A Review of Water Hammer

Mohamed S. Ghidaoui
email: [email protected]
Theory and Practice
Ming Zhao Hydraulic transients in closed conduits have been a subject of both theoretical study and
email: [email protected] intense practical interest for more than one hundred years. While straightforward in terms
of the one-dimensional nature of pipe networks, the full description of transient fluid flows
Department of Civil Engineering, The Hong Kong pose interesting problems in fluid dynamics. For example, the response of the turbulence
University of Science and Technology, structure and strength to transient waves in pipes and the loss of flow axisymmetry in
Hong Kong, China pipes due to hydrodynamic instabilities are currently not understood. Yet, such under-
standing is important for modeling energy dissipation and water quality in transient pipe
flows. This paper presents an overview of both historic developments and present day
Duncan A. McInnis research and practice in the field of hydraulic transients. In particular, the paper dis-
Surface Water Group, Komex International Ltd., cusses mass and momentum equations for one-dimensional Flows, wavespeed, numerical
4500 16th Avenue, Suite 100, N. W. Calgary, solutions for one-dimensional problems, wall shear stress models; two-dimensional mass
Alberta T3B 0M6, Canada and momentum equations, turbulence models, numerical solutions for two-dimensional
problems, boundary conditions, transient analysis software, and future practical and re-
search needs in water hammer. The presentation emphasizes the assumptions and restric-
David H. Axworthy tions involved in various governing equations so as to illuminate the range of applicabil-
163 N. Marengo Avenue, #316, Pasadena, ity as well as the limitations of these equations. Understanding the limitations of current
CA 91101 models is essential for (i) interpreting their results, (ii) judging the reliability of the data
email: [email protected] obtained from them, (iii) minimizing misuse of water-hammer models in both research and
practice, and (iv) delineating the contribution of physical processes from the contribution
of numerical artifacts to the results of waterhammer models. There are 134 refrences cited
in this review article. 关DOI: 10.1115/1.1828050兴

1 Introduction reality, his remarks apply equally to our technological ability to

deal with a revised or more complex view of the physical uni-
Thus the growth of knowledge of the physical aspect of reality
verse. It is in this condition that the field of closed conduit tran-
cannot be regarded as a cumulative process. The basic Gestalt
sient flow, and even more generally, the hydraulic analysis, de-
of this knowledge changes from time to time . . . During the
sign, and operation of pipeline systems, currently finds itself.
cumulative periods scientists behave as if reality is exactly as
The computer age is still dawning, bringing with it a massive
they know it except for missing details and improvements in
development and application of new knowledge and technology.
accuracy. They speak of the laws of nature, for example, which Formerly accepted design methodologies, criteria, and standards
are simply models that explain their experience of reality at a are being challenged and, in some instances, outdated and revised.
certain time. Later generations of scientists typically discover Computer aided analysis and design is one of the principal mecha-
that these conceptions of reality embodied certain implicit as- nisms bringing about these changes.
sumptions and hypotheses that later on turned out to be incor- Computer analysis, computer modeling, and computer simula-
rect. Vanderburg, 关1兴 tion are somewhat interchangeable terms, all describing tech-
Unsteady fluid flows have been studied since man first bent niques intended to improve our understanding of physical phe-
water to his will. The ancient Chinese, the Mayan Indians of Cen- nomena and our ability to predict and control these phenomena.
tral America, the Mesopotamian civilizations bordering the Nile, By combining physical laws, mathematical abstraction, numerical
Tigris, and Euphrates river systems, and many other societies procedures, logical constructs, and electronic data processing,
throughout history have developed extensive systems for convey- these methods now permit the solution of problems of enormous
ing water, primarily for purposes of irrigation, but also for domes- complexity and scope.
tic water supplies. The ancients understood and applied fluid flow This paper attempts to provide the reader with a general his-
principles within the context of ‘‘traditional,’’ culture-based tech- tory and introduction to waterhammer phenomena, a general com-
nologies. With the arrival of the scientific age and the mathemati- pendium of key developments and literature references as well as
cal developments embodied in Newton’s Principia, our under- an updated view of the current state of the art, both with respect to
standing of fluid flow took a quantum leap in terms of its theoretical advances of the last decade and modeling practice.
theoretical abstraction. That leap has propelled the entire develop-
ment of hydraulic engineering right through to the mid-twentieth 2 Mass and Momentum Equations for
century. The advent of high-speed digital computers constituted
another discrete transformation in the study and application of One-Dimensional Water Hammer Flows
fluids engineering principles. Today, in hydraulics and other areas, Before delving into an account of mathematical developments
engineers find that their mandate has taken on greater breadth and related to waterhammer, it is instructive to briefly note the societal
depth as technology rapidly enters an unprecedented stage of context that inspired the initial interest in waterhammer phenom-
knowledge and information accumulation. ena. In the late nineteenth century, Europe was on the cusp of the
As cited in The Structure of Scientific Revolutions, Thomas industrial revolution with growing urban populations and indus-
Kuhn 关2兴 calls such periods of radical and rapid change in our tries requiring electrical power for the new machines of produc-
view of physical reality a ‘‘revolutionary, noncumulative transi- tion. As the fossil fuel era had not begun in earnest, hydroelectric
tion period’’ and, while he was referring to scientific views of generation was still the principal supply of this important energy
source. Although hydroelectric generation accounts for a much
Transmitted by Associate Editor HJS Fernando. smaller proportion of energy production today, the problems asso-

