure
pulsations can be reduced by slanting the leading edge of the diffuser vanes in the
circumferential direction with respect to the trailing edge of the impeller blades. The
trailing edge of the impeller blades then sweeps gradually across the leading edge of
the diffuser vanes instead of overlapping exactly at a certain moment. On small
multistage pumps sometimes a few widely spaced diffuser vanes strongly inclined in
the tangential direction are positioned relatively close to the impeller exit. Their
purpose is to attempt to slow down the high-speed flow immediately after the impeller
exit and recover some pressure. The radial component of velocity in the diffuser is
imposed by the continuity equation, the ratio of the flow rate, and the available cross-
sectional area. Diffusion is accomplished primarily by reducing the tangential
component of the velocity.
The direction of the diffuser leading edge and the diffuser throat size, the cross-
sectional area between two vanes perpendicular to the velocity, must match the
magnitude and direction of the velocity approaching from the impeller exit. The velocity
may slow down somewhat in the radial vaneless space in proportion to the increased
radius: C3 = C2(R2/R3) while maintaining its direction. On the other hand, the diffuser
throat velocity CQ3, assuming straight vanes and sidewalls, is given by the expression
where AS is the diffuser throat area, R3 the radius at the diffuser inlet, B3 the width of
the diffuser, and /33 the diffuser inlet angle. The expected pressure recovery in the
vaned diffuser follows the traditional diffuser experience. The effectiveness of
diffusers consisting of widely spaced airfoil-shaped vanes is more difficult to estimate.
General airfoil theory might provide a measure of the turning of the velocity, and
corresponding diffusion, that could be expected.
Multistage pumps combine the vaned diffuser with return-flow passages, which
conduct the flow to the inlet of the next stage. The purpose of the return passages is to
turn the swirling flow in the axial direction. They don't necessarily slow down the fluid.
In some pumps the impeller absolute exit velocity is approximately equal to the inlet
velocity. Under such conditions no opportunity exists to slow down the flow and
recover pressure. No diffuser is needed. In return passages only limited pressure
recovery can be expected, concentrated near the inlet section of the diffuser. A complex
three-dimensional flow is generated in the subsequent return bend, which is not suitable
to slow the flow and recover pressure. The flow velocity in the return bend is usually
kept constant and is turned gradually in the axial direction.
At off-design flow rates the impeller exit velocity is not aligned with the diffuser
inlet, and leading edge separation occurs. Also, a mismatch appears between the
approach velocity and the throat velocity, which results in a loss. Less head is lost if the
flow has to accelerate into the diffuser throat than if the flow has to slow down. The
flow has to accelerate at flow rates above the design flow rate and has to slow down at
flow rates below the design flow rate. Theoretically, the diffuser should be correctly
matched at design flow conditions.

VOLUTE
The impeller often discharges directly into the volute, which is a spiral-shaped flow
passage usually of circular or trapezoidal cross section (Hagelstein et al. 1999) (Fig.
7.12). Its cross section increases gradually around the impeller periphery, starting from
the volute tongue and ending at the volute throat. The volute tongue directs the total
flow, collected from around the impeller, through the throat to the pump exit flange. A
diffuser may or may not exist between the volute throat and the exit flange, depending
on the available space and the flow velocities.
Figure 7.12 Volute.
The gradually increasing volute flow cross sections are calculated from the flow rate
and from an average velocity at the volute cross section center. The usual flow model
assumes that the impeller exit tangential velocity decreases in proportion to the radius to
maintain constant angular momentum. The volute cross section A as a function of the
circumferential angle 0 counted from the tongue becomes
This expression also determines the needed volute throat cross section.
When the flow rate is greater than the design flow rate, the flow generally accelerates
in the volute, and the pressure tends to decrease in the circumferential direction. When
the flow rate is less than the design flow rate, the volute velocity tends to decrease, and
the pressure increases circumferentially around the impeller. Because of this pressure
increase or decrease, a transverse pressure force appears on the impeller at off-design
conditions, which must be resisted by the pump bearings (Agostinelli et al. 1960,
Csanady 1962, Lorett and Gopalakrishnan 1986, Iversen et al. 1960, Schwarz and
Wesche 1978).
The static pressure distribution at the volute periphery can easily be checked
experimentally because the volute castings are usually provided with threaded draining
holes, at which pressure measuring instruments can be attached. As in the case of the
vaned diffuser, a radial space is left between the impeller periphery and the volute
tongue. The circle centered on the center of rotation and touching the volute tongue, the
base circle, usually has a diameter 10 % greater than that of the impeller. At flow rates
less than the design flow rate, the volute tongue deflects some of the flow approaching
the volute throat, which passes between the volute tongue and the impeller and returns
into the volute. If the space between the tongue and the impeller is very large, too much
flow returns unnecessarily, and losses increase. If the space is too small, strong
pressure fluctuations at the blade passing frequency can occur.
In some self-priming pumps a hole is drilled from beyond the volute throat across the
tongue back into the volute, to recirculate the fluid until the pump is fully primed. The
detailed geometry of the volute and the size and location of the hole are usually
determined by experiment. They indicate the amount of suction lift the pump can handle
during priming.

INCIDENCE
The flow separates at the leading edge of the impeller blade if the direction of the
relative velocity W, does not align with the leading edge, as shown in Fig. 7.13.
Correct pump design usually assures correct alignment between the velocity vector and
the blade at the inlet at the best efficiency flow rate. At flow rates greater than design
flow rate the flow angle fF,-measured from the meridional or radial direction-is
smaller than the blade angle at the inlet #,, and a separated flow region appears on the
pressure side of the blade. At flow rates below design flow rate, separation occurs on
the suction side of the blade. The term incidence designates the difference between the
flow angle and the blade angle: 13FI - #i. The separated region on the pressure side of
the blade becomes unstable. Flow oscillations and rapid mixing ensue. Separation on
the suction side remains stable and can persist far into the impeller.
The blockage from leading-edge separation restricts the flow passage cross section
and therefore increases the local velocity. The head loss incurred when the separated
region mixes with the main stream can be considered a sudden expansion loss and can
be calculated from the velocities at separation and downstream, where complete mixing
has taken place (Cornell 1962). The corresponding loss calculation procedure is
presented in Chapter 10.
Particular attention needs to be paid to calculation of the blade angles at the leading
edge. Since the inlet flow of centrifugal pumps turns from the axial direction to the
radial direction, the relative flow velocity W, and its direction vary along the leading
edge. The flow pattern is three-dimensional, and a threedimensional flow calculation
would be desirable. A simplified calculation procedure is presented below and used in
the INLET computer program in the Appendix.
The real flow conditions at the impeller blade inlet at incidence are often much more
complicated than those of the two-dimensional separation model. If the flow velocity
meets the leading edge at a slant, a tip vortex can develop in the separated region along
the leading edge, which trails away toward the hub (Fig. 7.14). High-specific-speed
pumps are particularly sensitive to inlet conditions because the through flow velocities
are relatively high and the inlet diameter is also relatively large. The circumferential
velocity on the shroud far exceeds the circumferential velocity at the hub. Meridional
velocities are high on the shroud and low on the hub because of the curvature of the
streamlines. Since flow losses vary with the square of the velocity, high velocities
indicate potentially high losses. The most critical conditions are encountered on the
shroud streamline.
Figure 7.13 Inlet blade incidence: (a) increased flow rate, Q > Qo; (b) reduced flow
rate, Q<Qo

CAVITATION
Chapter 9 is devoted to the subject of cavitation in pumps. Only a brief overview is
presented here.
The high local velocity, near the blade leading edge, corresponds to low static
pressure, and cavitation could appear. Under cavitating conditions the absolute local
static pressure falls below the vapor pressure of the fluid, and steam or gas bubbles
appear. The evolving gas, and corresponding volume increase, can choke down the flow
and prevent further increase of the flow rate. The head-flow characteristics of the pump
then end at this maximum flow rate.
The collapse of the cavitation bubbles, and the reentering fluid jet, impinging on the
blade surface, can generate very high local pressures and can destroy the blade
material. Such cavitation damage must be avoided and remains a critical issue in pump
design, selection, and operation. Manufacturers' catalogs provide information on the
capability of the pump to tolerate low inlet pressures. Plots usually show the net
positive suction head required (NPSHR) by the pump as a function of flow rate. In
cavitation test the pressure at the inlet flange of the tested pump is lowered until the
head delivered by the pump drops off by 3 %, due to cavitation. NPSHR is defined as
the absolute total pressure at the impeller inlet poi, less the vapor pressure of the fluid
p p, divided by the specific weight of the fluid:

Figure 7.14 Incidence and separation on slanted leading edge: leading-edge vortex.
Empirical data were compiled by the Hydraulic Institute on the NPSHR requirement of
typical pumps, which can be summarized by the expression
Here h,,p is the head corresponding to the vapor pressure of the fluid, about 1 ft in
terms of absolute pressure head for water at standard conditions. The specific speed
NS is in English units, Q (gpm), N (rpm), and H (ft).
Industry standards (ASME, API, ISO, Hydraulic Institute) prescribe test procedures
for the experimental determination of NPSHR. These generally base the definition of
NPSHR on that pressure head at the pump inlet at which the pump head drops off by 3
%. However, cavitation noise and cavitation damage usually precede the head drop as
the suction pressure decreases.
In Chapter 9 we present analytical models rather than empirical correlations for

estimating the conditions for the onset of cavitation in pumps. The analytical models
lead to the calculation of the local absolute static pressures near the blade leading
edges, where cavitation is most likely to appear first. Cavitation is expected to appear
when the local absolute static pressure reaches the vapor pressure of the fluid. Knowing
the flow conditions at the pump inlet, when cavitation sets in, an analytical expression
for estimating the NPSHR required by the pump can be derived from fluid mechanic
fundamentals.
When specifying or matching pumps to a certain application, the net positive suction
head available (NPSHA) from the system is calculated by estimating the total absolute
pressure available from the system at the pump inlet, pol. The NPSHA then becomes
This value of NPSHA is then compared with the value of NPSHR from the pump
manufacturer's catalog. The NPSHA values must be greater than the NPSHR values.
In practice, the appearance of cavitation or vapor bubbles depends on a variety of

factors. Absorbed or dissolved air, gas, or fine bubbles, for example, may exist in the
fluid, which evolve or expand with decreasing pressure long before the vapor pressure
of the fluid would be reached and boiling could start. Symptoms of cavitation and
malfunctioning of the pump can already appear at pressures above the vapor pressure of
the fluid. Such malfunctioning may occur, for example, when the inlet piping is not tight.
Because of the negative pressure in the inlet, air can then be aspirated into the inlet
through the leakage paths. Therefore, the NPSHA values available from the system
should be generally higher than the NPSHR value required by the pump as listed by the
manufacturer. The temperature of the fluid affects the vapor pressure and may also lead
to premature cavitation. Hot or boiling fluids may require pressurization of the pump
inlet. Fluids other than water have vapor pressures that require a different NPSHR
value than that needed with water.

DISK FRICTION ON THE IMPELLER

Practical hydraulic pump design must consider all hydraulic forces on the impeller.
The circumferential, fluid frictional force on the impeller, disk friction, absorbs shaft
power and affects the overall efficiency of the pump. The calculation procedure for
estimating the corresponding power loss is given in Chapter 10.
Clearance spaces must exist between the stationary housing and the rotating hub and
shroud faces of the impeller. In centrifugal pumps the external faces of the impeller on
the hub and shroud side resemble a disk that is rotating in a stagnant fluid (Fig. 7.15).
The fluid at the disk surface rotates with the circumferential velocity of the disk. The
fluid shear at the rotating disk tends to accelerate the fluid volume in the clearance,
whereas shear at the stationary housing wall tends to slow it down. Consequently, the
fluid volume in the clearance between the impeller face and the housing will rotate with
a velocity about half that of the circumferential velocity of the impeller. Since the disk
friction torque increases with the fifth power of the diameter, small changes in the
diameter have a disproportionately large effect on the disk friction torque. The outer
periphery of the disk contributes substantially to the total friction force.
Figure 7.15 Flow velocity profiles in clearance between rotating impeller face and
housing.

The circumferential velocity of the fluid in the clearance changes rapidly near the
disk, perpendicular to the surface, over a certain, small distance, the boundary layer. A
similar boundary layer exists on the housing wall. If the clearance between the impeller
and the housing is much greater than the thickness of these boundary layers, the
clearance width does not influence the magnitude of the friction force, the disk friction,
on the impeller. While the boundary layer thickness, and more generally the flow pattern
in the clearance space, can be affected by the leakage flow through the clearance space,
an approximate value of the boundary layer thickness can be obtained. It can be
estimated to be a low multiple of (v/w) 1 ~2, which in water at 1800 rpm amounts to a
few thousands of an inch (Schlichting 1960). The disk friction losses of the pump will
be minimized if the clearance width, between the impeller faces and the housing walls
is made much larger than the boundary layer thicknesses.
The fluid rotating near the impeller face generates a radial pressure gradient because
of the centrifugal acceleration. This radial acceleration induces a radial, secondary
flow outward. A corresponding radial flow exists near the housing wall, which is
directed inward. Therefore, fluid circulates radially out along the impeller face and
radially in along the housing wall. In low-specific-speed pumps producing high head
and low flow rate, the secondary flow flowing radially out along the impeller face can
become significant compared with the flow rate of the pump. Since this secondary flow
rotates with the circumferential speed of the impeller, it carries high energy. It was
suggested that on small pumps, a portion of the head of this secondary stream should be
added to the head produced by the pump (Hamkins 1984). The outside faces of the
impeller would then act as separate pumps contributing to the output of the flow inside
the impeller.
Special pump configurations have been proposed and tested, the impellers of which
consist of closely spaced parallel disks mounted on the shaft. The fluid is entrained by
the fluid friction on the disks without blades (Hasinger and Kehst 1963, Roddy et al.
1987). Such disk pumps found special applications with heavily contaminated and
viscous fluids, for example.
If considerable leakage flow passes through the clearance space between impeller
and housing, it can cause a change in the average rotational velocity of the fluid volume
in the clearance (Rohatgi and Reshotko 1974).
In the turbulent flow regime, which is the normal operating range of centrifugal
pumps, the shear force per unit surface area of the disk must be proportional to the
density p and to the square of the circumferential velocity of the disk, w2r2. The surface
area of the disk is approximately proportional to the square of its radius R2. The
moment of this force, M, with a moment arm of R, produces a resisting torque on the
pump shaft, which will be given by an expression of the form
By substituting consistent units, it can be verified that the coefficient Cm is

dimensionless. Experiments have shown that, if moment M represents the resisting
torque on both sides of a disk, the coefficient C,,, takes on values from 0.006 to 0.01
with increasing roughness of the disk surface (Nece and Daily 1960). In practice (as
will be shown), an appropriate value of the coefficient Cm should be established from
matching the head-flow curve of an existing, typical pump with a performance
calculation program. Protruding bolts and unevenness will affect the coefficient
(Zimmermann et al. 1986). Operating on a fluid other than water will also require the
use of a different coefficient, since the Reynolds number will be affected by the fluid
viscosity.

AXIAL THRUST ON THE IMPELLER

The rotating fluid volume in the clearance between the impeller and the housing results
in a radially increasing pressure distribution. A pressure force is exerted on the front,
the shroud, and on the back, the hub, of the impeller. The resultant of these two
opposing forces must be resisted by the thrust bearing of the pump and therefore needs
to be calculated (Due 1966, Guelich et al. 1989; Fig. 7.16). Assuming that both
clearances communicate with the impeller exit, the same impeller outlet pressure also
exists in the clearances at the outer radius. The static pressure at the impeller exit p2,
above the static pressure at the inlet p I, can be calculated either in terms of the
theoretical head and the absolute velocities, or in terms of the circumferential and
relative velocities if the rothalpy equation is used.
Figure 7.16 Axial pressure forces and thrust on impeller.

where R2 is the outer radius, and r is the radius at which the pressure is evaluated.
The local pressure above the inlet pressure on the impeller face at a radius r
becomes
The inner radius RI on the front, the shroud, of the impeller corresponds to the radius of
the wear ring or front sealing sleeve. The rear, hub cavity usually extends to the radius
of the pump shaft, which has a much smaller radius than that of the wear ring. The
pressure force on the rear side of the impeller would then be greater than on the front,
and without some special provision, their difference might represent an excessive
thrust bearing load. Therefore, on larger pumps, a sealing sleeve is made on the hub
also, which has the same radius as the wear ring, as shown in Fig. 7.17. Balancing
holes are drilled from the pump inlet through the hub, near the shaft, to allow for
leakage past the rear sealing sleeve. The balancing holes must be large enough to
establish pump inlet pressure in the rear clearance space near the shaft, in the balance
chamber, and to assure that the major pressure drop occurs across the sealing sleeve. If
the axial pressure forces on the impeller are not entirely balanced, the force on the
thrust bearing will increase with the impeller exit static pressure and will generally
increase with decreasing flow rate.

Figure 7.17 Balancing of axial thrust on impeller: leakage flow.
Some impellers are provided with shallow radial or inclined ridges or vanes on their
back side which produce a low pressure in the clearance space near the shaft. Such
expeller blades, or pump-out blades, are frequently found on slurry pumps or pumps
handling contaminated water. The expeller blades reduce the pressure that acts on the
shaft seals. Sometimes, clean water is fed to the other side of the shaft seal to create a
leakage flow into the pump and avoid an outward flow of the pumped liquid, which
might carry contaminants into the seals. Naturally, expeller blades also change the axial
pressure balance of the impeller, which must be taken into account.

LEAKAGE
The flow pattern in the front clearance space of the pump, between the shroud and the
housing, is complicated by the fact that on the outer periphery, it connects with the fluid
at the impeller exit, and at the inner periphery, with the pump inlet. A sleeve with close
clearance, a wear ring at the pump inlet, restricts the leakage flow from the pump exit
back to the inlet (Fig. 7.17). This leakage flow is a drain on the high-energy flow
delivered by the pump, and therefore reduces the pump efficiency. The leakage flow
rate must be deducted from the flow rate through the impeller to arrive at the actual, net
flow rate delivered by the pump. Leakage flow is usually kept to 1 or 2 % of the total
flow and may or may not be negligible.
Leakage calculations assume that the pressure drop across the wear ring clearance
alone governs the leakage flow rate q. A simple orifice flow equation will suffice to
estimate the leakage flow rate:
In this expression Cd is the discharge coefficient, R,,. the wear ring radius, c the
clearance width, and h,,, the head drop across the wear ring, which can be calculated
from the expression derived previously for estimating the axial pressure force on the
impeller:
Here h2 - h, corresponds to the static head rise across the impeller, calculated using
Bernoulli's equation from the theoretical head Hth and the velocity heads of the
absolute inlet and exit velocities, C, and C2. The impeller exit radius is R2. The
discharge coefficient, Cd, depends on the clearance length, and decreases from 1.0 to
0.5 as the ratio of the clearance length and clearance width increases to about 80
(Bilgen et at. 1973).
Obviously, the smallest possible clearance c is desirable to minimize leakage losses.

It is limited by the shaft runout, or eccentricity, which in turn depends on the
manufacturing tolerance stackup and on the shaft deflection. Because of its close
clearance, the wear ring can act as a water-lubricated bearing and can generate
considerable bearing forces, which must be considered in shaft dynamic calculations.
Recommended clearance sizes can be found in the standards of the Hydraulic Institute,
ASME, ISO, and API.

REFERENCES
Agostinelli, A., Nobles, D., Mockridge, C. R. (1960): An Experimental Investigation
of Radial Thrust in Centrifugal Pumps, ASME Journal of Engineering for Power,
April, pp. 120-126.
Bayly, B. J., Orszag, S. A. (1988): Instability Mechanisms in Shear-Flow Transition,

Busemann, A. (1928): Das Forderverhaltniss radialer Kreiselpumpen mit

logarithmischspiraligen Schaufeln, Zeitschrift fur Angewandte Mathematik and
Mechanik, Vol. 8, October, pp. 372-384.
Cornell, W. G. (1968): The Stall Performance of Cascades, Proceedings of the 4th

U.S. National Congress on Applied Mechanics, ASME, New York, pp. 12911299.
Csanady, G. T. (1962): Radial Forces in a Pump Impeller Caused by a Volute Casing,

ASME Journal of Engineering for Power, October, pp. 337-340.
Dallenbach, F. (1961): Aerodynamic Design and Performance of Centrifugal and

Mixed Flow Compressors, SAE Paper 268 A, January, pp. 9-13.
Dean, R. C., Jr., Senoo, Y. (1960): Rotating Wakes in Vaneless Diffusers, ASME
Journal of Basic Engineering, September, pp. 563-574.
Due, H. F., Jr. (1966): An Empirical Method for Calculating Radial Pressure
Distribution on Rotating Disks, ASME Journal of Engineering for Power, April,
188-196.
Foa, J. W. (1958): Pressure Exchange. ASME Applied Mechanics Review,

December, pp. 655-657.
Foa, J. W. (1973): Cryptosteady-Flow Energy Separation, ASME Paper 73-FE24.
Guelich, J., Florjancic, D., Pace, S. E. (1989): Influence of Flow Between Impeller
and Casing on Part-Load Performance of Centrifugal Pumps, in Pumping Machinery,
ASME FED Vol. 81, pp. 227-235.
Hagelstein, D., Hillewaert, K., Van den Braembussche, R. A., Egenda, A., Keiper, R.,
Rautenberg, M. (1999): Experimental and Numerical Investigation of the Flow in a
Centrifugal Compressor Volute, ASME Paper 99-GT-79. Also Transactions of the

ASME.
Hamkins, C. (1984): Correlation of One-Dimensional Centrifugal Pump Performance

Analysis Method, ASME Paper 84-WAFM-10.
Hasinger, S. H., Kehrt, I. G. (1963): Investigation of the Shear-Force Pump, ASME

Journal of Engineering for Power, July, pp. 201-207.
Iversen, H. W, Rolling, R. E., Carlson, J. J. (1960): Volute Pressure Distribution,

Radial Force on the Impeller, and Volute Mixing Losses of a Radial Flow
Centrifugal Pump, ASME Journal of Engineering for Power, April, pp. 136144.
Johnston, J. E, Rothe, P. H. (1967): Effects of System Rotation on the Performance of

Two-Dimensional Diffusers, ASME Journal of Fluids Engineering, September, pp.
422-430.
Johnston, P. J., Halleen, R. M., Lezius, D. K. (1972): Effects of Spanwise Rotation on

the Structure of Two-Dimensional Fully Developed Channel Flow, Journal of Fluid
Mechanics, Vol. 56, No. 3, pp. 533-557.
Lorett, J. A., Gopalakrishnan, S. (1986): Interaction Between Impeller and Volute of

Pumps at Off-Design, ASME Journal of Fluids Engineering, March, pp. 12-18.
McDonald, G. B., Howard, J. H. G. (1973): The Computation and Utilization of

Busemans Analysis of Potential Flow in an Impeller, ASME Paper 73-GT-45.
Nece, R. E., Daily, J. W. (1960): Roughness Effects on Frictional Resistance of

Enclosed Disks, ASME Journal of Basic Engineering, September, pp. 553562.
Roddy P. J., Darby, R., Morrison, G. L., Jenkins, P. E. (1987): Performance

Characteristics of a Multiple-Disk Centrifugal Pump, ASME Journal of Fluids
Engineering, March, pp. 51-57.
Rohatgi, U., Reshotko, E. (1974): Analysis of Laminar Flow Between Stationary and
Rotating Disks with Inflow, NASA Report CR-2356, February.
Schlichting, H. (1960): Boundary Layer Theory, McGraw-Hill, New York, pp. 86,
179.

Schwarz, D., Wesche, W. (1978): Radial Thrust on Double-Entry Single-Stage

Centrifugal Pumps, Sulzer Technical Review, 2/1978, pp. 63-68.
Tuzson, J. (1977): Stability of a Curved Free Streamline, ASME Journal of Fluids

Engineering, September, pp. 603-605.

Fluid Engineering, September, pp. 463-467.
Wiesner, F. J. (1967): A Review of Slip Factors for Centrifugal Impellers, ASME

Journal of Engineering for Power, October, pp. 558-572.
Wislicenus, G. F. (1965): Fluid Mechanics of Turbomachinery, Vol. 2, Dover, New

York, p. 626.
Zimmermann, H., Firsching, A., Dibelius, G. H., Ziemann, M. (1986): Friction Losses
and Flow Distribution, for Rotating Disks with Shielded and Protruding Bolts,
ASME Paper 86-GT-158.


BOUNDARY LAYERS
IN PUMPS
BASIC CONCEPT
In fluid machinery, handling water, shear forces are small and localized. Shear forces
appear when the total energy of the fluid is different on neighboring streamlines. If the
streamlines are straight and parallel, the velocities will be different. A velocity
gradient will exist perpendicular to the streamlines. In this case the effect of viscosity
manifests itself in a shear stress, force per unit area, which is proportional to the
dynamic viscosity p and to the velocity gradient, the velocity difference over a unit
distance perpendicular to the streamlines. A shear stress will also appear if turbulent,
unsteady flow fluctuations are present that transfer energy between the streamlines.
Such a turbulent shear stress will also be proportional to the velocity gradient.
However, the constant of proportionality, taking the place of the dynamic viscosity,
will not be a material constant but will depend on the local intensity of turbulent
fluctuations and may vary throughout the flow.
Shear forces become important only when high-velocity gradients exist perpendicular
to the streamlines. In centrifugal pumps handling water, such locations can be found
primarily near the flow passage walls. The fluid in the immediate proximity of the wall
does not move. The velocity parallel to the wall increases with distance from the wall
and reaches the main flow velocity at a small distance from the wall, as shown in Fig.
8.1. Therefore, fluid shear and the corresponding phenomena are restricted to a small
layer near the wall. Generally speaking, these small layers came to be called boundary
layers.
Since these shear effects are restricted to small regions of the fluid, scientists had the
idea that fluid shear and corresponding energy transfer between streamlines needed to
be considered only near passage walls, leaving the rest of the fluid unaffected. Since the
layer thickness was small and the flow velocity was essentially parallel to the wall,
important simplifying assumptions could be made in the fundamental equations of fluid
mechanics. Such simplifications in turn led to important analytical solutions. The
principal boundary layer assumptions are that the variation of the flow velocity along
the wall is negligibly small compared with the velocity variations perpendicular to the
wall, and that the static pressure does not vary perpendicular to the wall.

Boundary layers affect the flow in pumps in two respects. Shear forces at the wall
produce friction and a loss of energy, which is reflected in the energy conversion
efficiency of the pump. Boundary layers also restrict the cross section of the flow
passage. Such a blockage should be taken into account when the relationship between
flow rate and flow velocity is established. Before the advent of boundary layer
calculations, empirical data and relationships were accumulated in the pump industry to
account for these effects. Since these guidelines were based on extensive testing, they
remain valid and are still widely used.
The general approach of boundary layer calculations consists of starting with an

ideal flow calculation which assumes that the velocity at the wall is equal to the
velocity at large in the flow. Having an estimate of the velocity at a large distance from
the wall, the equations incorporating the assumptions of boundary flow are applied near
wall to calculate the shear at the wall and the blockage that the boundary layer
produces. A correction can then be applied to the main flow, which takes into account
the presence and the effects of the boundary layer (Comolet 1994, Schlichting 1960,
White 1991).

TURBULENCE
Turbulence consists of unsteady velocity fluctuations in the flow. Such fluctuations can
transfer energy between neighboring streamlines. If the energy on neighboring
streamlines is different, if a shear flow exists, turbulent fluctuations can act in a fashion
similar to that of the kinematic viscosity of the fluid. Simplified calculations often use
an eddy viscosity which replaces the viscosity coefficient.
Figure 8.1 Boundary layer.
Under highly viscous flow conditions, flow fluctuations tend to damp out rapidly, and
the energy of fluctuations dissipates by viscosity. In the case of low viscosity, turbulent
fluctuations persist and are convected far downstream by the flow. The relative
importance of viscous forces compared with inertia forces is expressed in the form of a
Reynolds number, formed by the product of a typical velocity and a typical length
dimension divided by the kinematic viscosity of the fluid. Low Reynolds numbers
correspond to viscous flow conditions, whereas at high Reynolds numbers, inertia
forces predominate.
Turbulent fluctuations originate with shear layers that become unstable and spill
random fluctuations into the flow. Shear layers typically start at a stagnation point on a
solid body immersed in the flow or on the boundaries of a closed flow passage. The
typical velocity entering into the Reynolds number is the velocity difference between the
free stream and the body or wall. Since the velocity at the wall is zero, the velocity
entering into the Reynolds number is the relative velocity with respect to the wall. The
typical length dimension characterizes the geometry of the flow, the length or width of
the object, or the flow passage width. Obviously, the size of the flow oscillations must
be limited by the passage walls, for example. The typical length determines the scale of
the flow pattern.
The difficulty in calculating the effect of turbulence is that the local intensity of
turbulence, and therefore the apparent viscosity or eddy viscosity, depends on the
turbulence generated upstream from the location that is being considered. Aircraft flying
in quiet air can encounter practically no turbulence. Great difficulties were met in
building such wind tunnels, for testing airplane components, which would have
sufficiently low levels of turbulence. The upstream turbulence entering a pump, for
example, might be quite different depending on whether the pump inlet is submerged in a
quiet pool or is preceded by a long pipe. It has been shown experimentally, and also
theoretically, that upstream turbulence disturbances are amplified, even in a uniform
parallel flow, when the Reynolds number exceeds a critical value, typically about 3000.
The usual turbulence model assumes that initially relatively large scale fluctuations are
generated, which break up into smaller eddies, and finally dissipate as they are
convected downstream. The major portion of the fluctuating energy, which corresponds
to the energy loss of the flow, is contained in the large eddies.
Turbulent boundary layer calculations usually assume one of two upstream

conditions: Either no turbulence is assumed to be present upstream, or turbulence
conditions corresponding to well-established pipe flow are assumed. In a long pipe or
uniform flow passage the phenomena of turbulence generation and dissipation reach
equilibrium. The distribution of velocities and turbulent fluctuations remains the same at
all passage cross sections and are consequently, well defined. The expressions for
calculating turbulent boundary layer thicknesses, or blockage, and wall shear assume
well-established turbulent pipe flow. They correspond to the generally accepted pipe
flow friction loss correlations used throughout industry.

BOUNDARY LAYER THICKNESS

In the simplest case that can be imagined, a flat plate is positioned parallel to a flow of
uniform velocity U (Schlichting 1960). For an order-of-magnitude estimate of the
boundary layer thickness, the technique of dimensional analysis will be applied,
whereby typical values of all the variables affecting boundary layer flow are combined
into dimensionless groups, and a relationship between these groups is stipulated on
dimensional grounds. In the case of the semi-infinite flat plate, we are missing a typical
length dimension. The only reasonable assumption is that the boundary layer thickness
depends on the distance from the leading edge of the flat plate L, shown in Fig. 8.2.
Since the boundary layer cannot exist ahead of the leading edge where it begins, it is
reasonable to assume that the boundary layer thickness S will gradually grow from the
leading edge on. The forces acting in the fluid are the viscous and inertia forces. The
corresponding variables are the dynamic viscosity z, having the dimensions (force x
time/length2), and the density p, having the dimensions (force x time2/ length4).
Combining these variables into dimensionless groups it is found that the boundary layer
thickness divided by the plate length 8/L must be a function of the Reynolds number
which is formed from the free-stream velocity U, the plate length L, and the kinematic
viscosity v = /p. The kinematic viscosity of water, at standard ambient conditions, is
v = 10.9 x 10-6 ft2/sec.
Indeed, analytical solutions of the equations of viscous fluid flow with boundary layer
simplifications show that for low Reynolds number, viscous flow conditions, the
boundary layer thickness on such a flat plate is given by the relationship

Figure 8.2 Boundary layer on flat plate.
Since the velocity in the boundary layer approaches the free-stream velocity gradually
and asymptotically, as the distance from the wall increases, the question arises as to
how the boundary layer thickness should be defined. In the expression above, the
boundary layer thickness b corresponds to the distance from the wall where the
velocity reaches 99% of the free-stream velocity U.
A more practical definition of the boundary layer thickness is the displacement

thickness = Sd, or blockage, which corresponds to the flow cross-sectional area
restriction caused by the boundary layer (Fig. 8.3). For viscous flow it is given by
In the case of high Reynolds number, turbulent flow, where unsteady flow fluctuations
transfer energy between streamlines and are responsible for shear, a similar
dimensionless relationship applies:
Comparing the two expressions for the viscous and turbulent boundary layer
thicknesses, it appears that the transition from viscous flow to turbulent flow occurs at a
Reynolds number of Re = 3.2 x 105 = 320,000. In reality, the transition from viscous to
turbulent flow conditions depends on upstream disturbances in the flow and occurs over
a range of Reynolds numbers. Therefore, the numerical value derived above cannot be
taken rigorously. In the case of a turbulent boundary layer the displacement thickness
becomes even smaller: bd = 6/8. As an example, water flowing under turbulent
conditions (Re = 1.35 x 107) at a velocity of 10 ft/sec (3 m/s), would take 15 ft (4.6 m)
of length to develop a boundary layer thickness 6 = 2.5 in. (6.35 cm), and a
displacement thickness bpi = 0.3125 in. (0.8 cm).

Figure 8.3 Displacement thickness of boundary layer.
The boundary layer development calculated above would be expected to apply to the
blades of impellers and vanes of vaned diffusers. An added complication might arise if
the flow velocity approaching the leading edge of the blade is not in a perpendicular
plane to the leading edge, as in the case of swept wings, for example. In such a case a
velocity component will exist along the leading edge which will convect boundary layer
fluid along the leading edge, usually toward the hub of the impeller. Three-dimensional
boundary layer calculations have been developed because of the interest in flow
predictions for swept airplane wings. However, no closed-form results can be obtained,
and only high-speed computers can lead to the solution of specific cases (Humphreys
1981, Tassa et al. 1982).
Boundary layer assumptions break down near the leading edge of blades or vanes if
the velocity of the approaching flow is not aligned with the blade. The flow arrives at
an angle to the direction of the leading edge, and the flow separates immediately at the
leading edge. The flow can be modeled by assuming a separated wake region and a free
streamline or shear layer at its boundary. The wake thickness can be calculated
assuming an ideal flow pattern and using conformal transformations (Cornell 1962). The
subsequent stability and rate of mixing of the shear layer on the wake boundary will
depend on the direction of the acceleration perpendicular to the blade surface.
Typically, the mixing will be accelerated if separation takes place on the side of the
blade facing the direction of rotation-the pressure side and will be slowed on the
trailing, suction side. Consequently, pressure-side leading-edge separation, at flow
rates greater than design flow, may reattach, but suction-side separation, at flow rates

less than design flow, will tend to persist.
Separation on swept leading edges shows further peculiarities, because a leading-

edge vortex may appear in the separated region immediately behind the leading edge,
which will trail away along the leading edge, usually toward the hub side of pump
impellers. Such a vortex changes entirely the configuration of the separated region.

BOUNDARY LAYERS IN CLOSED FLOW PASSAGES

The flat plate boundary layer calculation also applies to closed passages, as for
example the inlet of a pipe, as long as the pipe diameter or distance to the opposite
wall of the flow passage is large compared with the thickness of the boundary layer.
The boundary layer "does not know" that there is a wall at a relatively great distance.
When the pipe diameter or passage width becomes comparable with the length L along
the wall, it starts to influence the boundary layer development. In the dimensional
analysis approach the passage width becomes the characteristic length dimension, and
the Reynolds number based on the passage diameter or width becomes dominant.
In the case of a flat plate the fluid friction energy dissipated at the wall is taken at the
expense of the kinetic energy of the fluid in the boundary layer. The layer has to grow,
and more and more slow-moving fluid accumulates in the boundary layer. In a closed
passage a steady condition must be reached, since the boundary layers cannot grow
indefinitely. Therefore, in a closed passage the wall friction energy loss is compensated
by a pressure loss along the passage. When the boundary layer thickness becomes
significant compared with the passage width, the velocity profile across the passage
gradually converges to a steady configuration. At low Reynolds numbers, under viscous
flow conditions, the fully developed velocity profile will be parabolic (Fig. 8.4). Under
turbulent conditions the velocity in the central portion of the flow passage will be
almost uniform, and near the walls the velocity will drop off over a small distance, a
steady boundary layer.
In a circular pipe, at high Reynolds numbers, the steady, local flow velocity u
becomes a function of the distance from the wally toward the center of the pipe at a
radius R where the centerline velocity is U:
The exponent n depends on the Reynolds number and takes on the value of n = 6 at Re =
4 x 103, n = 7 at Re = 104, and reaches n = 10 at Re = 3 x 106. The ratio of the mean
velocity u,,,, which would correspond to the flow rate divided by the pipe cross
section and the maximum centerline velocity U, becomes

Figure 8.4 Velocity profiles in a circular pipe in the viscous and turbulent flow range.
It will be noted that this ratio increases with the Reynolds number from 0.791 at n = 6
to 0.865 at n = 10. The difference between the calculated value and 1 corresponds to
the fraction of the flow passage cross section that is taken up by boundary layer
blockage.

SKIN FRICTION
From the point of view of pumping machinery analysis and design, the significance of
boundary layer calculations is that they allow a theoretically based estimate of wall
friction losses in flow passages. By definition, the viscous shear stress i, shear force
per unit area, at the flow passage wall is proportional to the velocity gradient at the
wall. In the case of a flat plate, an exact analytical solution of the viscous fluid flow
equations, with boundary layer simplifications, led to the following expression of the
local shear stress, where the Reynolds number is based on the length from the leading
edge of the flat plate:
In the case of turbulent flow in a pipe, the local shear stress at the wall becomes
In this expression the Reynolds number is based on the pipe radius. The - a exponent
was derived for the case when the turbulent velocity distribution in the pipe correlates
with the seventh power, n = 7, of the Reynolds number. Extrapolating to velocity
distributions at higher Reynolds numbers, the exponent in the expression of the wall
shear would become -2/(n + 1). These shear stress values correspond to the well-
known pipe friction coefficients.
BOUNDARY LAYERS ON ROTATING DISKS

The sides of a rotating radial impeller can be considered rotating disks, provided that
the width of the space between the impeller faces and the stationary housing is much
larger than the boundary layer. Consequently, this case has a great importance for pump
design and operation. Figure 8.5 illustrates the tangential and radial velocity
distribution in the clearance space between the rotating impeller face and the housing
wall.
It so happens that an exact solution of the complete fluid mechanic equations, the
Navier-Stokes equations, can be obtained for a disk rotating in an infinite fluid if the
Reynolds number based on the disk radius, the rotational speed, and the dynamic
viscosity of water remains small, less than 50,000 (Schlichting 1960, p. 83). The
angular velocity co of the disk and the kinematic viscosity v govern the thickness of the
boundary layer, which must have the form (w/v)1/2 to balance the dimensions. The
boundary layer thickness corresponds to the distance from the disk where the tangential
velocity becomes half of the disk velocity. In this particular case the tangential velocity
at great distance from the disk is assumed to be zero. At the same distance from the disk
the radial velocity induced by the rotation of the disk reaches a maximum and becomes u
= 0.180r(u. The total radial flow generated by the disk of radius R is Q =
0.4437rR2(vw)112. For these laminar conditions the disk friction torque from one side
of the disk becomes
The coefficient CM stands for the disk friction torque in dimensionless form. Its
numerical value does not depend on the type of measurement system used, English or
SI.
Figure 8.5 (a) Tangential and (b) radial velocity profiles in the clearance space
between rotating impeller face and stationary housing.
For turbulent conditions boundary layer calculations become more complex. The
boundary layer thickness 6, the flow rate Q, and the torque coefficient CM are found to
be
The expressions above are for a disk in an infinite fluid. If the disk is enclosed in a
housing with a side clearance s, which is smaller than the boundary layer, the moment
for one side of the disk becomes for laminar conditions:
If the clearance is a multiple of the boundary layer, the moment coefficients for laminar
and turbulent flow become
Several detailed studies have been devoted to the case when the rotating disk is
enclosed in a housing (Bakke et al. 1973, Bodonyi and Stewartson 1975, Chanaud 1971,
Cooper 1973, Maroti et al. 1960, Rohatgi and Reshotko 1974, Schilling et al. 1974,
Senoo and Hayami 1976). In this case the tangential velocity, gradually decreasing
perpendicular to the disk, approaches a value that is about half of the circumferential
speed of the disk midway between the disk and the housing wall. A boundary layer also
exists at the wall, as described in connection with the disk friction loss calculation in
Chapter 7. The analytical expressions above remain valid except for the numerical
coefficients, which may take on other values, depending on the geometrical details.
Very detailed experimental torque measurements have been made on enclosed

rotating disks which closely conform to the theoretical results. These data are being
used in pump design and performance prediction to estimate disk friction torque on
impellers.

BOUNDARY LAYERS IN VANELESS DIFFUSERS

Centrifugal pumps often use vaneless diffusers at the exit of the impeller if sufficient
radial space is available. They consist of parallel radial walls, and unlike vaned
diffusers, are very tolerant to variation in flow rates, since in the absence of vanes, no
incidence losses can occur. The flow from the impeller exit enters the vaneless diffuser
with a certain tangential and radial velocity which slows down toward the diffuser exit
at a greater radius.
A loss-free fluid flow analysis predicts velocities based on three fundamental

equations. The radial variation of the tangential velocity is governed by the principle of
conservation of angular momentum. Consequently, the tangential velocity is expected to
decrease in inverse proportion to the radius:
The radial velocity must satisfy the continuity equation and maintain constant flow rate
through consecutive flow cross sections, the area being given by the product of the
diffuser widths and the circumference at any radius: V,= Q/27[rs. Consequently, the
radial velocity also would be expected to decrease in proportion to the radius.
The pressure can be obtained from the equation of radial force balance, which states
that the incremental radial pressure increase will be equal to the centrifugal force, the
fluid density multiplied by the square of the tangential velocity and divided by the
radius:
Since the equation of conservation of energy, Bernoulli's equation, is an alternative to

the basic equations, the pressure can also be calculated from the changes of the
tangential and radial velocity heads. In loss-free flow the increase in pressure head
must correspond to the decrease in tangential and radial velocity heads. This ideal
flow pattern serve to define the free-stream conditions of a boundary layer calculation
(Gopalakrishnan 1993, Jansen 1964, Stanitz 1954).
When boundary layers and viscous or turbulent shear are present, as shown in Fig.
8.6, the torque from the tangential wall shear forces will oppose the conservation of
angular momentum. If well-established turbulence levels are assumed, the tangential
velocity distribution on either side of the vaneless diffuser will follow the configuration
measured in pipe flow. Such a turbulence level assumption might be questioned,
because the flow leaving the impeller is highly turbulent. Depending on the blade
loading at the impeller blade tips, the relative velocity on the pressure side of the
blades is lower than on the suction side, which results in a periodic absolute velocity
fluctuation at the impeller exit. Whatever the turbulence level and the corresponding
shear stresses might be, qualitatively speaking, the maximum centerline tangential
velocity in the vaneless diffuser, as well as the average velocity and angular momentum,
will gradually decrease with the radius.
Figure 8.6 Radial and tangential velocity profiles in a vaneless diffuser.
A different situation is met in the radial direction. The average radial velocity must
maintain a constant flow rate despite increasing boundary layer blockage. In addition,
the radial pressure increase on the centerline, where high tangential velocities prevail,
will be greater than near the walls, where the tangential velocity approaches zero. Such
a difference in pressure rise must be compensated by secondary flows radially inward
near the walls, opposing boundary layer flow. Consequently, the radial boundary layer
will thicken considerably, and the point will be reached when the boundary layers from
the two walls merge at the center of the flow passage. No or little pressure recovery can
be expected beyond this point.
Calculations (Gopalakrishnan 1993) supported by experimental data (Jansen 1964)

show that depending on the impeller exit flow angle, or the ratio of radial to tangential
velocities at the diffuser inlet, vaneless diffusers with radius ratios, ratios of exit to
inlet radii beyond about 1.5 to 2.0 do not recover additional pressure head and will not
improve the efficiency of a centrifugal pump.

BOUNDARY LAYERS INSIDE ROTATING IMPELLERS

Boundary layer development inside impellers resembles the conditions in a vaneless
diffuser provided that calculations are based on the relative tangential and radial
velocities measured with respect to the walls of the impeller (Fig. 8.7). However,
there is a major, important difference between the boundary layers on stationary walls
in a swirling flow and those in the impeller. Because of the Coriolis acceleration, a
circumferential pressure gradient exists. The pressure decreases in the tangential
direction from the side of the blades facing in the direction of rotation toward the back
side of the next blade. The pressure rise is proportional to the Coriolis acceleration,
which is equal to the product of the angular velocity co and the radial velocity W,-.
Since the relative velocity vanishes in the boundary layers on the hub and shroud flow
passage walls-unlike in the main flow-little or no tangential pressure rise exists along
these walls. This lack of pressure rise must be compensated by secondary currents in
the boundary layers from the pressure side of the blade toward the suction side. These
currents carry boundary layer fluid from the pressure side of the blades to the suction
si de, where they accumulate and thicken the boundary layer. Because of these
secondary currents, arising from the action of the Coriolis acceleration, which will
exist only in rotating impellers but will not exist when the impellers do not rotate,
boundary layer thickening and separation will not appear on the pressure side of
impeller blades but on the suction side.
In addition to the effects noted above, the Coriolis acceleration also affects the
distribution of the turbulence intensity. It can amplify or dampen turbulent fluctuations,
depending on the local flow conditions. Therefore, the usual, fully developed pipe flow
turbulence assumptions cannot be accepted with confidence, and special corrections are
needed (Howard et al. 1980). Similar transverse acceleration effect, such as that
resulting from streamline curvature, can strongly influence boundary layer development
(Ellis and Joubert 1974, Irwin and Arnot Smith 1975, Johnston 1972, Litvai 1972, So
and Mellor 1973, Wollkind and DiPrima 1973).
It is important to note that to a first order of approximation, the flow in rotating

impellers can be calculated as if the impeller were stationary and would not rotate.
However, if real viscous flow conditions are to be taken into account, impeller rotation
and the action of the Coriolis acceleration must be introduced into the analysis. Ideal
loss-free flow calculations sometimes show that the flow would separate on the
pressure side of the blades. These calculations give a false picture of the flow pattern.
In reality, the flow tends to separate on the suction side of the blades.

Figure 8.7 Boundary layers in the rotating impeller; secondary flow in the boundary
layers due to Coriolis acceleration.
Boundary layers in rotating impellers present another peculiarity. In stationary flow

passages the fluid in the boundary layer has lost some, if not all, of its energy. It is
customary to refer to boundary layer fluid as stagnant, low-energy fluid. In rotating
impellers, boundary layer fluid does not necessarily have less energy than the main
stream. From the point of view of a stationary observer a fluid element in the impeller
has an energy proportional to its angular momentum, its absolute tangential velocity
multiplied by the radius from the center of rotation. In impellers with backward-curved
blades, the absolute tangential velocity of the fluid is generally less than the
circumferential velocity of the impeller itself. Indeed, the fluid in the impeller flows in
the direction opposite the sense of rotation with respect to the impeller. However,
boundary layer fluid which hardly moves with respect to the impeller, and practically
rotates at the same speed as the impeller, has a higher absolute tangential velocity than
that of the main stream. Consequently, boundary layer fluid in impellers with
backwardcurved blades has more energy than the main stream. Therefore, the boundary
layer fluid leaving the impeller can add to, instead of subtracting from, the theoretical
head, the energy that the impeller imparts to the flow.
Boundary layer flow from the blade pressure and suction faces will mix with the
main flow at the impeller exit. The boundary layers on the inner hub and shroud faces of
the impeller will produce trailing vortices at the impeller exit. Indeed, since the energy
imparted to the fluid by the impeller, the lift produced by the blades, will be different in
the boundary layers than in the main stream, streamwise vorticity will be spilled into the
flow, which will roll up into trailing vortices in the corners where the blades meet the
internal hub and shroud faces of the impeller. The trailing vortices become particularly
evident in slurry pumps, in which the particles carried by the vortices can wear deep
grooves into the impeller.

PRACTICAL PUMP DESIGN CALCULATIONS

Because of the great diversity and complexity of turbulent boundary layer flow in
pumps, very detailed, lengthy, and expensive calculations of uncertain validity are
often not warranted. Empirical data from pumps, similar to those being considered,
become extremely valuable. Dimensionless loss and blockage coefficients can be
derived from such test data. Great variations in geometrical scale or viscosity will be
reflected in the change of the respective Reynolds number, which enters into the
relationship connecting the dimensionless boundary layer thickness or shear stress to
the Reynolds number. If a valid qualitative model of the flow pattern can be
constructed, the quantitative results, the numerical values, can be calculated from
dimensionless correlations with empirical coefficients. Such experimentally derived
dimensionless coefficients have been accumulated over years by well-established
pump manufacturers. They represent the confidential technological capital of successful
companies.

REFERENCES
Bakke, E., Kreider, J. F., Kreith, F. (1973): Turbulent Source Flow Between Parallel
Stationary and Co-rotating Disks, ASME Journal of Fluid Mechanics, Vol. 58, Pt. 2,
pp. 209-231.
Bodonyi, R. J., Stewartson, K. (1975): Boundary-Layer Similarity Near the Edge of a

Rotating Disk, ASME Journal of Applied Mechanics, September, p. 584. Also
ASME Paper 75-WA/APM-3.
Chanaud, R. C. (1971): Measurements of Mean Flow Velocity Beyond a Rotating

Disk, ASME Journal of Basic Engineering, June, pp. 199-204.
Comolet, R. (1994): Mecanique experimentale desfluides, Tome II: Dynamique

desfluides reels, Turbomachines, Masson, Paris, pp. 177-227.
Cooper, P. (1973): Turbulent Fluid Friction of Rotating Disks, NASA Report

CR2274, July.

Ellis, L. B., Joubert, P. N. (1974): Turbulent Shear Flow in a Curved Duct, ASME
Journal of Fluid Mechanics, Vol. 62, Pt. 1, pp. 65-84.
Gopalakrishnan, S. (1993): Boundary Layer Growth in Vaneless Diffusers, in

Pumping Machinery, ASME FED Vol. 154, pp. 207-218.
Howard, J. H. G., Patankar, S. V., Bordynuik, R. M. (1980): Flow Prediction in

Rotating Ducts Using Coriolis-Modified Turbulence Models, ASME Paper 80-
WA/FE-1.
Humphreys, D. A. (1981): Three-Dimensional Wing Boundary Layer Calculated with

Eight Different Methods, AIAA Journal, Vol. 19, No. 2, pp. 232-234.
Irwin, H. P. A. H., Arnot Smith, P. (1975): Prediction of the Effect of Streamline

Curvature on Turbulence, Physics of Fluids, Vol. 18, No. 6, pp. 624-630.
Jansen, W. (1964): Steady Flow in a Radial Vaneless Diffuser, ASME Journal of

Basic Engineering, September, pp. 607-619.
Johnston, J. P. (1972): The Suppression of Shear Layer Turbulence in Rotating

Systems, ASME Paper 72-FE-B.
Litvai, E. (1972): Prediction of Velocity Profiles for Turbulent Boundary Layers on

the Blading of Radial Impellers, Proceedings of the Conference on the Fluid
Mechanics of Fluid Machinery, Budapest, Hungary, pp. 771-782.
Maroti, L. A., Deak, G., Kreith, F. (1960): Flow Phenomena of Partially Enclosed
Rotating Disks, ASME Journal of Basic Engineering, September, pp. 539552.
Rohatgi, U., Reshotko, E. (1974): Analysis of Laminar Flow Between Stationary and
Rotating Disks with Inflow, NASA Report CR-2356, February.
Schilling, R., Siegle, H., Stoffel, B. (1974): Stromung and Verluste in drei wichtigen
Elementen radialer Kreiselpumpen, Stromungsmechanik and Stromunmgsmaschinen,
No. 16, July, pp. 1-46.
Schlichting, H. (1960): Boundary Layer Theory, McGraw-Hill, New York, 1960.
Senoo, Y., Hayami, H. (1976): An Analysis on the Flow in a Casing Induced by a

Rotating Disk Using a Four-Layer Flow Model, ASME Journal of Fluids
Engineering, June, pp. 192-198.
So, R. M. C., Mellor, G. L. (1973): Experiment on Convex Curvature Effects in

Turbulent Boundary Layers, ASME Journal of Fluid Mechanics, Vol. 60, Pt. 1, pp.
43-62.
Stanitz, J. D. (1954): One-Dimensional Compressible Flow in Vaneless Diffusers,

NACA-TN 2610.
Tassa, A., Atta, E. H., Lemmerman, L. A. (1982): A New Three-Dimensional

Boundary Layer Calculation Method, AIAA-82-0224.
White, R M. (1991): Viscous Fluid Flow, McGraw-Hill, New York.
Wollkind, R., DiPrima, R. C. (1973): Effect of a Coriolis Force on the Stability of

Plane Poiseuille Flow, Physics of Fluids, Vol. 16, No. 12, pp. 2045-2057.


CAVITATION IN PUMPS
DESCRIPTION
Cavitation has been defined as the phenomenon when the local absolute static pressure
of the fluid, somewhere in the flow, drops below the vapor pressure of the fluid and
vapor bubbles appear (Arndt 1981; Brennen 1994; La Houille Blanche 1988, 1992;
Avellan et al. 1995). As the bubbles are convected downstream, they reach a location
where the pressure increases again above the vapor pressure. They collapse and create
a sudden, very high pressure pulse which can damage the material of a nearby flow
passage wall. In centrifugal pumps the lowest pressure usually appears on the impeller
blades at the inlet, especially at flow rates higher than the design flow rate, when the
flow separates on the leading, pressure side of the blades.
Pump designers wish to know the conditions for the appearance of cavitation in
pumps and the extent of the resulting damage. Prediction of these items remains difficult
because of the complexity of the phenomena and because several factors affecting
cavitation in a certain application remain unknown, such as the air content of the water
handled by the pump. Commercially available pumps must be able to accommodate such
uncertainties and operate satisfactorily over a range of application conditions. Exact
numerical calculations, executed for nominal operating conditions, have uncertain
general validity, and an uncertainty margin must be allowed. Therefore, only
approximate methods of calculation, supported by empirical data, can be offered here.
In particular, two phenomena concern pump designers and users. Cavitation damages
the impeller and cannot be tolerated for long. Extensive cavitation and significant
amounts of vapor will impair operation of the pump. This second issue is discussed in
detail in Chapter 15.
Pump designers and users have been concerned with cavitation primarily when
pumping cold water. Most experimental data and guidelines were taken with and apply
to water. However, water has very unusual physical properties affecting cavitation. It
has considerably higher latent heat of evaporation and surface tension than those of
other liquids. Indeed, it has been noted that cavitation of other liquids is different and
less damaging than cavitation of water (Salemann 1959). Cavitation with water is not
typical. Since the formation of vapor bubbles involves thermodynamic phenomena, mass
and heat transfer to the bubble, several physical properties of the liquid other than vapor
pressure affect the cavitation process. It can be assumed that the onset of cavitation, the
first appearance of vapor bubbles, indeed depends primarily on the vapor pressure of
the liquid. The phenomena of further evolution and collapse of the bubbles, of cavitation
damage, and of pump performance with a significant amount of vapor involve other
fluid properties and differ for different kinds of fluids. The issues presented in this
chapter are restricted to those dealing with cavitation of cold water.
Unfortunately, the inception of cavitation, the first appearance of vapor bubbles, only
produces some ill-defined noise which cannot provide a distinct indication for field
applications. Practical considerations demanded that the appearance of cavitation be
defined by distinct deterioration of the pump performance, which corresponds to an
advanced state of vapor bubble formation. The onset of cavitation in centrifugal pumps
is determined during tests, while gradually lowering the absolute inlet pressure of the
pump. The generally accepted condition, which signals the appearance of cavitation, is
when the pump head decreases by 3% compared with its value at high inlet pressures.
However, cavitation bubbles appear much sooner with decreasing inlet pressure, and
serious damage can also be done. The appearance of the bubbles coincides with the
appearance of cavitation noise, a crackling sound, which sounds somewhat like the
noise when a fluid just about starts to boil. Attempts have been made recently to
empirically correlate the intensity of this noise with cavitation damage. (Guelich 1989
a,b). Only correlations based on the fluid mechanics of the flow are presented here.

NET POSITIVE SUCTION HEAD

A centrifugal pump requires a certain total head hol at the impeller inlet in order to
avoid the absolute static pressure at the inlet or on the impeller blades near the inlet
dipping below the vapor pressure of the fluid, causing cavitation to appear. The vapor
pressure of the fluid, pvp, is a material property which depends on the temperature.
When the absolute pressure in the fluid, the pressure measured above absolute vacuum,
falls below the vapor pressure, vapor bubbles appear and the fluid starts to boil.
Absolute vacuum corresponds to a pressure of 14.7 psi, or a water column pressure of
33.8 ft (10.3 m) below standard ambient pressure. Therefore, pump inlet pressure
calculations must be expressed in absolute pressure, psia, measured from absolute
vacuum. Ambient pressure then becomes 14.7 psia in terms of absolute pressure.
The vapor pressure of water depends on the temperature; at 70F it is 0.3629psia

(2500N/m2) and at 85F it is 0.5958psia (4100 N/m2), only approximately 1 ft above
absolute vacuum, which corresponds to 0 psia. Organic fluids can have very different
vapor pressures (Salemann 1959). If the static wall pressure pi is measured at the pump
inlet with a gage that measures the pressure in psig relative to ambient conditions, the
atmospheric pressure, pat,,,-about 14.7 psia-must be added to arrive at the absolute
value of the static pressure p1 in psia. The relative static pressure pi (psig) can be
negative-a suction with respect to ambient pressure-but the absolute pressure will
always be positive.
To estimate the pressure needed by the pump at the inlet, the total, absolute pressure
p01 psia must be calculated, which amounts to the sum of the static pressure p, (psia)
and dynamic pressure of the absolute inlet velocity CI :
The capability of a pump to avoid cavitation is expressed by the net positive suction
head required (NPSHR) by the pump, which is established by pump manufacturers by
tests and is usually shown in pump catalogs. Standard testing procedures (ASME. ISO,
API, Hydraulic Institute) prescribe that the pressure at the inlet flange of the pump be
lowered until the head produced by the pump declines by 3%. The NPSHR value is then
based on the absolute total pressure pot psia at the pump inlet minus the absolute vapor
pressure of the fluid pp. The definition of the net positive suction head required

(NPSHR) by the pump becomes
Flow velocities even higher than at the pump inlet flange can be reached on the
blades of the rotating impellers. The local velocity head can be higher and the absolute
static pressure can be lower than at the inlet flange. The impeller blades at the inlet of
pumps are usually designed to admit the flow into the impeller without any change in
velocity. Therefore, at design conditions the pressure on the blade leading edge will be
essentially the same as at the inlet flange. However, at flow rates different from the
design flow rate, especially at higher flow rates, the incoming flow will not be aligned
with the direction of the blades, the flow will separate at the blade tips as shown in Fig.
9.1, and a separated flow region will restrict the flow passage cross section between
the blades. The flow velocity will have to accelerate to maintain the flow rate. Higher
velocities and lower pressures would be expected on the blade leading edge. Indeed,
experiments show that the inlet head needed to avoid cavitation, the NPSHR of the
pump, increases with increasing flow rate. Pump performance plots in manufacturers'
catalogs usually show the experimentally determined NPSHR of the pump as a function
of the flow rate. The experimentally measured NPSHR is that value for which the pump
head drops 3% compared with the pump head at the same flow rate but with ample inlet
head. Experiments are performed at standard ambient temperatures and pressures.
However, cavitation bubbles and damage can appear at inlet heads higher than those
corresponding to the 3% head drop.
A dimensionless cavitation coefficient 6 can be defined as the ratio of the net

positive suction head required by the pump, the total inlet head less the vapor pressure
head of the fluid, divided by the velocity head of the circumferential velocity at the
pump inlet U1:

This cavitation number has been used in the past by pump designers to obtain some
estimate of the likelihood of cavitation. Empirical data have been published which
correlate specific speed with the cavitation coefficient (Stepanoff 1957, Sulzer 1989).
It applied to the scaling of similar pumps, in particular reflecting the effect of rotational
speed on NPSHR. Analytical models are presented below for estimating the conditions
at the onset of cavitation in the pump and for arriving at an estimate of the NPSHR
required by the pump.

CAVITATION AT DESIGN CONDITIONS
Figure 9.1 Cavitation bubble.
At design conditions the absolute velocity on the blades at the inlet is the same as the
velocity at the pump inlet flange C1. This assumes that an allowance was made for the
blockage of the blade thickness when the inlet cross-sectional area was calculated.
Blade blockage is neglected below. At the onset of cavitation the absolute inlet static
pressure becomes equal to the vapor pressure of the fluid; the NPSHR equals the inlet
velocity head:
Using the relationships
the impeller inlet diameter can be expressed as
Consequently, if the consistent units N (rpm) and Q (ft3/sec) are used, the numerical
coefficient becomes a dimensionless pure number, and the expression for the NPSHR
in feet takes the form
Since the velocities can differ from the average velocity along the leading edge, the
numerical coefficient may differ from the value above. Cavitation may occur first near
the shroud, for example, where higher velocities may exist. However, an approximate
estimate can be obtained simply by assuming an average flow velocity (Vlaming 1989).
In analogy to the specific speed, a suction specific speed S is used in industrial

practice to provide a measure of the suction capability of a pump. Retaining consistent
units, the suction specific speed also remains dimensionless.
By comparing with the expression of NPSHR above, the suction specific speed can be
identified with
The theoretical derivation of the NPSHR above corresponds to the first inception of
cavitation. The suction specific speed values in catalogs derive from experiments and
correspond to the 3% head drop value and include the flow rate in gpm. Therefore,
they can differ considerably from the calculated values.
In general industrial practice, the NPSHR is sometimes expressed in terms of the

specific speed NS., in the form
The expression must have this particular form to balance the dimensions, as can easily
be verified. The coefficient Cc depends on the inlet blade angle /3I . It can be
dimensionless if the dimensionless specific speed is used and the speed, flow rate, and
head are expressed in consistent units. It will not be dimensionless if the specific speed
is expressed in the usual English units.
Here the specific speed N, is in English units, Q (gpm), H (fl), and N (rpm). The
numerical constant is not dimensionless and will be different if SI units are used. It
should be noted that NPSHR does not really depend on the pump head H, which
appears in this expression only to compensate for the dependence of the specific speed
on the head. When the NPSHR is expressed as a function of the shaft speed N and the
flow rate Q, the head H disappears.
The correlation offered by the Hydraulic Institute is based on average inlet designs
generally encountered in the industry and gives adequate estimates for the application of
pumps, but a more detailed calculation of the inlet velocities will be required for pump
design. The correlation, and the data of the Hydraulic Institute, are based on a pump
head drop of 3%, water at a temperature of 85F, and standard ambient pressure. For
pump applications the NPSHR estimate must be conservative, because cavitation
damage appears long before the head starts to drop off noticeably. Cavitation and NPSH
requirements have been investigated experimentally operating on hot water and liquids
other than water (Salemann 1959).

CAVITATION AT OFF-DESIGN CONDITIONS

At off-design conditions the inlet velocity is not aligned with the blades, and the flow
separates at the blade leading edge, forming a separated wake region at a constant
pressure ps, as discussed in Chapter 7 and in connection with incidence losses
included in performance calculations where the flow conditions are illustrated. The
flow adjusts gradually, and the relative flow velocity W, becomes uniform and parallel
to the blade surfaces. Since no losses occur up to this location, the sum of the static and
velocity heads must still equal the total head and must satisfy the equation of
conservation of energy. If Wsm and Wst are the meridional and tangential components
of the relative separated flow velocity WS , Ws, is also the absolute meridional
velocity, and U,-W, is the absolute tangential velocity. The energy equation, Bernoulli's
equation, can be written in terms of the absolute velocity and static pressure at the inlet
C, and pi, and the same variables corresponding to the separated flow:
The NPSHR value required by the pump will be obtained when the pressure in the
separated region Ps becomes equal to the vapor pressure of the fluid, pvp. The NPSHR
value equals the total inlet head, less the vapor pressure head, which corresponds to
this condition:
Since the relative velocity is parallel to the blade which makes an angle #' with the
meridional direction, W, = WS sin /3,, the NPSHR becomes
The relationship between the inlet relative velocity WI and the velocity at separation
W, depends on the size of the separated region and the flow passage blockage it
creates. The ratio of these two velocities can be obtained from an expression that was
derived for ideal, loss-free separated flow on the blade leading edge for a blade angle
/'1 and a flow angle NFI (Cornell 1962):
The NPSHR value is then given in terms of the inlet relative velocity W1 by
Similar and much more detailed calculations of the separated flow region on the blade
leading edges have been done, especially in conjunction with the development of
inducers (Brennen 1994).

Example
Let us assume an inlet blade angle of Q1 = 65 and a flow rate 40% higher than the
design flow rate Q/Qo = 1.4. The tangent of the inlet flow angle is the ratio of the
circumferential velocity UI and the meridional velocity C1, and since the meridional
velocity is proportional to the flow rate, it can be written
These values apply to the blade leading edge at the shroud where cavitation typically
first appears, and assume a correct alignment of the flow velocity with the blade angle
at design conditions. Similar calculations can be executed for other locations along the
leading edge.
Therefore, the flow angle at the flow rate Q = 1.4Qo becomes RFI = 56.86. The
quantities entering into the calculation of the velocity ratio A = WI/WS are cos RFI =
0.54665, cos(2/3F1 - #I) = 0.6080, cos F'1 = 0.4226. Therefore, A = 0.67407. Since C,
/ U, = 1 /tan fFI, the expression for the net positive suction head required becomes
These results apply to the impeller inlet at the shroud and assume that the blade leading
edge is radial, since the expression was derived from a two-dimensional model. In the
case of a slanted leading edge, boundary layer development becomes more complex,
and a leading edge vortex might appear. Also, some designers allow a slight positive
incidence at design conditions.
An experimental correlation based directly on the inlet velocities has been published
by Gongwer (1941). It is intended to predict the NPSHR value at correct incidence.
Here the axial or meridional inlet velocity C, was equal to the average velocity
calculated from the ratio of the flow rate and the inlet cross section. In reality, because
of the curvature of the flow, the meridional velocity at the shroud, where cavitation first
appears, is higher than the average value. There is also the issue of whether the
blockage from blade thickness was taken into account. Consequently, this
experimentally determined relationship may incorporate some effect of a velocity higher
than the average meridional velocity that would typically have been encountered in the
pumps from which the data were derived. Theoretically, at the design point the NPSHR
would be expected to be equal to the velocity head of the local, meridional velocity
A similar expression has been published more recently (Turton 1994, p. 24) in the form
where C, and U, are in ft/sec. The Hydraulic Institute recently issued guidelines for
rating centrifugal pumps while taking the available NPSHA into account (Hydraulic
Institute 1998).
At a certain flow rate below the design flow rate, inlet recirculation may set in which
completely alters the flow pattern along the leading edge. The twodimensional leading-
edge flow separation model loses its validity. Generally, less suction head is needed at
reduced flow, and the danger of cavitation is reduced. However, a leading-edge vortex
may appear with very low pressures at the centerline of the vortex, which may cause
unexpected cavitation.

CAVITATION DAMAGE CORRELATION

Quantifying erosion damage estimates has recently received much attention (Hattori et
al. 1998). A major research effort was initiated not too long ago by the Electric Power
Research Institute to find quantitative measures of cavitation damage in centrifugal
pumps (Guelich 1989 a,b). Detailed pump tests and a simultaneous visual observation
of the actual cavitation bubbles led to the conclusion that cavitation damage could be
correlated with the length of the cavitation bubble. The material removal rate ER was
found to be proportional to the following variables:
The coefficient of proportionality is C, = 0.6 x 10-6 in3/lbf-sec (2.3 x 10-12 m3/N s)

for the suction side and C, = 0.3 x 10-4 in3/lbf-sec (1.1 x 10-10 m3/N s) for the
pressure side, provided that the variables are normalized with the following reference
values of the variables: cavity length LcavR = 0.4 in. (1 cm), tensile strength of the
blade material Rm = 150,000 psi (1000N/mm2), which was for ferritic steel in surface
water, speed of sound in the vapor aR = 4888 ft/sec (1490 m/s), density of the vapor
pvpR = 0.00108 lbm/ft3 (0.0173 kg/m3), gas content of water OCR = 24 ppm, typical
for surface water. The exponent n was 2.83 for the suction side and 2.6 for the pressure
side.
In the study above the cavity length divided by the inlet diameter, Lcav/Di, was found
to generally decline from 0.2 at a = 0.2 to 0.5, to 0.005 at u = 0.5 to 2.0, depending on
the flow rate as a fraction of the design flow rate. The blade material properties needed
further corrective factors beyond the tensile strength of the reference ferritic material.
Austenitic steels were found to be more resistant by a factor of 1.7. The weight loss
varied from 0.00002 to 0.00004 lb/hr (0.0001 to 0.0002 N/h). Obviously, the data show
large scatter, and the principal value of the correlation resides in the designer's ability
to scale cavitation data derived from tests of existing pumps to new designs.
A correlation has been established (Sloteman 1995) between the difference of the
cavitation number based on the available NPSHA value and the NPSH value at which a
3% head loss is observed, UA - a3%, and the cavity length divided by the blade spacing
at the inlet circumference, Lcav/(JrD,/Z). Cavitation damage was also correlated with
cavitation noise, which promises to be an important diagnostic tool (Guelich 1989 a,b;
Guelich et al. 1993). However, this noise correlation does not enter into pump design.
INDUCERS
Cavitation on the impeller blades at the inlet can be avoided or mitigated if the
pressure of the approaching fluid is raised by some means. In critical cases an inducer
ahead of the impeller inlet can achieve such a pressure rise (Fig. 9.2). Inducers make
practical pump operation possible when the fluid actually boils or very nearly boils, as
for example in cryogenic rocket pumps, aircraft fuel pumps (Rohatgi 1995), or when
pumping some organic fluids which are near boiling in chemical processes.
Government research on rocket technology and hydrofoils has imparted a great impetus
to inducer research (Acosta 1973, Brennen 1994, Cooper 1967).
Inducers consist of axial flow rotors with spiral-shaped blades (Fig. 9.2). A design
procedure can be found in a recent article by Rohatgi (1995). Designers use only three
or four blades inclined 7 to 14 to the circumferential direction. The leading edge must
be as sharp as possible. The axial or meridional velocity C, is kept low, and a slight
positive incidence is preferred. Low axial velocities require large inlet cross sections,
larger than the impeller inlet would normally have. Therefore, pumps using inducers and
intended for very low suction heads must be custom designed. They need an
exceptionally large inlet cross section for the inducer. The flow passage cross section
decreases at the centrifugal impeller inlet to increase the meridional velocity and results
in reasonable impeller inlet blade angles.

VIBRATION
Cavitation often appears together with noise and vibration (Brennen 1994). Twophase
flow of liquid and gas mixtures is well known for the potential of generating pressure
fluctuations. However, since cavitation usually appears first at off-design conditions,
in which flow separation and recirculation can exist even in the absence of cavitation,
the causes and effects of vibration and cavitation remain difficult to separate. Flow
separation, vortices, and mixing of the fluid in the impeller are intimately connected
with the cavitation process, but also exist in the absence of cavitation.
Figure 9.2 Inducer.
Pressure fluctuations and vibration have an important effect on rotor dynamics, on the
sizing of the pump bearings, and on the corresponding bearing life. The magnitude and
duration of bearing loads resulting from vibrations cannot be estimated accurately. An
additional safety margin must be added to the anticipated bearing load when frequent
operation at off-design conditions are anticipated or under cavitation. Violent vibrations
might even bring about a catastrophic bearing overload. At present in design
engineering calculations, the only practical approach to account for such upset
conditions is to add a sufficient safety margin. Standards of the Hydraulic Institute, ISO,
ASME, and API prescribe permissible vibration levels. Rotating equipment and bearing
manufacturers often offer guidelines to estimate the service factors appropriate for
certain applications. Vibrations can be measured during pump tests, and their
anticipated long-term effect on bearing life can be estimated.

REFERENCES
Acosta, A. J. (1973): Hydrofoils and Hydrofoil Craft, Annual Review of Fluid
Mechanics, Vol. 5, pp. 161-184.
Arndt, R. E. A. (1981): Cavitation in Fluid Machinery and Hydraulic Structures,

Avellan, E, et al. (1995): La cavitation: mecanismes physiques et aspects

industrielles, Presses Universitaires de Grenoble, Grenobles, France.
Brennen, C. E. (1994): Hydrodynamics of Pumps, Oxford University Press, New

York, 1994.
Cooper, P. (1967): Analysis of Single and Two-Phase Flow in Turbo-Pump Inducers,

ASME Journal of Engineering for Power, October, pp. 577-588.

Gongwer, C. A. (1941): A Theory of Cavitation Flow in Centrifugal-Pump Impellers,

ASME Transactions, January, Vol. 63, pp. 29-40.
Guelich, J. F. (1989): Guidelines for Prevention of Cavitation in Centrifugal

Feedpumps, EPRI Paper GS-6398, November.
Guelich, J. F. (1989): Cavitation Erosion in Centrifugal Pumps, Chemical Engineering

Progress, November, pp. 68-73.
Guelich, J. F., Cother, A., Martens, H. J. (1993): Cavitation Noise in Jet Cavitation
Tests and Pumps, in Pumping Machinery, ASME FED Vol. 154, pp. 1-6.
Hattori, S., Mori, H., Okada, T. (1998): Quantitative Evaluation of Cavitation

Erosion, ASME Journal of Fluids Engineering, March, pp. 179-186.
Hydraulic Institute (1998): Centrifugal and Vertical Pumps for NPSH Margin,
ANSI/HI 9.6.1, HI, Parsippany, N. J.

Rohatgi, U. S. (1995): Sizing of Aircraft Fuel Pump, ASME Journal of Fluids

Salemann, V. (1959): Cavitation and NPSH Requirements of Various Liquids, ASME

Journal of Basic Engineering, June, pp. 167-180.
Sloteman, D. P. (1995): Avoiding Cavitation in the Suction Stage of High Energy

Pumps, World Pumps, September, pp. 40-48.
Stepanoff, A. J. (1957): Centrifugal and Axial Flow Pumps, Wiley, New York.
Reprinted by Krieger Publishing, Malabar, FLa., 1993, pp. 250-253.
Sulzer Brothers Ltd. (1989): Centrifugal Pump Handbook, Elsevier Applied Science,
New York, p. 28.

York, 1994.
Vlaming, D. J. (1989): Optimum Impeller Inlet Geometry for Minimum NPSH

Requirement for Centrifugal Pumps, in Pumping Machinery, ASME FED Vol. 81,
pp. 25-28.

10

PERFORMANCE
CALCULATION
CALCULATION PROCEDURE
A calculation procedure to estimate the theoretical performance of a pump is an
indispensable tool in pump design. Performance needs to be known, not only at the
rated, best efficiency point, but also off design. Pump specifications often impose
special requirements, such as head at shutoff, maximum power demand, rate of head
rise to assure stability, and so on. A good pump design process requires trial-and-error
iteration, a check on anticipated performance with a trial geometry, and progressive
approximation to the optimal design configuration. The best hydraulic design does not
necessarily correspond to the best commercial pump product. Compromises are
unavoidable.
Excellent pump performance calculation procedures have been developed and

published (Balje 1981, Church 1944, Guelich 1999, Hamkins 1984, Ida 1979, Jansen
and Sunderland 1990, Peng and Jenkins 1983, Pfleiderer 1961, Stepanoff 1965, Traupel
1988, Turton 1994), and several computer programs are commercially available. A
computer program developed by NASA and available from COSMIC (Veres 1994)
relies heavily on empirical correlations compiled from rocket pump data and does not
identify the various loss mechanisms. Generally, the more accurate and detailed these
calculations are, the greater the number of input variables needed: not only the desired
head, flow rate, and rotational speed, but also the details of the geometrical description
of the impeller and housing. The labor and cost required to prepare and enter the input
variables limits the practical use of computer programs for simple, inexpensive pumps.
A simple and fast calculation procedure with minimal input requirements, is

presented here (Tuzson and Iseppon 1997). The mathematical expressions, programmed
in the computer code LOSS3 and shown in the listing in the Appendix, are presented on
the following pages. Flow conditions are calculated on an average streamline through
the pump. It should be particularly helpful for modifying or improving existing designs,
provided that it is first adjusted to match the performance of a similar, existing pump. In
the process of matching the performance of an existing pump, the various loss
coefficients are determined. Many of these loss coefficients depend on the existing
manufacturing practices of the pump manufacturer and would be expected to remain the
same for new designs. Such factors might, for example, be the machining and casting
surface finishes in use.
Since FORTRAN listings do not permit variables with subscripts, and to minimize
the need for two sets of symbols for the same variables, the equations in this chapter are
often presented in terms of the symbols used in the computer program. In some instances
the use of two symbols for the same variable cannot be avoided. However, symbols are
clearly defined, and a full nomenclature is given with the computer program listing in
the Appendix. The expressions in the program listing include conversion factors since
the code was intended for numerical calculations in the English system. Use with the SI
system would require a review and modification of these conversion factors. As an
alternative, a conversion from SI to English, and vice versa, can be added at the
beginning and end of the program.
Designers can easily implement and use this calculation procedure on simple
personal computers. The Appendix contains a three-page FORTRAN listing, LOSS3.
The mathematical expressions of the listing can also be incorporated into larger,
proprietary computer programs. Since the procedure and program listing are simple,
users can and should understand the physical phenomena behind the calculations. The
calculation can be executed by hand, but the computer can do it faster. Every calculation
step can be checked by hand. If the results do not make sense, reasonable adjustments
should be made in the input. The designer and user should use practical judgment to
override, if necessary, the course of action suggested by the program. It should be kept
in mind that pump efficiency depends not only on the hydraulic design but also on the
mechanical inefficiency, for example.
The effect of individual variables, such as number of impeller blades, inlet diameter,
and impeller exit width (Fig. 10.1) on pump performance is complex and often
contradictory, depending on the value of other variables. Their effect on performance is
difficult to predict from memory or from empirical rules, because of the complexity of
their interdependence. A simple computer program can quickly show, if not the exact
numerical magnitude, at least the trend of their effect. This calculation procedure is
intended for radial or mixed flow pumps with return passages or volutes, in a range of
specific speeds from 500 to 5000 calculated in English units, 0.2 to 2.0 in
dimensionless units, and flow rates from 100 to 5000 gpm, equivalent to 0.006 to 0.3
m3/s.
Depending on the particular case, the calculated head and efficiency might approach
the measured value to within 2%, which is usually adequate since pump test data are
rarely more accurate or reproducible than 1%. The greatest accuracy can be expected
near best efficiency, where the losses are the smallest. The basic concept of the
calculation consists of estimating the energy imparted by the pump impeller to the fluid,
the theoretical head, and of subtracting all head losses to arrive at the net output head
for a given flow rate. To calculate efficiency, the power absorbed by disk friction, the
bearings, and inlet recirculation are added to the hydraulic power demand of the
impeller. The flow rate is adjusted for leakage.
The various flow phenomena corresponding to the losses and velocities calculated in
the computer program are discussed in Chapter 7. The mathematical expressions used in
the program are presented in the present chapter. The computer program requires input
of the design flow rate, rotational speed, geometrical variables of the pump, and several
loss coefficients, as shown on the printout of the design example. The head, efficiency,
losses, leakage, and several velocities are printed out for six selected flow rates across
the operating range. The Appendix lists the nomenclature and specific instructions for
executing the computer program.