Applied Mechanics Reviews Copyright © 2005 by ASME JANUARY 2005, Vol. 58 Õ 49

ciated with controlling the flow of water through penstocks and tion laws across a jump 共shock兲 关11兴. These conditions are ob-
turbines remains an important application of transient analysis. tained either by directly applying the conservation laws for a con-
Hydrogeneration companies contributed heavily to the develop- trol volume across the jump or by using the weak formulation of
ment of fluids and turbomachinery laboratories that studied, the conservation laws in differential form at the jump.
among other things, the phenomenon of waterhammer and its con- Allievi 关9,10兴 developed a general theory of water hammer
trol. Some of Allievi’s early experiments were undertaken as a from first principles and showed that the convective term in the
direct result of incidents and failures caused by overpressure due momentum equation was negligible. He introduced two important
to rapid valve closure in northern Italian power plants. Frictionless dimensionless parameters that are widely used to characterize
approaches to transient phenomena were appropriate in these early pipelines and valve behavior. Allievi 关9,10兴 also produced charts
developments because 共i兲 transients were most influenced by the for pressure rise at a valve due to uniform valve closure. Further
rapid closure and opening of valves, which generated the majority refinements to the governing equations of water hammer appeared
of the energy loss in these systems, and 共ii兲 the pipes involved in Jaeger 关12,13兴, Wood 关14兴, Rich 关15,16兴, Parmakian 关17兴,
tended to have large diameters and the flow velocities tended to be Streeter and Lai 关18兴, and Streeter and Wylie 关19兴. Their combined
small. efforts have resulted in the following classical mass and momen-
By the early 1900s, fuel oils were overtaking hydrogeneration tum equations for one-dimensional 共1D兲 water-hammer flows
as the principal energy source to meet society’s burgeoning de-
mand for power. However, the fascination with, and need to un- a2 ⳵V ⳵H
⫹ ⫽0 (2)
derstand, transient phenomena has continued unabated to this day. g ⳵x ⳵t
Greater availability of energy led to rapid industrialization and
urban development. Hydraulic transients are critical design factors ⳵V ⳵H 4
⫹g ⫹ ␶ ⫽0 (3)
in a large number of fluid systems from automotive fuel injection ⳵t ⳵x ␳D w
to water supply, transmission, and distribution systems. Today,
long pipelines transporting fluids over great distances have be- in which ␶ w ⫽shear stress at the pipe wall, D⫽pipe diameter, x
come commonplace, and the almost universal development of ⫽the spatial coordinate along the pipeline, and t⫽temporal coor-
sprawling systems of small pipe diameter, high-velocity water dis- dinate. Although Eqs. 共2兲 and 共3兲 were fully established by the
tribution systems has increased the importance of wall friction and 1960s, these equations have since been analyzed, discussed, red-
energy losses, leading to the inclusion of friction in the governing erived and elucidated in numerous classical texts 共e.g., 关20–23兴兲.
equations. Mechanically sophisticated fluid control devices, in- Equations 共2兲 and 共3兲 constitute the fundamental equations for 1D
cluding many types of pumps and valves, coupled with increas- water hammer problems and contain all the physics necessary to
ingly sophisticated electronic sensors and controls, provide the model wave propagation in complex pipe systems.
potential for complex system behavior. In addition, the recent
knowledge that negative pressure phases of transients can result in 2.2 Discussion of the 1D Water Hammer Mass and Mo-
contamination of potable water systems, mean that the need to mentum Equations. In this section, the fundamental equations
understand and deal effectively with transient phenomena are for 1D water hammer are derived. Special attention is given to the
more acute than ever. assumptions and restrictions involved in various governing equa-
2.1 Historical Development: A Brief Summary. The prob- tions so as to illuminate the range of applicability as well as the
lem of water hammer was first studied by Menabrea 关3兴 共although limitations of these equations.
Michaud is generally accorded that distinction兲. Michaud 关4兴 ex- Rapid flow disturbances, planned or accidental, induce spatial
amined the use of air chambers and safety valves for controlling and temporal changes in the velocity 共flow rate兲 and pressure 共pi-
water hammer. Near the turn of the nineteenth century, researchers ezometric head兲 fields in pipe systems. Such transient flows are
like Weston 关5兴, Carpenter 关6兴 and Frizell 关7兴 attempted to develop essentially unidirectional 共i.e., axial兲 since the axial fluxes of
expressions relating pressure and velocity changes in a pipe. Fri- mass, momentum, and energy are far greater than their radial
zell 关7兴 was successful in this endeavor and he also discussed the counterparts. The research of Mitra and Rouleau 关23兴 for the lami-
effects of branch lines, and reflected and successive waves on nar water hammer case and of Vardy and Hwang 关25兴 for turbulent
turbine speed regulation. Similar work by his contemporaries water-hammer supports the validity of the unidirectional approach
Joukowsky 关8兴 and Allievi 关9,10兴, however, attracted greater at- when studying water-hammer problems in pipe systems.
tention. Joukowsky 关8兴 produced the best known equation in tran- With the unidirectional assumption, the 1D classical water ham-
sient flow theory, so well known that it is often called the ‘‘fun- mer equations governing the axial and temporal variations of the
damental equation of water hammer.’’ He also studied wave cross-sectional average of the field variables in transient pipe
reflections from an open branch, the use of air chambers and surge flows are derived by applying the principles of mass and momen-
tanks, and spring type safety valves. tum to a control volume. Note that only the key steps of the
Joukowsky’s fundamental equation of water hammer is as fol- derivation are given here. A more detailed derivation can be found
lows: in Chaudhry 关20兴, Wylie et al. 关23兴, and Ghidaoui 关26兴.
Using the Reynolds transport theorem, the mass conservation
a⌬V 共‘‘continuity equation’’兲 for a control volume is as follows 共e.g.,
⌬ P⫽⫾ ␳ a⌬V or ⌬H⫽⫾ (1) 关20–23兴兲
g

where a⫽acoustic 共waterhammer兲 wavespeed, P⫽ ␳ g(H⫺Z)

⫽piezometric pressure, Z⫽elevation of the pipe centerline from a
⳵
⳵t 冕 ␳ d᭙⫹
cv
冕 cs
␳ 共 v"n兲 dA⫽0 (4)

given datum, H⫽piezometric head, ␳ ⫽fluid density, V⫽ 兰 A udA where cv⫽control volume, cs⫽control surface, n⫽unit outward
⫽cross-sectional average velocity, u⫽local longitudinal velocity, normal vector to control surface, v⫽velocity vector.
A⫽cross-sectional area of the pipe, and g⫽gravitational accelera- Referring to Fig. 1, Eq. 共4兲 yields
tion. The positive sign in Eq. 共1兲 is applicable for a water-hammer
wave moving downstream while the negative sign is applicable
for a water-hammer wave moving upstream. Readers familiar with
the gas dynamics literature will note that ⌬ P⫽⫾ ␳ a⌬V is obtain-
⳵
⳵t 冕 x⫹ ␦ x

x
␳ Adx⫹ 冕 cs
␳ 共 v"n兲 dA⫽0 (5)

able from the momentum jump condition under the special case The local form of Eq. 共5兲, obtained by taking the limit as the
where the flow velocity is negligible in comparison to the length of the control volume shrinks to zero 共i.e., ␦ x tends to
wavespeed. The jump conditions are a statement of the conserva- zero兲, is

50 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

 where ␥ ⫽ ␳ g⫽unit gravity force, ␣ ⫽angle between the pipe and
the horizontal direction, ␤ ⫽ 兰 A u 2 dA/V 2 ⫽momentum correction
coefficient. Using the product rule of differentiation, invoking Eq.
共7兲, and dividing through by ␳ A gives the following nonconser-
vative form of the momentum equation:
⳵V ⳵V 1 ⳵ 共 ␤ ⫺1 兲 ␳ AV 2 1 ⳵ P ␶ w␲ D
⫹V ⫹ ⫹ ⫹g sin ␣ ⫹
⳵t ⳵x ␳A ⳵x ␳ ⳵x ␳A
⫽0 (11)
Equations 共8兲 and 共11兲 govern unidirectional unsteady flow of a
Fig. 1 Control volume diagram used for continuity equation compressible fluid in a flexible tube. Alternative derivations of
derivation Eqs. 共8兲 and 共11兲 could have been performed by applying the
unidirectional and axisymmetric assumptions to the compressible
Navier-Stokes equations and integrating the resulting expression
with respect to pipe cross-sectional area while allowing for this
⳵ 共 ␳ A 兲 ⳵ 共 ␳ AV 兲
⫹ ⫽0 (6) area to change with pressure.
⳵t ⳵x In practice, the order of magnitude of water hammer wave
Equation 共6兲 provides the conservative form of the area-averaged speed ranges from 100 to 1400 m/s and the flow velocity is of
mass balance equation for 1D unsteady and compressible fluids in order 1 to 10 m/s. Therefore, the Mach number, M⫽U 1 /a, in
a flexible pipe. The first and second terms on the left-hand side of water-hammer applications is often in the range 10⫺2 –10⫺3 ,
Eq. 共6兲 represent the local change of mass with time due to the where U 1 ⫽longitudinal velocity scale. The fact that MⰆ1 in wa-
combined effects of fluid compressibility and pipe elasticity and ter hammer was recognized and used by Allievi 关9,10兴 to further
the instantaneous mass flux, respectively. Equation 共6兲 can be re- simplify Eqs. 共8兲 and 共11兲. The small Mach number approxima-
written as follows: tion to Eqs. 共8兲 and 共11兲 can be illustrated by performing an order
of magnitude analysis of the various terms in these equations. To
1 D ␳ 1 DA ⳵ V 1 D␳A ⳵V this end, let ␳ 0 aU 1 ⫽water hammer pressure scale, ␳ 0 ⫽density of
⫹ ⫹ ⫽0 or ⫹ ⫽0 (7) the fluid at the undisturbed state, and T⫽ ␨ L/a⫽time scale, where
␳ Dt A Dt ⳵x ␳ A Dt ⳵x
L⫽pipe length, X⫽aT⫽ ␨ L⫽longitudinal length scale, ␨ ⫽a
where D/Dt⫽ ⳵ / ⳵ t⫹V ⳵ / ⳵ x⫽substantial 共material兲 derivative in positive real parameter, ␳ f U 21 /8⫽wall shear scale, and f
one spatial dimension. Realizing that the density and pipe area
⫽Darcy-Weisbach friction factor T d ⫽radial diffusion time scale.
vary with pressure and using the chain rule reduces Eq. 共7兲 to the
The parameter ␨ allows one to investigate the relative magnitude
following:
of the various terms in Eqs. 共8兲 and 共11兲 under different time
1 d ␳ D P 1 dA D P ⳵ V 1 D P ⳵V scales. For example, if the order of magnitude of the various terms
⫹ ⫹ ⫽0 or ⫹ ⫽0 in the mass momentum over a full wave cycle 共i.e., T⫽4L/a) is
␳ d P Dt A d P Dt ⳵x ␳ a 2 Dt ⳵x
(8) desired, ␨ is set to 4. Applying the above scaling to Eqs. 共8兲 and
共11兲 gives
where a ⫺2 ⫽d ␳ /d P⫹( ␳ /A)dA/d P. The historical development
and formulation of the acoustic wave speed in terms of fluid and ␳0 D P* ⳵V*
⫹ ⫽0 or
pipe properties and the assumptions involved in the formulation ␳ Dt * ⳵x*