IMPELLER WORK INPUT

The energy imparted by the impeller to the fluid, the theoretical hydraulic head Hth,
designated DHTH in the computer program, is calculated from Euler's equation, which
corresponds to the conservation of angular momentum. No prerotation is assumed here.
When prerotation is present, the flow approaching the impeller rotates in the direction
of, or opposite to, the sense of rotation of the impeller. The angular momentum of the
fluid entering the impeller would then have to be subtracted from, or added to, the
angular momentum of the fluid leaving the impeller, in order to calculate the theoretical
hydraulic head. Without prerotation the theoretical hydraulic head is given by the
expression
Figure 10.1 Nomenclature of impeller and diffuser geometry.

where
If significant blockage, due to blades, boundary layers, or separation is anticipated at

the impeller exit, the effective impeller width B2 can be made somewhat smaller (but
never larger) than the geometrical width. In mixed flow impellers the meridional
velocity W,,,2 is inclined toward the rotational axis, and the impeller width B2 is to be
taken perpendicular to the direction of the velocity even if the exit leading edge is
slanted. The exit angle /2 should be measured in the plane of the exit velocity vector
W2 and its meridional projection W2. If the housing restricts the impeller exit, as in
some vertical, multistage pumps, the slip coefficient can become smaller than the value
given by Wiesner's correlation, and the head will be reduced.

FLOW LOSSES
Incidence losses, DQIN12, at the pump inlet are calculated by assuming a leading-edge
separation, and a sudden expansion loss, when the separated flow mixes. These flow
patterns at incidence, illustrated in Fig. 10.2, have been described in previous
chapters. Calculation of the extent of the separated region and the corresponding
velocity increase follows a potential flow model, solved by conformal transformation
(Cornell 1962). The inlet flow angle /3FI, BEI in the listing, is calculated from tan F'FI
= U1/CI, and the blade angle is #I, BEBI in the listing, both measured from the
meridional direction. The following equation of the potential flow calculation
determines the ratio 2 of inlet velocity WI to the velocity at separation W.:
Assuming a sudden expansion loss beyond the separated region, and substituting Wss =
W1 cos NFI/cos NI, the expression for the head loss becomes
Here W, stands for the velocity far downstream from the separated region, where the
separated flow has been mixed and the velocity becomes uniform.
The expression above, which is programmed into the performance calculation

computer program, was derived for a broad range of blade cascade configurations. A
simpler form will give essentially identical results. It assumes that the loss is
proportional to the square of the difference between the tangential component of the
inlet velocity C,, and the circumferential velocity U, :

Figure 10.2 Impeller inlet incidence and separation.
Therefore, the inlet incidence loss calculation does not include any empirical loss
coefficient. The calculation assumes a radial leading edge and an axial inlet velocity.
In the case of a slanted leading edge, the inlet blade angle is to be adjusted artificially
to achieve minimum incidence loss at the design flow rate. For calculating accurate
inlet flow conditions, a separate, preferably threedimensional calculation procedure
should be used. The computer program INLET in the Appendix offers a simple
approximation.
The wall friction or skin friction losses (DQSF12 and DQSF23) in the impeller and
diffuser or volute follow the standard pipe friction model (Hydraulic Institute 1998).
Since the flow passage cross sections are irregular, a hydraulic radius and average flow
velocities are used. The friction coefficient can be adjusted but has a default value of
0.005. The hydraulic radius of the impeller blade passages is obtained by dividing the
passage cross sectional area by half of the circumference. The geometrical
measurements are taken at the impeller exit: B, the blade height, becomes the passage
width, and the circumference, mD2, divided by the number of blades, XNB, and
multiplied by the cosine of the blade angle at the exit, BEB2, becomes the distance
perpendicular to the blade surface:
The blade passage length is taken to be (D2 - D1)/(2 cos BEB2). Finally, assuming an
average relative flow velocity (W2 + W1)/2, the friction head loss becomes
A diffusion loss (DQDIF) needs to be taken into account, since separation invariably
appears in the impeller at some point. The program assumes that when the ratio of the
relative velocity at the inlet W, and the outlet W2 exceeds a value of 1.4, a portion of
the velocity head difference is lost:
if W,/W2>1.4.
Volute head loss (DQIN23) results from a mismatch of the velocity leaving the
impeller and the velocity in the volute throat. If the velocity approaching the volute
throat is larger than the velocity at the throat, the velocity head difference is lost
(Gopalakrishnan and Lorett 1986, Worsted 1963). The program calculates the velocity
approaching the volute throat by assuming that the velocity leaving the impeller Ct2
decreases in proportion to the radius because of the conservation of angular momentum:
C3 = Ct2(D2/D3). The volute throat velocity CQ3 is calculated from the flow rate and
the volute throat cross-sectional area:
To calculate the diffuser loss (DQVD) the program estimates the pressure recovery
coefficient based on the area ratio and applies an adjustable loss coefficient CVD. In
pumps the diffuser accounts for the greatest head loss, much influenced by the detailed
design of the diffuser. Therefore, estimating diffuser losses is particularly difficult
(Reneau et al. 1967).
Some designers assume, as a rule of thumb, that half of the diffuser inlet velocity head
is lost.
The disk friction loss (DFH) calculation follows the well-known expression, the
default value of the loss coefficient CDF being 0.005 (Nece and Daily 1960). The disk
friction power is divided by the flow rate and appears as a parasitic head, to be added
to the theoretical head when the shaft power demand is calculated.
A parasitic power demand also results from the appearance of inlet recirculation
(DRECH) (Tuzson 1983). The program uses an expression of the type
where D1 is the inlet diameter, Qo is the design flow rate.
Unfortunately, the recirculation loss coefficient depends on the piping configuration

upstream of the pump in addition to the geometrical details of the inlet, and therefore is
difficult to estimate. Impellers with relatively large inlet diameters (usually encountered
in high-specific-speed pumps) are the most likely to recirculate. The program contains a
default value of 0.005 for the loss coefficient, but verification of such a value on similar
pumps is advisable. The parasitic power of recirculation is also divided by the volume
flow rate, like the disk friction power, in order to be converted into a parasitic head.

The leakage flow rate is deducted from the impeller flow rate when the efficiency is
calculated. Usually, it only amounts to 1 or 2% of the total flow rate. The calculated
value is very approximate and serves to give some idea as to whether leakage losses
are significant. The wear ring clearance width depends on the manufacturing tolerances
and becomes relatively small in large pumps. Shaft deflection must also be taken into
account when the least possible clearance width is estimated.
Finally, the pump head is calculated by subtracting from the theoretical head all the
flow losses:
The efficiency is calculated by dividing the pump head by the theoretical head, to
which the disk friction head and the recirculation head have been added, and
multiplying this ratio by the ratio of the output flow rate and the sum of the output flow
and the leakage flow rate:

CALIBRATION ON TEST DATA

The performance calculation computer program will not give satisfactory results unless
it has been calibrated on a similar pump. The loss coefficients depend on the design
and manufacturing practices of the pump manufacturer. For example, the skin friction
coefficient will depend on the machining and casting surface finishes. Detailed
information on typical surface finishes can be found in an article by Osterwalder and
Hippe (1982). However, the most practical method to establish the specific values of
these loss coefficients is to compare and adjust the calculation results to the tested
performance of a similar, existing pump. Since the design and manufacturing
procedures of the new pump are likely to be the same as those of the old pump, the loss
coefficients are also likely to be the same.
The calibration process, the adjustment of the calculation results by changing the loss
coefficients, also teaches the user the effect of each loss mechanism on the overall
performance of the pump. An intuitive feel develops for recognizing specific features of
the pump performance just by looking at the head-flow and efficiency curves. Low
efficiency at certain flow rates, or features of the headflow curve, can be connected
with specific loss mechanisms. The difference between the performance details of low-
and high-specific-speed pumps becomes apparent.
It is indispensable to first estimate correctly the energy input by the impeller.

Subsequent losses can only dissipate, can never add, energy. The first step in matching
calculation to tested performance consists of verifying the work input, the head H
divided by the efficiency q of the pump (Fig. 10.3). The work input head multiplied by
the flow rate corresponds to the power demand of the pump. As shown on the head
versus flow rate plot of Fig. 10.3, the curve of the work input becomes parallel and
almost identical to the theoretical head line at large flow rates, beyond the design flow
rate. The work input head consists of the sum of the theoretical head (DHTH), the
recirculation head (DRECH), and the disk friction head (DFH), which also includes
mechanical losses, such as bearing losses. The head corresponding to these various
losses decreases rapidly with increasing flow rate. Therefore, at high flow rates the
work input becomes practically identical with the theoretical head. The theoretical head
is sometimes called the ideal hydraulic head, which is the energy per unit volume that
the impeller imparts to the fluid.

Figure 10.3 Calculated head-flow curve of a centrifugal pump.
The impeller exit blade angle 12 and the exit width B2 determine the slope of the line
corresponding to the theoretical head, as can be seen from its mathematical expression.
Pivoting around the intersection point at shutoff, and tangent to the work input at high
flow rates, the slope of the line is easily found and adjusted. There is some justification
for deviating from the exact, geometrical measurements of the variables. The average
exit blade angle is difficult to measure and does not always correspond to the value
shown on drawings. The effective impeller exit width can be made smaller (but never
larger) if there is reason to suspect some blockage, due to blockage of blade thickness,
boundary layers, or flow separation in the impeller.
As to the effect of losses (Fig. 10.4), skin friction and diffuser losses increase with
the square of the flow rate and predominate at high flow rates. Therefore, these loss
coefficients should be adjusted when the calculated head deviates from the experimental
head at high flow rates. The volute loss depends on the impeller exit velocity, which
increases with decreasing flow rate. Therefore, the volute loss coefficient should be
adjusted when the calculated head deviates from the experimental head at low flow
rates. Incidence losses increase on either side of the design flow rate but have no
adjustable loss coefficient. The computer program assumes a radial leading edge. The
incidence losses are often fictitious, and the inlet blade angle should be adjusted to
achieve minimal losses at the design flow rate-unless a bad incidence is suspected at
the design flow rate-to make the calculation agree with the test data. The disk friction
and recirculation losses, which also include the mechanical losses of the bearings,
affect only the efficiency, not the head.
Figure 10.4 Pump head losses.

RESULTS
The printout of the computer program shows the flow rate, head, efficiency, individual
losses, and velocities at various locations in the pump. Unfortunately, testing only
informs about the overall performance of the pump. Measurement of the flow pattern
and velocities inside the pump, especially in the impeller, presents great difficulties
and would be prohibitively expensive in the case of most pumps. If a satisfactory match
is achieved between tested and calculated performance, there is reason to believe not
only that the losses are being assessed correctly but that the calculated average
velocities also approximate well the real flow velocities. Since high velocities usually
signal high losses, the location of potential improvements can be identified. Knowing
the local velocities provides additional information toward an understanding of the
behavior of the flow inside the pump.
The design and development of pumps with exceptional performance requires

alternating steps of experiment and analysis. An execution of the performance
calculation during or immediately following an experiment can reveal the details of
shortcomings and suggest opportunities for further improvement in the next iteration
cycle.
REFERENCES
Balje, O. E. (1981): Turbomachines, Wiley, New York.
Bilgen, E., Boulos, R., Akgungor, A. C. (1973): Leakage and Frictional

Characteristics of Turbulent Helical Flow in Fine Clearance, ASME Journal of
Fluids Engineering, December, pp. 493-497.
Church, G. (1944): Centrifugal Pumps and Blowers, John Wiley, New York, 1944.

U.S. National Congress on Applied Mechanics, ASME, New York, pp. 129 11299.
Gopalakrishnan, S., Lorett, J. A. (1986): Interaction Between Impeller and Volute of

Pumps at Off-Design Conditions, ASME Journal of Fluids Engineering, March, pp.
12-18.
Guelich, J. E (1999): Kreiselpumpen, Springer-Verlag, New York.

Hamkins, C. (1984): Correlation of One-Dimensional Centrifugal Pump Performance

Analysis Method, ASME Paper 84-WAFM-10.
Hydraulic Institute (1998): Pipe Loss Standards, HI, Parsippary, N.J.
Ida, T. (1979): Analysis of Scale Effects on Centrifugal Pumps, Science Reports of

the Research Institute.for Engineering, Kanagawa University, Kanagawa, Japan,
August.
Jansen, W, Sunderland, P. B. (1990): Off-Design Performance Prediction of

Centrifugal Pumps, in Fluid Machinery Components, ASME FED Vol. 101, pp. 1-9.
Nece, R. E., Daily, J. W. (1960): Roughness Effects on Frictional Resistance of

Enclosed Rotating Disks, ASME Journal of Basic Engineering, September, pp. 553-
562.

Peng, W. W, Jenkins, P. E. (1983): Hydraulic Analysis on Component Losses of

Centrifugal Pumps, in Performance Characteristics of Hydraulic Turbines and
Pumps, ASME FED Vol. 6, pp. 121-125.
Pfleiderer, C. (1961): Die Kreiselpumpen, Springer-Verlag, Berlin, 1961.
Reneau, L. R., Johnston, J. P, Kline, S. J. (1967): Performance and Design of Straight,

Two-Dimensional Diffusers, ASME Journal of Basic Engineering, March, p. 141.
Reprinted by Krieger Publishing, Malabar, Fl., 1993.
Traupel, W. (1988): Thermische Turbomaschinen, Springer-Verlag, Berlin, 1988.

York, 1994.
Tuzson, J. (1983): Inlet Recirculation in Centrifugal Pumps, in Performance

Characteristics of Hydraulic Turbines and Pumps, ASME FED Vol. 6, pp. 195-200.
Tuzson, J. and Iseppon, A. (1997): Centrifugal Pump Design Teaching Tool, ASME

Pumping Machinery Symposium, FEDSM97-3370.
Veres, J. P. (1994): Centrifugal and Axial Pump Design and Off-Design Performance
Prediction, NASA Technical Memorandum 106745, COSMIC Program LEW-
16173.
Wiesner, F. J. (1967): A Review of Slip Factors for Centrifugal Impellers, ASME

Journal of Fluids Engineering, October, pp. 558-572.
Worsted, R. C. (1963): The Flow in Volutes, Proceedings of the Institution of

Mechanical Engineers, Vol. 177, No. 31, pp. 843-875.

11

PUMP DESIGN
PROCEDURE
PUBLISHED DESIGN PROCEDURES

Several pump design procedures are described in the literature, including methods
covered in the books of Balje, Church, Guelich, Pfleiderer, Stepanoff, Traupel, and
Turton cited in Chapter 10. Others have been published in engineering journals (e.g.,
Peck 1968, Salisbury 1983, Spring 1984) and in government reports (e.g., Oak Ridge
1972). These methods are mentioned here for comparison with the procedure offered in
this book. These design procedures concern themselves primarily with design point
operation. Without a calculation to predict performance across the operating range, the
loop of design iteration cannot be closed. Development would then proceed
immediately to expensive fabrication and testing at considerable risk. The design
procedure advocated here relies on the performance calculation computer program to
verify that design specifications have been reached.
CHOICE OF IMPELLER DIAMETER, EXIT WIDTH,

AND EXIT BLADE ANGLE
The design procedure presented here will lead to a first, trial pump configuration. It
will choose the values of the input variables needed by the performance calculation
computer program. Execution of the performance calculation will show whether the
desired characteristics and efficiency have been attained. Modifications can then be
tested to optimize the design. After a satisfactory overall design has been achieved,
detailed design of the components-the impeller blading, volute configuration, or return
passage vanes-can proceed.

Note that in English units the diameter is usually in inches and the tip speed is desired
in ft/s, which introduces a conversion factor of 12.
To eliminate the effect of the losses for the time being, a target efficiency q will be
assumed. The pump specifications often prescribe such an efficiency target. The
prescribed efficiency can be checked against the efficiency derived from the plot in Fig.
11.1, which shows typical, industrywide efficiencies as a function of flow rate and
specific speed. The Hydraulic Institute (1994) also offers an efficiency prediction
method. Since the design head, flow rate, and speed are given, the specific speed can be
calculated. Desired efficiencies higher than those given by the plot would not be easy to
achieve, and would require an extensive combined analytical and experimental
development process.
By dividing the given design head Ho by the efficiency q, the theoretical head Hch
can be calculated, which does not depend on the losses, and which determines how
much energy the impeller has to impart to the fluid. To facilitate the choice of
subsequent variables, the dimensionless plot of Fig. 11.2 is introduced, which shows
the theoretical head, represented by the head coefficient 0, against the flow rate,
represented by the flow coefficient 0. These coefficients are related to the head rise
across the pump H, flow rate Q, efficiency q, tip speed U2, diameter D2, and exit width
B2 of the pump by the expressions
gH >7 U2
Care must be taken when comparing these head and flow coefficients with those
appearing in other publications, because some may be defined differently. Sometimes
head coefficients do not take the efficiency into account and are based only on the net
head rise of the pump. Sometimes, the head is normalized with U2/2g instead of U2/g
as is done above.
The definitions used here are chosen because they connect the theoretical head or
energy input of the pump directly to the absolute tangential velocity at the impeller exit
and the flow rate with the radial velocity at the impeller exit. The dimensionless head

coefficient corresponds to the absolute tangential velocity at the impeller exit C12
divided by the tip speed U2. The dimensionless flow coefficient corresponds to the
radial velocity at the impeller exit Cr2 divided by the tip speed U2:

The plot shown in Fig. 11.2 actually reproduces the velocity diagram at the impeller
exit in a dimensionless form. The tangential direction is along the vertical axis, and the
meridional direction along the horizontal. As the flow rate changes, the operating point
of the pump moves along the straight line of constant angle 132:

Figure 11.2 Head and flow coefficient diagram. The slip coefficient shown is for six
blades.
The velocities, C,2 proportional to the theoretical head, and Cm2 proportional to the
flow rate, fully determine the pump performance.
Several empirical correlations have been recommended in the literature between the
head coefficient and the specific speed. According to these the head coefficient would
generally decline, approximately linearly, from about 0 = 0.75 at N, = 1000 to 0 = 0.45
at Nt. = 4000 (Rey et al. 1982). Evidently, the head coefficient cannot be larger than the
slip coefficient.

The expression of the theoretical head above shows immediately whether the desired
head can be achieved with the tip speed and the head coefficient chosen. The exact
impeller diameter D2 can then be chosen to give the necessary tip speed U2:
It should be mentioned that some pump designers assume a prerotation in the flow
approaching the impeller (Spring 1983). Then the inlet velocity would have a
tangential component C,I and the angular momentum would not be zero at the inlet. The
corresponding head C,I UI /g would then have to be deducted from the theoretical head.
No such prerotation will be assumed here.
The expression of the flow rate determines the effective impeller exit width B2. The
geometrical impeller width can be made a few percent larger to account for blockage
from the blades, from boundary layers, or from separation:
Consistent units must be used:
Typical values for the head coefficient are 0 = 0.4 to 0.7, and for the flow coefficient
are 0 = 0.05 to 0.2. For high-specific-speed pumps, lower head coefficients and higher
flow coefficients are appropriate.
The choice of the head and flow coefficients from the plot of Fig. 11.2 also
determines the impeller exit blade angle /32, which in a typical pump is 68 from radial
or meridional and 22 from tangential. It corresponds to the angle of the inclined,
straight line, which represents the operating line of the pump. The line intersects the
vertical axis at a value that is numerically equal to the slip coefficient. If the number of
impeller blades is assumed-for example, six-a slip coefficient can be calculated from
Wiesner's correlation, and the intersection of the operating line and the axis on the plot
of Fig. 11.2 can be checked or modified. The line in the plot reproduces the expression
for calculating the absolute tangential velocity at the impeller exit in a dimensionless
form:
The number of impeller blades can be selected on a trial basis, usually six, give or take
one. The effect of more or fewer blades can be explored easily with the performance
calculation computer program. More blades guide the flow better, increase the slip
coefficient, and therefore increase the head somewhat. On the other hand, friction
losses are increased, and the blockage produced by the blade thickness at the inlet
could become excessive.

IMPELLER INLET DIAMETER

The value of the impeller inlet diameter D, is usually selected to minimize the inlet
relative velocity. Low velocities favor low losses (Vlaming 1989). Low inlet relative
velocities minimize the diffusion, the ratio of inlet to exit velocity W1 / W2, in the
impeller, which may lead to flow separation if excessive.
Theoretically, an optimal inlet diameter exists, as a trade-off between two opposing

trends. The vectorial sum of the circumferential velocity U, and the axial velocity at
design conditions C, adds up to the relative inlet velocity W1. (It is assumed here that
no prerotation exists.)
U, coD, = 2
Assuming a radial leading edge and an axial inlet velocity without swirl, the
circumferential velocity U, at the shroud increases with the inlet diameter, while the
axial inlet velocity C, decreases with the square of the inlet diameter for a given flow
rate. The inlet velocity C, is given by the flow rate divided by the inlet cross-sectional
area. An allowance has to be made for blockage from the blade thicknesses at the inlet,
which is usually more critical than at the impeller exit. These opposing trends balance
and make the inlet relative velocity a minimum, for a certain optimal inlet diameter
This optimal inlet diameter corresponds to an inlet blade angle NI of 54.73 from the
axial direction, or 35.26 from the tangential direction, at the shroud. The rotational
speed U, is then (2)'/z = 1.414 times the inlet velocity CI. The relative velocity
becomes W, = 1.225 U1 , and the velocity head corresponding to the relative velocity
can be related to the shaft speed N rpm and the flow rate Q gpm, or the specific speed
N, in English units and the head H ft by substituting the optimal diameter from the
expression above:

The expression must have this particular form, to balance the dimensions, as can easily
be verified. However, the coefficient of proportionality above implies the use of
English units: rpm, gpm, and feet.
If pump specifications require a low NPSHR value, then minimizing the inlet relative
velocity may not have first priority, and the possibility of cavitation must be examined
in detail. Selection of the inlet diameter, blade angles, and location of the leading edge
of the blades must proceed by trial and error. A trial design is assumed. The velocity
along the leading edge can be calculated with the procedure described in Chapter 12 for
inlet blade angle selection, which was programmed in the computer code INLET.
Knowing the inlet geometry and inlet velocities, the NPSHR value can be calculated
from the procedure described in Chapter 9. Of even greater importance than the NPSHR
at design conditions might be the NPSHR off design, at the greatest flow rate the pump
is likely to encounter. The flow will probably start to cavitate first at the shroud inlet.
As will be seen, the manufacturing requirements imposed on the detailed blade design
may require that the trial inlet geometry be changed. Cavitation calculations must then
be repeated for the new geometry.

INLET BLADE ANGLE

As discussed above, the choice of the inlet diameter D, and the velocities at the inlet C,
and U, also determine the inlet blade angle #I. However, the inlet blade angle usually
does not remain constant along the leading edge and must be calculated at several
locations. In the case of a radial leading edge, the circumferential velocity decreases
with decreasing radius along the leading edge in proportion to the radius, while the
axial velocity remains the same. Therefore, the inlet blade angle F'1, measured from
the axial direction, becomes smaller toward the axis of rotation of the impeller. The
blades become steeper.
For a first, approximate design, an inlet blade angle J31 of 55 from the meridional
direction, or up to 65 on low-specific-speed pumps, should be adopted on the mid-
streamline. The chosen axial inlet velocity usually conforms to the velocity in the
standard inlet piping, which is typically sized for a velocity of C1 = 5 to 25 ft/sec and is
given in piping handbooks. Execution of the performance calculation computer program
with a trial inlet angle will show if incidence losses have been minimized at the design
flow rate. Artificial adjustments can be made in the computer program input. The actual,
detailed inlet design should be the subject of a separate effort and must be combined
with the detailed design of the impeller blades, as described below. After a first trial, a
correction should also take into account the blockage due to the blade thickness at the
inlet. If the length of the leading edge is s, and Z blades, t thick, are present, the inlet
area in the calculation must be reduced by the amount of sZt/(cosfi,).
In practice, the leading edge of the blades is not radial but is swept back, as shown in
Fig. 11.3. The flow becomes three-dimensional. Because of the curvature of the
streamlines, the meridional inlet velocity becomes higher at the shroud than on the hub.
Often, the leading edge is not perpendicular to the direction of the inlet velocity.
However, the inlet blade angle must still be measured in the plane of the relative
velocity W, and its meridional projection C, = W,,,. Obviously, the velocities and the
blade angle will vary along the leading edge. The numerical value of the blade angle
must be calculated separately at three locations at least: the hub, midspan, and shroud.
Impeller blade drawings usually show the projection of the blade on a radial (r-)
plane, which is perpendicular to the axis of rotation. The blade angle on the drawing,
ar], measured from the radial direction, differs from the actual blade angle, /f1,
measured from the meridional direction. The meridional direction corresponds to the
tangent of the streamline in the axial-radial (r-z) plane, shown in the impeller cross
section in Fig. 11.3. If the meridional direction makes an angle a with the radial
direction, the actual blade angle is given by the relationship
Figure 11.3 Impeller inlet velocity distribution.
Therefore, the actual blade angle, measured from the meridional direction, is larger
than the blade angle in the projection.
An exact, axisymmetric potential flow calculation of the velocities and streamlines

would show that the local pressure would increase and velocity would decrease
perpendicular to the streamlines, away from the center of curvature of the streamlines,
due to the centrifugal acceleration. The smaller the radius of curvature of the streamline,
the greater will be the velocity change. High velocities will prevail at the shroud and
low velocities at the hub. A sharp curvature at the shroud will induce particularly high
velocities and should be avoided.
In the absence of a potential flow computer calculation, an approximate allowance

can be made for the decrease of the velocity toward the hub, and increase toward the
shroud, from an average velocity at the mid-streamline, which can be calculated from
the flow rate and the flow passage cross section. If Va and Vh are the velocities on two
neighboring streamlines spaced at a distance of sah and having an average radius of
curvature Rah, the ratio of velocities is given approximately by
Subdividing the space between the hub and shroud contours by several streamlines,
approximate values for the velocities can be calculated by applying stepwise the
expression above from one streamline to the next (Fig. 11.3). The resulting velocities
can be verified by calculating the flow rates between individual streamlines, or through
these streamtubes. The sum of the individual flows should add up to the total flow rate.
Such a simplified calculation procedure has been incorporated into the code INLET,
presented in the Appendix.
The selection of the inlet blade angles and the location of the leading edge may have
to be revised when the entire impeller blade is designed from inlet to outlet.
Manufacturing considerations come into play, which may require some drastic
compromises. However, blade angles in excess of 75 from meridional, or less than 15
from tangential, should be avoided. Such extreme angles would be encountered on the
shroud. The relative velocities tend to become large, and the blade passage cross
sections would become narrow. At extreme blade angles, blockage from leading-edge
separation, at off-design flow rates, would restrict the flow passages even more and can
lead to cavitation or choking of the flow.

VANED DIFFUSER AND CROSSOVER INLET

Considerable kinetic energy remains in the flow leaving the impeller, a portion of
which should be converted to pressure head with a diffuser (Fig. 11.4). In multistage
pumps the flow leaving the impeller must be returned to the inlet of the next stage, and
some pressure recovery is desirable in the process. The inlet section of the return
passage often acts as a diffuser. The inlet leading edge of the diffuser, or of the return
passage, must align with the direction of the absolute velocity leaving the impeller C2.
The exit velocity and its components in the radial or meridional and tangential
direction are known from the impeller exit velocity diagram. The kinetic energy or
head of the flow, and the desired diffuser vane angle f3 measured from the meridional
direction, are given by
In these calculations the effective width B2 should be assumed smaller than the actual
geometrical measurement, to allow for blockage. If the diffuser closely follows the
diffuser, the diffuser inlet throat must accommodate the flow rate at the exit velocity
C2, which determines the diffuser width, or vane height B3:

Figure 11.4 Vaned diffuser.
Note that consistent units must be used in these expressions, and a conversion factor
may be necessary, since velocities are usually given in ft/sec, the impeller geometry in
inches, and the flow rate in gpm [Q(in3/sec) = Q(gpm) (231 in3/gall)(lmin/60sec)].
The diffuser width B3 can be corrected for vane blockage. If there are Z3 vanes, t
thick, the width becomes
The inlet of the vaned diffuser follows closely the impeller exit, the radial distance
being perhaps 5% of the impeller radius. In the case of a return passage, more space is
left, and a correction needs to be made when calculating the inlet velocity, as explained
below.
The number of diffuser or return passage vanes usually exceeds by one the number of
impeller blades. If the number of blades were the same, all blades would pass the vanes
at the same moment, the corresponding pressure fluctuations would be synchronized,

and the blade passing noise and pressure fluctuations would be amplified. The intensity
of the blade passing noise also depends on the spacing between the impeller and the
diffuser inlet.
A radial vaned diffuser can consist of passages with straight, diverging walls,
inclined to the radial direction as shown in Fig. 11.5. The diffuser area ratio usually
remains below 2. As an alternative, the diffuser passages can be formed by airfoils, or
slightly curved, circular partitions, arranged in a circle. The location immediately after
the impeller exit, where the velocity is high, offers the greatest potential for pressure
recovery. Even a modest diffusion can have a significant effect. If the radial space is
limited, widely spaced, thin, curved partitions or airfoils are used close to the impeller
exit, to recover some velocity head.
Figure 11.5 Vaned diffuser configurations.
Return passages in vertical column pumps are typically oriented in the axial
direction, and the diameter of their centerline at the inlet D3 is larger than the impeller
diameter D2. Therefore, the tangential component of the impeller exit velocity
decreases in inverse proportion to the diameter, conforming to the principle of
conservation of angular momentum. The return passage inlet velocity C3 becomes

Often a wide space remains between the impeller exit and the return passage inlet. It
can be assumed that the fluid rotates in this space with a constant angular momentum.
This space merges with the clearance space between the hub and shroud faces of the
impeller and the housing.
Since an imposed outer diameter often limits the radial space in multistage vertical
pumps, the impeller exit can come close to the outer housing wall, and the flow must
turn sharply to enter the return passage. The meridional velocity component is usually
small; therefore, the turn does not result in significant losses. However, the proximity of
the housing wall can restrict the impeller exit, which results in more slip than would be
expected, and less head. On some radial pump designs the rear, hub face of the impeller
exit is cut back to facilitate the turn into the return passage. The impeller exit then slopes
backwards. The effective impeller diameter becomes somewhat smaller, but an optimal
compromise may still be obtained. At higher specific speeds, mixed flow impellers are
used with a slanted exit, in which the flow has a significant axial component.

VOLUTE DESIGN
The volute, as the diffuser, transforms some of the kinetic energy of the flow coming
from the impeller into pressure energy by slowing down the flow velocity. The
inefficiency of this diffusion process, in the components following the impeller, usually
accounts for the major part of the pump losses. Therefore, the volute design must be an
integral part of the pump design and should not be an afterthought.
The volute consists of a spiral-shaped passage surrounding the impeller exit, and
collects the flow from around the impeller periphery (Fig. 11.6). The volute cross
section, of circular or trapezoidal shape, increases gradually from the volute tongue to
the volute throat, where the flow exits from the pump casing. The circle, centered on the
axis of rotation and tangent to the volute tongue, is called the base circle. Its diameter is
made about 10% greater than the impeller diameter. At flow rates below the design
flow rate, some of the flow returns into the volute, passing between the impeller and the
tongue, instead of exiting through the volute throat. Therefore, the space between the
impeller and the tongue must not be made too small.
In some designs, radial space permitting, a radial vaneless space, with parallel
walls, exists between the impeller exit and the volute. In this case the volute tongue is
located at a distance more than 10% of the impeller radius. The width of the vaneless
space is made slightly larger than the impeller exit width, and only a small clearance
remains between the impeller rim and the vaneless space walls at the inlet. The purpose
of the vaneless space is to recover some pressure by slowing the flow, due to an
increase in the radius, as explained below.
Ideally, the volute cross section should accommodate the flow, which leaves the
impeller without any velocity change. This consideration implies that the cross-
sectional area A should increase in proportion to the circumferential angle 0, measured
from the tongue. However, since the center of the cross-sectional area moves out
radially to increasingly larger radii, the circumferential, or tangential, velocity of the
fluid will gradually decrease, in agreement with the principle of conserving its angular
momentum. If the center of the volute throat, the final cross section above the tongue, is
at a radius of D3/2, the ideal volute crossection can be calculated from the expression

Figure 11.6 Volute.
where
Note that consistent units must be used since the flow rate is usually in gpm, the
velocity in ft/sec, and the area in int.
In practice, the volute cross-sectional area A should be made greater by 15 to 25%

than the ideal cross section, because losses slow the velocity and boundary layers build
up. Also, it is easier to modify volute casting patterns to reduce than to increase the
volute cross-sectional area. Volute design also depends on any diffuser and pressure
recovery, which is expected to follow the volute throat. If the velocity in the volute
throat has the same magnitude as the velocity at the pump exit flange, no further diffusion
and pressure recovery can be expected. On the other hand, if an effective diffuser is to
follow the volute, it makes sense to conserve kinetic energy and keep volute velocities
relatively high in the volute. If the volute losses already slow down the velocity to its
value at the exit flange, further diffusion, or attempt of pressure recovery, is futile. If
there is no intention to recover the velocity head at the pump exit, it may as well be lost
in the volute. The exit flange of standard process pumps is often located vertically on
the centerline of the impeller, with a sharp bend between the volute throat and the exit
flange. No diffusion and pressure recovery can be expected beyond the volute throat.

MECHANICAL DESIGN
The standards of the Hydraulic Institute, ISO, ASME (B73), and API (610) prescribe
appropriate inlet and outlet pipe and flange sizes, overall housing configurations,
sufficiently strong mounting arrangements, wear ring clearance sizes, and unbalance
and vibration limits. Some of these standard measurements must be adopted directly,
without questions, to satisfy particular customers. Other portions of the standards only
prescribe the forces that the structure must withstand and leave it to the designer to
select the configuration and to ascertain sufficient strength by stress calculations.
A general description of the mechanical design procedures, in particular stress

calculations and rotor dynamic calculations, would far exceed the scope of this book
and cannot be given here. However, this omission should not be construed to belittle
their importance-on the contrary. Reliability and durability, which depend critically on
sound mechanical design, have received great emphasis in recent years. They have
become the most important concern of pump users, especially in the chemical and
power generation industry, where service and maintenance costs over the life of the
pump often exceed its first cost. A brief introduction to pump rotor dynamics is offered
in Chapter 13.
Much research has been done on rotor dynamics at the Turbomachinery Laboratory of
Texas A&M University or at ROMAC of the University of Virginia, for example
(Barrett et al. 1978, Childs 1993, Corbo and Malanoski 1998, Vance 1988).

IMPELLER BLADE DESIGN

The desired suction head and correct incidence at the pump inlet determine the inlet
blade angle N1, and the desired pump head determines the exit blade angle #2. The
major dimensions of the impeller-inlet diameter D1, exit diameter D2, exit width B2-
have been fixed. These choices have been verified by executing the performance
calculation computer program. The detailed contour of the hub and shroud and the
blade angle distribution from inlet to exit remain to be designed.
The impeller blade angle distribution will have to be determined on at least three
streamlines: the hub, shroud, and the mean streamlines shown in Fig. 11.7. In reality the
flow rarely follows these paths but can twist around from the hub to the pressure or
suction side of the blades, on to the shroud, or vice versa. Still, in a well-designed
pump, the energy imparted to the fluid is about the same on all streamlines, and at least
as an approximation, it can be assumed that the streamlines are interchangeable. Indeed,
one of the guidelines of blade angle design is to impart the same energy on each
streamline.
High-speed computer programs can calculate the flow velocities in great detail
throughout the impeller. In particular, they can calculate the angular momentum rC,
which is a measure of the energy added to the fluid at various locations along the blade.
Ideal fluid mechanic theory predicts that the ideal blade angle distribution would be
such that the angular momentum did not change perpendicular to the streamlines in the
meridional (r-z) plane (Tuzson 1993). The exact location of several streamlines and the
velocities on them would have to be calculated with the computer. Such a calculation is
often impractical, and the great effort would not be warranted in the case of most
pumps. Instead, streamlines are assumed to follow the hub, shroud, and the mean path
through the impeller. Sometimes, calculations are also made along the pressure and
suction sides of the blades. The length of the streamlines are estimated, and the blades
are designed to produce the same energy increase over the same length fraction of each
streamline. Practical blade shapes require a compromise between the ideal conditions,
the flow losses, and practical manufacturing considerations. The final design results
from several iterations. After each iteration the design is checked for inlet incidence and
cavitation, for appropriate exit conditions-in particular exit recirculation in mixed flow
pumps-and above all, for ease and cost of manufacturing. The final compromise should
conform to general industrial practices, or perhaps, even exceed these. Exceptional
performance requires a combined analysis, design, and testing program, consisting of
several iterations.

The best approach to determine the hub and shroud contours is to prepare a
preliminary layout drawing. In radial impellers, when the inlet diameter Di is about half
or less of the impeller diameter D2, the shroud inlet radius of curvature should be about
half of the inlet radius. As a first approximation one can assume that the meridional
velocity, already estimated at the inlet and exit with the performance calculation
program, declines slowly along the mean streamline. The needed approximate flow
cross section and width B from hub to shroud, perpendicular to the meridional velocity
vector, can be calculated from the meridional velocity W. and the flow rate Q at a given
radius r:
Figure 11.7 Impeller streamline locations for velocity and blade coordinate
calculations.

It will be noted that along a slanted leading edge, the increase in the tangential velocity
due to the increasing radius can compensate, at least partially, for the increase in
meridional velocity due to the curvature of the streamlines. The meridional blade angle
is not the angle shown on the usual view of the axial projection of the drawings, as
explained above.
In the case of mixed flow pumps, when the flow path in the impeller is inclined
toward the axis of rotation, the hub and shroud contours are often straight lines. There
can be a curvature at the shroud inlet to provide a smooth transition. Ideally, the radius
should be constant along the exit, the trailing edge of the blades. However, a slanted exit
is often used, especially in mixed flow pumps. It can lead to exit recirculation, as
discussed in Chapter 14.