冉 冊
are discussed in Sec. 3.
The momentum equation for a control volume is 共e.g., 关20– ␳0 ⳵ P* ⳵ P* ⳵V*
⫹MV * ⫹ ⫽0 (12)
23兴兲: ␳ ⳵t* ⳵x* ⳵x*

兺F ext⫽
⳵
⳵t 冕
cv
␳ v᭙⫹ 冕
cs
␳ v共 v"n兲 dA (9)
⳵V*
⳵t*
⫹MV *
⳵V*
⳵x*
⫹M
1 ⳵ 共 ␤ ⫺1 兲 ␳ AV * 2 ␳ 0 ⳵ P *
␳A ⳵x*
⫹
␳ ⳵x*
Applying Eq. 共9兲 to the control volume of Fig. 2; considering g␨L ␨L f
gravitational, wall shear and pressure gradient forces as externally ⫹ sin ␣ ⫹ M ␶ w* ⫽0 (13)
Ua D 2
applied; and taking the limit as ␦ x tends to zero gives the follow-
ing local form of the axial momentum equation: where the superscript * is used to denote dimensionless quantities.
Since MⰆ1 in water hammer applications, Eqs. 共12兲 and 共13兲
⳵␳ AV ⳵ ␤ ␳ AV ⳵P
2
become
⫹ ⫽⫺A ⫺ ␲ D ␶ w⫺ ␥ A sin ␣ (10)
⳵t ⳵x ⳵x
␳0 ⳵ P* ⳵V*
⫹ ⫽0 (14)
␳ ⳵t* ⳵x*
⳵ V * ␳ 0 ⳵ P * g␨ L
⳵t*
⫹
␳ ⳵x*
⫹
Ua
L f
sin ␣ ⫹ ␨ M ⫹ ␨
D 2
Td
L/a 冉 冊
␶ w* ⫽0.
(15)
Rewriting Eqs. 共14兲 and 共15兲 in dimensional form gives
1 ⳵ P ⳵V
⫹ ⫽0 (16)
␳a2 ⳵t ⳵x
⳵V 1 ⳵ P ␶ w␲ D
⫹ ⫹g sin ␣ ⫹ ⫽0 (17)
⳵t ␳ ⳵x ␳A
Fig. 2 Control volume diagram used for momentum equation Using the Piezometric head definition 共i.e., P/g␳ 0 ⫽H⫺Z), Eqs.
derivation 共16兲 and 共17兲 become

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 51

 g␳ 0 ⳵ H ⳵ V 3 Water Hammer „Acoustic… Wave Speed
⫹ ⫽0 (18)
␳a2 ⳵t ⳵x The water hammer wave speed is 共e.g., 关8,20,23,33,34兴兲,

⳵V ␳ 0 ⳵ H ␶ w␲ D 1 d ␳ ␳ dA
⫹g ⫹ ⫽0 (19) ⫽ ⫹ (25)
⳵t ␳ ⳵x ␳A a2 d P A d P

The change in density in unsteady compressible flows is of the The first term on the right-hand side of Eq. 共25兲 represents the
order of the Mach number 关11,27,28兴. Therefore, in water hammer effect of fluid compressibility on the wave speed and the second
problems, where MⰆ1, ␳ ⬇ ␳ 0 , Eqs. 共18兲 and 共19兲 become term represents the effect of pipe flexibility on the wave speed. In
fact, the wave speed in a compressible fluid within a rigid pipe is
g ⳵H ⳵V obtained by setting dA/d P⫽0 in Eq. 共25兲, which leads to a 2
⫹ ⫽0 (20) ⫽d P/d ␳ . On the other hand, the wave speed in an incompressible
a2 ⳵t ⳵x
fluid within a flexible pipe is obtained by setting d ␳ /d P⫽0 in
⳵V ⳵ H ␶ w␲ D 共25兲, which leads to a 2 ⫽Ad P/ ␳ dA.
⫹g ⫹ ⫽0 (21) Korteweg 关33兴 related the right-hand side of Eq. 共25兲 to the
⳵t ⳵x ␳A
material properties of the fluid and to the material and geometrical
which are identical to the classical 1D water hammer equations properties of the pipe. In particular, Korteweg 关33兴 introduced the
given by Eqs. 共2兲 and 共3兲. Thus, the classical water hammer equa- fluid properties through the state equation d P/d ␳ ⫽K f / ␳ , which
tions are valid for unidirectional and axisymmetric flow of a com- was already well established in the literature, where K f ⫽bulk
pressible fluid in a flexible pipe 共tube兲, where the Mach number is modulus of elasticity of the fluid. He used the elastic theory of
very small. continuum mechanics to evaluate dA/d P in terms of the pipe
According to Eq. 共15兲, the importance of wall shear, ␶ w , de- radius, thickness e, and Young’s modulus of elasticity E. In his
pends on the magnitude of the dimensionless parameter ⌫ derivation, he 共i兲 ignored the axial 共longitudinal兲 stresses in the
⫽ ␨ LMf /2D⫹ ␨ T d /(L/a). Therefore, the wall shear is important pipe 共i.e., neglected Poisson’s effect兲 and 共ii兲 ignored the inertia of
when the parameter ⌫ is order 1 or larger. This often occurs in the pipe. These assumptions are valid for fluid transmission lines
applications where the simulation time far exceeds the first wave that are anchored but with expansion joints throughout. With as-
cycle 共i.e., large ␨兲, the pipe is very long, the friction factor is sumptions 共i兲 and 共ii兲, a quasi-equilibrium relation between the
significant, or the pipe diameter is very small. In addition, wall pressure force per unit length of pipe Dd P and the circumferential
shear is important when the time scale of radial diffusion is larger 共hoop兲 stress force per unit pipe length 2ed ␴ ␪ is achieved, where
than the wave travel time since the transient-induced large radial ␴ ␪ ⫽hoop stress. That is, Dd P⫽2ed ␴ ␪ or dp⫽2ed ␴ ␪ /D. Using
gradient of the velocity does not have sufficient time to relax. It is the elastic stress-strain relation, dA⫽ ␲ d ␰ D 2 /2, where d ␰
noted that T d becomes smaller as the Reynolds number increases. ⫽d ␴ ␪ /E⫽radial 共lateral兲 strain. As a result, Ad P/ ␳ dA⫽eE/D ␳
The practical applications in which the wall shear is important and and
the various ␶ w models that are in existence in the literature are
discussed in Sec. 4. Kf
If ⌫ is significantly smaller than 1, friction becomes negligible 1 ␳ ␳ ␳
and ␶ w can be safely set to zero. For example, for the case L 2⫽ ⫹ or a 2⫽ (26)
a Kf e KfD
⫽10,000 m, D⫽0.2 m, f ⫽0.01, and M⫽0.001, and T d /(L/a) E 1⫹
⫽0.01 the condition ⌫Ⰶ1 is valid when ␨Ⰶ4. That is, for the case D eE
considered, wall friction is irrelevant as long as the simulation The above Korteweg formula for wave speed can be extended
time is significantly smaller than 4L/a. In general, the condition to problems where the axial stress cannot be neglected. This is
⌫Ⰶ1 is satisfied during the early stages of the transient 共i.e., ␨ is accomplished through the inclusion of Poisson’s effect in the
small兲 provided that the relaxation 共diffusion兲 time scale is smaller stress-strain relations. In particular, the total strain becomes d ␰
than the wave travel time L/a. In fact, it is well known that ⫽d ␴ ␪ /E⫺ ␯ pd ␴ x /E, where ␯ p⫽Poisson’s ratio and ␴ x ⫽axial
waterhammer models provide results that are in reasonable agree- stress. The resulting wave speed formula is 共e.g., 关17,23兴兲
ment with experimental data during the first wave cycle irrespec-
tive of the wall shear stress formula being used 共e.g., 关29–32兴兲. Kf
When ⌫Ⰶ1, the classical waterhammer model, given by Eqs. 共20兲 ␳
and 共21兲, becomes a 2⫽ (27)
KfD
g ⳵H ⳵V 1⫹c
eE
⫹ ⫽0 (22)
a2 ⳵t ⳵x where c⫽1⫺ ␯ p/2 for a pipe anchored at its upstream end only,
⳵V ⳵H c⫽1⫺ ␯ 2p for a pipe anchored throughout from axial movement,
⫹g ⫽0 (23) and c⫽1 for a pipe anchored with expansion joints throughout,
⳵t ⳵x which is the case considered by Korteweg 共i.e., ␴ x ⫽0).
which is identical to the model that first appeared in Allievi 关9,10兴. Multiphase and multicomponent water hammer flows are com-
The Joukowsky relation can be recovered from Eqs. 共22兲 and mon in practice. During a water hammer event, the pressure can
共23兲. Consider a water hammer moving upstream in a pipe of cycle between large positive values and negative values, the mag-
length L. Let x⫽L⫺at define the position of a water hammer nitudes of which are constrained at vapor pressure. Vapor cavities
front at time t and consider the interval 关 L⫺at⫺ ⑀ ,L⫺at⫹ ⑀ 兴 , can form when the pressure drops to vapor pressure. In addition,
where ⑀ ⫽distance from the water hammer front. Integrating Eqs. gas cavities form when the pressure drops below the saturation
共22兲 and 共23兲 from x⫽L⫺at⫺ ⑀ to x⫽L⫺at⫹ ⑀ , invoking Leib- pressure of dissolved gases. Transient flows in pressurized or sur-
nitz’s rule, and taking the limit as ⑀ approaches zero gives charged pipes carrying sediment are examples of multicomponent
water hammer flows. Unsteady flows in pressurized or surcharged
a⌬V sewers are typical examples of multiphase and multicomponent
⌬H⫽⫺ (24) transient flows in closed conduits. Clearly, the bulk modulus and
g
density of the mixture and, thus, the wave speed are influenced by
Similarly, the relation for a water hammer wave moving down- the presence of phases and components. The wave speed for mul-
stream is ⌬H⫽⫹a⌬V/g. tiphase and multicomponent water hammer flows can be obtained

52 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

 Fig. 3 Velocity profiles for steady-state and af-
ter wave passages

by substituting an effective bulk modulus of elasticity K e and an ␳ f 共 t 兲兩 V共 t 兲兩 V共 t 兲

effective density ␳ e in place of K f and ␳ in Eq. 共27兲. The effective ␶ w 共 t 兲 ⫽ ␶ ws ⫽ (28)
8
quantities, K e and ␳ e , are obtained by the weighted average of the
bulk modulus and density of each component, where the partial where ␶ ws (t)⫽quasi-steady wall shear as a function of t.
volumes are the weights 共see, 关23兴兲. While the resulting math- The use of steady-state wall shear relations in unsteady prob-
ematical expression is simple, the explicit evaluation of the wave lems is satisfactory for very slow transients—so slow, in fact, that
speed of the mixture is hampered by the fact that the partial vol- they do not properly belong to the water hammer regime. To help
umes are difficult to estimate in practice. clarify the problems with this approach for fast transients, con-
Equation 共27兲 includes Poisson’s effect but neglects the motion sider the case of a transient induced by an instantaneous and full
and inertia of the pipe. This is acceptable for rigidly anchored pipe closure of a valve at the downstream end of a pipe. As the wave
systems such as buried pipes or pipes with high density and stiff- travels upstream, the flow rate and the cross-sectionally averaged
ness, to name only a few. Examples include major transmission velocity behind the wave front are zero. Typical transient velocity
pipelines like water distribution systems, natural gas lines, and profiles are given in Fig. 3. Therefore, using Eq. 共28兲, the wall
pressurized and surcharged sewerage force mains. However, the shear is zero. This is incorrect. The wave passage creates a flow
motion and inertia of pipes can become important when pipes are reversal near the pipe wall. The combination of flow reversal with
inadequately restrained 共e.g., unsupported, free-hanging pipes兲 or the no-slip condition at the pipe wall results in large wall shear
when the density and stiffness of the pipe is small. Some ex- stresses. Indeed, discrepancies between numerical results and ex-
amples in which a pipe’s motion and inertia may be significant perimental and field data are found whenever a steady-state based
include fuel injection systems in aircraft, cooling-water systems, shear stress equation is used to model wall shear in water hammer
unrestrained pipes with numerous elbows, and blood vessels. For problems 共e.g., 关25,30,32,37,38兴兲.
these systems, a fully coupled fluid-structure interaction model Let ␶ wu (t) be the discrepancy between the instantaneous wall
needs to be considered. Such models are not discussed in this shear stress ␶ w (t) and the quasi-steady contribution of wall shear
paper. The reader is instead directed to the recent excellent review stress ␶ ws (t). Mathematically
of the subject by Tijsseling 关35兴.
␶ w 共 t 兲 ⫽ ␶ ws 共 t 兲 ⫹ ␶ wu 共 t 兲 (29)
␶ wu (t) is zero for steady flow, small for slow transients, and sig-
nificant for fast transients. The unsteady friction component at-
4 Wall Shear Stress Models tempts to represent the transient-induced changes in the velocity
It was shown earlier in this paper that the wall shear stress term profile, which often involve flow reversal and large gradients near
is important when the parameter ⌫ is large. It follows that the the pipe wall. A summary of the various models for estimating
modeling of wall friction is essential for practical applications that ␶ wu (t) in water hammer problems is given below.
warrant transient simulation well beyond the first wave cycle 共i.e.,
large ␨兲. Examples include 共i兲 the design and analysis of pipeline 4.2 Empirical-Based Corrections to Quasi-Steady Wall
systems, 共ii兲 the design and analysis of transient control devices, Shear Models. Daily et al. 关39兴 conducted laboratory experi-
共iii兲 the modeling of transient-induced water quality problems, ments and found ␶ wu (t) to be positive for accelerating flows and
共iv兲 the design of safe and reliable field data programs for diag- negative for decelerating flows. They argued that during accelera-
nostic and parameter identification purposes, 共v兲 the application of tion the central portion of the stream moved somewhat so that the
transient models to invert field data for calibration and leakage velocity profile steepened, giving higher shear. For constant-
detection, 共vi兲 the modeling of column separation and vaporous diameter conduit, the relation given by Daily et al. 关39兴 can be
cavitation and 共vii兲 systems in which L/aⰆT d . Careful modeling rewritten as
of wall shear is also important for long pipes and for pipes with
high friction. L ⳵V
K u ⫽K s ⫹2c 2 (30)
V2 ⳵t
4.1 Quasi-Steady Wall Shear Models. In conventional
transient analysis, it is assumed that phenomenological expres- where K u ⫽unsteady flow coefficient of boundary resistance and
sions relating wall shear to cross-sectionally averaged velocity in momentum flux of absolute local velocity and K s ⫽ f L/D⫽ steady
steady-state flows remain valid under unsteady conditions. That is, state resistance coefficient. Daily et al. 关39兴 noted that the longi-
wall shear expressions, such as the Darcy-Weisbach and Hazen- tudinal velocity and turbulence nonuniformities are negligible and
Williams formulas, are assumed to hold at every instant during a K u ⬇K⫽F/ ␳ AV 2 /2⫽unsteady flow coefficient of boundary resis-
transient. For example, the form of the Darcy-Weisbach equation tance, where F⫽2 ␲ DL ␶ w ⫽wall resistance force. Therefore, Eq.
used in water hammer models is 共Streeter and Wylie 关36兴兲 共30兲 becomes

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 53

 Fig. 4 Pressure head traces obtained from
models and experiment

␳ f V 2 c 2␳ D ⳵ V Theoretical investigations aimed at identifying the domain of

␶ w⫽ ⫹ (31) applicability of Eq. 共32兲 have appeared in the literature. For ex-
8 4 ⳵t
ample, Carstens and Roller 关43兴 showed that Eq. 共32兲 can be de-
Denoting c 2 by k and ␳ f V 2 /8 by ␶ ws reduces Eq. 共31兲 to the rived by assuming that the unsteady velocity profiles obey the
following: power law as follows:
k␳D ⳵V
␶ w ⫽ ␶ ws ⫹
4 ⳵t
(32)