MANUFACTURING REQUIREMENTS
Manufacturing methods profoundly affect impeller blade design. Prototype impellers
are sometimes machined. With advanced, numerically controlled machine tools, just
about any shape can be produced if cost is not an issue. Massproduced impellers are
most often cast. The metal casting and mold preparation process imposes severe
limitations on the blade design. Often, it is not feasibility but cost which decides the
use of a particular casting process. In the mold preparation process a wooden model is
made of the impeller and its blades, casting sand is packed around the impeller model,
and then the wooden model must be pulled from the mold, leaving a cavity that is an
exact replica of the form to be cast. The geometrical shape of the blades must be such
that it should be possible to pull them from the mold. Evidently, twisted blade shapes
would be impossible to remove.
When the blade shape does not depend on the axial z coordinate, the simplest blade
configurations are two-dimensional blades. The surface of all blades consists of axial
generator lines. In the radial plane the blades are generally spiral shaped and are fully
defined by the radial r and circumferential 0 coordinates. The trace of the intersection of
the blade with the hub and shroud surfaces overlap when viewed from the axial
direction. Impeller models with such blades can be pulled from the mold in the axial
direction, in one piece, with a simple movement which is easy to mechanize and
relatively inexpensive to implement. Note that the leading edge need not necessarily be
in the axial direction, nor need the hub and shroud contours be radial.
In a more complex mold configuration the blades are pulled out individually, in a
direction inclined to the axial direction. This approach offers much greater flexibility
for blade design but is more expensive to produce, requiring that each blade be handled
separately. The surface of each blade consists of straight-line generators, all of which
are inclined in the same direction, at an angle to the axial direction. The blades need to
be slightly tapered in the direction in which they are pulled, which is usually toward the
hub.
Finally, separate molds can be prepared for individual blades of quite arbitrary
shape. The molds divide into two parts, one for each face of the blade, and the blades
do not need to be pulled out. The individual blade molds can be assembled for casting.
However, handling of the individual blade molds makes the process expensive. Also,
the individual blade molds of the assembly can shift during casting, increasing the reject
rate.

Exceptionally, investment (or lost wax) casting can be used when high accuracy is
desired at any cost. A wax model of the impellers or blades is embedded in casting
sand and the wax model is melted out, leaving a cavity for the metal. Manufacturing
techniques more often used recently, such as stamping or polymer molding, impose
special requirements which cannot be described in detail here. However, the design
tools offered here are suitable for the design of such pumps-only the feasibility criteria
differ.

BLADE COORDINATES
A list of coordinates-the radius r, axial distance z, and circumferential angle 0on
several streamlines determines the complex surface of the blade shape. These
coordinates are defined on drawings by showing the intersection curves with the hub,
the shroud, and also a curve along the mid-streamline. In impellers with
twodimensional blades, the hub and shroud lines coincide, the axial coordinate z is
constant, and the other two coordinates suffice to define the blade shape. One of the
variables, usually r or z, is selected as the independent variable. Its extent, from inlet to
exit, is divided into 10 or 20 equal increments. Before the value of the other variables
can be calculated at each station, the desired stepwise increases in the angular
momentum rC, along the streamlines need to be specified.
The incremental change in the angular momentum corresponds to the pressure

difference exerted on the blade, the blade loading. Ideally, it should increase gradually
from the inlet, reach a maximum at the middle of the blade length, and taper off at the
exit (Tuzson 1993). Heavy blade loading at the inlet may increase inlet losses. Heavy
blade loading at the exit may create exit losses, pressure fluctuations, and noise.
Therefore, the angular momentum is gradually incremented from the inlet to its exit
value, R2 Cr2. The local blade angle, measured from the radial direction or from the
meridional direction in mixed flow impellers, can be calculated from the following
geometrical relationship, illustrated in Fig. 11.8:
where the angular momentum rC, is a stepwise-increasing fraction of its value at the
impeller exit:
The slip u appears only toward the exit, and gradually decreases from 1, at a distance
from the periphery along the blade equal to the blade spacing to its value at the exit. If
one were to assume an eddy relative to the impeller between the blades, which causes
the slip, its effect would extend over a distance on the order of the blade spacing. The
value of the meridional or radial relative velocity W,,, varies at the inlet from hub to
shroud, and has been calculated when the inlet angles were determined. After the inlet
bend, after the flow has turned radial-or approximately radial in mixed flow impellers
the relative meridional flow velocity can be assumed to be uniform across the flow
passage and can be calculated from the flow rate and the passage cross-sectional area,
corrected for blade and boundary layer blockage. It should be apparent that a computer
program, calculating the meridional velocity in the impeller, would facilitate the
design. However, since several iterations and adjustments are needed to arrive at a
satisfactory blade shape, the informed judgment of the designer remains indispensable
and may override theoretical considerations.
Figure 11.8 Local blade geometry and blade angle.
Having the blade angle fi distribution along the blade, the increments of the variables
4r, Az, and AO can be calculated from
where

Note that the angle must be in radians: 0 radians = 0 degrees (71/180). The tabulated
blade coordinates are suitable for computer processing and can be transferred to
plotting programs, which not only allow viewing of the blade shapes from different
directions but also uncover any numerical error and unevenness of the shapes. These
blade calculations are relatively simple but repetitive and are best executed with an
appropriate computer program. Examples of such a programs are presented in Chapter
12. Since the calculations are simple, they can easily be checked by hand.
The resulting blade contour should change gradually from inlet to exit. It is not
unusual to find that the inlet and exit blade angles are about the same, and consequently,
a constant blade angle can be maintained along the blade. Such radial, constant-angle,
logarithmic blades allow closed-form analytical flow calculations. The blade shape
calculated in this manner may not be practical. The blade wrap, the cumulative
circumferential angle from inlet 01 to outlet, 02, may be excessive. It should not exceed
100 to 120. If the hub line is different from the shroud line, the blade wrap may also be
different for each. The leading or trailing edge of the blade will then be inclined from
the axial direction. Some slant can be tolerated, but excessive inclination produces
small angles between the hub or shroud surfaces and the blade. Ideally, the blades
should be approximately perpendicular to the hub and shroud surfaces.
Finally, a check is necessary to determine whether the blade can be pulled from the
casting mold. With a computer plotting program, such a check can be performed by
plotting both the blade curve on the hub and the curve on the shroud onto the same graph
in three dimensions, and by turning the image until the two curves overlap. If the two
curves cannot be made to overlap, it will be impossible to pull the blade from the mold.
Some allowance can be made for the increasing blade thickness from shroud to hub, the
casting draft, which will be applied. The casting technique dictates the blade thickness.
Sufficient thickness must be provided for the passage of molten metal flow, which fills
the cavity left by the withdrawn blade pattern. Often, especially for mixed flow pumps,
the curves cannot be made to overlap. In such cases-or if the blade is not satisfactory for
some other reason-a revision will be needed. A study of the first trial design will
suggest possible modifications which can be tried in the next cycle of iteration.
Typically, several iterations will be needed until a satisfactory blade shape is found.
Not only the angular momentum distribution, but also the hub and shroud profile as well
as the location of the leading edge, may have to be changed. High-specific-speed mixed
flow impellers are considerably more difficult to design than radial impellers.
Compromises are necessary, and extreme specifications may be unfeasible or too
expensive, because complex casting procedures would be required.


COMPUTER CALCULATION OF BLADE COORDINATES

Simple, two-dimensional blades for radial impellers are often specified on drawings,
in the view of axial projection, by successive arcs of circles, as shown in Fig. 11.9.
The blade angle fl, here measured from the radial direction, is specified at certain radii
r. The radius of curvature R, of the arc of circle joining the locations on the two radii
r and rh, where the arc of circle forms the angles 90 - fla and 90 - fib with the
tangential direction, can be calculated from
This relationship is derived by applying the cosine rule to the triangles formed by the
radius, the radius of curvature, and the enclosed angle, which is 90 - /i.
Modern computer drafting programs allow easy construction of the blade curves and
can define the centers of curvature, as well as the circumferential locations, B, where
the arcs of circles join. The program listing BLADE2 can be implemented and executed
to calculate these variables. The computer programs IMPEL, MIXED, and BLADE2
presented in the Appendix can help design impellers and bladed return passages. The
programs are not deterministic and do not lead directly to a practical blade
configuration. Blade design requires a compromise, and the informed judgment of the
designer determines, whether the criteria have been sufficiently well met. But before the
criteria can be applied, the trial blade geometry must be fully developed. The value of
the computer program resides in executing the detailed calculations fast and allowing
several iterations and trial designs to be examined in rapid succession.

Figure 11.9 Blade shape defined by arcs of circles.
The IMPEL computer program is used to design blades that produce uniform angular
momentum distribution, and therefore uniform energy addition, on all streamlines. The
mean streamline coordinates in the meridional (r-z) plane-the flow rate, the inlet and
exit meridional velocities, and the angular momentum at the exit-are specified in the
input file. The program calculates the meridional flow passage width B and the hub and
shroud r and z coordinates from the flow rate and the meridional velocity, which is
assumed to change linearly, in equal increments, between the inlet and exit stations. The
local value of the angular momentum, following a sinusoidal distribution from zero at
the inlet to its specified value at the exit, determines the circumferential coordinates of
the streamlines. The angular momentum being defined, division by the radius, gives the
absolute tangential velocity, and subtracting the rotational speed, the relative tangential
velocity is obtained. The tangent of the local blade angle /3 corresponds to the ratio of
the relative tangential and the meridional velocities. Once the local blade angle is
known, the circumferential, wrap angle (02 - 01) can be calculated by adding up all its
increments 4O, calculated from
Figure 11.10 Hub and shroud coordinate calculation along a blade generator line in the
meridional plane.
In the output file the blade coordinates, angular momentum, and blade angles on shroud,
mean, and hub are printed out.
The MIXED program designs blade surfaces generated with parallel straight lines,
inclined to the axial direction, and starting from a reference streamline. The generator
lines and geometry are shown on projections in the r-z and r-O planes in Figs. 11.10 and
11.11, respectively. Parallel generator lines assure that the blade shape can be pulled
from the casting mold in the direction of the generators. This direction can be chosen
arbitrarily, but preferably should be approximately perpendicular to the hub and shroud
surfaces near the middle of the blade. The angular momentum increase, the energy
addition, is specified on the reference streamline. However, on the other streamlines the
angular momentum distribution will generally differ. The blade angles along the hub and
shroud streamlines are determined by the requirement of parallel generator lines, and a

particular angular momentum distribution cannot be imposed. This is where the

judgment and skill of the designer come into play, to ascertain that the energy addition
on other streamlines, although different, remains acceptable.
Figure 11.11 Hub and shround coordinate calculation along a blade generator line in the
circumferential plane.
The computer program listing MIXED in the Appendix contains the actual calculation
procedure. As shown in Fig. 11.10, first a flow passage width B in the meridional (r-z)
plane is calculated, perpendicular to the meridional velocity W-. Next the angle
between the perpendicular to the mean streamline and the chosen direction of pull in the
meridional plane are determined. The hub and shroud coordinate points in the
meridional plane will lie in the chosen pull direction, at a distance from the mean
streamline point, which corresponds to the half passage width B/2 divided by the cosine
of the angle just calculated.
The coordinates in the (r-z) plane being determined, the calculation of the
circumferential angle remains. As shown in Fig. 11.11, this time the circumferential
angles of the hub and shroud coordinate points, which lie along the projection of the
chosen pull direction in the circumferential (r-O) plane, need to be determined. The hub
and shroud coordinate locations in this projection lie at the known hub and shroud radii
along the line in the direction of pull passing through the coordinate points. Some
trigonometry and application of the cosine rule implement the numerical calculation,
which can be followed using the program listing in the Appendix.
The criterion of parallel generating lines is conservative. Skilled casting workers can

gently twist and rotate the blade pattern while they are pulled from the mold and can get
away with slight deviations from the ideal shape. The taper of the blade thickness, the
casting draft, and a fillet at the hub surface can also help. The blade is usually pulled
toward the hub. A casting trial will be made before production begins. Close
cooperation between the designer and the casting shop can avoid the need for later
modifications.
Design of the leading and trailing edges requires a further compromise. The
directions of the leading and trailing edge of the blade usually do not coincide with the
direction of the generators, as shown in Fig. 11.12. The computer output can be
modified so that other streamlines on the blade surface, than the reference streamline, do
not start or end at the first and last stations of the blade calculation. Coordinate stations
can be added to the hub or shroud lines beyond the ones calculated by the program, and
the number of stations need not be the same on all streamlines for the purposes of
preparing engineering drawings. Hand adjustments are required. These allow certain
blade angle distributions at the leading and trailing edges of the hub and shroud
streamlines, which do not affect the other streamlines and which may help approximate
the desired criteria. The computer program MIXED presents blade starting points,
which are along the straight-line generators. The corresponding blade surface needs to
be tailored and modified by hand to accommodate slanted leading or trailing edges. The
inlet blade angles along the leading edge, calculated by the MIXED program, will not
coincide with the values previously calculated, and the trailing edge blade angles may
also differ. Adjustments can be made, for example, on the hub streamline if the effective
length of the hub streamline, from leading to trailing edge, is longer than the effective
length of the shroud streamline. The hub streamline will emerge from the mold before
the rest of the blade, and the blade angles on both ends can be chosen without affecting
the rest of the blade.
An approximate estimate of the meridional velocity W,,,1 distribution at the blade

leading edge can be obtained by executing the INLET computer program. Ideally,
complete, polar, axisymmetic partial differential equations should be solved, which,
however, require that the hub and shroud contours be defined with very detailed smooth
lists of coordinates. The partial differential equations operate with the higher
derivatives of the variables. In numerical calculations the first derivatives are
approximated by the finite differences of the variables, and the second derivatives, by
the differences of these differences. Slight errors or inaccuracies from the finite
difference approximation and from the rounding off of the numbers can propagate and
amplify throughout the calculated flow field, leading to poor convergence of the
iterations. For example, the radii of curvature of the streamlines, entering into the
calculations, depend on the higher derivatives of the streamlines, the location of which
themselves are unknown and need to be calculated.
The approximate calculation procedure incorporated into the code INLET assumes
that a preliminary layout of the meridional cross section of the impeller has been
prepared, and that, therefore, estimates of the radii of curvatures of the hub and shroud
contours in the neighborhood of the inlet are available. The input file INLET.DAT
needs to contain the coordinates of the hub and shroud streamlines, the coordinates of
the leading edge on hub and shroud, the radii of curvature of the hub and shroud
contours, and the flow rate. Equally spaced streamline locations are defined across the
inlet cross section, and their radii of curvature are calculated from the hub and shroud
radii of curvature by linear interpolation. The meridional velocity, from one streamline
location to the next, increases by the factor (1+ distance between streamlines /radius of
curvature), as explained above. The program prints the calculation results, the values of
t he meridional velocity WM 1 across the inlet cross section, into the same file
INLET.DAT. The possibility of cavitation at the inlet must be reconsidered in the light
of higher shroud velocities, and a new value of NPSHR should be calculated.
Cavitation appears first at the shroud inlet.
Figure 11.12 Blade adjustments at the leading and trailing edges.
The computer program MIXED can also be used to design return passages with
stationary vanes for multistage vertical pumps. Obviously, there is no rotation w = 0,
and the absolute tangential velocity C, must gradually decrease from its value at the inlet
to the outlet. The vanes at the exit must point in the axial direction to eliminate any swirl
that may enter the next stage. The meridional contour takes on an "S" shape, from axial
at the inlet to axial again at the exit, near the shaft. In the case of low-specific-speed
pumps, when the flow passages are relatively narrow, the return passage can be made
radial and simple twodimensional vanes become possible. The inlet leading edge is
sometimes at an extreme slant, reaching over the impeller exit, along the outer housing
wall.
The blade shapes must somehow be communicated to the patternmaker. In computer-

based drafting, the blade coordinates, r, z, and 0 are specified on the hub and shroud
contour. The coordinates are listed on the impeller drawings but should preferably be
transferred to the patternmaker by the computer disk, because data input labor can then
be minimized and numerical errors avoided. Older practice, still often used, requires
that several blade cross sections be shown on the impeller drawings. The cross sections
are taken in planes defined by a constant axial coordinate, z = constant, the location of
successive cuts being at axial coordinates z incremented by equal amounts. The
intersection of such a plane with the blade surface is defined by the curve relating the
radius r to the circumferential angle 0. The patternmaker builds up the blade model from
layers of parallel equal-thickness boards, which are cut to the required profile, and
stacked. The assembly is then sanded smooth.
The computer program CUT presented in the Appendix executes such cuts. The blade
shapes are defined by the intersections of the generator lines and the hub and shroud
surfaces, which are the hub and shroud coordinates given in the input file. The first step
in the calculation procedure consists of determining the r coordinates corresponding to
the chosen z coordinates in the merdional r-z plane. The r coordinate is given by the
intersection of the z plane and the projection of the generator line between the hub and
shroud points. In the circumferential r-0 plane the circumferential 0 coordinate is given
by the intersection of the generator line projection between hub and shroud points and
the circle of the previously calculated radius r. The actual angle is calculated using the
sine rule. Calculation details are contained in the program listing.
When the blade design has been completed satisfactorily, the entire design process
should be repeated, and key variables should be checked by hand calculations to
ascertain that the simplifications that have been made in the design process have not
distorted the geometry, that no mistakes were made, and that the design will indeed
perform satisfactorily.


REFERENCES
Barrett, L. E., Gunter, E. J., Allaire, P. E. (1978): Optimum Bearing and Support
Damping for Unbalance Response and Stability of Turbomachinery, ASME Journal
of Engineering for Power, January, pp. 89-94.
Childs, D. (1993): Turbomachinery Rotordynamics, Wiley, New York.
Church, G. (1944): Centrifugal Pumps and Blowers, Wiley, New York.
Corbo, M. A., Malanoski, S. B. (1998): Pump Rotordynamics Made Simple,

Proceedings of the 15th International Pump User Symposium, Texas A&M
University, College Station, Texas, pp. 167-204.
Hydraulic Institute (1994): Centrifugal Pump Standards, HI, Parsippary, N.J.
Oak Ridge (1972): Design and Procurement Guide for Primary Coolant Pumps in
Light- Water-Cooled Nuclear Reactors, Government Report ORNL-TM- 3956,
November.
Peck, J. F. (1968): Design of Centrifugal Pumps with Computer Aid, Proceedings of

the Institution of Mechanical Engineers, Vol. 183, Pt. 1, No. 17, pp. 321351.
Pfleiderer, C. (1961): Die Kreiselpumpen, Springer-Verlag, Berlin.
Rey, R., Kermarec, Y., Guiton, E, Vullioud, G. (1982): Etude statistique sur les
caracteristiques a debit partiel des pompes centrifuges et sur la determination
approximative du debit critique (Statistical Study of Partial Flow Characteristics of
Centrifugal Pumps and of Approximate Determination of the Critical Flow), La
Houille Blanche, No. 2/3, pp. 107-120.
Salisbury, A. G. (1983): Current Concepts in Centrifugal Pump Hydraulic Design,

Proceedings of the Institution of Mechanical Engineers, Vol. 197A, October, pp.
221-231.
Spring, H. (1983): A Comprehensive One-Dimensional Streamline Theory Based on

Euler, with Pfleiderer's Modification for Slip and Hydraulic Efficiency, Suitable for
Centrifugal Pumps with Prerotation, in Performance Characteristics of Turbines and
Pumps, ASME FED Vol. 6, pp. 141-147.

Spring, H. (1984): The Impeller Inlet Diameter Method to Justify Proper Selection for
Pumps, ASME Paper 84-WA/FM-1.
Turton, R. K. (1994): Rotordynamic Pump Design, Cambridge University Press, New

York.

Fluids Engineering, September, pp. 463-467.
Vance, J. M. (1988): Rotordynamics of Turbomachinery, Wiley, New York.
Vlaming, D. J. (1989): Optimum Impeller Inlet Geometry for Minimum NPSH

Requirement for Centrifugal Pumps, in Pumping Machinery, ASME FED Vol. 81,
pp. 25-28.

12

DESIGN EXAMPLES
PRELIMINARY DESIGN AND PERFORMANCE ESTIMATE

The objective of this chapter is to illustrate the design process with numerical
examples. Since a first design trial is verified with the help of the LOSS3 performance
prediction computer program, and since the computer programs in the Appendix are
used to calculate the detailed blade geometry, in many instances the symbols of the
FORTRAN listings are also used in the text, with the expectation that this approach
will facilitate interpretation of the computer printouts. In the text the symbols are
preceded by their verbal description. A full nomenclature can be found in the
Appendix, and often in the listings themselves.
As an example, the design of a pump producing 100 ft (30.5 m) head at 2000 gpm
(0.126 m3/s) when driven at 1770 rpm is considered. It is a vertical type with a mixed
flow impeller and combined diffuser and return passage. The foregoing specifications
correspond to a specific speed of 2500 in English units. The efficiency of similar
commercially available pumps would be 86% based on the specific speed and flow
rate. Assuming a head coefficient 0 on the order of 0.5, the relationship between head
Ho and tip speed shows that a tip speed of about 86 ft/sec (26.32 m/s) might be
appropriate. At 1770 rpm such a tip speed would be reached with an I 1-in.-diameter
impeller.

Assuming a flow coefficient 0 of 0.15, an impeller width B2 of 1.4 in. would result in
a flow rate of 2000 gpm.
As a start, the optimum inlet diameter, corresponding to an inlet blade angle (31 of
54 from meridional, should be calculated from
The exit blade angle turns out to be about 65 from meridional or 25 from tangential,
whether read from the head-flow coefficient plot or-assuming a slip coefficient of
0.82-calculated from
The return passage average diameter D3 should be 12.75 in. (0.324 m) which is 14%
more than the impeller diameter, and its vane height B3 should be 1.0 in. (2.54 cm),
somewhat smaller than the impeller exit width. In these calculations the flow passage
widths correspond to the effective width available to the flow. The geometrical width
may have to be somewhat larger to allow for blade and boundary layer blockage.
These measurements must be checked against the available space on a layout. Diameter
limitations may be imposed by the maximum permissible outer diameter of the casing.
As a first try, the diffuser inlet angle can be taken equal to the angle of the flow leaving
the impeller, which is given by
To complete the input values for the performance calculation computer program
LOSS3, six impeller blades and seven diffuser vanes are assumed. The loss
coefficients should match the loss coefficients of a similar pump on which the program
should have been calibrated previously.
The printout of the performance calculation computer program LOSS3 in Fig. 12.1,
with the foregoing values of the input variables, shows the following:
1. The maximum efficiency, only 81.21%, is at 1500 gpm (0.0946m 3/S), where
the diffuser approach velocity C3 = 43.01 ft/sec (13.14 m/s) and the diffuser
throat velocity CQ3 = 41.81 ft/sec (12.74 m/s) are reasonably well matched.
The diffuser throat velocity at 2000 gpm (0.126 m3/s) is excessive: CQ3 =
55.75 ft/sec (17.0 m/s), compared with the approach velocity C3 = 38.64 ft/sec
(11.78 m/s). The high diffuser throat velocity also accounts for the high diffuser
loss DQVD = 15.15 ft (4.6 m). Therefore, the diffuser throat needs to be opened
up, which is accomplished by decreasing the inlet angle /33 =BETA3 from
73.3 to about 70.0, and increasing the diffuser width B3 from 1.0 to 1.2 in.
(2.54 to 3.05 cm).
2. The head is low at 2000 gpm: DH = 83.22 ft (25.36 m). Therefore, the impeller
diameter D2 could be increased from 11.00 in. to 11.60 in. (0.28 to 0.295 m),
but the exit blade angle #2 = BETA2 is increased to 68, and the impeller exit
width B2 is changed to 1.300 in. (3.3 cm).
3. The inlet incidence is perfect, the loss DQIN12 being 0.01 ft at 2000 gpm, the
design point. However, the inlet absolute velocity Cl = 29.08 ft/sec (8.86 m/s)

is high and could cause cavitation problems, if that were an issue. According to
the expression given for design conditions, the NPSHR value would be

Figure 12.1 Printout of the performance calculation program LOSS3 for the first
iteration for a pump design example.

Correct incidence will still be maintained if the inlet diameter is made D1 =6.5 in.
(0.165m), and the inlet blade angle #( = BETA! = 68.0.
The printout of the second iteration, in Fig. 12.2, with corrected input values, shows
that the maximum efficiency, which increased to 86.24%, is now at 2000 gpm (0.126
m3/s) where the diffuser velocities C3 and CQ3 are equal. The head is 99.72 ft
(30.4m), and the absolute inlet velocity CI dropped to 19.34 ft/sec (5.89 m/s), from
which an NPSHR value of 5.8 ft (1.77 m) could be calculated. The performance
calculation code LOSS3 does not estimate NPSHR. These values meet the design
specifications.

Figure 12.2 Printout of the performance calculation program LOSS3 for the second
iteration for a pump design example.
Additional runs, with slight modifications, can ascertain whether optimal conditions
have been reached. It must be kept in mind that the absolute value of the quantities
calculated by the LOSS3 code can differ by 1 or 2% from actual values measured in
tests, even if the loss coefficients have been checked against the test data from a similar
pump. The program does not make allowance for boundary layer and blade thickness
blockage, which would require that the actual geometrical measurements of the flow
passages be somewhat larger than the calculated effective cross-sectional areas. Still,
trial runs can show in which direction changes in input variables affect the calculated
values, even if the absolute values are in doubt. Evidently, other, perhaps more detailed
performance prediction codes could also be used instead of the LOSS3 code, but
usually they will require a greater number of input variables, which would need to be
chosen before the program can be executed.

INLET VELOCITY CALCULATION

A simple estimate of the meridional velocity WM1 distribution across the inlet cross
section can be obtained by executing the INLET computer program. The input file
1NLET.DAT contains coordinates r and z of the hub and shroud contours, their radii of
curvature near the inlet, the coordinates of the leading edge on hub and shroud, and the
flow rate. In this example a radial impeller configuration was assumed.
The calculation example gives the radii of curvature RHOH = 4.75 = 12 cm, and
RHOS = 1.0 = 2.54 cm, and a flow rate Q = 1000 gpm (0.063 m3/s). The calculated
meridional velocities across the inlet cross section, printed in file INLET.DAT and
shown in Fig. 12.3, range from 6.374 ft/sec (1.942 m/s) on the hub to 20.620 ft/sec
(6.285 m/s) on the shroud. Evidently, the high local velocity on the shroud will have to
be taken into account in designing the blades if correct incidence at design conditions is
to be achieved. The shroud velocity will also affect the appearance of cavitation and the
NPSHR value of the pump. New values of NPSHR, with updated shroud velocity
estimates, should be prepared. Cavitation will appear first at the shroud inlet.
Figure 12.3 Input and output file of the INLET computer program, calculating the inlet
velocity distribution for a pump impeller design example.
The designer must exercise good judgment as to whether to accept very high shroud
velocities. In the case of a very small shroud radius of curvature, especially if the flow
decelerates rapidly on the shroud beyond the inlet, flow separation from the shroud

contour should be suspected. Since the relative flow velocity also has a strong
tangential component, separation can take the form of a leading-edge vortex, especially
at off-design conditions at incidence. A very complex separated region and embedded
trailing vortex may appear along the shroud and blade corner which defy accurate
analysis and mathematical description. Preferably, strong curvature on the shroud
should be avoided.
The average of these calculated meridional velocities, 9.93 ft/sec (3.027 m/s),
compares with a value of 10.2 ft/sec (3.109 m/s) calculated from the flow rate and the
inlet cross-sectional area as if the velocity were uniform. The close agreement shows
that in this case the flow passage cross sections can be estimated sufficiently accurately
from an average meridional velocity and the flow rate, even when the streamlines are
curved.

DETAILED THREE-DIMENSIONAL BLADE DESIGN

Since the values of the preliminary variables have been determined, detailed design of
the impeller blades can begin. The blade coordinate calculation code IMPEL requires
the input values of 21 coordinate points of the mean streamline in the meridional (r-z)
plane through the impeller. The flow rate, rotational speed, inlet velocity, meridional
exit velocity, absolute tangential velocity, and slip are available from the printout of
the performance calculation program LOSS3. In addition, the blade thickness T needs
to be specified. The blade thickness only enters into the calculation of a blockage when
the meridional velocity is estimated from the geometrical cross section of the flow
passage. Blade thickness will be on the order of 0.125 to 0.250 in. (3.175 to 6.35 mm)
and should be selected in consultation with the casting manufacturer, since it has to
allow a satisfactory flow of the molten metal.
As expected, the printout of the IMPEL program (Fig. 12.4) shows that indeed the
angular momentum distribution on hub, mean, and shroud is the same. The wrap angle
from inlet to exit varies from 82 on the hub to 100 on the shroud. A three-dimensional
plot of the blade coordinates would, however, show that hub and shroud lines cannot be
made to overlap, regardless of the direction from which they are viewed. Therefore, the
blades could not be pulled from a mold, and individual blade molds would be the only
practical casting method. The individual blade molds would then have to be assembled
and joined to form the impeller mold.
If the blades are to be pulled individually from an impeller casting mold, the
computer program MIXED would have to be used. The direction of the pull, which will
also define the straight-line generators, connecting corresponding points on hub and
shroud, must then be chosen. In the circumferential direction the direction of pulling is
made to coincide with the circumferential angle of the tenth coordinate point of the mean
streamline. In the meridional (r-z) plane the direction of pulling is defined by the tangent
of the angle with the axial direction TNA. The direction of pulling should be
approximately perpendicular to the mean streamline at the tenth coordinate point. For
pure radial flow, the direction of pulling would be in direction of the z axis, the axis of
rotation; therefore, TNA = 0.0. All blades of a radial impeller can be pulled together in
a single action. The mean streamline direction does not need to be radial for the blades
to be pulled together in the axial direction.

Figure 12.4 Output file of the computer program IMPEL, calculating blade coordinates
for a pump impeller design example, having the same angular velocity distribution on
all streamlines.
In the case of the design example considered here, the mean meridional streamline
was chosen to be a straight line inclined from the radial direction. Coordinate points
were defined by subdividing the distance between inlet and exit in 20 equal increments,
0.1375 in. (0.35 cm) in the r direction and 0.077 in. (0.2 cm) in the z direction. The
inclination of the direction of pulling to the z axis was selected as TNA = tan 20.
The computer program also allows a choice of angular momentum distribution on the
mean streamline (Fig. 12.5). A sinusoidal distribution is preferred which starts and ends
gradually, in order to unload the blade leading and trailing edges. An exponent of the
distribution function XN can modify the shape of the curve. An exponent smaller than 1
increases blade loading toward the front; an exponent greater than 1 increases loading
toward the end of the blade.
Figure 12.5 Angular momentum distribution function from pump inlet to exit.
The printout of the computer program MIXED (Fig. 12.6), executed with an exponent
XN = 1.0 and with the input above, shows (1) the three polar blade coordinates r, z,
and 0 for each streamline, the shroud or tip, the mean, and the hub, (2) the angular
momentum Cr distribution along the three streamlines, and (3) the local blade angles Q.
The following observations can be made:

Figure 12.6 Input and output files of the computer program MIXED, calculating blade
coordinates for a mixed flow pump impeller design example having parallel straight-
line blade surface generator lines.
1. An obvious but easily overlooked requirement is that the sense of rotation of the
impeller be correct. The sense of the blade can be adjusted by counting the
circumferential angle 0 either clockwise or counterclockwise.
2. The circumferential wrap angles, about 80 on the shroud and about 110 on the
hub, are reasonable. At the inlet the blade is inclined with respect to the
meridional plane but is only slightly slanted at the exit since the circumferential
angles on hub and shroud differ by only 4.
3. The angular momentum distribution shows that there is a positive incidence at

the shroud and a negative incidence at the hub. At the exit about 15% more
energy is added to the fluid on the shroud and 15% less on the hub. Leading- and
trailing-edge adjustments would be needed if equal-energy addition on all
streamlines were desired. The streamline curvature at the inlet will alleviate the
incidence problem somewhat. The meridional inlet velocity will be higher than
the average at the shroud and lower at the hub, which will reduce the incidence
in both locations. A detailed inlet flow calculation should be made at this point
to determine the actual meridional velocities on hub and shroud. Local blade
angle changes can be made by thinning the blade on the pressure or suction side.
At a blade angle of 68, a 20/ 1000 in., thickness change over a length of 1 in.
(0.5 mm over 25 mm) corresponds to a 1 change in the blade angle. Actual
blade angles are very difficult to measure, especially since the effective angle
should be taken along the streamline, the exact direction of which is usually
unknown. The front and exit ends of the blade can be adjusted on the hub by
adding coordinate points. If the blade is pulled toward the hub, additional blade
length on the hub does not affect the shroud.
4. It is important to remember that the blade angle /3 is measured from the

meridional direction, not the radial direction.
The blade curve shown on a projection onto a plane perpendicular to the axial
direction, the r-O plane, as customary on engineering drawings, does not form an
angle /3 with the radial direction, but an angle /3r. The following relationships
exist between the meridional blade increment Am, the radial increment Ar, and the
circumferential increment rAO:
In these expressions the circumferential angle increment AO must be measured

in radians, not degrees.
5. Alternative blade designs can be executed with different angular momentum

distributions XN, with different directions of pulling TNA, including an axial
pull TNA = 0.0, and with different mean streamline coordinates. The closer the
inclination of the blade surface generating lines to the axial direction, the less
will be the difference in angular momentum on the streamlines. Ideally, the
blades should deliver uniform angular momentum distribution at the impeller
exit, but often a compromise will be necessary. The designer's judgment will
decide whether the compromise is acceptable and satisfactory.
Plots of the blade coordinates calculated here, which are available in the file
MIXED.GRF and can be transferred to a plotting program, show that the blade
shapes are smooth and error free (Fig. 12.7). They can be made to overlap when
rotated and viewed from the direction in which they would be pulled. For the
purposes of preparing engineering drawings, the intersections of the blade surface
with planes perpendicular to the z axis, r-6 planes, can be prepared with the help
of the CUTS computer program. The input file CUTS.DAT, containing the blade
coordinates on the hub and shroud in free form, can be prepared, for example, by
editing the file MIXED.DTA from the MIXED computer program. The file needs
to be renamed and the text deleted. Four cuts are calculated, equally spaced
between the shroud inlet location and the hub exit. The coordinates r, z, and 0 of
the six cuts are listed in the output file CUTS.OUT (Fig. 12.8) and are suitable for
transfer to a plotting program.

Figure 12.7 Plots showing impeller blade coordinates at the hub and shroud for the test
case of Fig. 12.6.

TWO-DIMENSIONAL BLADE DESIGN

At lower specific speeds, a two-dimensional blade shape becomes possible. The blade
surface does not depend on the axial z coordinate and can be fully defined by the radial
r and circumferential 0 variables. Such a blade shape becomes possible if the flow
velocities along the hub and the shroud are essentially the same. The blade coordinates
can still be calculated with the MIXED computer program by making the blade surface
generator lines parallel to the axial direction, and setting TNA = 0.0. The blade surface
is formed with axial generator lines, and the blade can be pulled in the axial direction.
Figure 12.8 Sample printout of the computer program CUTS, calculating sections of
blades in consecutive circumferential planes for the test case of Fig. 12.6.
After executing the MIXED program with the input as above and TNA = 0.0, the
printout in Fig. 12.9 shows identical radial coordinates for the corresponding points of
the three streamlines. A slight difference appears in the angular momentum distribution
and in the blade angles because the meridional velocity changes from inlet to outlet and
the hub and shroud streamlines are not parallel to the mean streamline in the meridional
(r-z) plane view. The blade angles are measured along the streamlines, as before.
Figure 12.9 Printout of the computer program MIXED, calculating blade coordinates for
a mixed flow pump impeller design example having axial blade surface generator lines.
Figure 12.10 Printout of the computer program BLADE, calculating radii and centers of
curvature for defining by arcs of circles the blades of a radial flow pump impeller
design example.
If the impeller flow passage and the streamlines are substantially radial, a very
simple blade shape can be specified by executing the BLADE2 code, which defines the
blade by segments of arcs of circles. The input file BLADE.DAT requires the inlet and
outlet angles and five radii at which the arcs of circles join. The output, shown in Fig.
12.10, lists the radii, the blade angles, which are incremented from inlet to exit in
proportion to the radial increments, the circumferential angle, the radius of curvature,
and the coordinates of the centers of curvature of the arcs of circles. In this case the
blade angles calculated are indeed the angles in the plane (r-6), perpendicular to the
axial direction.
The computer program assumes an axial, constant-radius blade leading edge. In

reality the leading edge in practical impellers can be chosen inclined. The radius at the
shroud inlet is larger than the radius at the hub inlet. Therefore, the blade surface should
be calculated starting at the hub inlet radius, and the blade should be trimmed at the
shroud. Since the blade angle at the hub inlet does not affect the shroud inlet, it can be
chosen at will. However, when the radius of the blade surface reaches the radius of the
shroud inlet, the blade angle must take on the value that will give correct incidence at
the shroud inlet. As before, the leading edge of the blade needs to be adjusted. Even in
purely radial impellers, the flow enters the impeller eye in the axial direction and turns
toward the radial direction in the impeller. Therefore, the meridional velocities at the
hub and shroud will not be the same. They must be calculated separately, and the blade
angles at the inlet must be adjusted to give correct incidence. In low-specificspeed
pumps, relatively low velocities exist at the impeller inlet, and incidence losses remain
relatively low, which can be ascertained from the performance calculation program
printout. Therefore, in this case, less ideal conditions can be tolerated at the impeller
inlet.

RETURN PASSAGE DESIGN

Assuming that a return passage follows the mixed flow impeller (Fig.12.11) which was
calculated previously, the computer program MIXED can again be used to calculate the
diffuser vane coordinates. The sense of direction of the calculation needs to be
reversed. The return passage exit, at the entrance of the next stage, becomes the first
station to be calculated with the positive direction of the axial, z coordinate pointing
backwards with respect to the direction of the flow. The return passage inlet, near the
impeller exit, becomes the last station to be calculated. The input variables must be
assigned accordingly.
The coordinates of the mean streamline, r and z, which need to be input, will follow
an S-shaped curve starting axially at the average radius of the impeller inlet, and ending
again axially at the return passage radius D3/2 which has been determined in the
performance calculation with the code LOSS3. The zcoordinate points correspond to 20
equal increments. The radial coordinate remains constant for two or three stations at the
inlet and exit, and follows a constant slope, for example, 45, in between. The slope
changes need to be rounded off. The program MIXED smoothes the input data with a
simple averaging over three stations.
As to the other input variables: the flow rate Q = 2000 gpm (0.126 m3/s), rotational
speed N = 0.0 rpm, the vane thickness T = 0.25 in. (6.35 mm), number of vanes XNB =
7, inlet meridional velocity WM 1 = 19.34 ft/sec (5.98 m/s), and exit meridional
velocity WM2 = 13.35 ft/sec (4.07 m/s) are available from the performance calculation.
The absolute tangential exit velocity CT21 must be calculated from the impeller exit
tangential velocity CT2, given in the printout of the performance program, by accounting
for the increase in the radius from impeller exit to return passage inlet:

Figure 12.11 Return passage.
The inclination of the direction, in which the vane would be pulled from the casting
mold, GAMAREF = 45, was chosen approximately perpendicular to the mean
streamline. The exponent of the tangential velocity distribution XN = 3 will load the
vanes more at the end, where the flow enters the return passage. No slip would be
expected in stationary passages: SIGMA21 = 1.0.
The printout of the MIXED code and the plot of the r and z coordinates, with the
foregoing input for the return passage in Figs. 12.12 and 12.13, may show that the first,
trial meridional streamline input could have used some modifications at both vane ends.
A generous fillet at the hub and large radius of curvature at the shroud would streamline
the contours. Slight modifications would not be expected to affect the flow velocities,
especially since boundary layers and blockage have been neglected. The vane wrap
angle varying from 50 to 75 is reasonable but will change if the vane ends are trimmed.
The tangential velocity changes gradually, and little difference exists from one
streamline to the next. The local vane angles could use some smoothing, but all reach
70 at the end, where the flow from the impeller enters, in agreement with the value that
was calculated with the performance program. The three-dimensional view shows that
the hub and shroud streamlines indeed overlap if viewed from the direction in which
they would be pulled.