The formulations of Daily et al. 关39兴 shows that coefficient

u 共 x,r,t 兲 共 2n⫹1 兲共 n⫹1 兲
V 共 x,t 兲
⫽
2n 2 冉 冊
1⫺
r
R
1/n
(33)

c 2 ⫽k is a measure of the deviations, due to unsteadiness, of the

wall shear and momentum flux. Therefore, k generally depends on where n⫽7 for Reynolds number Re⫽105 and increases with
x and t. This remark is supported by the extended thermodynam- Reynolds number, r⫽distance from the axis in a radial direction,
ics approach used by Axworthy et al. 关30兴. Figure 4 clearly illus- R⫽radius of the pipe. An unsteady flow given by Eq. 共33兲 de-
trates the poor agreement between model and data when using Eq. scribes flows that exhibit slow acceleration and does not allow for
共32兲 with a constant value of k. flow reversal 共i.e., does not contain inflection points兲. In fact, Eq.
The experimental data of Daily et al. 关39兴 show that k⫽0.01 共33兲 cannot represent typical water hammer velocity profiles such
for accelerating flows and k⫽0.62 for decelerating flows. On the as those found in Vardy and Hwang 关25兴, Silva-Araya and
other hand, the research of Shuy 关40兴 led to k⫽⫺0.0825 for ac- Chaudhry 关37兴, Pezzinga 关38,44兴, Eichinger and Lein 关45兴 and
celerating flows and k⫽⫺0.13 for decelerating flows. In fact, Ghidaoui et al. 关46兴. The theoretical work of Carstens and Roller
Shuy’s data led him to conclude that unsteady wall friction in- 关43兴 shows only that Eq. 共32兲 applies to very slow transients in
creases in decelerating flows and decreases in accelerating flows. which the unsteady velocity profile has the same shape as the
This result contradicts the previously accepted hypothesis, steady velocity profile. Unfortunately, the Carstens and Roller
namely, that unsteady wall friction decreases in decelerating flows 关43兴 study neither supports nor refutes the possibility of using Eq.
and increases in accelerating flows. Shuy 关40兴 attributed the de- 共32兲 in water hammer problems.
crease in wall shear stress for acceleration to flow relaminariza- The theoretical work of Vardy and Brown 关47兴 shows that Eq.
tion. Given its controversial conclusion, this paper generated a 共32兲 can be derived for the case of an unsteady pipe flow with
flurry of discussion in the literature with the most notable remarks constant acceleration. In addition, they show that this model is
being those of Vardy and Brown 关41兴. approximately valid for problems with time dependent accelera-
Vardy and Brown 关41兴 argued that Shuy’s results should not be tion as long as the time scale of the transient event greatly exceeds
interpreted as contradicting previous measurements. Instead, the the rising time, which is a measure of time required for the vor-
results indicated that the flow behavior observed in Shuy’s experi- ticity diffusion through the shear layer. Their work also warns
ments may have been different from the flow behavior in previous against using Eq. 共32兲 for problems with time dependent accelera-
experiments. Vardy and Brown 关41兴 put forward the time scale tion induced by transient events with time scales smaller than the
hypothesis as a possible explanation for the different flow behav- rising time (i.e., L/aⰆT d ).
ior between Shuy’s 关40兴 experiments and previous ones. They also Axworthy et al. 关30兴 found that Eq. 共32兲 is consistent with the
observed that, while Shuy’s experiments dealt with long time theory of Extended Irreversible Thermodynamics 共EIT兲 and sat-
scales, previous measurements dealt with much shorter time isfy the second law of thermodynamics. In addition, the EIT deri-
scales. Vardy and Brown 关41兴 provided insightful and convincing vation shows that unsteady friction formulas based on instanta-
arguments about the importance of time scale to the flow behavior neous acceleration such as Eq. 共32兲 are applicable to transient
in unsteady pipe flows. In fact, the stability analysis of Ghidaoui flow problems in which the time scale of interest 共e.g., simulation
and Kolyshkin 关42兴 concurs with the time scale hypothesis of time兲 is significantly shorter than the radial diffusion time scale of
Vardy and Brown 关41兴. Moreover, the stability analysis shows vorticity. Using the vorticity equation, Axworthy et al. 关30兴
that, while other experiments belong to the stable domain, the showed that for such short time scales, the turbulence strength and
experiments of Shuy belong to the unstable domain. structure is unchanged 共i.e., ‘‘frozen’’兲, and the energy dissipation

54 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

behind a wave front is well represented by the degree of shift in the unsteady friction when the flow is accelerated (V ⳵ V/ ⳵ t⬎0)
the cross-sectional mean value of the velocity 共i.e., dV/dt) and and small correction when the flow is decelerated (V ⳵ V/ ⳵ t⬍0)
the cross-sectional mean value of V, itself. 关50兴.
The time scale arguments by Vardy and Brown 关47兴 and Ax- Utilization of the models presented in this section requires a
worthy et al. 关30兴 represent two limit cases: very slow transients reliable estimate of the parameter k. The data of Brunone et al.
and very fast transients, respectively. In the former case, there is 关31兴, Daily et al. 关39兴, and others show that k is not a universal
enough mixing such that the acceleration history pattern is de- constant. An empirical method for estimating this parameter was
stroyed, only the instantaneous acceleration is significant to the proposed by Brunone et al. 关52兴 by fitting the decay of measured
wall shear stress. In the latter case, the pre-existing flow structure pressure head history. Moody diagram-like charts for k were de-
is frozen, there is no additional acceleration history developed veloped by Pezzinga 关44兴 using a quasi-two-dimensional turbu-
except that of instantaneous acceleration. The Axworthy et al. lence model. Vardy and Brown 关47兴 provided a theoretically-based
关30兴 argument represents a water hammer flow situation where the expression for determining the coefficient k. This expression was
acceleration behaves like a pulse, say, the flow drops from a finite successfully applied by Bergant et al. 关52兴 and Vitkovsky et al.
value to zero in a short period. 关55兴. Although the charts of Pezzinga 关44兴 and the formula of
An important modification of instantaneous acceleration-based Vardy and Brown 关47兴 are theory-based, their reliability is limited
unsteady friction models was proposed by Brunone and Golia by the fact that they rely on steady-state-based turbulence models
关48兴, Greco 关49兴, and Brunone et al. 关50,51兴. The well known to adequately represent unsteady turbulence. It should, however,
Brunone et al. 关50兴 model has become the most widely used be stressed that modeling turbulent pipe transients is currently not
modification in water hammer application due to its simplicity and well understood 共see Sec. 9兲.
its ability to produce reasonable agreement with experimental The mechanism that accounts for the dissipation of the pres-
pressure head traces. sure head is addressed in the discussion by Ghidaoui et al. 关46兴.
Brunone et al. 关50兴 incorporated the Coriolis correction coef- They found that the additional dissipation associated with the in-
ficient and the unsteady wall shear stress in the energy equation stantaneous acceleration based unsteady friction model occurs
for water hammer as follows: only at the boundary due to the wave reflection. It was shown that
after n c complete wave cycles, the pressure head is damped by a
⳵H 1 ⳵V ␩⫹␾ ⳵V factor equivalent to 关 1/(1⫹k) 兴 2n c .
⫹ ⫹ ⫹J s ⫽0 (34)
⳵x g ⳵t g ⳵t
4.3 Physically Based Wall Shear Models. This class of un-
where ␩ ⫽difference from unity of the Coriolis correction coeffi- steady wall shear stress models is based on the analytical solution
cient, J s ⫽( f 兩 V 兩 V)/2gD ⫽steady-state friction term, of the unidirectional flow equations and was pioneered by Zielke
( ␾ /g) ( ⳵ V/ ⳵ t)⫽difference between unsteady friction and its cor- 关56兴. Applying the Laplace transform to the axial component of
responding steady friction. In Eq. 共34兲, the convective term is the Navier-Stokes equations, he derived the following wall shear
dropped as the Mach number of the flow is small in water hammer expression for unsteady laminar flow in a pipe:
problems.
A constitutive equation is needed for ␩ ⫹ ␾ . Brunone et al.
关50兴 proposed
␶ w共 t 兲 ⫽
4 ␯␳
R
V共 t 兲⫹
2 ␯␳
R 冕 t⳵V
0 ⳵t⬘
共 t ⬘ 兲 W 共 t⫺t ⬘ 兲 dt ⬘ (39)