VOLUTE DESIGN EXAMPLE

For the sake of illustrating the design procedure, the principal dimensions of a volute
will be chosen for the impeller delivering 2000 gpm (0.126 m3/s) at 100 ft (30.5 m)
head and rotating at 1770 rpm, designed previously. The impeller has an 11.6-in.
(0.295-m) diameter D2 and an exit width of B2 = 1.3 in. (3.3 cm). The diameter of the
base circle tangent to the volute tongue will be made about 10% larger than the
impeller, about 12.75 in. (0.324 m). Pressure fluctuations at the tongue when the
individual blades pass by will decrease with increasing base circle diameter.
The volute throat area is determined by matching the velocity coming from the
impeller exit with the throat velocity calculated from the flow rate. At design
conditions, Q = 2000 gpm (0.126 m3/s), the absolute tangential velocity leaving the
impeller has been calculated: CT2 = 40.42 ft/sec (12.32 m/s). To calculate the volute
throat cross-sectional area, let us assume for a first estimate that the velocity at the
throat has slowed down because of the increased radius to CQ3 = 36.75 ft/sec (11.20
m/s). The throat area A, assumed to be circular with diameter d, would then be
calculated from

Figure 12.12 Printout of the computer program MIXED, calculating the vane
coordinates of the return passage of a mixed flow pump impeller design example.

Figure 12.13 Plots showing the hub and shroud coordinates of the return passage vanes
of Fig. 12.12.

Assuming a circular volute flow passage cross section, the throat area diameter will
bed = 4.7 in. (12.0 cm). However, the center of the throat area will move to a radius of
(base circle diameter + d)/2, and the velocity approaching the throat will slow down in
inverse proportion to the distance from the impeller center. Therefore, the throat
velocity will diminish to
Therefore, the corrected area will become
Depending on the actual shape of the throat cross section, which could approach the
rectangular or trapesoidal, and the location of the average streamline, slightly different
throat areas would be calculated. The calculated area corresponds to an effective flow
cross section. In practice, the geometrical area should be 15 to 25% greater because of
boundary layer build up in the volute and because wall friction and sudden expansion
losses tend to slow the velocity. A larger volute throat will tend to shift the best
efficiency point to higher flow rates and can reduce diffusion losses at the pump exit,
which, however, will depend on the effectiveness of any diffuser, which may follow
the volute.
The volute flow cross sections at other locations around the circumference of the
impeller vary in proportion to the circumferential angle 0 measured from the tongue:
The width of the volute in the axial direction immediately at the impeller exit should
ideally be slightly larger than the impeller exit width, with close clearances between
the impeller sides and the volute walls. However, high efficiencies have been reached
with much wider widths. An obvious but easily overlooked issue in volute design is
the sense of rotation of the impeller.

Mechanical design considerations strongly influence volute design. The volute

passages, being under high hydraulic pressure, exert strong forces on the casing
structure, and some of the passage walls remain weakly supported. Industrywide
standards often prescribe the location of the inlet and exit flanges, mounting bolts, and
external loads, which the bolts have to support. Manufacturing, assembly, and
installation consideration also come into play.

CONCLUDING REMARKS
The simple design tools offered here-the computer programs LOSS3, INLET, IMPEL,
MIXED, CUTS, BLADE2-can lead to a first trial design, which can then be refined at
will. The data may serve as input to more detailed and refined calculation procedures
and computer programs. The degree of refinement, and the time spent on them, will be
governed by the value of the pump. Hydraulic design is only a modest part of the
overall design effort. It must be supplemented by rotor dynamic and stress analysis
calculations, and finally, by a manufacturing cost analysis. These other requirements
may impose constraints on the hydraulic design and make a new hydraulic design
iteration necessary.

13

ROTOR DYNAMICS
OF PUMPS
FUNDAMENTALS OF SHAFT VIBRATIONS

A general overview of pump rotor dynamics can be found in the books of Vance
(1988), Childs (1998), and the publication of Corbo and Malanoski (1998), for
example. Only a brief overview is given here.
Any mass held in position elastically is capable of oscillations if its position is

disturbed. Because of force balance, the inertia force, the mass m times acceleration,
must be balanced by the spring force, the spring rate k times the displacement. The
spring constant k has the dimension of force per unit length. On dimensional grounds the
frequency of oscillation f must be given by the following proportionality, the constant of
proportionality depending on the detailed configuration of the system:
f ( k li/z m)
When an outside excitation has the same frequency as the natural frequency of the
system, resonance is encountered. The amplitude of oscillations increases
progressively. The maximum amplitude can be limited by friction or some other
energy-dissipating force that extracts energy from the oscillating system. Operation at,
or passage through, the critical speed becomes possible if sufficient damping maintains
deflections sufficiently low (Barrett et al. 1978). Otherwise, the amplitude of
oscillations will increase until the system destroys itself.
In the simplest case of a rotating system, the lateral deflection of the shaft will result
in a restoring spring force, and the mass of the rotor, usually much greater than the mass
of the shaft, will stand for the mass in the expression above (Childs 1993, Vance 1988).
The rotational frequency, w/2it, will excite the oscillation. The system reaches its
critical speed when the rotational frequency equals the natural frequency of the system.
The magnitude of the exciting force will be in proportion to the unbalance of the rotor.
Ideally, if rotors could be perfectly balanced, no excitation could take place.
Unfortunately, some unbalance always remains. Standards by the Hydraulic Institute,
ISO, ASME, and API prescribe the permissible unbalance of various rotating
assemblies.
Static balancing can be accomplished by placing the rotating assembly on knife edges
or by mounting on low-friction bearings, and by letting the assembly rotate to its
equilibrium position under the effect of gravity. The direction of unbalance, which is the
eccentric position of the center of gravity of the rotor with respect to the axis of rotation,
then points downward. Adding small weights, by trial and error, in the opposite, upper
position, a balanced, neutral equilibrium state can be reached in which the rotor remains
in any angular position without moving.
When the rotating mass is long, an unbalance moment can still remain from eccentric
mass distributions on its ends, which tend to produce a longitudinal bending torque on
the shaft. Such unbalance moments can also be produced by large-diameter rotors,
which are not mounted perpendicular to the rotational axis or by hydrodynamic forces
on wobbling impellers. Balancing machines and procedures exist in industry to detect
and correct not only static unbalance, but also unbalance torques resulting from
eccentric mass distributions. Shaft bow or distortion can also excite oscillations, and
these act in a manner similar to unbalance. Depending on the relative directions of
bowing and unbalance, their combined effect can give rise to responses that can be
different from those of unbalance only.
For the purposes of an analysis, the instantaneous transverse deflections of the shaft
should be separated into two components, corresponding to the two perpendicular
directions of the force components and displacement variables. Ideally, vibrations in the
two directions are independent. However, hydrodynamic forces in bearings and in close
clearances, such as seals, introduce a coupling between the two directions which must
therefore be considered jointly. Indeed, the deflection in a sleeve bearing, which
produces an eccentric position of the shaft with respect to the bearing bore, results in a
bearing force that is not directed opposite to the deflection as an ordinary spring would,
but also produces a force component in the transverse direction. Bearing supports, on
the other hand, can deflect elastically, producing forces directly opposed to, and
proportional to, the deflection. Such a support flexibility is equivalent to greater
flexibility of the shaft.
A further complication is introduced by the fact that complex rotating systems, having
several rotors or long shafts, for example, have more than one critical speed. The
system has several modes or configurations of oscillation, each having a different
natural frequency. Such natural frequencies can be detected on existing assemblies with
vibration-measuring instruments but are difficult to predict exactly during the design of
the pump assembly. The numerical rotor dynamic analysis of such complex systems
requires the use of modern computer-aided calculations.
In pump design and operation, shaft vibrations affect two issues. Excessive shaft
deflections may cause metal-to-metal contact in bearings, wear rings, or close
clearances that could damage the pump. Excessive shaft oscillations may generate
vibrations that may overload the bearings-rolling element bearings also-and may even
cause structural damage. Consequently, it is important to estimate the shaft deflections
and vibration amplitudes when designing pumps. Shaft oscillation at or near the critical
speed can be limited by sufficient damping at the bearings or their support (Barrett et al.
1978).
Other important mechanisms exciting oscillations below the first critical speed are
intermittent, once per revolution, rubbing, or loss of contact at bearings or seals (Childs
1993). The oscillations appear at an exact fraction of the running speed, usually one-half
but at times one-third or one-fourth of the speed. Ball bearings, for example, remain stiff
as long a contact is maintained between the balls and the races. However, unless the
bearings are preloaded, some clearance always exists, and contact can be lost
temporarily and periodically during each revolution.
An important distinction must be made between shaft deflections that are

synchronous, always pointing in the same direction with respect to the rotating shaft, and
deflections that rotate with respect to the shaft. Unbalance and shaft bowing excite
synchronous deflections. Oscillations excited by other instabilities, such as rubbing, are
not synchronous. The corresponding deflections rotate with respect to the shaft, flex the
shaft periodically, and create alternating bending stresses in the shaft. Internal friction
can be produced between the bending shaft and an impeller pressed onto the shaft,
which can also affect shaft dynamics (Childs 1993).
Several mechanisms can cause instability and excite oscillations at frequencies

below the rotational frequency of the shaft (shaft whirl) if the assembly is operated
above its first critical speed (Ehrich and Childs 1984). Such might be the case for high-
speed boiler feed pumps or rocket fuel pumps but are less likely to be encountered in
general industrial applications.

SHAFT AND IMPELLER SUPPORTED ON TWO BEARINGS

An impeller on a shaft, supported by two bearings, constitutes the simplest
configuration, which is encountered frequently. Two cases are illustrated here (Fig.
13.1): an overhung impeller corresponding to an end suction pump, and an impeller and
shaft assembly supported by bearings on either side, corresponding to a double suction
pump. Ideally, the bearings can be assumed rigid, which may be the case for preloaded
ball bearings, for example. If, in addition, the effect of the front wear ring forces and
the damping effect of the water surrounding the impeller is neglected, the system can be
modeled by the concentrated mass of the impeller, including the water contained in it,
and the bending spring rate of the shaft. A first, approximate estimate of the natural
frequency of the impeller and shaft assembly can be calculated. Usually, the natural
frequency is well above the frequency of the shaft rotation. If the shaft were of uniform
cross section, the shaft spring rate for overhung impellers would be given by

Figure 13.1 Rotor dynamic model of (a) end suction and (b) double suction pumps.

In this expression, E is the elastic modulus of the shaft material, I the moment of inertia
of the shaft, L1 the overhang, L2 the distance between bearings, and d the shaft
diameter.
For impellers supported by bearings on either side, for example for double suction
pumps,
If the water-filled impeller mass is m, the natural frequency f will be given by
Of perhaps greater importance than the critical speed will be the deflection at the front
wear ring at the nominal shaft speed. It is assumed that an unbalance is present, which
acts as if the center of gravity of the impeller were at a small distance e from the axis
of rotation. The mass of the impeller, assumed to be concentrated at its center of
gravity, rotates eccentrically and results in a radial centrifugal force resisted by the
spring force of the shaft. Well below the critical speed, the direction of the centrifugal
force is in the direction of the eccentricity e and the deflection y, and rotates
synchronously, always pointing in the same direction with respect to the shaft. The
deflection at the impeller can then be calculated from the equality of the radial
centrifugal force and the spring force of the shaft. The centrifugal force is mw2(e + y),
and the resisting spring force is Ay. The following expression gives the deflection:
w2 = k m
It can be observed that if the rotational speed w becomes equal to the critical speed w,
the deflection becomes infinite, assuming that no damping is present (Fig. 13.2). If the
shaft stiffness k is much greater than mw2-the rotational speed remains far below the
critical speed-the expression simplifies to

The deflection is proportional to the unbalance eccentricity and the ratio of the
rotational speed and the critical speed squared.
The deflection at the front wear ring may be slightly larger or smaller than the
deflection at the impeller center of gravity, where y is measured, depending on the
spacing of the bearings and location of the impeller. Evidently, the clearance of the
wear ring must be larger than the shaft deflection. Large deflections and large
clearances are undesirable because they result in large leakage.
The numerical values given by the expressions above approximate only rarely the
real values and give at best a very rough estimate. Sleeve bearings deflect under load
and add to the shaft flexibility. Hydrodynamic bearings and seals introduce other force
components which change the ideal values of the critical speeds and deflections. An
impeller that is not mounted perpendicular to the shaft, or fluid forces resulting from a
wobbling impeller, introduce twisting moments on the rotating assembly that depend on
the rotational speed and influence the critical speed of the pump.

REFERENCES
Barrett, L. E., Gunter, E. J., Allaire, P. E. (1978): Optimum Bearing and Support
Damping for Unbalance Response and Stability of Turbomachinery, ASME Journal
of Engineering for Power, January, pp. 89-94.
Childs, D. (1993): Turbomachinery Rotordynamics, Wiley, New York.
Corbo, M. A., Malanoski, S. B. (1998): Pump Rotordynamics Made Simple,

Proceedings of the 15th International Pump Users Symposium, Texas A&M
University, College Station, Texas, pp. 167-204.
Ehrich, F., Childs, D. (1984): Identification and Avoidance of Instabilities in High

Performance Turbomachinery, Mechanical Engineering, May, pp. 66-79.
Figure 13.2 Shaft deflection due to unbalance as a function of speed for simple rotor
dynamic models.
Vance, J. (1988): Rotordynamics of Turbomachinery, Wiley, New York.

14

INLET AND EXIT

RECIRCULATION
RECIRCULATION PHENOMENA
Pump users are particularly concerned about the minimum flow rate that can be
tolerated in a pump without risking damage. Reduced flow can result in pressure
fluctuations, cavitation, and overheating (Cooper 1988; Gopalakrishnan 1988; La
Houille Blanche 1980, 1982, 1985; Vlaming 1989). Exit recirculation can also
influence the head-flow curve of the pump near shutoff and affect stability. Therefore,
these phenomena deserve great attention.
At off-design conditions, especially at flow rates less than those for which the pump
was designed, the flow often does not follow the impeller passage wall surfaces, and
different flow patterns appear. Three flow phenomena can be distinguished more
frequently: inlet recirculation, separation in the impeller, and exit recirculation. As the
name indicates, in the case of inlet recirculation the flow, which already entered the
impeller, turns around and flows back into the inlet pipe. In the case of exit
recirculation, the flow, which has left the impeller, reverses and flows back into the
impeller.
These phenomena have long been observed and studied (Guelich et al. 1993; Jansen
1967; Pfleiderer 1961; Sen 1979; Spannhake 1934; Von Karman Institute 1978). An
understanding of the underlying causes was made difficult by the fact that these
phenomena often appear combined, simultaneously, and interact. Complex interactions
and instabilities may appear (Kaupert et al. 1996). However, as will be seen, their
direct causes reside in the local geometry of the impeller. They can appear separately.
Preferably they should be modeled independently from each other.

Figure 14.1 Inlet and exit recirculation flow patterns in an axial flow impeller at a
reduced flow rate.
Inlet and exit recirculation become clearly evident in axial flow impellers or blade
rows, as shown in Fig. 14.1. At reduced flow rates the main flow enters the blade row
near the hub and follows an S-shaped path from the inlet near the hub to the exit near the
periphery, where it leaves. Part of the entering flow splits off, turns around, and flows
back into the inlet pipe near the periphery. Part of the flow, having left the blade row,
turns around in the exit pipe and reenters the impeller near the hub. In centrifugal pumps,
recirculation is most likely to appear, when the inlet diameter is large compared with
the impeller diameter. Such proportions exist, in particular, in high-specific-speed
mixed flow pumps, an extreme case being, precisely, axial flow impellers.
From an application engineering point of view, it has been desirable for pump users
to derive statistically valid practical guidelines for the appearance of recirculation in
typical commercially available pumps. Recirculation can result in severe flow
oscillations and cavitation, which might erode the impeller, shorten bearing life, and
reduce reliability. Such guidelines have been established by surveying typical pump
populations (Fraser 1981). However, these empirical correlations do not establish a
cause-and-effect relationship between specific geometrical features of the pump and
recirculation. Consequently, they are not suited to provide guidelines for design and do
not help to calculate the effect of recirculation on pump performance.
Although the actual phenomena are complex, simplified models are presented here
that will allow their effect on pump performance to be calculated. A connection will be
made between some features of the pump geometry and the appearance of these flow
patterns.


INLET RECIRCULATION
To develop and illustrate a calculation model for inlet recirculation, let us assume for
the time being that the leading edge is radial, and that at design flow rates the inlet
blades are perfectly aligned with the direction of the inlet flow (Tuzson 1983). When
the flow rate is reduced, separation appears on the suction side of the blades. The
stagnant fluid in the separated region rotates with the impeller and is subject to the
centrifugal acceleration. The static pressure in the separated region increases radially
in proportion to the square of the radius:
This pressure also prevails on the separation streamline. Since the pressure remains
constant at a given radius along the separated region, the relative velocity along the
separated surface must also remain constant, as can be seen from the rothalpy equation
if the foregoing value of the pressure is substituted:
The relative velocity is the vectorial sum of the tangential and meridional velocity
components:
Since the quantity w2r2 increases with the radius and the relative velocity W remains
constant, the meridional velocity C,,, must decrease with the radius. The separated
region becomes progressively wider with increasing radius, and gradually blocks the
inlet. Eventually, the point can be reached when the meridional velocity becomes zero
at the greatest radius, at the shroud. This occurrence signals the onset of inlet
recirculation.
For most practical pumps this flow rate is slightly below the design flow rate.
However, slight recirculation near the shroud has little effect on performance. As will
be seen, the parasitic power drain on the pump increases with a higher power of the
decreasing flow rate and becomes significant only at generally lower flow rates.
Because of the gradual appearance of inlet recirculation, it is difficult to designate a
very specific flow rate below which dangerous pump deterioration can occur.
Once inlet recirculation has set in, the flow pattern in the pump inlet changes
completely. The flow entering the impeller near the hub can still be assumed uniform
and of equal energy. Such an assumption is supported by detailed measurements in the
inlet of several mixed flow pumps (Hureau et al. 1993; Tanaka 1980; Tuzson 1983). It
is also assumed that the meridional flow component in this portion of the inlet flow
retains the magnitude, which will give correct incidence. Therefore, it is not the inlet
velocity that is reduced when the flow rate becomes smaller but the effective cross-
sectional area of the inlet. The remaining geometrical cross section of the inlet is taken
up by the recirculating flow. Figure 14.2 illustrates such a flow pattern for an impeller
with a slanted leading edge.
Figure 14.2 Inlet recirculation in a centrifugal pump at reduced flow rate. (From Tuzson
1983.)
If in the central portion of the inlet the same inlet velocity Ci is assumed along a
straight, slanted leading edge, the outer radius ra of the flow, which passes through the
impeller, is given by the relationship
where rsh is the shroud radius, rh the hub radius, and Qo the flow rate at design
conditions. The remaining portion of the inlet, where the flow recirculates, can be
regarded as a separate pump with an exit near the shroud and an inlet at the smaller
radius ra. The pressure Ap and flow rate q of such an imaginary pump would be given
by expressions of the form
These expressions must have the form shown because of dimensional reasons, as can
easily be verified. The axisymmetic geometry demands that the radii appear squared.
The expression for the power absorbed by recirculation will be proportional to the
product of the pressure and the flow:
This relationship is obtained by substituting the expression for rQ given above. The
incremental head due to recirculation, AHr, which would have to be added to the
theoretical head of the pump when the efficiency is calculated is obtained by dividing
the power of recirculation by the mass flow rate:
Unfortunately, the coefficient of proportionality C depends not only on the geometry of

the pump, but also on the inlet piping configuration (Guesnon and Lapray 1985). For
example, the flow returning into the inlet pipe can impart rotation to at least a portion
of the flow reentering the pump. It will make a difference whether the pump is
connected directly to a large tank or is preceded by a long pipe. The most practical
method to determine the numerical value of the coefficient is to match the performance
of an existing similar pump with the performance calculation computer program. The
computer program LOSS3 contains a default value of 0.0005 for the coefficient, which
implies certain conversion factors and should be verified from case to case.
This model of inlet recirculation leaves the flow through the impeller well behaved.
The flow enters without incidence. The average relative velocity at the inlet W, remains
constant. However, flow separation in the impeller can still appear at some point with
reduced flow rate. If the streamline through the impeller from inlet to exit is relatively
short, inlet recirculation can affect the exit flow conditions. Especially in mixed flow
pumps, if exit recirculation also appears, the two phenomena can merge and result in
very complex flow patterns.
Advanced impeller flow computer programs can calculate recirculating flow, but
sometimes require that special inlet conditions be imposed (Graf 1993). Normally,
uniform parallel flow is assumed upstream of the pump as the inlet boundary condition.
Provision must be present for modeling leading-edge separation and boundary layer
development. Some powerful programs incorporating advanced turbulence models can
successfully calculate recirculating flow (Gopalakrishnan et al. 1995).
It will be noted that the proposed expression assumes that recirculation begins
immediately below the design flow rate. In reality, the onset of recirculation depends
strongly on the detailed design of the inlet, inlet blade angle distribution, location of the
leading edge of the blades, and blade loading near the inlet. For example, if the
direction of the flow meets the leading edge at a slant, a sweptwing effect may appear.
The separated region will consist of a leading-edge vortex trailing along the leading
edge, which will be swept along by the velocity component parallel to the leading edge.
The presence of such a vortex can change entirely the conditions in the separated region,
where no stagnant fluid will remain. The leading-edge vortex may convey the low
momentum fluid of the separated region downstream, through the pump, and may retard
the appearance of recirculation. The simple expression presented above cannot be
expected to account for such details of the flow. Much more detailed geometrical data
input would be needed to estimate the effect of leading-edge vortices.
Entirely different conditions also exist when substantial cavitation takes place. If the
separated region behind the leading edge of the blades contains vapor bubbles, its bulk
density will be low and the pressure will not increase in the radial direction along the
leading edge. On the boundary-the separated streamline or stream surface-of vapor-
filled separated regions, it will be the absolute velocity that will remain constant. In the
case of intermittent cavitation, the flow pattern may switch back and forth between the
normal flow configuration and a recirculating flow pattern. Violent flow oscillations
would be expected and would be reflected by surges in power demand. These are the
conditions that must be avoided in pump operation. These dangers explain the particular
concern of pump operators with recirculation.

EXIT RECIRCULATION
Exit recirculation, when the flow returns into the impeller, appears at reduced flow
rates in mixed flow pumps, which often have a slanting exit. Exit recirculation has been
recognized as one of the phenomena limiting the operation of pumps at reduced flow
rates (Gopalakrishnan 1988; Guelich et al. 1993). Determining its onset is of great
interest to pump users.
The underlying cause of exit recirculation has given rise to much speculation.
Measurements and visualization of the flow at the impeller exit show confused, unsteady
fluctuations and strong interactions with the diffuser vanes or the volute, which are
difficult to interpret. On the other hand, it is well known that ideal, axisymmetric,
rotating shear flows-vortices, for example-can have stagnant or recirculating cores near
the axis of rotation (Batchelor 1967; Keller and Egli 1985; Strscheletzky 1961; Tuzson
1993). The stability theory of rotating shear flows predicts that separate fluid regions of
different energy can persist, with little mixing, if the low-energy fluid remains near the
axis of rotation. One can therefore assume that a somewhat similar flow pattern can also
exist at the pump exit when less energy is imparted to the fluid on the streamlines near
the hub of the impeller. Ideal recirculating flow conditions are complicated in pumps by
the unsteadiness of the flow and by the presence of diffuser vanes or the volute.
Evidently, flow separation in the impeller can result in uneven angular momentum
distribution at the impeller exit and can also precipitate exit recirculation (Jaberg and
Hergt 1989). Still, the theoretical concept of nonuniform angular momentum distribution
can serve as a basis for a calculation model that will reproduce the main features of exit
recirculation and will point to its causes.
The appearance of exit recirculation depends on the energy imparted to the fluid on
different streamlines at the pump exit, which in turn is intimately related to the exit
blade angles (Fig. 14.3). If the blade trailing edge is inclined to the axial direction, the
flow on parallel streamlines leaves the impeller at different radii. The circumferential
velocity U2 varies along the trailing edge. The tangential velocity Cr2 would also have
to vary at design conditions if the same energy were to be imparted to the flow on all
streamlines. If the meridional velocity G2 remains the same on all streamlines, the exit
blade angle /32 would also have to vary along the trailing edge. The blade angle is
measured here from the meridional direction. These relationships derive from the
expression of the theoretical head Hch of the pump at the design flow rate and apply
separately on each streamline:

where Cm02 = Qo/(impeller exit area).
Since U2 = wr varies along the trailing edge and Cm02 remains the same, for Hch to
remain constant, the blade angle 132 must change and compensate for the variation. If
HO designates the head at the design flow rate Q0, the relationship between the exit
blade angle #2 and the radius along the trailing edge can be derived from the equations
above:
Figure 14.3 Onset of exit recirculation in a mixed flow pump at reduced flow rate.
This theoretical expression prescribes that particular impeller exit blade angle
distribution as a function of the radius that will give uniform energy input to the fluid on
all streamlines at the impeller exit at the design flow rate. Actual geometrical blade
angle distributions may differ from this ideal. However, high efficiency at design flow
rate requires that the energy input be uniform. Therefore, the blades of most good
pumps will conform approximately to this relationship, or at any rate, produce uniform
energy input.
In general, it can be observed that the blade angle Q2 has to become larger
(measured from the meridional direction) as the radius becomes greater. In mixed flow
pumps the shroud radius is usually larger than the hub radius. Therefore, the exit blades
must be more backward swept at the shroud than at the hub. At offdesign point
operation, when the flow rate Q is reduced, the blade angles being fixed, less energy
will be imparted to the flow along the hub than along the shroud, and a shear flow will
leave the pump, which is ultimately responsible for exit recirculation. The meridional
velocity C2 cannot remain constant along the trailing edge, and eventually reverses at
the hub.
To formulate a mathematical expression for the meridional velocity C,,,2 along the
trailing edge, the energy equation will first be used to express the local static pressure
p2 at the impeller exit. The energy increase through the pump pU2Cr2 is added to the
pressure at the inlet p 1, and the kinetic energy of the tangential and meridional velocity
components at the trailing edge is subtracted.
The radial force balance equation predicts that the static pressure increases radially
because of the centrifugal acceleration of the swirling fluid leaving the impeller (Fig.
14.4):
The pressure can be eliminated from these two equations by taking the derivative of the
expression for the pressure p2 with respect to the radius and setting it equal to the
second equation. This substitution results in a differential equation in terms of the
velocities:
Substituting expressions for the circumferential velocity U2, tangential velocity C12,
and blade angle 112, an equation is obtained containing only two variables, the radius r
and meridional velocity C2. The equation cannot be solved analytically, in closed
form, but a finite difference solution is possible by substituting ratios of finite
increments instead of the derivatives. The trailing-edge length is subdivided into equal
increments, and the meridional velocity is calculated at each station.
The reduced flow rate, at which recirculation starts, is obtained by calculating the
meridional velocity at several locations along the trailing edge. Flow rate increments
are calculated from the product of the meridional velocity and the corresponding,
incremental exit cross-sectional areas, which are finally added together to arrive at the
total flow rate. Calculation starts at the hub exit radius, where the exit meridional
velocity becomes zero when recirculation starts. The velocity is then incremented
stepwise along the trailing edge. The computer program listing EXREC for executing
the calculation is given in the Appendix.
The flow rate at which recirculation starts, and the corresponding meridional
velocity profile, appear to correspond to measured values (Gopalakrishnan et al. 1995;
Tuzson 1999). The calculation assumes that the impeller produces uniform angular
momentum at the exit at the design flow rate. If the exit conditions of the actual pump
differ from the assumed form, the calculations will be in error. The onset of exit
recirculation becomes evident from the head-flow curve of the pump. The curve, which
approaches the horizontal or even shows a maximum with decreasing flow rate,
suddenly starts rising again at the particular flow rate, when recirculation starts as
shown in Fig. 14.5. The gradual rise of the head with decreasing flow rate can be
explained by the fact that the average effective streamline at which the flow leaves the
pump gradually increases with decreasing flow, since recirculation takes increasingly
more space near the hub. A greater effective impeller radius generally results in higher
head.
An approximate shape of the head-flow curve of a pump can be sufficiently well

calculated by considering only an average streamline passing through the impeller. In
mixed flow impellers the theoretical head usually declines only slowly with increasing
flow rate. Under these conditions the actual pump head, which is obtained when the
losses are deducted from the theoretical head, often shows a maximum at a reduced
flow rate. The operating range below this flow rate, corresponding to the maximum
head, is unstable. Indeed, it was shown that a positive slope of the head-flow curve
leads to flow oscillations and surging. However, if exit recirculation sets in at or before
the maximum is reached with decreasing flow rate, the head starts rising again and a
positive slope can be avoided. The onset of exit recirculation does not depend on the
conditions on the mean streamline and cannot be calculated with a mean streamline
performance prediction program such as the LOSS3 program. Exit recirculation
depends on the geometrical configuration of the impeller exit, the slanted trailing edge
of the blades, and the blade angle distribution. By tailoring the exit appropriately, the
onset of exit recirculation can be controlled, and the pump head-flow characteristics can
be adjusted to avoid a positive slope and the corresponding instability.
Figure 14.4 Impeller exit flow conditions in a mixed flow pump.
An important conclusion from the agreement between this calculation procedure and
measurements is that exit recirculation results from uneven angular momentum
distribution and energy input at the impeller exit. Therefore, the calculation model
above has a firm theoretical basis and can be further refined, if desired, with powerful
computer programs that can estimate the angular momentum distribution at the impeller
exit, or can calculate the exit meridional velocity distribution directly at off-design
conditions.

REFERENCES
Figure 14.5 Mixed flow pump head-flow curve with exit recirculation.
Batchelor, G. K. (1967): Fluid Dynamics, Cambridge University Press, New York,

pp. 543-555.

Texas A&M 5th International Pump User Symposium, Houston, Texas, pp. 177-192.
Fraser, W. H. (1981): Recirculation in Centrifugal Pumps, ASME Winter Annual

Meeting, Washington, D.C., November 15-20.
Gopalakrishnan, S. (1988): A New Method for Computing Minimum Flow,

Proceedings of the Texas A&M 5th International Pump User Symposium, Houston,
Texas, pp. 41-47.
Gopalakrishnan, S., Cugal, M., Ferman, R. (1995): Experimental and Theoretical

Flow Field Analysis of Mixed Flow Pumps, 2nd International Conference on Pumps
and Fans, Tsinghua University, Beijing, October 17-20.
Graf, E. (1993): Analysis of Centrifugal Impeller BEP and Recirculating Flows:

Comparison of Quasi-3D and Navier-Stokes Solutions, in Pumping Machinery
Symposium, ASME FED Vol. 154, pp. 236-245.

Guelich, J. F., Bolleter, U., Simon, A. (1993): Feedpump Operation and Design
Guidelines, Sulzer Brothers Ltd., Winterthur, Switzerland. Also McCloskey, EPRI
Publication TR-102102, T. Electric Power Research Institute, Palo Alto, Calif.,
June.
Guesnon, H., Lapray, J. F. (1985): Influence de la geometrie des structures situees a

('aspiration d'une pompe sur les courbes caracteristiques a debit partiel, La Houille
Blanche, No. 5, pp. 387-404.
Hureau, F., Kermarec, J., Stoffel, B., Weiss, K. (1993): Study of Internal
Recirculation in Centrifugal Impellers, ASME Pumping Machinery Symposium,
FED Vol. 154, pp. 151-157.
Jaberg, H., Hergt, P. (1989): Flow Patterns at Exit of Radial Impellers at Part Load
and their Relation to Head Curve Stability, in Pumping Machinery, ASME Vol. 81,
pp. 213-225.
Jansen, W. (1967): Flow Analysis in Francis Water Turbines, ASME Journal of

Engineering for Power, July, pp. 445-451.
Kaupert, K. A., Holbein, P, Staubli, T. (1996): A First Analysis of Flow Field

Hysterisis in a Pump Impeller, ASME Journal of Fluids Engineering, December, pp.
685-691.
Keller, J. J., Egli, W. (1985): Force- and Loss-Free Transition Between Flow States,
Journal ofApplied Mathematics and Physics, ZAMP, Vol. 36, pp. 854889.
La Houille Blanche (1980): Comportement dynamique des turbomachines

hydrauliques, No. 1/2.
La Houille Blanche (1982): Fonctionnement des turbomachines a debit partiel, No.

2/3.
La Houille Blanche (1985): Fonctionnement des turbomachines a regime partiel, No.

5.
Pfleiderer, C. (1961): Die Kreiselpumpen, Springer-Verlag, Berlin, pp. 408-416.
Sen, M. (1979): Inlet Flow in Centrifugal Pumps at Partial Delivery, Government

Publication N81-10437, Von Karman Institute, Rhode Saint Genese, Belgium, June
20-21.
Spannhake, W. (1934): Centrifugal Pumps, Turbines and Propellers, MIT Press,

Cambridge, Mass., pp. 244-249.
Strscheletzky, M. (1961): Kinetisches Gleichgewicht der Innenstromungen

Incompressibler Flussigkeiten, VDI-Z Fortschritt Berichte, Vol. 7, No. 21, pp. 113-
132.
Tanaka, T. (1980): An Experimental Study of Backflow Phenomena in High Specific

Speed Propeller Pumps, ASME Paper 80-FE-1980.
Tuzson, J. (1983): Inlet Recirculation in Centrifugal Pumps, ASME Symposium on

Performance Characteristics of Hydraulic Turbines and Pumps, 1983 Winter Annual
Meeting, Boston, FED Vol. 6, pp. 195-200.

Fluids Engineering, September pp. 463-467.
Tuzson, J. (1999): Impeller Exit Recirculation, Paper FEDFM99-6863, ASME Fluids

Engineering Conference, San Francisco, Summer.
Vlaming, D. J. (1989): Optimum Inlet Geometry for Minimum NPSH Requirements for
Centrifugal Pumps, in Pumping Machinery, ASME FED Vol. 81, pp. 25-28.
Von Karman Institute (1978): Off-Design Performance of Pumps, Lecture Series, Vol.
3, No. 1, March 6-8.

15

TWO-PHASE FLOW IN
PUMPS
TWO-PHASE FLUID BULK PROPERTIES

Although pumps are designed to handle incompressible fluids, at times gases and
vapors appear in the liquid and must pass through the pump. In power plant pumps,
water near boiling conditions may release water vapor if the pressure of the water is
lowered. Process pumps also often handle various organic liquids near boiling and
must survive upset conditions by handling liquid-gas or vapor mixtures. Refrigeration
typically operates with organic fluids near boiling conditions. Oil well pumps
encounter crudes under high pressure, saturated with a variety of dissolved gases,
which evolve as the fluid is brought to the surface. Nuclear power plant coolant loss
can be feared if because of accidental excess temperatures, the cooling water flashes
into vapor. The capability of centrifugal pumps to tolerate and handle liquid-gas
mixtures can be an important factor in their application.
Several possible cases can be distinguished regarding the behavior of the twophase
fluid. In one case, gases, which are insoluble in the liquid, may enter the pump with the
liquid. In another case, gases originally dissolved in the liquid may come out of solution
because of lowered pressures. In still other cases, the liquid itself may vaporize
because of lowered pressures. These different cases must be distinguished because the
properties characterizing the fluid, in particular the density or specific volume of the
fluid, are related differently to the pressure and temperature. The complexity of two-
phase flow demands that simplifying assumptions be made to arrive at a manageable
mathematical model. The assumptions will differ depending on the particular case.
Insoluble gases in the liquid phase or noncondensable gases can be modeled by

assuming that the pressure-density relationship of the gas fraction follows the ideal gas
law at a constant temperature. Isothermal compression prescribes that the specific
volume v = 1 /p vary in inverse proportion to the absolute pressure p. The constant-
temperature assumption remains valid if the gas is present in the form of small bubbles,
and any temperature difference between the gas and the liquid can be equalized
instantaneously because of the relatively large surface of the bubbles. The assumption
breaks down if the gas accumulates in certain locations of the flow passages.
Accelerations perpendicular to the streamlines will tend to separate the phases. The rate
of separation, the velocity of the bubbles relative to the fluid, will depend on the bubble
size. Surface tension presents an additional complicating factor. High surface tension
tends to help agglomerate bubbles. Surface energy is reduced if small bubbles with a
large aggregate surface accumulate into large bubbles with less total surface area. Small
amounts of surface-active agents or even organic impurities, which affect the surface
tension, can play an important role. The fluid may appear in the form of a foam, which
tends to resist the breakup of bubbles. The following relations define the fluid
properties under isothermal conditions:
In these expressions R is the gas constant, T the absolute temperature in Rankin degrees
(Fahrenheit + 459.7), and pt and pg designate the density of the liquid and the gas. The
symbol a stands for the volume fraction of the gas, the void fraction, which itself
depends on the gas density or specific volume.
To put things into perspective, a simple calculation would show that uniform
spherical bubbles in a rectangular array would touch at a void fraction of 7r/6 = 0.52.
At a void fraction of 0.2, the width of the liquid layer separating bubbles would amount
to about 30% of the bubble diameter. Evidently, coalescence of bubbles could hardly be
avoided at void fractions above 0.2, and bubble interactions would be expected to
become important well below this limit.
Being compressible, sonic speed may appear in liquid-gas mixtures. The relatively
high compressibility of the bubbles and the high inertia of the liquid phase results in a
relatively low speed of sound, which can be calculated from the relationship between
bulk density and the pressure (Leung 1996). Vaporization of a single-substance liquid
introduces thermodynamic effects. Most fluids have a sizable heat of vaporization and
cool down when vapors evolve unless heat is supplied from the outside. Consequently,
isothermal conditions cannot be taken for granted. Heat transfer between bubbles and
the liquid may still be instantaneous, but the heat capacity of the liquid phase must be
taken into account.
The case of water mixed with air or steam has great importance in industrial
applications and has been investigated intensely. However, care must be taken when
experimental results with water are applied to other fluids, because of the exceptional

properties of water. Its heat of vaporization, about 1000 Btu/lbm, exceeds the heat of
vaporization of other liquids by a factor of 2 or more. Also, water has the highest
surface tension among liquids and has the greatest tendency to form large bubbles and to
separate the phases.
Two-phase flow conditions in pumps are often compared with similar flows in
pipes. It has been recognized that regions of different gas and liquid mixture
configurations prevail in pipe flow, depending on the void fraction. Generally, water-
air mixtures can be considered to form a homogeneous flowing fluid up to void fractions
of about 0.2. Beyond this limit a bubbly flow regime appears, which exists up to a void
fraction of about 0.6. At even higher void fractions, separated annular flow has been
observed in pipes until the liquid fraction becomes so small that only a fine mist is
formed. Starting with the bubbly flow regime, the two phases may move with different
velocities. The ratio of velocities, the phase slip, will depend on the interfacial shear
force between the phases. These correlations have been based on pipe flow and assume
that typically encountered pipe flow turbulent mixing prevails and that no transverse
accelerations exits that would tend to separate the phases. In the presence of transverse
accelerations, encountered in pumps, the gas and liquid phases would tend to separate,
and with the reduction of the interfacial area between the phases, interaction between
them would be minimized. The bubbly flow regime could then be expected to disappear
in pumps and separated flow may take its place.
In these various flow regimes, different kinds of interactions and drag forces between
the phases exist, depending primarily on the degree of dispersion of the gas in the
liquid. In the bubbly flow regime, interfacial surface and shear between the phases
increases with decreasing bubble size. The bubble size itself depends on the intensity of
mixing brought about by the turbulence in the flow, which is difficult to estimate.
Analytical studies fall back on empirical correlations, which are valid only in the range
for which they have been developed. These drag forces link the motion of the gas and
liquid phases and influence the relative velocity between the phases. A difference in
velocity between the phases can bring about an accumulation or reduction in the local
gas volume, and therefore can affect the local value of the void fraction. In the presence
of transverse acceleration, the gas and liquid phases would tend to separate. Obviously,
the particular void fraction at which the transition from one regime to the other takes
place depends on the physical properties of the liquid. In substances that tend to foam,
the transition occurs at higher void fractions (Lea and Bearden 1982).