␩ ⫹ ␾ ⫽k 1⫺a冉 ⳵V
⳵x
冒 冊 ⳵V
⳵t
(35)
where t ⬘ ⫽a dummy variable, physically represents the instanta-
neous time in the time history; ␯ ⫽kinematic viscosity of the fluid;
W⫽weighting function
or in terms of wall shear stress 2 2
W 共 t 兲 ⫽e ⫺26.3744共 ␯ t/R 兲 ⫹e ⫺70.8493共 ␯ t/R 兲 ⫹e ⫺135.0198共 ␯ t/R
2
兲

␶ w ⫽ ␶ ws ⫹
4 冉
k␳D ⳵V
⳵t
⫺a
⳵V
⳵x 冊 (36)
2
⫹e ⫺218.9216共 ␯ t/R 兲 ⫹e ⫺322.5544共 ␯ t/R
␯t
2
兲

Equation 共36兲 provides additional dissipation for a reservoir-pipe- for ⬎0.02

R2
valve system when the transient is caused by a downstream sud-
den valve closure. The pressure head traces obtained from the
models and experiment are plotted in Fig. 4. It is shown that
although both the Darcy-Weisbach formula and Eq. 共32兲 with con-
W 共 t 兲 ⫽0.282095 冉 冊
␯t
R2
⫺1/2
⫺1.25000⫹1.057855 冉 冊
␯t
R2
1/2

stant k cannot produce enough energy dissipation in the pressure

head traces, the model by Brunone et al. 关50兴 is quite successful in
producing the necessary damping features of pressure peaks, veri-
for
␯t
R2
␯t ␯t
⬍0.02⫹0.937500 2 ⫹0.396696 2
R R 冉 冊 3/2

fied by other researchers 关29,52–55兴.

Slight modifications to the model of Brunone et al. 关50兴, which
renders this model applicable to both upstream and downstream
⫺0.351563 冉 冊
␯t
R2
2
(40)

transients, were proposed in 关44兴 and in 关52兴. In particular, Pezz- The first term on the right-hand side of Eq. 共39兲 represents the
inga 关44兴 proposed steady-state wall shear stress ␶ ws and the second term represents
the correction part due to the unsteadiness of the flow ␶ wu . The
冋
␩ ⫹ ␾ ⫽k 1⫹sign V 冉 冊 冒 册
⳵V ⳵V
⳵x
a
⳵x
⳵V
⳵t
(37)
numerical integration of the convolution integral in Eq. 共39兲 re-
quires a large amount of memory space to store all previously
calculated velocities and large central processing unit 共CPU兲 time
and Bergant et al. 关53兴 proposed to carry out the numerical integration, especially when the time
step is small and the simulation time large. Trikha 关57兴 used three
冋
␩ ⫹ ␾ ⫽k 1⫹sign共 V 兲 a 冏 冏冒 册
⳵V
⳵x
⳵V
⳵t
(38)
exponential terms to approximate the weighting function. The ad-
vantage of using exponential forms is that a recursive formula can
easily be obtained, so that the flow history can be lumped into the
The dependence of ␩ ⫹ ␾ on x and t as well as the flow accel- quantities at the previous time step. In this way, only the calcu-
eration is consistent with the theoretical formulations in 关30兴 and lated quantities at the previous time step needs to be stored in the
关39兴. In addition, the form of ␩ ⫹ ␾ gives significant correction for computer memory, and there is no need to calculate the convolu-

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 55

tion integral from the beginning at every time step. This reduces
the memory storage and the computational time greatly. In Suzuki
et al. 关58兴, for ␶ ⬍0.02, the summation is calculated in a normal
way; for ␶ ⬎0.02, the recursive formula similar to that of Trikha
关57兴 is used, since each of the five terms included in the weighting
function is exponential. Although Zielke’s formula is derived for
laminar flow, Trikha 关57兴 and others 关29,52兴 found that this for-
mula leads to acceptable results for low Reynolds number turbu-
lent flows. However, Vardy and Brown 关47兴 warned against the
application of Zielke’s formula outside the laminar flow regime,
but did note that the error in applying Zielke’s formula to turbu-
lent flows diminishes as the duration of the wave pulse reduces.
Vardy et al. 关59兴 extended Zielke’s approach to low Reynolds
number turbulent water hammer flows in smooth pipes. In a later
paper, Vardy and Brown 关60兴 developed an extension of the model
of Vardy et al. 关59兴 that was applicable to high Reynolds number
Fig. 5 Weighting function for different Reynolds numbers
transient flows in smooth pipes. In addition, Vardy and Brown
关60兴 showed that this model gives results equivalent to those of
Vardy et al. 关59兴 for low Reynolds number flows and to those of
often negligibly small and pipe length far exceeds flow develop-
Zielke 关56兴 for laminar flows. That is, the Vardy and Brown 关60兴
ment length. The validity of assumptions such as that the flow
model promises to provide accurate results for Reynolds numbers
remains axisymmetric 共stable兲, that the eddy viscosity is indepen-
ranging from the laminar regime to the highly turbulent regime.
dent of time, and that its shape is similar to that in steady flow, is
This model has the following form:
discussed later in the paper 共see Secs. 6 and 7兲.

␶ w共 t 兲 ⫽ ␳ f
V 共 t 兲 兩 V 共 t 兲 兩 4 ␯␳
8
⫹
D 冕
0
t
W 共 t⫺t ⬘ 兲
⳵V
⳵t⬘
dt ⬘ (41)
Understanding the connection between Eq. 共32兲 and the physi-
cally based unsteady wall friction models proposed by Zielke 关56兴
and Vardy and Brown 关60兴 further illuminates the limitations of
where instantaneous acceleration, unsteady wall friction models as de-
scribed in the previous section. In particular, it is evident from
W 共 t 兲 ⫽ ␣ exp共 ⫺ ␤ t 兲 / 冑␲ t; ␣ ⫽D/4冑␯ ; Eqs. 共39兲 and 共41兲 that Eq. 共32兲 is recovered when the acceleration
is constant. In addition, plots of W in Fig. 5 show that for flows
␤ ⫽0.54␯ Re␬ /D 2 ; ␬ ⫽log共 14.3/Re0.05兲 with large Reynolds number, this function is very small every-
and Re⫽Reynolds number. Similar to Zielke’s model, the convo- where except when ␯ t/R 2 approaches 0, that is, when t ⬘ ap-
lution nature of Eq. 共41兲 is computationally undesirable. An accu- proaches t in Eq. 共41兲. The region where W(t⫺t ⬘ ) in Eq. 共41兲
rate, simple, and efficient approximation to the Vardy-Brown un- becomes significant and provides a measure of the time scale of
steady friction equation is derived and shown to be easily the radial diffusion of vorticity T d . If the acceleration varies
implemented within a 1D characteristics solution for unsteady slowly in the region where W(t⫺t ⬘ ) is significant, it is clear that
pipe flow 关32兴. For comparison, the exact Vardy-Brown unsteady Eqs. 共39兲 and 共41兲 can be accurately approximated by Eq. 共32兲.
friction equation is used to model shear stresses in transient tur- This is simply an alternative way to state that Eq. 共32兲 is accept-
bulent pipe flows and the resulting water hammer equations are able when the acceleration is not constant as long as the time scale
solved by the method of characteristics. The approximate Vardy- of the flow disturbance far exceeds the time scale of radial diffu-
Brown model is more computationally efficient 共i.e., requires 61 -th sion of vorticity across the shear layer. Moreover, it is obvious
the execution time and much less memory storage兲 than the exact that Eq. 共32兲 is a good approximation to Eqs. 共39兲 and 共41兲 when
Vardy-Brown model. Both models are compared with measured t is small, as the integral interval is so small that the integrand can
data from different research groups and with numerical data pro- be considered as a constant. Furthermore, the time interval where
duced by a two-dimensional 共2D兲 turbulence water hammer W(t⫺t ⬘ ) is significant reduces with Reynolds number, which
model. The results show that the exact Vardy-Brown model and shows that Eq. 共32兲 becomes more accurate for highly turbulent
the approximate Vardy-Brown model are in good agreement with flows.
both laboratory and numerical experiments over a wide range of
Reynolds numbers and wave frequencies. The proposed approxi- 5 Numerical Solutions for 1D Water Hammer Equa-
mate model only requires the storage of flow variables from a tions
single time step while the exact Vardy-Brown model requires the The equations governing 1D water hammer 共i.e., Eqs. 共20兲 and
storage of flow variables at all previous time steps and the 2D 共21兲兲 can seldom be solved analytically. Therefore, numerical
model requires the storage of flow variables at all radial nodes. techniques are used to approximate the solution. The method of
A summary of the assumptions involved in deriving Eqs. 共39兲 characteristics 共MOC兲, which has the desirable attributes of accu-
and 共41兲 is in order. The analytical approach of Zielke 关56兴 in- racy, simplicity, numerical efficiency, and programming simplicity
volves the following assumptions: 共i兲 the flow is fully developed, 共e.g., 关20,23,61兴兲, is most popular. Other techniques that have also
共ii兲 the convective terms are negligible, 共iii兲 the incompressible been applied to Eqs. 共20兲 and 共21兲 include the wave plan, finite
version of the continuity equation is used 共i.e., the influence of difference 共FD兲, and finite volume 共FV兲 methods. Of the 11 com-
mass storage on velocity profile is negligible兲, and 共iv兲 the veloc- mercially available water hammer software packages reviewed in
ity profile remains axisymmetric 共i.e., stable兲 during the transient. Sec. 12, eight use MOC, two are based on implicit FD methods,
In order to extend Zielke’s approach to turbulent flows, Vardy and and only one employs the wave-plan method.
Brown 关60兴 made two fundamental assumptions in relation to the
turbulent eddy viscosity in addition to assumptions 共i兲 through 5.1 MOC-Based Schemes. A significant development in
共iv兲. First, the turbulent kinematic viscosity is assumed to vary the numerical solution of hyperbolic equations was published by
linearly within the wall shear layer and becomes infinite 共i.e., a Lister 关62兴. She compared the fixed-grid MOC scheme—also
uniform velocity distribution兲 in the core region. Second, the tur- called the method of fixed time interval—with the MOC grid
bulent eddy viscosity is assumed to be time invariant 共i.e., frozen scheme and found that the fixed-grid MOC was much easier to
to its steady-state value兲. Assumptions 共i兲, 共ii兲, and 共iii兲 are accu- compute, giving the analyst full control over the grid selection and
rate for practical water hammer flows, where the Mach number is enabling the computation of both the pressure and velocity fields

56 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

in space at constant time. Fixed-grid MOC has since been used interpolate either in space or in time. It is claimed that the reach-
with great success to calculate transient conditions in pipe systems back Holly-Preissmann scheme is superior to the classical Holly-
and networks. Preissmann method. An interesting discussion of this work, pub-
The fixed-grid MOC requires that a common time step (⌬t) be lished in Bentley 关68兴, showed that the solution obtained by the
used for the solution of the governing equations in all pipes. How- classical Holly-Preissmann method when the time step equals
ever, pipes in the system tend to have different lengths and some- m⌬t is identical to that obtained by the reachback in space Holly-
times wave speeds, making it impossible to satisfy the Courant Preissmann 共i.e., the foot of the characteristic line is extended
condition 共Courant number C r ⫽a⌬t/⌬x⭐1) exactly if a com- back more than one time step until it intersects the space-line兲
mon time step ⌬t is to be used. This discretization problem can be with m reachbacks and a time step value of ⌬t. The only differ-
addressed by interpolation techniques, or artificial adjustment of ence is that the reachback approach produces m⫺1 extra interme-
the wave speed or a hybrid of both. diate solutions at the cost of more computational time.
To deal with this discretization problem, Lister 关62兴 used linear Sibetheros et al. 关69兴 showed that the spline technique is well
space-line interpolation to approximate heads and flows at the foot suited to predicting transient conditions in simple pipelines sub-
of each characteristic line. Trikha 关57兴 suggested using different ject to simple disturbances when the nature of the transient behav-
time steps for each pipe. This strategy makes it possible to use ior of the system is known in advance. The most serious problem
large time steps, resulting in shorter execution time and the avoid- with the spline interpolation is the specification of the spline
ance of spatial interpolation error. This increased flexibility comes boundary conditions.
at the cost of having to interpolate at the boundaries, which can be The authors point out that the selection procedure was a ‘‘trial
a major source of error when complex, rapidly changing control and error’’ one involving many possibilities. This ‘‘flexibility’’
actions are considered. suffers from the curse of ‘‘permutability,’’ i.e., in a complex sys-
Wiggert and Sundquist 关63兴 derived a single scheme that com- tem the number of permutations and combinations of spline
bines the classical space-line interpolation with reachout in space boundary conditions can become enormous. Moreover, in many
interpolation. Using Fourier analysis, they studied the effects of multipipe applications it is not accuracy that directly governs the
interpolation, spacing, and grid size on numerical dispersion, at- selection of the time step, but the hydraulically shortest pipe in the
tenuation, and stability. These researchers found that the degree of system. Since the most successful spline boundary conditions nec-
interpolation ␰ decreases as the ratio of the wavelength of the kth essarily involve several reaches, application of the method be-
harmonic L k to the reach length ⌬x increases. As a result, both comes problematic in short pipes. It would appear to require much
numerical dissipation and dispersion are improved. These conclu- smaller time steps simply to apply the method at all. Other nec-
sions are not surprising for several reasons. First, every interpola- essary conditions for the success of spline schemes are: 共i兲 the
tion technique can be expected to produce better results for wave dependent variable共s兲 must be sufficiently smooth, 共ii兲 the compu-
components with larger wavelengths. Second, for a fixed time step tation of the derivatives at internal nodes must be accurate, and
⌬t, larger values of n imply smaller values of ⌬x and vice versa, 共iii兲 the formulation of the numerical and/or physical derivative
since n⌬x represents the total length of the reachout on one side. boundary conditions must be simple and accurate. Conditions 共i兲
Consequently, this scheme generates more grid points and, there- and 共iii兲 are a problem in water hammer analysis because the
fore, requires longer computational times and computer storage. boundary conditions are frequently nonlinear and complex, and
Furthermore, an alternative scheme must be used to carry out the the dependent variables may be discontinuous.
boundary computations. Karney and Ghidaoui 关70兴 developed ‘‘hybrid’’ interpolation
The reachback time-line interpolation scheme, developed by approaches that include interpolation along a secondary character-
Goldberg and Wylie 关64兴, uses the solution from m previously istic line, ‘‘minimum-point’’ interpolation 共which reduces the dis-
calculated time levels. The authors observed that reachback time- tance from the interpolated point to the primary characteristic兲,
line interpolation is more accurate than space-line interpolation and a method of ‘‘wave path adjustment’’ that distorts the path of
for the same discretization. This is a subjective comparison be- propagation but does not directly change the wave speed. The
cause, as the degree of temporal interpolation ␰ varies from 0 to 1, resulting composite algorithm can be implemented as a preproces-
the degree of spatial interpolation ␣ is only allowed to vary from sor step and thus uses memory efficiently, executes quickly, and
1/(m⫹1) to 1/m. A fairer comparison would have been to also provides a flexible tool for investigating the importance of dis-
divide the distance step by m so that both ␰ and ␣ vary equally. In cretization errors in pipeline systems. The properties of the algo-
addition, Goldberg and Wylie 关64兴 assert that numerical errors are rithm are analyzed theoretically and illustrated by example in the
reduced by increasing m. This is somewhat misleading because, paper.
for a fixed ⌬x, increasing m means increasing the number of
computational steps 共i.e., reducing the effective time step ⌬t) 5.2 Other Schemes. The wave plan method 关71兴 is similar
which in turn generates finer interpolation intervals. Moreover, in to the MOC in the sense that both techniques explicitly incorpo-
cases where the friction term is large and/or when the wave speed rate wave paths in the solution procedure. However, the wave plan
is not constant, reaching back in time increases the approximation method requires that flow disturbance functions such as valve
error of these terms. curves be approximated by piecewise constant functions. That is,
Lai 关65兴 combined the implicit, temporal reachback, spatial flow disturbances are approximated by a series of instantaneous
reachback, spatial reachout, and the classical time and space-line changes in flow conditions. The time interval between any two
interpolations into one technique called the multimode scheme. consecutive instantaneous changes in flow conditions is fixed. The
Depending on the choice of grid size (⌬t,⌬x) and the limit on the piecewise constant approximation to disturbance functions implies
maximum allowable reachbacks in time m, this scheme may func- that the accuracy of the scheme is first order in both space and
tion as either of the methods or a combination of any two meth- time. Therefore, fine discretization is required for achieving accu-
ods. Numerical errors were studied using a mass balance ap- rate solutions to water hammer problems.
proach. Stability conditions were derived from Von Neumann The wave plan method ‘‘lumps’’ friction at the center of each
analysis. The multimode scheme gives the user the flexibility to pipe. In particular, friction is modeled using a disturbance func-
select the interpolation scheme that provides the best performance tion, where the form of this function is determined using the ‘‘ori-
for a particular problem. fice analogy.’’ This disturbance function is friction approximated
Yang and Hsu 关66,67兴 published two papers dealing with the by piecewise constant functions. The modeling of friction as a
numerical solution of the dispersion equation in 1D and 2D, re- series of discrete disturbances in space and time generates small
spectively. The authors propose reaching back in time more than spurious waves. In general, with small values of friction, these
one time step and then using the Holly-Preissmann method to would be observed only as low-amplitude noise on the main tran-