EXPERIMENTAL PUMP PERFORMANCE DATA

The anticipated possibility of a nuclear reactor coolant flow accident triggered
intensive research on the performance of coolant circulating pumps with water and
steam or air mixtures. The electric power generation industry supported several full-
sized and reduced model tests (Fujie and Yamanouchi 1985; Kamath and Swift 1982;
Kim 1983; Mikielewicz et al. 1978; Minato et al. 1985; Runstadler 1976; Wilson
1977; Winks 1977). Detailed data can be found in the original publications. Two-phase
pump flow technology was reviewed, data reduction and scaling issues were
considered, and empirical correlations were developed (Rundstadler 1976; Wilson
1977). These investigations and their results were summarized in an article by Furuya
(1985).
Data were taken at design conditions and were presented in the form of a normalized
head and normalized torque. The head was normalized by dividing with the head on
pure liquid. The torque was also corrected for density by dividing it with the bulk two-
phase density at the pump inlet. These corrected values were plotted against the void
fraction as shown in Figs. 15.1 and 15.2 (Furuya 1985; Kamath and Swift 1982).
The normalized head is seen to remain close to 1, from a void fraction of zero, which
corresponds to pure liquid, to about 0.2. At higher void fractions it gradually declines to
about 0.1 at a void fraction of 0.8. Void fractions beyond 0.8 correspond practically to
pure gas and represent conditions in a compressor rather than a pump. The normalized
torque also starts from I and drops suddenly to between 0.5 and 0.7 at a void fraction of
0.2. From then on it remains approximately constant up to void fraction values even
beyond 0.8.

Figure 15.1 Two-phase density-corrected torque ratio of a centrifugal pump as a

function of inlet void fraction. (From Kamath and Swift 1982.)
Similar experiments were performed in Japan on behalf of the nuclear power

generation industry (Noghrehkar et al. 1995). Analysis results conforming with test data
show a sudden drop of normalized head at void fractions of 0.2 or slightly higher. The
test results are similar to those in Figs. 15.1 and 15.2. Tests and analysis executed in
Germany for an axial nuclear reactor cooling pump attribute the head deterioration to an
increase in the impeller slip coefficient resulting from a "displacement of centre of mass
due to phase separation" in the impeller (Schneider and Winkler 1988). The relative
velocity of the liquid phase moving to the leading face of the blade, in the direction
against the sense of rotation, is claimed to diminish the effective tip speed of the
impeller. A change in the void fraction at the impeller exit would also contribute to
head deterioration. Both of these factors would depend on the inlet void fraction. The

experimental head data resemble those of Figs. 15.1 and 15.2.
Entirely different kind of experiments were described by Lea and Bearden (1982).
The study intended to address the problem of pumping crude containing dissolved gases
from deep oil wells. Tests on water and air, as well as with mixtures of kerosene and
C02, measured the performance of small vertical pump stages across the operating
range. Some of these test data are shown in Fig. 15.3. On air-water mixtures at design
conditions and low void fractions up to about 11% air, the head decreases in proportion
to the bulk density. At 14% air, the head drops to about 10% of its value on water. On
kerosene and carbon dioxide mixtures, which were observed to foam, complete head
loss was not reached until void fractions of 40 to 50%. In particular, only about 20%
head drop was measured with 50% gas when the inlet pressure was raised to 400 psig,
which amounted to 10 times the pressure rise through a single stage. The head tends to
decline rapidly with decreasing flow rate below the rated flow, as shown in Fig. 15.3.
However, the corresponding positive slope of the pump characteristics induced
instability and surging in every instance. The compressibility of the fluid further
enhances the instability. At flow rates higher than the design flow rate, choking
phenomena appear which are the equivalent of choking in centrifugal compressors and
result from sonic flow at the impeller inlet. At choking the characteristics drop suddenly
to zero head and preclude operation at higher flow rates. Indeed, two-phase fluids,
being compressible, can experience supersonic flow. The compressibility of the gas
bubbles and the high inertia of the liquid phase can combine to yield relatively low
values of the speed of sound.

Figure 15.2 Two-phase density-corrected head ratio of a centrifugal pump as a function

of inlet void fraction. (From Kamath and Swift 1982.)

Figure 15.3 Two-phase head-flow curve of centrifugal pump with 15% by volume CO2
in fuel oil: (a) K-70 pump; (b) C-72 pump. (From Lea and Bearden 1982. Copyright
SPE.)
The tests with kerosene and carbon dioxide were performed on three different pump
designs: on low-specific-speed radial stages and on high-specific-speed radial and
mixed flow stages. The nominal head and flow ratings of the radial and mixed flow
stages were the same. Therefore, their specific speeds were also the same. Tests
showed that the mixed flow design handled gas-liquid mixtures better. The explanation
might be that radial designs rely more on producing a static pressure rise through the
impeller, while mixed or axial flow pumps produce an energy increase mostly in the
form of a high-velocity head at the impeller exit. With a two-phase fluid it is primarily
the static pressure rise through the impeller that is lost. Consequently, high-specific-
speed mixed flow stages would be expected to suffer less.
For the eight low-specific-speed radial stages, rated at a nominal 50 gpm and 35 ft
per stage, performance curves are published for inlet pressures of 50, 100, and 400 psi
and for gas fractions up to 50%. The curves at high inlet pressure show the least
deterioration. These performance trends can be explained by the reduction in flow rate
in consecutive stages due to the gradual compression of the gas phase. At low inlet
pressures the volumetric flow rate in the last stage is reduced to about 70% of the inlet
flow rate, assuming isothermal compression of the gas. Solubility of CO2 in kerosene
would be expected to reduce the volumetric flow rate even further. At high inlet
pressures the last stage still handles close to 90% of the inlet flow rate. With reduced
flow velocities, separation of the phases would proceed faster. Satisfactory pump
operation with such surprisingly high void fractions is possible only because CO2 in
kerosene tends to foam and resists phase separation better than do water and air or
steam.
Important tests have also been performed measuring the pressure recovery in
diffusers with two-phase flow (Hench and Johnston 1971). The general trend of the
pressure recovery coefficient follows the trend of liquid diffusers and appears closer to
a sudden expansion with a maximum near an area ratio of 2, than an ideal diffuser. Little
performance deterioration was found up to void fractions of 0.38, but pressure recovery
dropped at higher void fractions and vanished at a void fraction of approximately 0.7.
Pressure kept increasing beyond the diffuser exit, in the straight pipe section following
the experimental diffuser, indicating that considerable slip between the phases might
still have been present at the diffuser exit, and that in the following straight pipe section,
the faster-moving liquid compressed the gas fraction in the process of equalizing their
velocities.

CALCULATION MODELS
Several analytical and empirical correlation methods have been developed, especially
on behalf of the nuclear power industry, to represent mathematically the experimental
pump data taken with water-air or steam mixtures (Kim 1983). In most instances, a
factor, a head multiplier correlation, was developed, the expectation being that the
head on water could be multiplied by the multiplier to calculate the two-phase flow
head (Kamath and Swift 1982). The multiplier was expected to depend only on the
void fraction. In a similar formulation, a head loss ratio correlation was proposed
instead (Mikielewicz et al. 1978).
An unspoken assumption in the case of some of these correlations is that the flow
pattern with gas-liquid mixtures remains the same as with pure liquid: that "the
trajectory of the liquid be identical to that of bubbles" (Furuya 1985). Such may be the
case for low void fractions and small bubbles (Minemura and Murakami 1980).
However, beyond a void fraction of about 0.2, the phases separate and an entirely
different flow pattern appears in the impeller which bears no relationship to the flow
pattern on liquid water. Therefore, an entirely different model is required for higher
void fractions, which cannot be grafted onto the liquid flow pattern.
The simplest analytical model assumes that pump performance remains the same with
the two-phase fluid, except for a change in bulk density. The assumption implies that the
pressure rise through the pump remains insufficient to produce a significant volume
change in the gas phase while passing through the pump. Such is the case when the pump
pressure rise is small compared with the absolute pressure of the fluid. A further
condition requires that the gas be well dispersed in the fluid and that no significant
separation of the phases should take place in the pump. A low void fraction would
probably satisfy such a requirement. Validity of the model can be evaluated by
calculating the gas volume change from pump inlet to pump exit and estimating the
percentage velocity change from the changed volumetric flow at the pump exit.
At low void fractions and in the case of vapors that can condense or gases that can
dissolve in the liquid, the possibility exists that the pressure rise through the impeller
suffices to significantly reduce or even eliminate the gas volume (Noghrehkar et al.
1995). Even in noncondensable gases, a change in the void fraction from impeller inlet
to exit cannot be excluded. Therefore, a further correction to the foregoing model might
be to calculate the bulk density change, or the gas volume change, stepwise along the
streamline through the impeller from the local static pressure, assuming isothermal
conditions.
More elaborate models calculate the flow of gas and liquid phases separately.
Accelerations in the direction of the streamlines are taken into account and a slip, a
velocity difference between the phases, is stipulated, which can result in a gas holdup, a
local increase in the void fraction (Furuya 1985; Grison and Lauro 1979; Mikielewicz
et al. 1978; Minato et al. 1985; Murakami and Minemura 1983; Noghrehkar et al. 1995;
Zakem 1980, 1987). The velocity difference between the phases implies mutual
interaction between phases. Additional assumptions need to be introduced regarding the
drag forces between phases, which in turn depend on the bubble sizes. Since theoretical
bubble dynamics do not allow generally valid predictions, empirical data are
introduced which derive from test conditions similar to those of the intended
application, but may lack general validity. Bubble size, gas dispersion, and mixing in
the liquid should be a function of the turbulence. Specialized turbulence models exist for
two-phase flow which could be linked to the bubble motions (Lopez de Bertodano et al.
1994).
The test data in Fig. 15.1 lead to the important observation that beyond a void
fraction of 0.2, the density-corrected torque does not depend on the void fraction. This
fact implies that the theoretical head, the energy imparted to a unit volume of fluid by the
impeller, remains the same regardless of the void fraction. Once the flow separates, the
void fraction at the impeller exit is irrelevant. Therefore, the variation of the head with
the void fraction shown in Fig. 15.2 cannot result from a variation of the theoretical
head but must result from head losses. Such head losses can be present at the inlet or in
the volute and diffuser. Since no significant static pressure rise can take place in the
impeller in the presence of gas, the energy input must be present in the form of velocity
head at the impeller exit. Major head losses occur when velocity head is transformed
into pressure head by the diffuser. Diffuser losses probably account for the head losses
in two-phase pumps. The diffuser can consist of a high-speed jet exhausting into a
straight flow passage since even a sudden expansion can recover up to 50% of the jet's
kinetic energy.

FULLY SEPARATED FLOW MODEL

A drastic simplification consists of assuming that the liquid and gas completely
separate in the impeller. A liquid layer would then be present on the side of the blades
facing the direction of rotation (Fig. 15.4). The gas would collect on the suction side.
Since the gas region would extend from the inlet to the exit and could support
practically no radial pressure increase, the energy imparted to the fluid by the pump
would consist exclusively of the kinetic energy of the liquid at the pump exit. Applying
the rothalpy I equation to a streamline in the liquid layer at constant pressure (p2 =pt),
the exit relative velocity would become
Figure 15.4 Model of fully separated two-phase flow in a pump impeller.
The principle of conservation of angular momentum still applies and can be used to
calculate the absolute tangential velocity Ct2 at the impeller exit, noting that no slip
takes place since the flow in the liquid layer follows the blade.

Here Hth is the theoretical head, the energy imparted to the liquid per unit volume, and
f'2 is the blade angle measured from the radial in the direction opposite to the sense of
rotation. The symbol k stands for a coefficient reducing the relative velocity W in the
impeller as a consequence of losses.
The shaft torque T can be calculated from the product of liquid density pi, the
theoretical head Hth, and the liquid flow rate Q(1 - a), by dividing their product by the
shaft speed N:
It will be noted that when fully separated flow exists, the bulk density-corrected torque
T/(1 - a) will not depend on the void fraction a and will remain constant since the
theoretical head remains constant. Consequently, the power input to the pump will be in
proportion of the liquid flow rate into the pump.
It is also apparent that the bulk flow rate affects the relative velocity at the exit W2
only slightly. Only the inlet velocity Cl depends on the flow rate. It is usually much
smaller than the tip speed U2:
Consequently, when the phases separate in the impeller, the theoretical head will not
depend on the void fraction and will depend only slightly on the flow rate. This mode
of operation differs significantly from that with liquid, in which case the theoretical
head increases linearly with decreasing flow rate. One consequence of this change in
flow pattern is that the slope of the pump characteristics decreases and a maximum of
the head appears at a reduced flow rate. Pump operation below this flow rate becomes
unstable since the head-flow characteristic has a positive slope.
To calculate the net head H, the theoretical head Hth would need to be multiplied by
the efficiency q. As a first approximation it could be assumed that the losses result
entirely from diffusion losses, and that therefore the pump efficiency numerically equals
the diffuser pressure recovery coefficient, which is available from test results (Hench
and Johnston 1971). The liquid sheets leaving each blade at the impeller exit appear for
a stationary observer at the impeller exit to form alternating liquid and gas layers at the
inlet to the diffuser vanes or the volute tongue. They will strongly mix and produce a
homogeneous two-phase fluid at the entrance of the diffuser section. Therefore, well-
mixed fluid conditions exist at the diffuser inlet, and flow separation is unlikely at that
location. A good diffuser design may act as a two-phase ejector, which entrains gas
with a highvelocity liquid jet. Efficiencies would then vary and can be approximately
correlated for void fractions beyond 0.2 by the empirical expression which
approximately follows the experimentally found pressure recovery coefficient in a two-
phase diffuser (Hench and Johnston 1971):
Consequently, the head of the pump on two-phase flow will suddenly drop near a void
fraction of about 0.2, to a level of 0.5 to 0.7, because of the drop in the theoretical
head, and decay with void fraction toward zero at very high values because of a
declining pressure recovery coefficient, which is numerically equal to the pump
efficiency.
At off-design conditions the inlet flow contacts the leading edge of the blades, and
flow separation would be expected to occur. Some of the liquid will spread to the hub
and shroud surfaces and may not accumulate on the blade leading face. The possibility
of inlet recirculation may exist. With a separation of the liquid and gas phases at the
impeller inlet, the pump may lose its prime and stop pumping, or enter an irregular
surging mode of operation. Without liquid in the impeller, the pump cannot aspirate
fluid into the inlet. The appearance of a maximum on the head-flow characteristic of the
pump will contribute to the instability of the pumping system. All these phenomena have
been observed in pump tests with liquid-gas mixtures. The specific conditions when
these flow pattern appear depend on the particular design of the pump.

COMPARISON WITH TEST DATA

The fully separated two-phase pump flow model contains two empirical quantities: the
dimensionless loss coefficient k and the efficiency or diffuser pressure recovery
coefficient q = Cp. The loss coefficients can be determined from the extensive tests
performed on the nuclear reactor cooling pumps, in particular from Figs. 15.1. and
15.2. The specific test data used here come from the one-fifth scale model of the mixed
flow pump, having the specifications 252 ft (77 m) at 3500 gpm (0.221 m3/s) and 4500
rpm, corresponding to a specific speed of 4200. The model had a 6-in. (0.1525-m)-
diameter inlet flange and a nominal impeller diameter of 7.75 in. (0.2 m). The exit
blade angle was 65. Other geometrical dimensions can be scaled from the cross-
sectional drawing of the pump (Winks 1977, Fig. 4-8). A performance calculation
performed with these data will approximately reproduce the published head-flow
curve of the pump. The inlet velocity becomes CI = 40 ft/sec (12.2 m/s), the tip speed
U2 = 152 ft/sec (46.4 m/s), and the impeller exit relative velocity W2 = 157 ft/sec
(47.8 m/s). The experimental data show a density-corrected torque value that is 0.5 to
0.7 times the value on liquid. It remains practically constant from a void fraction of 0.2
to very high values. Analytical considerations show that at constant rotational speed,
torque is proportional to the absolute tangential velocity at the impeller exit, which in
this instance, for pure liquid, takes on the value of Cr2 = 53.4 ft/sec (16.3 m/s). On
liquid-gas mixtures it would be 0.5 to 0.7 times as much. These test results will agree
with the value calculated from the fully separated model if the loss coefficient were to
take on the value of k = 0.88 to 0.8. These values correspond to a velocity head loss in
the impeller amounting to 25 to 35% of the velocity head WW /2g, which does not
appear unreasonable.
The theoretical head Hth, proportional to the absolute tangential velocity at the
impeller exit CO, remains independent of the void fraction. Indeed, the separated liquid
layer on the blades leaves the impeller with the same velocity regardless of the void
fraction, which affects only the layer thickness, not its velocity. The gradual decline of
the actual head results entirely from diffuser losses, which also account for the drop in
efficiency. Using the two-phase diffuser pressure recovery data of Hench and Johnston
(1971), good agreement can be obtained with the test data. For void fractions from 0.2
to 0.6, a linear relationship appears to be appropriate between the pressure recovery
coefficient Cv, or the efficiency q, and the void fraction a:
Considering the drastic simplifications, the test data agree satisfactorily with the
calculation. The real flow pattern in the impeller is certainly much more complicated
than the fully separated flow model would lead us to believe. After all, it is only an
analytical model. However, it illustrates the extreme case and can bracket the actual
state in the impeller. The separation of gas bubbles from a homogeneous mixture can
certainly not go beyond total separation. This simple model illustrates and explains, at
least qualitatively, the peculiar phenomena that appear in centrifugal pumps when
operating on liquid and gas mixtures.

FILL-SPILL FLUID COUPLING

The application of the fully separated two-phase flow model to a fill-spill fluid
coupling is a most appropriate example (Uchiyama et al. 1981). A fluid coupling
consists of a close-coupled centrifugal pump and centripetal turbine with a common
axis of rotation as shown on Fig. 15.5. Power is supplied to the pump and is
transmitted to the turbine by fluid forces. Depending on the magnitude of the input
torque and the capacity of the coupling, some slip, a velocity difference between the
input and output shafts, exists. Since the same torque must act on the pump as on the
turbine, the slip corresponds to the power transmission inefficiency or loss. Torque
converters operate in a similar fashion, but torque amplification or reduction becomes
possible only if a stationary reaction element, a bladed stator, exists somewhere in the
fluid circuit (Brun and Flack 1996; Foster 1960; Lucas and Rayner 1970).
The flow takes on a toroidal shape and cycles through the pump and turbine
repeatedly. The power transmitted by the coupling from the pump to the turbine
corresponds to the angular momentum produced by the pump, or the head, multiplied by
the mass flow. Power transmission efficiencies above 90% can be achieved since the
flow never slows down, and no diffusion losses exist. These high efficiencies prove that
the inefficiency of centrifugal pumps results mostly from diffusion losses.

Figure 15.5 Fill-spill fluid coupling.
In fill-spill couplings, the torque capacity and the slip, the velocity difference
between the input and output shafts, is adjusted by admitting air to the center of the
toroidal flow. Because of the centrifugal acceleration the air remains separated from the
liquid in the center of the torus. The net flow cross section and, as a consequence, the
flow rate are reduced by the presence of the air, reducing the torque capacity of the
coupling in this manner.
It is to be noted that the interface between the air core and the liquid remains at
constant pressure. The pump adds energy to the fluid by increasing its kinetic energy. No
pressure rise can take place. The turbine is of the pure impulse type, converting the
kinetic energy of the fluid to shaft power. The flow pattern in the fill-spill coupling
shows that the presence of gas-liquid mixtures in a pump impeller does not reduce the
impeller's capability to impart energy to the fluid efficiently, that most energy losses
result from diffusion, and that the flow pattern takes on an entirely different
configuration than in an impeller filled with liquid alone.

REFERENCES
Brun, K., Flack, R. D. (1996): The Flow Field Inside an Automative Torque
Converter: Laser Velocimeter Measurements, SAE Technical Paper 960721.
Forster, H. J. (1960): Fottinger-Wandler und-Kupplungen fur Kraftfahrzeuge,

Automobil Industrie, April, pp. 53-83.
Fujie, H., Yamanouchi, A. (1985): A Study on Applicability of Similarity Rule to

Performances of Centrifugal Pumps Driven in Two-Phase Flow, Nuclear
Engineering Design, Vol. 85, pp. 345-352.
Furuya, O. (1985): An Analytical Model for the Prediction of Two-Phase

(Noncondensable) Flow Pump Performance, ASME Journal of Fluids Engineering,
Vol. 107, March, pp. 139-145.
Grison, P., Lauro, J. (1979): Comportement des pompes en regime diphasique, La

Houille Blanche, No. 5, pp. 405-411.
Hench, J. E., Johnston, J. P. (1971): Two-Dimensional Diffuser Performance with

Subsonic, Two-Phase, Air-Water Flow, ASME Paper 71-FE-20.
Kamath, P. S., Swift, W. L. (1982): Two-Phase Performance of Scale Models of a

Primary Coolant Pump, EPRI Paper NP-2578, September.
Kim, J. H. (1983): Perspective on Two-Phase Flow Modeling for Nuclear Reactor

Safety Analysis, ASME Cavitation and Multiphase Forum.
Lea, J. F., Bearden, J. L. (1982): Effect of Gaseous Fluids on Submersible Pump

Performance, Journal of Petroleum Technology, December, pp. 2922-2930.
Leung, J. C. (1996): On the Application of the Method of Landau and Lifshitz to Sonic
Velocities in Homogeneous Two-Phase Mixtures, ASME Journal of Fluids
Engineering, March. pp. 186-188.
Lopez de Bertodano, M., Lahey, H. T., Jr., Jones, O. C. (1994): Development of a k-c
Model for Bubbly Two-Phase Flow, ASME Journal of Fluids Engineering, March,
pp. 128-134.
Lucas, G. G., Rayner, A. (1970): Torque Converter Design Calculations, Automobile

Engineer, February, pp. 56-60.
Mikielewicz, J., Chan, T. C., Wilson, D. G., Goldfinch, A. L. (1978): A Method for
Correlating the Characteristics of Centrifugal Pumps in Two-Phase Flow, ASME
Journal of Fluids Engineering, December, p. 395.
Minato, A., Yamanouchi, A., Narabayashi, T. (1985): Estimation of Centrifugal Pump

Head in Steam-Water Two-Phase Flow, Journal of Nuclear Science and
Technology, May, pp. 379-386.
Minemura, K., Murakami, M. (1980): A Theoretical Study on Air Bubble Motion in a

Centrifugal Pump Impeller, ASME Journal of Fluids Engineering, Vol. 102,
Murakami, M., Minemura, K. (1983): Effects of Entrained Air on the Performance of

Horizontal Axial-Flow Pumps, ASME Journal of Fluids Engineering, Vol. 105,
Noghrehkar, G. R., Kawaji, M., Chan, A. M. C., Nakamura, H., Kukita, Y. (1995):
Investigation of Centrifugal Pump Performance Under Two-Phase Flow Conditions,
ASME Journal of Fluids Engineering, March, pp. 129-137.
Runstadler, P. W, Jr. (1976): Review and Analysis of State-of-the-Art of Multiphase

Pump Technology, EPRI Paper NP-159, February.
Schneider, K., Winkler, E J. (1988): Physical Model for Reactor Coolant Pumps,
Nuclear Engineering and Design, Vol. 108, pp. 99-105.
Uchiyama, K., Takagi, T., et al. (1981): Rotational Speed Fluctuation and Internal
Flow in a Variable-Filling Fluid Coupling, Bulletin of the JSME, Vol. 24, January,
pp. 109-116.
Wilson, D. G. (1977): Analytical Models and Experimental Studies of Centrifugal

Pump Performance in Two-Phase Flow, EPRI Paper NP- 170, January.
Winks, R. W. (1977): One-Third-Scale Air-Water Pump Program, Test Program and

Pump Performance, EPRI Paper NP-135, July.
Zakem, S. (1980): Determination of Gas Accumulation and Two-Phase Slip Velocity

Ratio in a Rotating Impeller, ASME Polyphase Flow and Transport Technology
Symposium, pp. 167-173.
Zakem, S. (1987): Correlation for Two-Phase Flow Pump Performance, ASME Paper
87-FE-5.

16

HIGH-VISCOSITY PUMPS
PERFORMANCE ESTIMATES AT HIGH VISCOSITY

Stepanoff offers correction factors to convert pump performance data from tests on
water to performance on viscous fluids in particular hydrocarbons. Since the
correction factors have been derived from test data they remain valid and can give
satisfactory results (Stepanoff 1957).
More recently, the Hydraulic Institute published graphical procedures for estimating
the effect of viscosity on the head, flow rate, and efficiency. Correction factors for head,
flow rate, and power are presented as functions of head, flow rate, and viscosity.
Higher viscosities than the viscosity of water are encountered, for example, in
hydrocarbon fluids. Standard testing procedures (ISO, API, ASME) also contain
accepted methods for correcting test data for the effect of viscosity (Lunzmann and
Kosyna 1993). Test data have been published by (Hergt et al. 1981), and recently,
calculation procedures have been given by Guelich (1999). The pump performance
calculation computer program LOSS3, presented in this book, offers an easy approach,
which requires only that some of the loss coefficients be adjusted for high viscosity
(Tuzson and Iseppon 1997).
A variety of measurement units are in use to express the numerical value of viscosity.
Dynamic p and kinematic v = p/p viscosity must be distinguished, p being the density of
the fluid. At standard ambient conditions the dynamic viscosity of water is 21.1 x 10-6
lbf sec/ft2 (1 x 10-3 N s/m2) and the kinematic viscosity is 10.9 x 10-6ft2/sec (I x 10-
6m2/s), the density being 1.939 lbf sect/ft4. Metric units use the centipoise (cP; 1/1000
N. s/m2) to express the dynamic viscosity, which is just about its value for water. The
viscosity of lubricating oils, for example, can be several hundreds of times that of water
and depends very strongly on the temperature.
Viscosity affects only certain type of losses in the pump-skin friction, disk friction,
and diffusion losses-designated as SKIN, DISK, and CVD in the performance
calculation computer program. The dimensionless coefficients, corresponding to these
losses, depend on the Reynolds number, as shown schematically in Fig. 16.1. The
Reynolds number is the product of the pertinent flow velocity and the typical
geometrical dimension divided by the kinematic viscosity, Re = VL/v. In each case the

coefficient remains approximately constant at high Reynolds numbers such as those

encountered in water pumps. With increasing viscosity the Reynolds number decreases
and reaches a transition range at a certain value. Below the transition range the loss
coefficients depend in inverse proportion on the Reynolds number and therefore, if the
geometry and flow velocity remain the same, increase in direct proportion to the
viscosity.
Mathematical expressions and plots for determining the numerical value of pipe
friction or wall friction coefficients have been presented in Chapter 4. The Reynolds
number dependence of loss coefficients and pump efficiency is discussed in great detail
in an article by Osterwalder and Hippe (1982). In the case of the skin friction loss
(SKIN), the average through-flow velocity and the flow passage diameter, or hydraulic
diameter, represent the characteristic variables entering into the Reynolds number. The
skin friction coefficient )L is available from handbooks and publications. Empirical
correlations have been given in earlier chapters of the book. The plot follows the
general trend described above. The numerical value of the coefficient at high Reynolds
numbers, and the Reynolds number at which the transition occurs, depend on the relative
roughness of the passage walls. The transition Reynolds number varies between 2000
and 10,000. Depending on the Reynolds number, the friction coefficient in the viscous
and turbulent ranges is given by
64 Re
In this expression k is the average roughness height as compared with the flow passage
size characterized by the diameter D (Schlichting 1960).

Figure 16.1 Logarithm of the friction coefficient versus logarithm of the Reynolds
number.
The coefficient C,,, of the disk friction loss DISK follows a similar trend (Nece and
Daily 1960). In this case the Reynolds number is based on the impeller tip speed U2 and
the impeller diameter D2. However, the side clearance space, between the impeller or
disk and the housing must be larger than the boundary layer that could be expected at the
viscosity and rotational velocity under consideration, as discussed below. The
coefficient remains approximately constant at high Reynolds numbers, beyond the
transition range around Re= 104 to 105 and has a value from 0.006 to 0.01. The viscous
range can be approximated by
In the performance calculation computer program the mechanical losses of the pump,
which result primarily from bearing friction, are lumped together with the disk friction
losses. Sleeve bearings already operate in the viscous flow regime. Therefore, bearing
losses will increase in direct proportion to the viscosity if sleeve bearings are used
and if the fluid passing through the pump lubricates the bearings. If rolling element
bearings are used or if the bearings are separately lubricated with lubrication oil, they
obviously remain unaffected by the viscosity of the fluid being pumped.
The coefficient of the diffusion losses (CVD) would also be expected to depend on
the viscosity. When viscous forces predominate, the kinetic energy of the fluid can be
neglected. Not much, if any, of the velocity head of the fluid can be transformed into
pressure head. In this case the Reynolds number should be based on the flow passage
hydraulic diameter and the flow velocity. At high Reynolds numbers, when inertia
forces predominate, the coefficient remains constant. As in pipe flow, the transition
could be expected at a Reynolds number of around 103, below which the loss
coefficient would increase in inverse proportion to the Reynolds number. Although
tabulated values or a graphical plot showing the loss coefficient against the Reynolds
number are not available, test data on water can be extrapolated to high-viscosity fluids
by assuming that the experimental coefficient would begin to rise approximately at the
transition Reynolds number. Proposed loss coefficients to be used in the performance
calculation computer program LOSS3 (Tuzson and Iseppon 1997) are shown in Table
16.1. Computer calculation results shown in Fig. 16.2 using loss coefficients adjusted
according to the guidelines above have been compared with test data and support the
approach suggested for estimating the effect of viscosity on pump performance. The
experimental data presented by Stepanoff (1993) show a similar trend.

DESIGN FOR HIGH VISCOSITY

The insignificance of inertia forces, compared with viscous forces, governs pump
design for high viscosity. The head delivered by the pump approximately equals the
static pressure produced at the impeller exit since the kinetic energy of the flow at the
impeller exit cannot be recovered. Diffusers are ineffective. It is found that all other
things remaining the same, the smaller the tangential velocity Ct2 at the impeller exit,
the less head the pump will produce but the greater will be the fraction in the form of
static head. Only efficiencies below about 50% can be expected. Therefore, ideally,
viscous pumps should be designed for low tangential velocity at the impeller exit C12;
consequently, for low head coefficients, (P = Ct2/U2, which implies strongly
backward-curved blades, corresponding to large exit blade angles $2.
The major losses in viscous pumps will come from the close clearances that exist in
sleeve bearings and wear rings. These close clearances affect pump performance by the
frictional torque they produce and by the leakage flow through the clearance. The
bearing torque T will increase in direct proportion to the dynamic viscosity , and the
angular velocity w and the length L in proportion to the square of the diameter D and in
inverse proportion to the clearance width c, as shown below. The leakage rate q will
increase with the diameter D, the pressure difference (P2 -pi), and the cube of the
clearance width c, but will decrease in direct proportion to the dynamic viscosity p and
length L. If the torque and leakage are to be minimized at the same time, the effect of
increased viscosity should be compensated by decreasing the length. The length of
sleeve bearings must be reduced only in proportion to the viscosity increase, since the
supporting bearing forces must be maintained. The clearance can also be opened to
reduce the torque, but this modification will increase the leakage rate
disproportionately.

Figure 16.2 Calculated head-flow curve of a centrifugal pump on high-viscosity fluids.
In high-viscosity pumps, attention must also be paid to the clearance spaces between
the impeller and the stationary housing. If the boundary layer on the rotating face of the
impeller is larger than the width of the clearance space, the disk friction torque will
depend on the clearance width and will generally be greater than otherwise. An
approximate estimate of the boundary layer thickness is a low multiple of (v/w)1/2
(Schlichting 1960, pp. 86, 178). Consequently, the side clearance in high-viscosity
pumps may have to be opened up. Changes in the direction and magnitude of the flow
velocity can be better tolerated in pumps for high viscosity, since the inertia and the

velocity head of the flow do not matter.

REFERENCES
Guelich, J. F. (1999): Pumping Highly Viscous Fluids with Centrifugal Pumps, World
of Pumps, August/September, pp. 30-34.
Hergt, P., et al. (1981): Verlustanalyse an einer Kreiselpumpe auf des Basis von
Messungen bei hoher Viscositat des Fordermediums, VDI Berichte, No. 242.
Liinzmann, H., Kosyna, G. (1993): Erforschung der Verfahrensparameter and

Ermittlung von Korrecturfactoren fur Pipeline-Kreiselpumpen beim Fordern von
Mineralien mit Erhoter Viscositat, DGMK Forschungsberichte, No. 479, February.
Nece, R. E., Daily J. W. (1960): Roughness Effects on Frictional Resistance of

Enclosed Disks, ASME Journal of Basic Engineering, September, pp. 553562.

Schlichting, H. (1960): Boundary Layer Theory, McGraw-Hill, New York, pp. 86,
178, 517.
Reprinted by Krieger Publishing, Malabar, Flo., 1993.

17

SAND AND SLURRY

EROSION IN PUMPS
APPLICATIONS AND REQUIREMENTS

Erosion from sand particles might be accidental in pumps handling surface water, such
as irrigation, dewatering or flood control pumps, pump-turbines in hydropower
installations, or pumps in wells handling fluids with suspended sand particles. In these
applications low solid particle concentrations are encountered. Erosion damage
accumulates over a considerable time period. One might also assume that the particles
are small, do not settle out easily, and remain suspended (Addie et al. 1996; Truscott
1972; Wellinger and Uetz 1955).
Dredging pumps must handle high concentrations of solids, sometimes consisting of

large pebbles. However, the particles, consisting primarily of beach sand, are well
rounded and less abrasive than freshly crushed material. Pumps applied in coal or ore
beneficiation processes encounter severe specialized requirements. Ore is crushed and
wet-ground down to small size in ball or rod mills, in order to produce separate ore and
rock particles. The outlet from the mills is fed to hydrocyclones by large slurry pumps.
These centrifugal separators pass on the particles, which are fine enough, to the ore
separation process, while returning large particles to the mills. The slurry pumps used
in these grinding circuits handle high-concentration slurries consisting of freshly
crushed, relatively fine slurry particles, which are very abrasive. The question is not
whether the pumps will wear out, but when. Pumps are sometimes rated by the total
tonnage of ore they can process before they wear out.
Since erosion from slurries accumulates progressively, calculations must estimate the
rate of material removal. In most slurry pump applications the monetary loss due to
pump failure consists of lost production rather than the cost of the pump, since the
process must be shut down when the pump fails. If the erosion rate can be estimated, the
pump can be changed out in time, during scheduled maintenance, instead during an
emergency. A second purpose of slurry erosion calculations could be to determine the
distribution of the erosion rate within the pump. Ideally, material should be removed by
erosion uniformly from all flow passage walls. An excessive localized wear would
result in an undesirable premature pump failure.


SLURRY EROSION MECHANISMS

In analogy to sand blasting, the dominant slurry erosion mechanism consists of impact
erosion (Fig.17.1). Particles generally move with the fluid but deviate from the
streamlines when they encounter accelerations or decelerations transverse to the
streamlines. The particles are heavier than the fluid and experience greater inertia
forces during acceleration or deceleration. When relatively large particles, in a liquid
jet at low concentration, are directed against a solid wall, they do not follow the fluid
and turn away but strike the wall. Their kinetic energy is dissipated in the destruction
of the wall surface and in material removal (Peterson and Winer 1980). The erosion
mechanism has been studied in the process of abrasive machining (Hashish 1984),
where a high-velocity water jet containing abrasive particles is used to cut metal. A
complex combined cutting and shallow-angle impact mechanism was stipulated.
Experiments under idealized conditions with single spherical balls striking

comparatively flat solid surfaces at different angles have determined the theoretical
energy dissipation and the remaining energy of the particles rebounding from the wall,
the restitution energy (Brach 1984, 1988, 1991). However, particles below a certain
size and kinetic energy, immersed in a liquid, do not strike the solid surface but
approach it asymptotically (Zenit et al. 1999). Experimental erosion rate data, particles
colliding with flat surfaces, have also been correlated and provide an empirical
framework for ranking materials of construction (Clark and Wong 1995).
Complications arise at high concentrations and small particle sizes. Most analytical
studies assume relatively low concentrations of relatively large particles. These
assumptions allow that interactions between particles be neglected, which facilitates the
calculations. Particle concentration reaches a limit at about 75% by volume, a void
fraction of 0.25, when maximum packing occurs; the particles touch each other, and the
fluid is contained in the interstices between the particles (Suzuki et al. 1981). Such beds
of closely packed particles accumulate near flow passage walls when the acceleration
is directed toward the wall. Under such conditions, particles impact the particle bed
and cannot reach the wall. Also, when particles become small, on the order of 100m
(0.1 mm or 4/ 1000 in.), in general the particle drag forces in water exceed inertia
forces, and the particles slow down considerably before reaching the wall (Zenit et al.
1999). Particle inertia decreases with the cube of the particle size, while particle drag
only drops off with the square of the particle size.

Figure 17.1 (a) Impact and (b) sliding erosion mechanisms from slurry particles.
Therefore, at high concentrations and small particles, a different erosion mechanism

must be assumed. If in the extreme, a thick particle bed would become entirely solid, it
would be pressed against the wall by its inertia, and would act as sand paper (Mulhearn
and Samuels 1962). The erosion mechanism would be scratching and plowing by
individual particles (Fig. 17.1). Such an erosion could be characterized as sliding bed
erosion (Chen et al. 1980). Between the two extremes of impact erosion and sliding
erosion, a transition will exist; when the particle bed is not entirely solid, the particles
can slide, roll, and tumble but are still pressed against the wall, which they indent and
plow. The pressure force against the wall will depend on the thickness of the particle
bed and on the acceleration against the wall.
Whichever erosion mechanism acts, the material removal rate has been found to be
approximately proportional to the local rate of power dissipation (Beckmann 1980;
Clark and Wong 1995; Maan and Broese Van Groenou 1977; Tuzson 1984). The
erosion depth would then be proportional to the local energy dissipated by the slurry
particles. Naturally, the material removal rate will depend on the slurry material and on
the material of the flow passage wall. The energy required to remove a unit volume of
wall material, called the specific energy of the particular slurry and wall material, has
to be determined by experiment. Generally speaking, the process whereby the molecular
bonds of materials are broken, not only in erosion but also in grinding or crushing
processes, is extremely complex and has given rise to several theories, since the
corresponding energy requirement is of great interest (Beke 1964). Alternative
correlations for the relationship between material removal and energy expenditure have
been proposed. However, the assumptions leading to the specific energy appear to
apply to slurry erosion. The specific energy allows calculation of the fluid mechanic
variables and particle motions to be linked to the material removal rate. If the local

power dissipated by the slurry particles at the wall can be calculated, the erosion depth
will also be known from the specific energy of the slurry and construction materials.
In addition to the overall flow pattern in pumps, trailing vortices can produce severe
local erosion (Tuzson 1999). When trailing vortices are stretched, the swirl velocity
and centrifugal acceleration within the vortex can considerably intensify, and the vortex
can grind out deep groves in the streamwise direction. Such trailing vortices are
difficult to detect in clean water, but their presence becomes very evident from the
erosion pattern of pumps from abrasive solid particles in water.

EROSION RATE ESTIMATES

The calculation of fluid flow velocities in a rotating impeller already presents
considerable difficulty. The presence and motion of solid particles only increases the
difficulty. Only computers offer the hope to achieve numerical results (Cader et al.
1993). The calculation of slurry particle velocities and concentrations in a pump
impeller usually starts with calculation of the pure liquid flow. If low concentration
can be assumed, and particle interactions and their effect on the liquid flow can be
neglected, particle velocities can be calculated from the inertia and drag forces, which
the fluid imposes on the particles. The concentration distribution is calculated from the
particle flux due to the local particle velocities, based ultimately on the principle of
conservation of mass. The concentration change in a fluid element results from the
difference between the inflow and outflow of particles. However, special conditions
must be imposed on the flow passage walls, where the particle concentration might
reach its upper limit of maximum packing (Tuzson 1984). The assumption of low
concentration loses its validity near the walls. Still, if the extent of the accumulation at
the wall does not affect the liquid flow pattern significantly, valid results can be
obtained. At high concentrations the analytical calculations become unmanageable.
Mathematically, feasible calculation models require several speculative assumptions,
which destroy the credibility of the calculation results.
Inertia and drag forces govern the motion of individual particles. A particle of
diameter d in a fluid flow subjected to an acceleration of a experiences an inertia force
F1, and a drag force Fd given by the relations
The density of the particle and the liquid are p, and pl. The difference between the
velocity of the fluid and the particle is V - Vp. The drag coefficient Cd depends on the
Reynolds number formed from the velocity difference, the particle size, and the fluid
kinematic viscosity v. In this case the Reynolds number takes on the form Re = (V -
Vp)d/v. The coefficient of proportionality in the laminar range becomes Cd = 24/Re.
For Reynolds numbers above 105, it is given by the expression Cd = 0.2923[l + (9.06/
Re)]2 and remains about constant at Cd = 0.15 beyond.

Since the inertia and drag forces are vector quantities and have three components in
the three directions of the coordinate system, which is being used, the acceleration and
velocities must be taken in the corresponding directions. Drag forces are opposed to the
direction of the velocity difference. Inertia forces act in the direction of the particle
acceleration. Direct particle interactions at higher concentrations generally tend to
increase the drag forces.
As a final step, the particle velocities, accelerations, and concentrations calculated

near the passage walls must be linked to the material removal rate. In the case of large
particles and low concentrations, impact erosion rate correlations can be used.
Boundary layer turbulence is sometimes invoked to estimate impact angles and
velocities (Chamkha 1994, Roco 1984). Turbulence calculations introduce further
complications into the calculations and often become impractical. On the other hand, in
the presence of small particles and high concentration, a sliding bed model is more
likely and use of the specific energy concept can provide material removal rate
estimates (Tuzson 1984). These complex calculations require a powerful computational
fluid dynamic, a CFD type of computer program, modified specifically for suspended
particles.

TYPICAL EROSION PATTERN IN SLURRY PUMPS

Some qualitative estimates can be made regarding the general erosion pattern in slurry
pumps without detailed flow calculations. Erosion will predominate where the local
acceleration is directed against the flow passage walls, where particles will
accumulate, and where high velocities prevail. The direction of the acceleration will
therefore be the guide to locate significant erosion. At the pump inlet, due to streamline
curvature in the meridional, (r-z) plane, the centrifugal acceleration will be directed
toward the hub. As the flow accelerates in the circumferential direction, the centrifugal
acceleration due to rotation comes into play. The resulting acceleration vector will
gradually turn in the axial and then in the radial direction, but still oriented toward the
hub. Slurry particles will tend to accumulate and slide along the hub.
On the blade-to-blade surface, accelerations in the radial and tangential directions

are given by the following expressions, provided that the flow velocities change slowly
in the circumferential direction. They correspond to the centrifugal and Coriolis
accelerations in a rotating cylindrical coordinate system:
The direction of the tangential acceleration a, is defined positive in the sense of

rotation. Since the right-hand side of the corresponding expression is negative, the
actual acceleration vector in the impeller points in the direction opposite to the sense
of rotation. The angle of the resulting acceleration vector, /3a is defined by
In pump impellers the absolute tangential velocity Cr is positive and increases with the
radius, and the radial velocity Cr is also positive. The angle fl, measured here from the
radial in the direction of rotation, is negative, as shown in Fig. 17.2. Therefore, the
acceleration vector is always inclined in the direction opposite to the sense of rotation.
According to the expression above, if the radial velocity Cr is small as in impellers
with strongly backward-curved blades, the acceleration vector points almost in the
radial direction, as shown in Fig. 17.2, and slurry particles would tend to accumulate

on the suction side of the blades. Conversely, if the radial velocity is large, as in
impellers with radial blades, for example, the acceleration vector is almost tangential,
and the slurry particles would tend to accumulate on the pressure side of the blade.
Severe slurry erosion may also result from trailing vortices (Tuzson 1999).
Tangential swirl velocities in vortices can spin up to very high values and generate high
centrifugal accelerations perpendicular to the axis of the vortex filament. Such vortices
can erode deep groves in the flow passage walls when solid particles are present in the
fluid. Trailing vortices, being localized phenomena, are usually not identified in pump
impeller flow calculations, and with few exceptions, no simple calculation methods or
analytical expressions have been developed. Trailing vortices are likely to appear at the
blade leading edges at incidence, at the corners of the blades at the impeller exit on the
hub and shroud side as shown in Fig. 17.3, and behind any protuberance at the outside
faces of the impeller (Tuzson 1999).
Figure 17.2 Particle acceleration and location of accumulation in a centrifugal pump

impeller: (a) radial blades; (b) backward-curved blades.
Blade tip vortices occur at incidence, when the flow meets the leading edge at a
slant, and the velocity, approaching the blade, has a component in direction of the
leading edge. The separated region behind the tip of the blade forms a vortex, which
trails away along the leading edge. In centrifugal pumps the leading edge usually slants
toward the hub. The tip vortex detaches from the blade at the hub and is swept through
the impeller along the hub. Near the point of detachment at the hub the vortex flow
accelerates in its length direction, the vortex is stretched, and its swirl velocity
increases. Intense erosion is often found near, or downstream of, the location where the
blade leading edge meets the hub. The remedy for such an erosion consists of avoiding
operation at off-design conditions. Large slurry pumps are often belt driven, which
offers an opportunity to match the pump performance to the load characteristics by
adjusting the rotational speed.
Trailing vortices at the corners of blades (Fig. 17.3), near the impeller exit, arise
from an uneven angular momentum distribution across the trailing edge. Such vortices
resemble wingtip vortices of aircraft, for example. A fundamental principle of fluid
mechanics predicts that a vortex sheet will be spilled into the flow at the trailing edge
of airfoils if the lift in the spanwise direction varies. Such a shear layer rolls up in
wingtip vortices. In pumps, such trailing vortices result from nonuniform flow
conditions at the blade trailing edges. Nonuniform conditions can be caused by the
particular impeller geometry and blade design, or by boundary layers and flow
separation along the hub and shroud surfaces. If the blades are heavily loaded near the
impeller exit, and consequently the pressure difference between the pressure and suction
sides is great, the trailing vortices near the impeller exit will accelerate and stretch,
resulting in high rotational velocities and intense erosion. Therefore, if possible, the
blades of slurry pumps should be unloaded near the exit, and flow separation in the
impeller should be avoided. Uniform angular momentum distribution characterizes good
pumps.

Figure 17.3 Impeller blade tip slurry erosion from trailing vortices.
Erosion from trailing vortices can also appear on the outer hub and shroud faces of a
slurry pump. The separation behind protruding bolts or indentations can generate
trailing vortices. An approximate extent of the erosion can be obtained from a
calculation model, which estimates the lift acting on the protuberance, since the lift is
related directly to the vortex intensity and angular momentum of the trailing vortex
(Tuzson 1999). The lift would then be estimated from the pressure forces acting on the
protuberance. An exact numerical value of the erosion rate cannot be expected, but an
order-of-magnitude estimate will become available. To minimize erosion,
protuberances, cavities, or any geometry that could create a separated flow region
should be avoided on the impeller of a slurry pump.

SLURRY EROSION TESTING

Erosion in slurry pumps can be minimized by selecting appropriate materials. The
erosion resistance of materials must be determined by laboratory tests. The erosion
rate in laboratory tests must be accelerated. Whereas in applications, significant
erosion takes months or years, measurable erosion must be accomplished in the
laboratory within hours. Two approaches are available to accelerate erosion rates:
The erosion intensity must be increased significantly, or the material removed must be
measured with extremely sensitive methods to minimize the need for deep erosion. An
increase in erosion intensity can only be used with confidence if the relationship
between material removal rate and intensity is well enough known over a sufficiently
large range of values to allow a scaling of the results. Generally, it is not certain
whether erosion rate increases exactly in direct proportion of the intensity of the forces
or magnitude of the velocities over several orders of magnitude. A very accurate
measurement of the material removed requires, that the original size and configuration
of the sample be precisely known, and that the conditions of erosion be uniform and
well defined.
Principal laboratory methods, specifically targeted at slurry erosion, are jet erosion
testers, slurry pots and the Coriolis erosion tester, which are shown in Fig. 17.4
(Pagalthivarthi and Helmly 1992; Tuzson et al. 1984; Tuzson and Clark 1998). In jet
erosion testing a high-velocity slurry jet strikes a flat specimen at some adjustable
angle. In slurry pots a specimen rod of circular cross section, fixed to the end of an arm,
is rotated around in a cylindrical container filled with slurry. In the Coriolis tester
slurry is introduced into the center cavity of a rotor and exits through two radial
channels (Tuzson et al. 1984). The specimen is mounted into the wall of the radial
channel facing the direction of rotation and is eroded by the slurry particles, which are
pressed against the specimen face by the Coriolis acceleration.
In jet erosion tests and slurry pot tests the amount of material removed is determined
by the weight loss. The samples are weighed before and after the test. In the case of the
Coriolis erosion test the erosion groove is traversed using a profilometer, and the local
erosion depth is determined to within a few microns. In jet erosion tests, the material
that accumulates on the specimen surface tends to interfere with the incoming particles.
The conditions of erosion vary on the specimen with the distance from the centerline of
the jet. The weight loss of the specimen corresponds to an average erosion over the
entire surface of the specimen. While maintaining the same test conditions, the erosion
resistance of different materials can be compared, and can be ranked; a quantitative
relationship between the conditions of erosion and the local material removal rate
remain only approximate (Mueller et al. 1978).
In general, somewhat similar comments also apply to slurry pot tests. However, in a
laboratory experiment (Wong and Clark 1993), local material removal was measured
precisely with a custom-built measuring instrument, and assuming a model of slurry
flow around the specimen, calculated results have been compared with test data. The
correlation was based on the impact erosion model and has been found to give good
agreement. However, results for particles smaller than about 100m tend to deviate
from the test data. In practice, slurry pots are only used to rank materials based on
weight loss.
In the Coriolis erosion tester (Tuzson 1984; Tuzson and Clark 1998), a relatively
small batch of slurry, from an overhead tank, passes through the rotor in one pass. At
rotor speeds of 6000 rpm a measurable groove is worn into the specimen within a few
minutes. From the rotational speed and the radial velocity of the slurry flow, the local
Coriolis force and sliding velocity can be estimated, which also defines the dissipated
energy. The local erosion depth is determined from profilometer measurements to
within a few microns. From these measurements the specific energy of the sample
material can be calculated. The well-defined test conditions and accurate wear
measurements allow closely reproducible results and good capability in distinguishing
between the erosion resistance of a variety of materials (Tuzson and Clark 1998; Xie et
al. 1999).
Figure 17.4 Laboratory methods targeted at slurry erosion: (a) jet erosion; (b) slurry
pot; (c) Coriolis tester.

REFERENCES
Addie, G. R., Pegelthivarthi, K. V., Visintainer, R. J. (1996): Centrifugal Slurry Pump
Wear Technology and Field Experience, ASME FED Vol. 236, pp. 703715.
Beckmann G. (1980): A Theory of Abrasive Wear Based on Shear Effects in Metal

Surfaces, Wear, Vol. 59, pp. 421-432.
Beke, B. (1964): Principles of Comminution, Akademiai Kiado, Budapest.
Brach, R. M. (1984): Friction, Restitution and Energy Loss in Planar Collision,

ASME Journal of Applied Mechanics, March, pp. 164-170.
Brach, R. M. (1988): Impact Dynamics with Application to Solid Particle Erosion,

International Journal of Impact Engineering, Vol. 7, No. 1, pp. 3753.
Brach, R. M., Editor (1991): Particle Surface Collisions: Erosion and Wear, in
Mechanical Impact Dynamics, Wiley, New York.
Cader, T., Masbernat, 0., Roco, M. C. (1993): Two-Phase Velocity Distributions and
Overall Performance of a Centrifugal Slurry Pump, in Pumping Machinery
Symposium, ASME FED Vol. 154, pp. 177-185. Also ASME Journal of Fluids
Chamkha, A. J. (1994): Boundary Layer Theory for a Particulate Suspension, ASME

Journal of Fluids Engineering, March, pp. 147-153.
Chen, T.-Y., Walawender, W. E, Fan, L. T. (1980): The Solid Flow Properties,

AIChE Journal, Vol. 26, No, 1, pp. 31-36.
Clark, H. M., Wong, K. K. (1995): Impact Angle Particle Energy and Mass Loss in
Erosion by Dilute Slurry, Wear, Vol. 186/187, pp. 454-464.
Hashish, M. (1984): A Modeling Study of Metal Cutting with Abrasive WaterJets,

ASME Journal of Engineering Materials and Technology, January, pp. 88-100.
Maan, N., Broese Van Groenou, A. (1977): Low Speed Scratch Experiments on
Steels, Wear, Vol. 42, pp. 365-390.
Mueller, J. J., Wright, I. G., Davis, D. E. (1978): Erosion Evaluation of Materials for
Service in Liquefaction Coal Slurry Feed Pumps, Proceedings of the 3rd

International Technical Conference on Slurry Transportation, Las Vegas, Nev.,
March 29-31, pp. 107-116.
Mulhearn, T. 0., Samuels, L. E. (1962): The Abrasion of Metals: A Model of the

Process, Wear, Vol. 5, pp. 478-498.
Pagalthivarthi, K. V, Helmly, F. W. (1992): Applications of Material Wear Testing to

Solids Transport via Centrifugal Slurry Pumps, in Wear Testing ofAdvanced
Materials, ASTM Paper STP 1167, pp. 114-126.
Peterson, M. B., Winer, W. O. (1980): Wear Control Handbook, ASME, New York.
Roco, M. C., Nair, E, Addie, G. R., Dennis, J. (1984): Erosion of concentrated

Slurries in Turbulent Flow, in Liquid-Solid Flows and Erosion Wear in Industrial
Equipment, New Orleans, La., February 12-16, ASME FED Vol. 13, pp. 69-77.
Suzuki, M., Makino, K., Yamada, M., Iinoya, K. (1981): A Study on the Coordination
Number in a System of Randomly Packed, Uniform-Sized Spherical Particles,
International Chemical Engineering, July, pp. 482-488.
Truscott, G. F. (1972): A Literature Survey on Abrasive Wear in Hydraulic

Machinery, BHRA, Cranefield, Tenn., October. Also Wear, Vol. 20, 1972, pp. 29-
50.
Tuzson, J. (1984): Laboratory Slurry Erosion Tests and Pump Wear Rate
Calculations, ASME Journal of Fluids Engineering, June, pp. 135-140.
Tuzson, J. (1999): Slurry Erosion from Trailing Vortices, Fluids Engineering Division
Summer Meeting, San Francisco, July 18-23, FEDSM99-7794.
Tuzson, J., Clark, H. M. (1998): The Slurry Erosion Process in the Coriolis Erosion
Tester, ASME Paper SM98-5144, Fluids Engineering Division Summer Meeting,
Washington, D.C., June.
Tuzson, J., Lee, J., Scheibe-Powell, K. A. (1984): Slurry Erosion Tests with
Centrifugal Erosion Tester, in Liquid-Solid Flows and Erosion Wear in Industrial
Equipment, New Orleans, La., February 12-16, ASME FED Vol. 13, pp. 84-87.
Wellinger, K., Uetz, H. (1955): Gleitverschleiss, Spulverschleiss, Strahlvers- chleiss

unter der Wirkung von kornigen Stoffen, VDI-Forschungsheft, Vol. 449, Pt. B, No.
21.
Wong, K. K., Clark, H. M. (1993): A Model of Particle Velocities and Trajectories in

a Slurry Pot Erosion Tester, Wear, Vol. 160, pp. 95-104.
Xie, Y., Clark, H. M., Hawthorne, H. M. (1999): Modeling Slurry Particle Dynamics
in the Coriolis Erosion Tester, Wear, Vol. 225/229, pp. 405-416.
Zenit, R., Joseph, G. G., Hunt, M. L. (1999): The Coefficient of Restitution for Liquid
Immersed Collisions, Proceedings of the 3rd ASME/JSME Joint Fluids Engineering
Conference, San Francisco, July, FEDSM99-7793.

18

EJECTOR PUMP SYSTEMS
SYSTEM DESCRIPTION
In many installations the suction head available from the application would be
insufficient to satisfy the requirement of the pump. Such cases are encountered, for
example, in well pumps where the water table is considerably below the surface.
Vertical pumps with a long shaft or with a submersible electric motor can be used to
lower the pump inlet below the water table level. Another, inexpensive simple solution
consists of boosting the pump inlet pressure with an ejector, which entrains the inlet
flow using a high-pressure side-stream from the pump. Ejectors are simple,
inexpensive, rugged, and have no moving parts. The arrangement also has the
advantage that the pump remains aboveground and is easily accessible (Elger et al.
1991; Hansen and Kinnavy 1965; Nece 1968; Sanger 1970; Stepanoff 1957).
A typical arrangement is shown in Fig. 18.1. The pump inlet head Ho is shown
positive for the purposes of calculation but can also be negative with respect to
ambient. The pump produces a total head rise H. The ejector is lowered to a depth L
below the surface and is submerged H2 below the water level. The ejector pumping
system delivers a net flow of Q2 at a total head of Hp + Ho to the user. The pump also
supplies a flow rate of Qi to the primary jet of the ejector, which produces a total
discharge head of Hd and a combined flow rate of (Qi + Q2) at the exit of the ejector.
Several cases can limit system operation. The static head at the mixing tube inlet
H,1 might fall below the vapor pressure of the water, and the ejector may cavitate.
When the output head of the ejector becomes insufficient to pressurize the pump inlet,
the inlet head Ho drops below the NPSHR of the pump, and the pump will cavitate.

Figure 18.1 Ejector pump system.
SYSTEM CHARACTERISTICS
With the help of Fig. 18.1, expressions can be written for the ejector primary total inlet
head H, and total delivery head Hd:
Here lossl and loss2 are the pipe friction losses in the descending and ascending pipes.
As will be seen, the characteristics of the ejector can be represented by the
dimensionless diagram relating a head ratio Rh and a flow rate ratio Rq with the ratio
of cross-sectional areas of the primary nozzle and the mixing tube R as a parameter.

In this expression, Hd - H2 is the head rise of the secondary stream and H, - Hd is the
head loss of the primary stream. The product of the two quantities represents the
fraction of the total energy transferred from the primary stream to the secondary stream,
and therefore stands for the efficiency of the ejector:
11 = RhRq

EJECTOR
A typical ejector, shown in Fig. 18.2, consists of a primary inlet nozzle exhausting
axially into a cylindrical mixing section several diameters long. The secondary stream
enters parallel through the annulus surrounding the primary nozzle. The high-speed
primary and low-speed secondary streams mix in the mixing tube. A diffuser following
the mixing tube slows the discharge stream and recovers some of its kinetic energy.
It can be assumed that the static pressure at the mixing tube inlet, where the primary
and secondary streams merge, is the same for both streams. This assumption may not
apply to unusual cases (Elger et al. 1991). It is conceivable that periodic vortices
appear in the shear layer separating the two streams, which distort the pressure
distribution. Still, assuming an identity of the pressures remains a good approximation.
Indeed, the two streams are straight and parallel; therefore, no transverse pressure
gradient can be present. The velocities of the two streams at that location will
correspond to the head difference between the total or stagnation head and the static
head at the mixing tube inlet and can be calculated from Bernoulli's equation for each
stream. They can also be expressed by the corresponding flow rates divided by the
cross-sectional areas. By the end of the mixing tube the two streams would have been
fully mixed. The total head will be the sum of the static and velocity head at the mixing
tube exit. A portion of the velocity head will be converted to static head in the diffuser
following the exit of the mixing tube. The analytical problem of defining the ejector
characteristics consists of finding an expression for the static head rise from the mixing
tube inlet to its exit (Hill 1967).
Since all momentum and pressure forces in the mixing tube act in the same direction,
the momentum equation, which is equivalent to the balance of forces in the axial
direction, can be applied to a control volume defined by the inlet and exit cross sections
and the cylindrical walls of the mixing tube. Wall friction forces will be neglected.
Only the pressure forces on the inlet and exit cross sections and the momentum of the
two streams arriving and of the mixed stream leaving need to be taken into account.
Also using the continuity equation, the static head rise from the mixing tube inlet to the
exit is finally given by

Figure 18.2 Ejector.
In this expression R is the ratio of the primary nozzle cross-sectional area and the
cross-sectional area of the mixing tube, the area ratio. The quantities Q1 /R and Q2/(1 -
R) correspond to the primary and secondary stream velocities. It will be noted that if
the secondary flow rate becomes zero, the expression corresponds to the head recovery
from a sudden expansion of the primary jet. It is also evident that the expression would
remain the same if the secondary flow were to enter through the central nozzle and the
primary jet through the periphery. The two streams are interchangeable, and only the
difference in their velocities enters into the expression. For the expression to be valid,
the mixing tube has to be long enough to assure that the exit velocity becomes uniform.
Having expressions for the static heads and all velocities at the mixing tube inlet and
exit, the ejector performance characteristics can be expressed in terms of the total heads
of the streams, in dimensionless form. In calculating the discharge total head, the
assumption is made that only 65% of the mixing tube exit velocity head can be counted;
35% is lost in the diffuser. Such an assumption leads to good agreement with published
results (Elger et al. 1991; Hansen and Kinnavy 1965). Writing the final form of the
ejector performance characteristics in terms of the dimensionless ratios Rh and Rq, the
expression becomes

A, = 1 2-k- 2R (I -R)1 -R A2=k+ 2R 1-R
B, =4-2k B2 = -4 + 2k
2(1 - R) C1 = -k - R = k 1 2(1 - R) C2 R2 + R
k = 0.65
This function is plotted in Fig. 18.3 for a range of variables that correspond to
practical applications. The ejector efficiency is not shown separately but can easily be
calculated from the product of the two dimensionless variables.
Figure 18.3 Ejector characteristics.

EJECTOR SPECIFICATIONS AND SELECTION

Application specifications will prescribe the net flow rate from the pump. Q2, and the
head H, + Ho. The setting depth L and submergence H2 will also be given.
The ejector geometry, in particular the nozzle and mixing tube diameters, and the total
flow rate required from the pump, can be determined with the help of the dimensionless
ejector performance characteristics. The head ratio will be a function of the application
variables and can be obtained by substituting the expressions for the variables Hd, HI,
and H2:
If the pipe friction head losses loss 1 and loss2 and the pump inlet head Ho are
relatively small compared with the setting depth L, the head ratio will roughly
correspond to the ratio of the depth of the water table, L - H2, and the pump head rise
Hp:
For a first trial design the pump inlet head Ho can be chosen zero and the pipe friction
losses can be set at a few feet. With this value of the head ratio, a first choice of the
area ratio R and the flow rate ratio Rq can be made from the ejector characteristics.
Optimal conditions will be found around Rh = 0.35 to 1.0 and Rq = 1.0 to 0.35, where
ejector efficiencies will be in the 30s, as can be seen from their product.
Knowing the net flow required from the pump Q2, the flow rate of the primary Qi,
and the total flow required from the pump, Q = Q, + Q2 can be calculated from the flow
ratio:
1 Q=Q2 1+- Rq Qz Qi= 2 Rq
These values of the flow rates in the descending and ascending pipe branches can be
used to calculate approximate values for the pipe losses loss l and loss2, and corrected
values for the area ratio and the flow rate ratio can be selected.
The exact diameter d of the primary nozzle of the ejector is determined by the
primary flow rate Q, and the flow velocity, which in turn depends on the head
difference (H, - H,,,,) between the head supplying the nozzle and the static pressure head
at the mixing tube inlet. Since the secondary flow rate Q2 will depend in turn on the
head difference (H2 - H,,,,) between the submergence and the static pressure head at the
mixing tube inlet, the primary nozzle area can be expressed in terms of the head
difference (H, - H2) by eliminating H1 I and introducing the area ratio R and the flow
rate ratio Rq:
The mixing tube diameter D follows immediately from D = d/R'12. The mixing tube
length is usually 8 to 10 times the mixing tube diameter. At this point the head at the
mixing tube inlet H,,,, can be checked to ascertain that no cavitation will take place:
With these values the major dimensions of the ejector have been defined.

EJECTOR SYSTEM PERFORMANCE MAP

Calculation of the entire performance characteristics of the ejector pump system may
have to proceed by two or more steps of approximation since the experimental ejector
characteristics and the pump head-flow curve can be too complex to be represented by
analytical expressions, and therefore are often plotted in the form of curves. Typical
pump and ejector performance curves, and their relationship in an ejector pump system,
are shown in Fig. 18.4.
Figure 18.4 Ejector operating diagram.
The primary nozzle diameter d and the area ratio R of the ejector have been selected.
In the approach presented here, the flow ratio Rq and the head ratio Rh will have been
chosen from the ejector characteristics corresponding to the area ratio R, and the
required pump head Hp and flow rate Q = Qi + Q2 will be determined. The following
three equations are used to eliminate Ho and H, and to arrive at an equation in terms of
the ejector ratios and the pump head-flow relationship:

By rearranging, the expression is brought to the form
If the ejector head and flow ratios are selected from the performance curve of the
ejector, the pump head divided by square of the flow rate can be calculated. The
separate values of the pump head and flow rate will depend on the characteristics of
the particular pump that is being used and can be determined from the headflow curve
of the pump.
If the losses can be assumed to be proportional to the square of the flow rates in the
respective pipes, the expression above can be modified to take into account the effect of
the flow rate variation on the losses:
The primary Q, and secondary flow rate Q2 can be calculated from the total flow rate
Q and the flow rate ratio Rq:

The practical operating range of the pump will have to remain between the extreme
operating points of the ejector, Hh = 0 and Hq = 0. It may not be possible to reach these
points because the pump inlet or the ejector inlet may cavitate. The corresponding
negative pressure heads must stay above the vapor pressure of the water and are given
by
The pump will operate over a limited portion of its total range. The pump flow rate Q
will still be considerable when the secondary flow rate Q2, which corresponds to the
net flow delivered by the system, becomes zero. Similarly, the pump head Hp will
remain high when the ejector ceases to pressurize the pump inlet.

DESIGN EXAMPLE
As an example, an ejector system will be calculated for the following input variables:
Hp + Ho = 130 ft (40 m), Ho = 0, Q2 = 10 gpm = 38.5 in3/s (0.000631 m3/s), L =
6041 (18.3m), H2 = 20 ft (6.1 m), lossl = 5 ft (1.5m), and loss2 = 5 ft (1.5 m).
The following variables are calculated in sequence from the relationships given
above. Since the head is given in feet and the flow rate in cubic inches per second, it is
expedient to convert the head to inches in numerical calculations. Obviously, any
measurement system can be used, SI or English, as long as consistent units are chosen.
The expressions here presented do not contain constants, which are not dimensionless.
The only exception might be the gravitational acceleration g = 32.2 ft/sect (9.8 m/s).
Note that the ejector efficiency is 37.5%, the highest for an area ratio of 0.3. With these
measurements the system geometry is fully defined. One operating point has been
determined on the head-flow curve of the pump, which should be close to the best
efficiency point of the pump. Since the flow rates are known, the descending and
ascending pipe sizes can be selected, and the head losses lossl and loss2 can be
verified. If they differ greatly from the assumed 5 ft (1.5 m), the design process can be
repeated. Obviously, the losses should be kept as low as possible.
To calculate performance at other operating points, the head-flow characteristics of

the pump would be needed. For the sake of illustration it will be assumed here that the
pump characteristics is given by the correlation
This equation corresponds to an inverted parabola with its origin at Q = 0.0, where the
shutoff pressure would be 150 ft (46 m). The operating point Hp - loss 1 - loss2) = 120
ft. and Q = 77 in3/sec (0.00 1262 m3/s) lies on the curve. By substituting the design
values of the head and flow rate, or using the expression containing the nozzle area and
the ejector head and flow rate ratios, one can verify that
To further verify performance, the pump inlet pressure Ho = 0, and the ejector inlet
static pressure H,,,i = 3.6 ft (1.08 m) can be calculated.
Considering now the extreme case when there is no secondary flow Rq = 0, the
ejector characteristics for the area ratio of R = 0.3 shows that the head ratio becomes
Rh = 0.8. The following variables are then calculated using the assumed pump head-
flow relationship:
Q2 = 0

Now taking an operating point at flow rates larger than the design flow rate, the ejector
head and flow ratios of Rh = 0.26 and Rq = 1.2 will be used:
The flow rate cannot be increased beyond this operating point because the ejector will
cavitate. The pump inlet pressure Ho remains acceptable provided that the pump has an
NPSHR value of at least -12 ft (-3.7m). It will be noted that the operating range of the
pump is limited by the ejector form Hp = 150 ft (46 m) and Q = 45.32 in3/sec) = 11.8
gpm (0.000743 m3/s), to Hp = 126.6 ft (38.58 m) and Q = 81.7 in3/sec = 21.2 gpm
(0.00 1339 m3/s = 1.33 L/s). While the delivery head of the system changes little, the
delivered flow rate Q2 varies from zero to 44.57 in3/sec = 11.6 gpm (0.00073 m3/s).

REFERENCES
Elger, D. F., McLam, E. T., Taylor, S. J., (1991): A New Way to Represent Jet Pump
Performance, ASME Journal of Fluids Engineering, September, pp. 439-444.
Hansen, A. G., Kinnavy, R. (1965): The Design of Water-Jet Pumps, ASME Paper 65-
WA/FE 31 and 32.
Hill, P. G. (1967): Incompressible Jet Mixing in Converging-Diverging Axisymmetric

Ducts, ASME Journal of Basic Engineering, March, pp. 210220.
Nece, R. E. (1968): The Suction Manifold as a Water Jet Pump, ASME Journal of
Basic Engineering, June, pp. 316-318.
Sanger, N. L. (1970): An Experimental Investigation of Several Low-Area-Ratio

Water Jet Pumps, ASME Journal of Basic Engineering, March, pp. 11-20.
Reprinted by Krieger Publishing, Malabar, Fla., 1993, pp. 402-424.

19

EVALUATION OF NEW
FLUID MACHINERY
CONCEPTS
DEVELOPMENT CRITERIA
In new product development, an early decision needs to be made as to whether a new
concept, incorporating new technology, is likely to lead to profitable business (Tuzson
1993). Ideally, incremental funding of research and development projects should be in
proportion to the probability of success. The magnitude of risk should govern the
commitment of funds. A major factor in the decision is the required development effort
and corresponding cost on one hand, and the risk or probability of success on the other.
Unfortunately little technical and economic information is available at the beginning of
a development process that would help in making such a decision.
The approach proposed here consists of considering separately various elements of

the concept or invention, and identifying those that are already in use in other
applications. New inventions often become feasible provided that the performance
range of certain components can be stretched beyond the range of its present use.
Elements that are not in commercial use or those whose performance range needs to be
extended significantly will require extensive technological development before work
can proceed to product development. Product development must not be undertaken until
the technology has been developed. To build hardware to "try out" an idea is rarely
cost-effective.
Technology development consist of the acquisition and accumulation of experimental

data and analytical design tools. Experiments, not tests, are performed for the specific
purpose of accumulating empirical data that will be used in the design process.
Analytical studies are pursued to arrive at valid design procedures. These tools must be
suitable to design lines of products with a variety of specifications. Theoretically, a
patent should teach the invention and enable those skilled in the art to design, in a
straightforward manner, a working and properly functioning device. However, a patent
is but a first step, a small-almost negligible-part of the development process leading to
profits. In practice, considerable work remains to be done before sufficient knowledge
is accumulated to allow the confident design of a practical product. This accumulation

of knowledge, which is usually company confidential and is not available at universities

or in the public domain, becomes the intellectual investment and capital of the
enterprise, which assures it a competitive commercial position for years in the future.
The real value, sold by outstanding companies, does not consists of the bare hardware
but of the quality assurance and technical support that these companies provide with the
product.
Product development consists of the application of analytical design tools and

empirical design data to the design and development of a particular product, designed to
specifications by a certain deadline within a projected budget. Specifications,
deadlines, and budgets are prescribed by business considerations, which are affected by
marketing, manufacturing, and financial factors beyond any technological issues.
The classification of fluid machinery usually includes, quite broadly, machines

imparting energy to, or extracting energy from, a fluid stream. The definition comprises
positive-displacement and fluid dynamic pumps and compressors, which are often
incorporated into power-generating systems such as engines or gas turbines or into
refrigeration and various chemical process systems. Even ejectors would be included.
However, in this instance, the designation will be restricted to fluid machines handling
incompressible fluids, in particular positivedisplacement or fluid dynamic pumps.

ENERGY TRANSFER
Fundamental analytical studies arrived at the conclusion that efficient energy transfer to
or from a fluid must involve an unsteady, periodic, cyclic process (Dean 1959). The
need for such a process is evident in positive-displacement pumps. However, pressure
fluctuations are also necessary in centrifugal and axial flow pumps. The rotating blades
necessarily produce a periodic passage of highand low-pressure regions when seen by
a stationary observer at the impeller exit. Only they can assure high efficiencies.
Fluid machinery does exist that transfers energy by steady viscous shear such as
ejectors (Cunningham 1974; Hansen et al. 1965; Huang et al. 1985; Sanger 1970) and
shear pumps (Craford and Rice 1974; Rice 1965; Truman et al. 1978). Shear pumps
consist of closely spaced, parallel disks mounted on a rotating shaft which entrain liquid
without blades, by the friction of the disk faces. However, the viscous friction required
for energy transfer in these devices implies viscous energy losses. Practical ejector
efficiencies rarely exceed 20%, and shear pump efficiencies remain considerably
below the efficiencies of centrifugal pumps. Only unusual requirements, such as low
noise or insensitivity to contamination, can justify their use. Ejectors have the advantage
of no moving parts.
A hydraulic ram appears to operate on the principle of ejectors. In these devices

slugs of a high-volume, low-head stream periodically compress batches of a lesser
volume stream to higher pressure by the ram effect (Iversen 1975; Young 1997). The
hydraulic ram cycles and uses the inertia of the driving fluid (not viscous forces) to
transfer energy. It accelerates and then suddenly stops the flow in a long pipe, producing
a short-duration, high-pressure pulse. Therefore, energy is transferred by periodic
pressure fluctuations, not by viscous shear.

SPECIFIC SPEED
Specific speed N, is a well-known figure of merit that characterizes the specifications
of the fluid machinery application and can suggest the most appropriate type of
machinery. It is dimensionless if consistent units, SI or English, are used; however, in
industrial practice using English units of measurements, it is expressed in units of rpm,
gpm, and feet.
In the English system, N is the shaft speed (rpm), Q the flow rate (gpm), and H the head
(ft). Low specific speeds correspond to low flow rates and high pressures, while high
specific speeds correspond to high flow rates and low pressures.
In turbomachinery selection and design, the specific speed of the application suggests
the most efficient configuration: centrifugal, mixed flow, or axial turbomachines,
depending on increasing specific speed. Best turbomachinery efficiencies correspond to
a specific speed of about 2300 in English units (rpm, gpm, and ft). Efficiency declines
rapidly with decreasing specific speed. Since the application sometimes specifies only
the flow rate and head, leaving the shaft speed open, efficiency can be improved by
choosing a shaft speed that changes the specific speed toward more optimal values.
Specific speed also governs the choice of hydraulic turbine configurations. At low
specific speeds, pure impulse Pelton turbines give the highest efficiencies; centrifugal
Francis turbines and axial Kaplan turbines are used at progressively higher specific
speeds (Streeter 1961).
These trends result from fundamental physical principles, not from an arbitrary
classification. At low specific speeds, high pressures, and low flow rates, leakage
flows, increasing with pressure levels, become significant compared with the flow rate.
Windage losses also become relatively important. At high specific speeds, high flow
rates, and low pressures, the frictional flow losses predominate. The particular design
configurations, optimal at the corresponding specific speed, minimize the prevailing
losses.
Use of the specific speed to identify suitable fluid machinery configurations is not
restricted to turbomachinery or fluid dynamic machinery (Balje 1962; Cartwright 1977).
The lower shaft speeds favor positive-displacement machinery, as do high pressures

and relatively low flow rates. The maximum head capability of turbomachinery is
primarily a function of the tip speed, the circumferential velocity of the impeller, and is
limited in commercial applications by the maximum speed of standard single-pole
electric induction motors. Higher pressures, especially at small flow rates, call for
screw, gear, or other rotary positivedisplacement pumps, or ultimately for reciprocating
piston pumps. Oil hydraulic pumps, handling pressures of up to 5000 psi and flow rates
of just a few gpm, need to be of the positive-displacement type, preferably piston pumps
(Tuzson 1978).
These analytical considerations are born out by the marketplace. Refrigeration and
shop-air compressors best illustrate this trend. Approximately constant pressure ratings
are set by industry standards, regardless of capacity. Flow rate is proportional to
capacity. In turn the specific speed increases with flow rate and capacity. The smallest-
capacity units sold commercially invariably use positivedisplacement piston or rotary
compressors, piston compressors being restricted to low speeds, on the order of 1000
rpm, and rotary compressors to electric motor speeds of 3600 rpm. At larger capacities,
typically for commercial applications, screw compressors predominate, the speed of
which is often geared up, since optimal efficiencies are reached at several thousand
rpm. In very large industrial units, centrifugal or even axial compressors, driven at high
speeds, are used. Small refrigeration systems have been proposed in the past using
centrifugal compressors. However, these would require very high shaft speeds, on the
order of 50,000 rpm, to reach sufficiently high specific speeds, which would assure
satisfactory efficiencies. To date, no such systems are commercially competitive.
Evidently, the specific speed of the application is a valuable guide to selection of the
optimal fluid machinery configuration.
The most consistent technological trend through centuries consisted of increasing

shaft speeds. An evident economic factor motivates the trend. If the shaft speed of a
fluid-handling machine can be increased without significant modification to the
hardware, the capacity of the device increases without additional cost and the price per
unit capacity is reduced. Fluid machinery prices generally increase with the capacity,
resulting in constant or slowly declining prices per unit capacity. Centuries ago, pumps
were slow-running reciprocating units. Earlier hand drives gave way to waterwheel
drives and were replaced by steam-driven linear reciprocating pumps in the early
nineteenth century. Centrifugal pumps did not become practical until higher-speed
drives appeared in particular electric motors.

FLUID MACHINERY TYPES

Fluid machinery types can be classed approximately in order of increasing specific
speed. The highest pressures are encountered in oil hydraulic machinery and some
chemical processes. Oil hydraulic applications typically require flow rates below
about 100 gpm. Slow-running hydraulic motors use radial piston or cam-actuated
piston devices, which can deliver very high torque (Olson 1966). Potentially high
leakage and very high bearing loads dictate the use of highviscosity lubricating oils.
Operation with water would lead to immediate seizure. These devices are reminiscent
of early radial aircraft engines, now obsolete. Similar engines have been proposed
even recently (Decher 1984; Murray 1990) but they have not been developed to the
point of commercialization.
Piston devices are the most popular, with a crank or connecting rod mechanism.
However, free piston (Braun and Schweitzer 1973; Huber 1988), Scotch yoke (Stiller
1990; Weiss 1986), and swashplate devices (Robertson 1981; Scott 1981; Weber 1978;
Yu and Lee 1986) have been proposed and tested, but are not commercially available.
Axial piston-, swashplate-, or wobble plateactuated devices did find mass-production
applications in automotive airconditioning compressors. They represent the preferred
configuration for highefficiency oil hydraulic pumps, often with variable capacity
(Olson 1966; Tuzson 1978).
A great variety of rotary positive-displacement fluid machines can be found on the

market (Nelik 1999). The German Vereins Deutscher Ingenieure (VDI) has published
periodic reviews on positive-displacement pumps (VDI 1975). For low flow rates and
moderately high pressures of a few hundred psi, gear pumps with fixed clearances are
preferred (Scheel 1970). Minimum clearance widths are limited by manufacturing
tolerance stackup, thermal expansion, and shaft runout. The gear tooth shapes can take
several forms, as for example in the Roots-type blower. Internal and external gear
arrangements can be distinguished. External gear types compete with piston pumps in
oil hydraulic applications when fixed flow rates are acceptable. They can reach
pressures of 1000 or 2000 psi with fixed clearances, and up to several thousand psi
when pressure-balanced side plates are used to minimize leakage. Most inexpensive oil
pumps, which form part of the integral lubrication system of engines, are of the internal
gear type.
The Wankel family type of machines forms an exceptional configuration of internal

gear arrangements (Yamamoto 1981). Leakage is minimized with sliding tip seals.
Recently developed ceramic materials have overcome earlier apex seal wear problems.
Wankel engines, mass produced by the millions by Mazda, represent the only engine
type that has successfully challenged piston engines in automotive applications. The
Wankel configuration has also been mass produced in a refrigeration compressor
application.
The sliding vane type of fluid machinery configuration tends to overcome the tip
leakage problem of fixed clearance devices (Karmel 1986; Peterson 1966). Lubrication
of the sliding vanes remains a problem, however, and calls for the use of viscous oils.
Such lubrication problems, especially at high temperatures, frustrate attempts at engine
applications (Bassett 1990; Ewing 1982; Pennock and Beard 1988). A variable
delivery configuration in which the outer cam can be displaced is on the market for oil
hydraulic applications. Simple inexpensive positive-displacement pumps having rotors
with integral vanes made of flexible rubber are on the market, but their limited life rules
out demanding industrial applications.
Double- (Adkins and Larson 1970; Kaneko and Hirayama 1985) or singlescrew
(Chan et al. 1981; Zimmern and Patel 1972) pumps and compressors run with fixed
clearances. With low-viscosity fluids, as in compressor applications, for example, high
speeds on the order of 4000 to 8000 rpm are needed to minimize the effect of leakage.
Pressure ratings remain a few hundred psi unless viscous fluids are handled or sealing
oil is injected into those in compressor applications. The popularity of these pumps has
increased because of recent advances in manufacturing tolerances and quality control.
The elongated configuration of internal screw pumps presents an advantage in certain
applications.
An unusual, very low specific speed pump is the pitot tube pump. It consists of a
rotating water-filled case and a stationary radial pipe ending in a nozzle that points in
the direction opposed to the sense of rotation on the periphery of the rotating case. The
fluid, moving with the peripheral speed of the case, strikes the nozzle and can produce a
maximum pressure equal to the stagnation pressure of the fluid, when no flow enters the
nozzle.
Regenerative or side-channel turbomachinery occupies an intermediate position

between positive-displacement and centrifugal machinery (Hollenberg and Potter 1979;
Sasahara et al. 1980; Senoo 1956; Wilson et al. 1955). These pumps can deliver
pressures in excess of those of centrifugal pumps and can remain competitive in
selected applications, especially at low flow rates, but have generally low efficiencies.
The unusual approach of partial admission was attempted to extend centrifugal pump
operation to low specific speeds (Wonsak 1963). In this device, intended for very low
flow rates, the inlet flow is channeled to only one or two blade passages of a rotating
radial impeller. At an appropriate circumferential location at the blade exit, a diffuser is
positioned that receives the flow from the particular flow passage that happens to be
filled with liquid. It is essentially a constantpressure impeller in which the fluid is
accelerated to high speed. The diffuser achieves a pressure increase by slowing the
fluid. Efficiency will correspond to the diffuser pressure recovery, which might be in
the range 50 to 70%, at most.
Centrifugal pumps reach acceptable efficiencies above a specific speed of about 500.
Optimal efficiencies correspond to specific speeds of about 2300. The tip speed of
centrifugal pumps, the product of impeller radius and the rotational speed, limits their
pressure rating. Higher pressures demand multistage pumps. In smaller commercial
units a single-pole synchronous motor speed of 3600 rpm is almost universal and results
in a pressure rise per stage on the order of 25 to 200 psi, depending on impeller size. In
process applications, variable-frequency drives are justified by better control. In larger
commercial or industrial centrifugal pumps, lower speeds tend to be used. An
extraordinary variety of special and custom-designed units have been developed in the
past. At extremely high power densities and very high speeds and pressures, rotor-
dynamic considerations begin to dominate feasibility (Cooper 1996).
Although radial inflow turbines are more efficient, inexpensive centrifugal pumps can
also be used as power recovery turbines (Nelik and Cooper 1984). Power can be
recovered in certain chemical processes and in water distribution systems by (instead of
throttling) converting the energy from the reduction of the pressure of a fluid stream to
electrical energy. Large hydroelectric water turbines are often designed to be reversible
and act as pumps for pumped storage systems. In these cases the power produced by
operating thermal power stations continuously during hours of low demand is stored by
pumping water up into reservoirs, from where it is let down to produce peaking power
during hours of high demand.
Mixed flow pump configurations, intermediate between centrifugal and axial flow
configurations, overlap with centrifugal pumps at the lower end of their specific speed
range. Mixed flow configurations are favored in the case of radial space limitations,
such as in vertical multistage pumps which must fit in a certain well casing size. The
factors affecting the choice of mixed flow pumps are similar to those for centrifugal
pumps except for certain peculiarities of their characteristics: potential instabilities at
reduced flow rates, for example.

Axial flow pumps become cost-effective at specific speeds of several thousand. They
are usually single-stage designs since the need for higher pressures would lead to the
choice of mixed flow pumps at better efficiencies rather than multistage axials. An axial
stage or an inducer with spiral-shaped blades is sometimes used in front of centrifugal
stages to minimize cavitation problems by producing a small pressure rise ahead of the
impeller inlet. Axial pumps are most suitable for conveying very large amounts of fluids
at very low pressures, such as flood control or dewatering applications.

FLUID MACHINERY COMPONENTS AND SYSTEMS

The feasibility and cost-effective functioning of fluid machinery depend on the entire
system, not only on the fluid machine alone. Elements of the system assembly are often
supplied by independent component manufacturers, all of whom are collectively held
responsible by users for undisturbed operation of the system. The rating and reliability
of the components or subassemblies must be matched to the fluid machine. Such
components are, for example, bearings, seals, lubrication systems, and drives.
Seals
Seals rank as the most frequent service item in centrifugal pumps. Their purpose
distinguishes two different kinds of seals. Seals preventing backflow from the impeller
exit to the inlet are called wear rings. Since leakage detracts directly from the useful
flow rate of a pump, minimizing leakage is indispensable. Wear rings, as their name
indicates, also act as water-lubricated bearings and even allow metal-to-metal contact
in case of emergency. Bearing forces from wear rings can significantly affect the rotor
dynamics of the pump assembly and must be considered in rotor dynamic analysis.
Equally important are the shaft seals, which prevent leakage of the fluid being
pumped from the pressure enclosure of the piping system and the intrusion of air into the
piping system when subatmospheric pressures exist in the pipes. Careful design assures
the balancing of the pressure forces on the mating stationary and rotating faces of the
shaft seal, to minimize the clearance space between the seal faces. On the one hand,
carbon or very hard materials are chosen to minimize friction and wear; on the other
hand, flexibility is provided to allow the seal faces to accommodate misalignment,
thermal distortion, and shaft runout. Proper alignment and installation of face seals are
prerequisites of proper functioning and are important factors in seal failure.
The issue of leakage becomes particularly important in the case of environmentally

harmful fluids. Absolute sealing can be assured when canned, hermetic electric motor
and pump assemblies are used, in which the motor itself is immersed in the fluid being
pumped and the fluid fills the air gap between the stator and the rotor. In centrifugal
pumps that handle water containing suspended materials, special provisions need to be
made to keep the suspended matter away from the shaft seal; otherwise, rapid
deterioration and damage to the seal will occur. Seal contamination is prevented by
continually flushing the seal with purified water. In some instances the contaminated
water is purified by tapping into the pump outlet and piping a small stream to a

hydraulic cyclone centrifugal separator, and from there to the vicinity of the seal.
In fixed-clearance positive-displacement pumps, manufacturing tolerance stackup,

thermal and mechanical deformation, and shaft runout limit the minimum clearance
width. A circular bore and piston that can be honed to very precise dimensions offers
the most favorable geometry for sealing, and also acts as a bearing. Piston rings and O-
rings have the most convenient geometry for best sealing. Self-adjusting pressure-
balanced side plates in high-pressure oil hydraulic pumps actively minimize the side
clearance width.

Bearings
The choice of the principal bearing types, hydrodynamic or rolling element, is dictated
by the application. The bearing life required is the most important factor. Rolling
element bearings have a well-defined life expectancy; hydrodynamic bearings do not.
The life of rolling element bearings is adequate for aircraft or automotive applications,
for example, which require approximately 2000 hours of operation, corresponding to
about 100,000 miles of road travel for cars. The life requirement of consumer
appliances typically remains below this level. Commercial machinery, including
electric motors and directly coupled pumps, fall into a category requiring about 25,000
hours of life, about three years of continuous operation at 8760 hours per year. Rolling
element bearings can still satisfy this requirement, provided that no side load is
applied to the bearings, as it would be in belt, chain, or gear drives, and therefore the
bearings only have to support the weight of the rotating parts. On the other hand, in
power-generating machinery or process machines in practically constant operation,
rolling element bearings will not last and sleeve bearings are mandatory. Power-
generating machinery is designed for 40-year life with, at most, annual scheduled
service. Process machinery applications often demand similar performance. Reliability
becomes of primary importance because lifetime maintenance and service costs exceed
the first cost in these applications.
Rolling element ball or roller bearings rely on rolling-contact Hertzian stresses to

support the load. These periodic stresses lead eventually to fatigue failure of the ball or
race. Slower-running and smaller rolling element bearings are lubricated with grease
and are sealed. At very high speeds a fine mist lubrication is needed, since excessive
oil or grease would increase the rolling friction and the heat generated. Consequently, a
separate, active lubrication system is needed for such applications. Bearing life is
proportional to the number of cycles or number of shaft revolutions, and typically
depend on the third power of the bearing load. Evidently, an increase in bearing load
reduces bearing life significantly. Successful applications with bearing lives in excess
of 20,000 hours are very rare.
Hydrodynamic or sleeve bearings rely on the viscosity of the lubricating oil to

produce a bearing force on the eccentrically positioned rotating shaft. These bearings
must either be flooded with oil or need an external source of oil lubrication. The
rotating shaft in a simple sleeve bearing must deflect to an eccentric position to produce
a bearing force. However, the bearing force does not oppose the deflection directly but
has a component in the transverse direction. These complex interactions between the
direction of the deflection and the corresponding force lead to complex rotor dynamic
effects and can even result in instabilities. Tilting pad hydrodynamic bearings have been
conceived to counteract some of these problems. Multiple pivoting pads, arranged
around the rotating shaft, self-adjust in response to the displacement and load applied.
In some machinery applications, externally pressurized bearings become possible,

which, however, require a separate lubrication oil pump, typically of the gear type. The
cost and complexity of an external lubricating system penalizes hydrostatic bearings.
Such a separate lubrication system is more likely to be justified for larger machines.
Hydrostatic pads can be made to support large loads and can be self-adjusting to assure
and maintain close running clearances.
Bearing forces also arise from rotor unbalance forces, which depend on the
unbalance or eccentricity tolerated in manufacturing, and increase with shaft speed. The
life of an electric motor, mounted on ball bearings and balanced for normal duty at 3600
rpm, will rapidly deteriorate when operated at or above 5400 rpm. Since the load that a
specific bearing has to support depends in a complex manner on the magnitude and
distribution of the rotating masses, only a detailed rotor dynamic analysis can estimate
the load bearing capability required from the bearing (Childs 1993).
During the last decade, electric bearings have been introduced, especially for larger
industrial machinery (Habermann and Brunet 1985; Maslen et al. 1989). Bearing forces
are produced, and the shaft is suspended without touching the bearing, by
electromagnetic action of coils, arranged around the bearing periphery. Electric
bearings have the advantage of precise control of the instantaneous bearing forces, and
diagnosis of any anomalous behavior. Need for a separate lubrication system, and
associated maintenance, is eliminated.
For very high rotor speeds of 50,000 to 100,000 rpm or higher, hydrodynamic
bearings operating on air or some other gas, sometimes also called foil bearings, have
been proposed and developed. Specialized applications, such as for dentist drills, exist,
but their range of applicability and economic significance is very limited (Turner 1992).

Drives
The most frequent pump driver is an electric motor with a rotational speed of 3600,
1800, 1200, 900, or 720 rpm. The pump impeller is either mounted directly on the
motor shaft or has a separate shaft and bearing assembly, in which case motor and
pump are mounted on a frame and their shafts are connected with a coupling. A special
procedure may be needed for startup, depending on the pump load. Very large electric
motors, for example in power generation applications, may only be started twice in
quick succession because of the danger of overheating.
When evaluating the cost and economic justification of an electric motor drive, the
cost of needed accessories must also be counted. Electric motors need switchgear,
starters, and circuit breakers, which sometimes cost as much as the motor itself. Larger
motors may need special high-voltage transmission lines and voltage transformers or
phase-correction equipment.
A variable-frequency variable- or high-speed electric motor drive for pumps may be

desirable for three reasons. The maximum pressure produced by centrifugal pumps is
limited primarily by the rotational speed. The need for complex, expensive multistage
pumps could be avoided if a high-speed driver were available. The second reason is
that better, more efficient control of the pump flow can be achieved with variable pump
speed. In typical process applications a remotely controlled throttling valve follows the
pump, which reduces the pump pressure by about 30% under normal operating
conditions, to allow a suitable margin to increase or decrease the flow rate. Such
throttling wastes 30% of the pumping power. Similar control can be achieved with
variable speed without any losses. The third reason for the use of variable-speed
motors is a better match of characteristics of the pump and the system load. Process
equipment characteristics often change with time because of the deterioration of the
equipment or because of modifications of the system. In such instances, the performance
of existing equipment can sometime be adapted to the changed conditions by changing
the rotational speed. For example, slurry pump wear can increase significantly if the
pump is not operated at its rated condition. A change in speed can readjust pump
performance to design conditions. Slurry pumps are often belt driven to provide an
opportunity for speed adjustment. The cost of variable-frequency converters approaches
$100 or more per kilowatt capacity, which is often prohibitive (EPRI 1985). The
limited voltage capability of electronic components makes smaller-capacity, relatively
lower voltage converters less expensive and more popular than high-powered units.
In remote or mobile applications such as irrigation, pumps are driven by internal

combustion engines, most often diesels or spark-ignited natural gas engines. The shaft
speed can be variable but is usually governed at a constant speed below about 2000
rpm. Diesel engines have relatively high shaft torque oscillations, which can damage
couplings or cause bearing damage. Use of an internal combustion engine implies the
necessity for several auxiliary systems: air filter, lubrication system, cooling water
pump and heat exchanger, fuel pump, fuel tank, fuel controls, spark plugs or glow plugs,
electric generator and associated electrical controls, batteries, starter motor, exhaust
system, and safety equipment. The collective reliability, durability, cost, and proper
functioning of all these components, regardless of their provenance, will determine the
service that the fluid machine system provides to the user.

COST
The greatest uncertainty surrounds the cost of a novel machine. A detailed estimate of
the manufacturing cost only becomes possible when detailed engineering drawings and
a parts list are available, and when quotes have been obtained for all purchased items.
Availability of this information implies that development has progressed to an
advanced state and that considerable funds have already been spent on development.
An early estimate of product cost is highly desirable.
An approximate first estimate of product cost can be obtained by calculating the

anticipated weight per unit power of the machine, and by comparing this figure of merit
with similar figures from existing competitive equipment. The underlying assumption is
that the manufacturing cost of equipment with similar ratings is approximately
proportional to the weight. This assumption is more likely to hold if an item is mass
produced and therefore the cost of material accounts for a major portion of the
manufacturing cost. However, new products can rarely be mass produced. Product
introduction and market development take time. Sometimes years go by before
sufficiently large sales volumes are reached. An early large capital investment in mass
production tooling can result in high interest charges and in a negative cash flow for
years, jeopardizing the timely recovery of the investment in development and startup.
Sometimes it is easier to introduce a new kind of product based on new technology into
a small but lucrative market segment with a custom-manufactured large capacity
product, and switch to mass production after a positive cash flow has been achieved. A
cost estimate based on weight per unit power has the advantage that commercial
catalogs usually list the power rating and also the weight of the products. Therefore, it
becomes relatively easy to establish the current status of commercially available
products.
If drawings and parts lists are available, a detailed manufacturing cost estimate can
be prepared. Unfortunately, a comparison with competitive products still remains
difficult because only their sales price would be known, which includes confidential
overhead charges, the amortization of development costs, sales cost, distributor markup,
and profit. To get a valid comparison, the cost of the competitive product must be
evaluated in exactly the same manner as the new product. Drawings and parts list must
be prepared for the competitive product, also.
Ultimately, the price of a new product must be justified by the benefit it provides to
the customer. Commercial applications usually require a payback period of about three
years and corresponding amortization write-off. Obviously, product life must exceed the
payback period. In these applications, purchase decisions are made by valuing low first
cost more than lower operating cost or power consumption cost. Low cost and moderate
efficiency are desired. Utility and process applications count on 20 or even 40 years of
life and a correspondingly longer amortization period. Consequently, greater emphasis
is given to low operating and service costs, which over many years can amount to a
multiple of the purchase price.
A frequent mistake made when evaluating the feasibility of new products is

comparing the present performance of commercial products with some future
performance of the new product. Present commercial products will also be improved in
future years. Often, manufacturers have already improved product prototypes on the
shelf but must delay their introduction until the investment costs of present products are
recovered. The obstacle to early introduction is not lagging technical development but
an economic consideration. The staying power of existing products must not be
underestimated.

REFERENCES
Adkins, R. W, Larson, C. S. (1970): Basic Geometric Method in Helical Lobe
Compressor Design, ASME Paper 70-WA/FE-23.
Balje, O. E. (1962): A Study on Design Criteria and Matching of Turbomachinery,

ASME Journal of Engineering for Power, January, pp. 83102, 103-114.
Bassett, H. E. (1990): Rotary Compressor for Heavy Duty Gas Service, U.S. patent
4,960,371, October 2.
Braun, A. T., Schweitzer, P. H. (1973): The Braun Linear Engine, SAE Paper 730185.
Cartwright, W. G. (1977): Specific Speed as a Measure of Design Point Efficiency

and Optimum Geometry for a Class of Compressible Flow Turbomachines, in
Scaling for Performance Prediction in Rotodynamic Machines, Institution of
Mechanical Engineers, New York, pp. 139-145.
Chan, Y., et al. (1981): The Hall Screw Compressor for Refrigeration and Heat Pump
Duty, International Journal of Refrigeration, September, p. 275.
Childs, D. (1993): Turbomachinery Rotordynamics Phenomena: Modeling and

Analysis, Wiley, New York.
Cooper, P. (1996): Perspective: The New Face of R&D, ASME Journal of Fluids
Engineering, December, pp. 654-664.
Craford, M. E., Rice, W. (1974): Calculated Design Data for the Multiple-Disk
Pumps Using Incompressible Fluid, ASME Journal of Engineering for Power, July,
pp. 274-282.
Cunningham, R. G. (1974): Gas Compression with the Liquid Jet Pump, ASME
Journal of Fluids Engineering, September, pp. 203-215, 216-226. Also March 1975,
p. 133.
Dean, R. C., Jr. (1959): On the Necessity of Unsteady Flow in Fluid Machines, ASME
Journal of Basic Engineering, March, pp. 24-28.
Decher, R. (1984): The Britalus Brayton Cycle Engine, ASME Paper 84-GT-258.

EPRI (1985): Adjustable Speed Drives: Directory-Manufacturers and Applications,

Electric Power Research Institute, Palo Alto, Calif
Ewing, P. (1982): The Orbital Engine, SAE Paper 820348.
Habermann, H., Brunet, M. (1985): The Active Magnetic Bearing Enables Opimum
Control of Machine Vibrations, ASME Paper 85-GT-221.
Hansen, A. G., Kinnavy, R., Jett, A. V. (1965): The Design of Water Jet Pumps,
ASME Paper 65-WA/FE-31 and 32.
Hollenberg, J. W., Potter, J. H. (1979): An Investigation of Regenerative Blowers and

Pumps, ASME Journal of Engineering for Industry, May, pp. 147-152.
Huang, B. J., et al. (1985): Ejector Performance Characteristics and Analysis of Jet
Refrigeration System, ASME Journal of Engineering for Gas Turbines and Power,
July, pp. 792-802.
Huber, R. (1988): Free Piston Engines (in French), Entropie, Vol. 24, No: 141, May,
pp. 3-24.
Iversen, H. W. (1975): An Analysis of the Hydraulic Ram, ASME Journal of Fluids

Kaneko, T., Hirayama, N. (1985): Study of Fundamental Performance of Helical

Screw Expander, Bulletin of JSME, September, pp. 1970-1977.
Karmel, A. M. (1986): A Study of the Internal Forces in a Variable-Displacement

Vane Pump, ASME Journal of Fluids Engineering, June, p. 227.
Maslen, E., Hermann, P., Scott, M., Humphris, R. R. (1989): Practical Limits to the
Performance of Magnetic Bearings, Journal of Tribology, April, pp. 331336.
Murray, J. (1990): Rotocam Engine, New Scientist, August 4, p. 33.
Nelik, L. (1999): Centrifugal and Rotary Pumps, CRC Press, Boca Raton, Fla., 1999.
Nelik, L., Cooper, P. (1984): Performance ofMulti-Stage Radial-Inflow Hydraulic

Power Recovery Turbines, ASME 84 Paper WA/FM-4.
Olson, J. (1966): What IS New in Piston Pumps, Proceedings of the 22nd National
Conference on Fluid Power, October, pp. 133-141.
Pennock, G. R., Beard, J. E. (1988): A Variable Stroke Engine, SAE Paper 880577.
Peterson, R. E. (1966): What IS New in Vane Pumps, Proceedings of the 22nd

National Conference on Fluid Power, October, pp. 127-132.
Rice, W. (1965): An Analytical and Experimental Investigations of Multiple-Disk

Turbines, ASME Journal of Engineering for Power, January, pp. 29-36.
Robertson, D. (1981): Revolver Barrel Engine: The New K-Cycle, Science &
Mechanics, Winter, pp. 58-61, 101.
Sanger, N. L. (1970): An Experimental Investigation of Several Low-Area-Ratio

Water Jet Pumps, ASME Journal of Basic Engineering, March, pp. 11-20.
Sasahara, T., Yamazaki, S., Tomita, Y. (1980): Researches on the Performance of the
Regenerative Type Fluid Machinery, Bulletin of the JSME, December, pp. 2047-
2054.
Scheel, L. F. (1970): A Technology for Rotary Compressors, ASME Journal of

Engineering for Power, July, pp. 207-216.
Scott, D. (1981): Inclined Axes Design Simplifies Piston Engines/Compressors,

Automotive Engineering, April, pp. 76-82.
Senoo, Y. (1956): A Comparison of Regenerative Pump Theories, Transactions of

ASME, July, pp. 1091-1102.
Stiller (1990): The Stiller-Smith I. C. Engine, 3rd International New Energy

Technology Symposium, Planetary Association for Clean Energy, Hull, Quebec,
Canada.
Streeter, V. L. (1961): Handbook of Fluid Mechanics, McGraw-Hill, New York.
Truman, C. R., Rice, W., Jankowski, D. F. (1978): Laminar Through-Flow of Varying

Quality Steam Between Corotating Disks, ASME Journal of Fluids Engineering,
June, pp. 194-200.
Turner, A. B. (1992): Externally Pressurized Air-Bearing System with Applicability

to Small Turbomachinery, ASME Paper 92-GT-382.
Tuzson, J. (1978): Literature Sources on Positive Displacement Pump Design,

Proceedings of the 34th National Conference on Fluid Power, November, pp. 399-
402.
Tuzson, J. (1993): Evaluation of Novel Fluid Machinery Concepts, ASME Fluids

Engineering Summer Conference, Washington, D.C., FED Vol. 154, pp. 383386.
VDI (1975): Verdrangungspumpen, VDI, Berlin, pp. 542-546. Also (1973): pp. 493-
496; (1972): pp. 399-403; (1969): pp. 838-840.
Weber, A. (1978): Hydrostatische Axiallager in Axialkolbenmaschinen der

Schrdgachsenbauart, VDI, Berlin, August, pp. 736-742.
Weiss, B. (1986): Advanced Technology Engine Receives Extension, Metalworking

News, August, p. 150.
Wilson, W. A., et al. (1955): A Theory of the Fluid-Dynamic Mechanism of

Regenerative Pumps, Transactions of ASME, November, pp. 1303-1316.
Wonsak, G. (1963): Die Stromung in einer partiell beaufschlagten radialen

Gleichdruckkreiselpumpe, Konstruktion, Vol. 15, No. 3, March, pp. 99-106.
Yamamoto, K. (1981): Rotary Engine, Sankaido Co. Ltd., Tokyo.
Young, B. (1997): Design of Homologous Ram Pumps, ASME Journal of Fluids

Yu, Z., Lee, T. W. (1986): Kinematic, Structural and Functional Analysis of Wobble-
Plate Engines, Journal of Mechanical Transmissions in Automotive Design, June,
pp. 226-236.
Zimmern, B., Patel, G. C. (1972): Design and Operating Characteristics of the

Zimmern Single-Screw Compressor, Purdue International Compressor Conference.

20

COMPUTER PROGRAMS
FOR PUMPS
COMPUTER CALCULATIONS
The computer programs discussed here are intended for use in solving flow equations
for pumps. Computer programs predicting pump performance are discussed in the
appropriate chapter. Computer programs for pump selection from commercial catalog
databases, for pipe friction loss calculations, for water hammer calculation, or for
empirical pump design formulas are not discussed here.
The actual fluid flow in pumps and related components is extremely complex.
Calculations are performed by assuming a model, which is a simplified, imaginary
configuration of the flow field. If the model is too simple, the calculation results will be
trivial. If the model is too detailed, the calculation becomes too complicated. Ideally, a
good model should be simple but still provide valid, useful calculation results. A
complex and refined model is sometimes expected to give more precise results, which
unfortunately, is not always the case. The complexity of the model should be matched to
the need or economic justification for the effort required to obtain a solution.
Computer programs available today only verify the flow pattern in an existing pump
geometry. The pump geometry is an input to the computer program and must be
available. The pump flow passages must be designed before an impeller, diffuser, or
volute flow calculation computer program can be executed. Only simplified,
approximate calculation procedures are presented here, which give adequate, practical
results. They are really hand calculations but can be done using faster computers. Pumps
have to operate satisfactorily over a range of conditions that are not defined precisely or
can change with time. In the face of such inherent uncertainties, the accuracy of
calculated results does not need to be pushed to unreasonable limits. In some instances,
or for more precise design calculations, these simple methods do not suffice: for
example, when the flow becomes strongly three-dimensional. The problem of predicting
the inlet flow pattern of centrifugal pumps presents such a case. Because of the
curvature of the streamlines, pressure and velocity changes in all directions. A solution
requires that the four basic partial differential equations be solved. Numerical methods
must then be used, which are easiest to solve on high-speed computers.

Before the advent of high-speed computers, flow equations were generally based on
the ideal potential flow model, which were solved either analytically, sometimes using
conformal transformations, or by stepwise integration, which had to be performed using
hand calculations or mechanical adding machines. Because of earlier limitations of
computer capabilities, an approximate quasithree-dimensional calculation method was
later used. It consists of calculating flow in a pump impeller separately in the
meridional hub-to-shroud and circumferential blade-to-blade directions, as shown in
Fig. 20.1. The two calculations are linked by assuming averaged blade-to-blade
conditions in the meridional calculations. Theoretically, calculations could be repeated
with progressive corrections. Modem computers can execute a direct solution of the
threedimensional, partial differential equations, the Navier-Stokes equations, provided
some means are included to handle turbulent fluctuations.
Two principal approaches are in use for executing complex flow calculations: the
finite difference and finite element methods. In finite difference calculations, derivatives
of the partial differential equations, which are based on infinitely small increments of
the variables, are replaced by finite increments-hence the name finite differences. In
finite element calculations, the forces acting on small volume elements of the flow are
balanced directly. In either case, the flow region of interest is covered with a network
of mesh points at which pressure and velocity are calculated by successive iterations.
Unfortunately, because of the implied approximations, the calculations often do not
converge, and despite many trial iterations, the desired solution is not approximated.
Special calculation techniques can overcome these problems of convergence. Usually, a
fine mesh with many mesh points, and lengthy iterations, are required. The calculation
accuracy increases in proportion to the number of mesh points; the calculation time
increases roughly with the square or cube of the number of mesh points. The
assumptions regarding the inlet velocity distribution and the exit conditions, which
constitute the boundary conditions of the mathematical model, also give trouble
sometimes, in particular when inlet or exit flow recirculation is present.

Figure 20.1 Quasi-three-dimensional velocity calculation in impeller.
Powerful computational fluid mechanic (CFM) computer programs are commercially

available (Douglass and Ramshaw 1994; Lakshminarayana 1991; McNally and Sokol
1985; Muggli et al. 1997; Schachenmann and Muggli 1993). As input they require a
precise numerical description of the flow boundaries, which determines the number of
mesh points required. Sharp curvatures of the boundaries require a particularly fine
mesh. It then becomes an economic consideration whether the time and expense of
preparing and running the computer calculation is warranted by the value of the pump to
be designed. Impeller calculations have been executed with several hundred thousand
mesh points running on a workstation overnight. Such calculations are warranted for
large, reversible pump-turbines, for example, which can have a performance penalty of
$1 million per percentage point deficiency.
A key issue in the calculation of real flow pattern remains the handling of turbulence.
Theoretically, an exact solution of the fundamental partial differential equations, the
Navier-Stokes equations, will describe the instantaneous flow pattern perfectly.
However, turbulent flow is unsteady, ever changing. Design requires that averaged,
approximately steady values be calculated. The details of instantaneous fluctuations are
of little interest. Consequently, some average representation of the turbulent fluctuations
is needed, which although artificial, will be suitable for design purposes. Several such
turbulence models have been proposed and are in use which can give satisfactory
results, provided that they have been calibrated on test cases of similar flow patterns
(Bradshaw et al. 1996; He and Walker 1995; Menter 1996).


REFERENCES
Bradshaw, P., Launder, B. E., Lumley, J. L. (1996): Collaborative Testing of
Turbulence Models, ASME Journal of Fluids Engineering, Vol. 118, June, pp. 243-
247.
Douglass, R. W, Ramshaw, J. D. (1994): Perspective: Future Research Directions in

Computational Fluid Dynamics, ASME Journal of Fluids Engineering, Vol. 116,
June, pp. 212-215.
He, J., Walker, J. D. A. (1995): A Note on the Baldwin-Lomax Turbulence Model,

ASME Journal of Fluids Engineering, Vol. 117, September, pp. 528-531.
Lakshminarayana, B. (1991): An Assessment of Computational Fluid Dynamic

Techniques in the Analysis and Design of Turbomachinery, ASME Journal of Fluids
Engineering, September, pp. 315-352.
McNally, W. D., Sokol, P. M. (1985): Review: Computational Methods for Internal

Flows with Emphasis on Turbomachinery, ASME Journal of Fluids Engineering,
Vol. 107, March, pp. 6-22.
Menter, E. R. (1996): A Comparison of Some Recent Eddy-Viscosity Turbulence

Models, ASME Journal of Fluids Engineering, Vol. 118, September, pp. 514519.
Muggli, E A., Eisele, K., Casey, M. V., Gulich, J., Schachenmann, A. (1997): Flow
Analysis in a Pump Diffuser, Part 2: Validation and Limitations of CFD for Diffuser
Flows, ASME Journal of Fluids Engineering, December, pp. 978985.
Schachenmann, A., Muggli, E (1993): Comparison of Three Navier-Stokes Codes

with LDA Measurements on an Industrial Radial Pump Impeller, in Pumping
Machinery Symposium, ASME FED Vol. 154, pp. 247-252.

APPENDIX
This appendix contains FORTRAN computer listings for the programs LOSS3, INLET,
IMPEL, MIXED, CUTS, BLADE, and EXREC. They are written and can be edited in
DOS. A math processor and a FORTRAN compiler are needed to execute the
programs, unless they are modified. The programs require that an input file, FILE.DAT,
be set up containing input variables in free ASCII form. The programs generate several
output files. The principal file, FILE.OUT, presents the calculated numerical data
identified by their symbols. The output FILE.GRF lists blade coordinates as well as a
reference circle corresponding to the impeller circumference, which are in free-form
ASCII and are suitable for transfer to a plotting program. They can be edited before
plotting. The individual calculation steps of the listings can be incorporated into and
customized for more advanced computer systems. The calculation procedures
contained in the computer programs are discussed in Chapters 11 and 12.
Calculations are done in English units. Pump geometry is in inches, velocities in

ft/sec, and flow rates in gpm. (1 in. = 2.54 cm, 1 ft = 12 in. = 30.48 cm = 0.3048 m, 1
gall = 231 in3 = 3785 cm3 = 3.785 L = 0.003785 m3.)
LOSS3 offers a very simple performance estimate for centrifugal and mixed flow
pumps. It is described in detail in Chapter 10. The program provides quick,
approximate performance estimates for alternative design trials. The calculation will
not give correct results unless the loss coefficients are first calibrated on a similar
pump. Reasonable adjustments in the input variables must be made by the designer to
account for details such as nonuniform inlet flow or boundary layer blockage, for
example. The output file LOSS.OUT prints the estimated head, efficiency, various head
losses, and velocities for six selected flow rates.
INLET uses a very simple calculation to estimate the meridional velocity along the
leading edge of impeller blades. Coordinates of the hub and shroud, the approximate
radius of curvature at hub and shroud, and the leading-edge coordinates on hub and
shroud need to be listed in the input file INLET.DAT, in addition to the flow rate.
IMPEL calculates blade coordinates on the hub and shroud with the same angular
momentum distribution on all streamlines.
MIXED calculates impeller blade coordinates on hub, shroud, and mean streamlines

for a blade shape that has inclined parallel straight-line generators conncting
corresponding coordinate points on hub and shroud. Consequently, these blade shapes
can be pulled individually from an impeller mold in the direction of the generators. The
mean streamline coordinates, the angular momentum at the exit and the flow rate, must
be listed in the input file MIXED.DAT. The calculated coordinates, the angular
momentum distribution and the local blade angles, are listed in the output file
MIXED.OUT. A plot of the coordinates transferred from the output file MIXED.GRF
can verify whether the blades will pull.
CUTS prepares intersections of the blade shapes with equally spaced planes
perpendicular to the axial direction. Such intersections are customarily shown on
engineering drawings of impellers. The input file CUTS.DAT can be transferred from
the MIXED blade coordinate calculation program.
BLADE calculates radii and centers of arcs of circles representing a simple two-
dimensional blade in an axial projection. Inlet and exit blade angles and the radii at
which the arcs are to join are specified in the input file BLADE.DAT.
MIXED is also used in an example to calculate the blade coordinates of the return
channel or bowl of a multistage vertical pump. The calculation proceeds from the exit to
the inlet, in the direction opposite to the direction of flow. The mean streamline
coordinates, the flow rate and velocities at the return passage inlet, need to be input.
EXREC calculates the meridional velocity distribution along the blade trailing edge
of mixed flow pumps at the onset of exit recirculation, and the corresponding flow rate.
It assumes that at rated conditions uniform angular momentum is produced at the
impeller exit.



































INDEX

Centrifugal Pump Design Tuzson 2000

Uploaded by

Copyright:

Available Formats

Centrifugal Pump Design Tuzson 2000

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Centrifugal Pump Design Tuzson 2000

Uploaded by

Copyright:

Available Formats

******Created by ebook converter - www.ebook-converter.

******ebook converter DEMO - www.ebook-converter.com*******

******ebook converter DEMO - www.ebook-converter.com*******

******ebook converter DEMO - www.ebook-converter.com*******

******ebook converter DEMO - www.ebook-converter.com*******

2. FLUID MECHANIC FUNDAMENTALS

Practical numerical calculations

Potential flow, rotational flow and shear

Curved and vortex flow

Coordinates relative to rotating axes

3. INSTABILITY, TURBULENCE AND SEPARATION

Instability and turbulence

******ebook converter DEMO - www.ebook-converter.com*******

4. PIPES, BENDS AND DIFFUSERS

Pipes and bends

Pump specifications and selection

6. GENERAL PUMP THEORY

Pump types and specific speed

7. PUMP FLUID MECHANICS

******ebook converter DEMO - www.ebook-converter.com*******

Impeller work input

Flow at impeller inlet and exit

Relative velocity in the impeller and rothalpy

Flow separation in the impeller

Jet-wake flow pattern

Disk friction on the impeller

Axial thrust on the impeller

8. BOUNDARY LAYERS IN PUMPS

Boundary layer thickness

Boundary layers in closed flow passages

******ebook converter DEMO - www.ebook-converter.com*******

Boundary layers on rotating disks

Boundary layers in vaneless diffusers

Boundary layers inside rotating impellers

Practical pump design calculations

Net positive suction head

Cavitation at design conditions

Cavitation at off-design conditions

Cavitation damage correlation

10. PERFORMANCE CALCULATION

Impeller work input

Calibration on test data

******ebook converter DEMO - www.ebook-converter.com*******

11. PUMP DESIGN PROCEDURE

Published design procedures

Choice of impeller diameter, exit width, and exit blade

Impeller inlet diameter

Inlet blade angle

Vaned diffuser and crossover inlet

Impeller blade design

Computer calculation of blade coordinates

12. DESIGN EXAMPLES

Preliminary design and performance estimate

Inlet velocity calculation

Detailed three-dimensional blade design

Two-dimensional blade design

Return passage design

******ebook converter DEMO - www.ebook-converter.com*******

Volute design example

13. ROTOR DYNAMICS OF PUMPS

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*

ebook converter DEMO - www.ebook-converter.com*