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 57

sient signal. It is also unclear as to how additional physics, such as This section discusses a number of methods employed by tran-
convolution-integral unsteady friction models, can be incorpo- sient modelers to quantify numerical dissipation and dispersion.
rated with the wave plan methodology.
Wylie and Streeter 关72兴 propose solving the water hammer 5.3.1 Von Neumann Method. Traditionally, fluid transient re-
equations in a network system using the implicit central difference searchers have studied the dispersion and dissipation characteris-
method in order to permit large time steps. The resulting nonlinear tics of the fixed-grid method of characteristics using the Von Neu-
difference equations are organized in a sparse matrix form and are mann 共or Fourier兲 method of analysis 关63,64兴. The Von Neumann
solved using the Newton-Raphson procedure. Only pipe junction analysis was used by O’Brian et al. 关79兴 to study the stability of
boundary conditions were considered in the case study. It is rec- the numerical solution of partial differential equations. The analy-
ognized that the limitation on the maximum time step is set by the sis tracks a single Fourier mode with time and dissipation by
frequency of the dependent variables at the boundaries. Two com- determining how the mode decays with time. Dispersion is evalu-
mercially available water hammer software packages use the four ated by investigating whether or not different Fourier modes travel
point implicit scheme 共see Sec. 12 Water Hammer Software兲. The with different speeds.
major advantage of implicit methods is that they are stable for There are a number of serious drawbacks to the Von Neumann
large time steps 共i.e., C r ⬎1 关65,72兴兲. Computationally, however, method of analysis. For example, it lacks essential boundary in-
implicit schemes increase both the execution time and the storage formation, it ignores the influence of the wave profile on the nu-
requirement and need a dedicated matrix inversion solver since a merical errors, it assumes constant coefficients and that the initial
large system of equations has to be solved. Moreover, for most conditions are periodic, and it can only be applied to linear nu-
problems, iterative schemes must also be invoked. From a math- merical models 关69,79– 81兴. To illustrate, the work by Wiggert
ematical perspective, implicit methods are not suitable for wave and Sundquist 关63兴, Goldberg and Wylie 关64兴, and others clearly
propagation problems because they entirely distort the path of shows that the attenuation and dispersion coefficients obtained
propagation of information, thereby misrepresenting the math- from the Fourier analysis depend on the Courant number, the ratio
ematical model. In addition, a small time step is required for ac- of the wavelength of the kth harmonic L k to the reach length ⌬x,
curacy in water hammer problems in any case 关23兴. For these and the number of reachbacks and/or reachouts, but does not de-
reasons, most of the work done on numerical modeling of hyper- pend on the boundary conditions. Yet, the simulation of boundary
bolic equations in the last three decades concentrated on develop- conditions and knowledge of how these boundary conditions in-
ing, testing, and comparing explicit schemes 共e.g., 关63,64,73兴兲. troduce and reflect errors to the internal pipe sections is crucial to
Chaudhry and Hussaini 关74兴 apply the MacCormack, Lambda, the study of numerical solutions of hydraulic problems. In short,
and Gabutti schemes to the water hammer equations. These three the Von Neumann method cannot be used as the only benchmark
methods are explicit, second-order 共in time and space兲 finite dif- for selecting the most appropriate numerical scheme for nonlinear
ference schemes. Two types of boundary conditions are used: 共i兲 boundary-value hyperbolic problems.
one characteristic equation and one boundary equation, or 共ii兲 ex- 5.3.2 L 1 and L 2 Norms Method. Chaudhry and Hussaini
trapolation procedure boundary condition. The second boundary 关74兴 developed L 1 and L 2 norms to evaluate the numerical errors
condition solution method adds one fictitious node upstream of associated with the numerical solution of the water hammer equa-
the upstream boundary and another downstream of the down- tions by the MacCormack, Lambda, and Gabutti schemes. How-
stream boundary. Using the L 1 and L 2 norms as indicators of the ever, the L 1 and L 2 method as they apply it can only be used for
numerical errors, it was shown that these second-order finite- problems that have a known, exact solution. In addition, these two
difference schemes produce better results than first-order method norms do not measure a physical property such as mass or energy,
of characteristics solutions for C r ⫽0.8 and 0.5. Spurious numeri- thereby making the interpretation of the numerical values of these
cal oscillations are observed, however, in the wave profile. norms difficult. Hence, the L 1 and L 2 norms can be used to com-
Although FV methods are widely used in the solution of hyper- pare different schemes, but do not give a clear indication of how
bolic systems, for example, in gas dynamics and shallow water well a particular scheme performs.
waves 共see recent books by Toro 关75,76兴兲, this approach is seldom
applied to water hammer flows. To the authors’ knowledge, the 5.3.3 Three Parameters Approach. Sibetheros et al. 关69兴
first attempt to apply FV-based schemes was by Guinot 关77兴. He used three dimensionless parameters to study various numerical
ignored the advective terms, developed a Riemann-type solution errors. A discussion followed by Karney and Ghidaoui 关82兴 and a
for the water hammer problem, and used this solution to develop closure was provided by the authors. Salient points from the dis-
a first-order-based FV Godunov scheme. This first-order scheme cussion and closure are summarized below.
is very similar to the MOC with linear space-line interpolation. At The attenuation parameter is intended to measure the numerical
the time of writing, a second paper by Hwang and Chung 关78兴 that dissipation by different interpolation schemes by looking at the
also uses the FV method for water hammer, has appeared. Unlike maximum head value at the valve. This parameter, however, un-
in Guinot 关77兴, the advective terms are not neglected in the work derestimates the numerical attenuation because the computation of
of Hwang and Chung 关78兴. Instead, they use the conservative form head and flow at the downstream end of the pipe uses one char-
of the compressible flow equations, in which density, and not acteristic equation and one boundary equation. The dispersion pa-
head, is treated as an unknown. The application of such a scheme rameter is intended to measure the numerical dispersion by differ-
in practice would require a state equation relating density to head ent interpolation schemes. This parameter is determined by
so that 共i兲 all existing boundary conditions would have to be re- asserting that the change in the wave shape is governed by the
formulated in terms of density and flow rather than head and flow, constant diffusion equation with initial conditions described by
and 共ii兲 the initial steady-state hydraulic grade line would need to the Heaviside function. Although this method allows a rudimen-
be converted to a density curve as a function of longitudinal dis- tary comparison of simple system responses, general conclusions
tance. At present, no such equation of state exists for water. Ap- cannot be drawn for a hyperbolic equation based on a diffusion
plication of this method would be further complicated at bound- equation. The longitudinal displacement parameter is intended to
aries where incompressible conditions are generally assumed to measure the extent by which different numerical schemes artifi-
apply. cially displace the wave front. However, this parameter only sug-
gests to what degree the interpolation method used is symmetri-
5.3 Methods for Evaluating Numerical Schemes. Several cally dispersive and says little about the magnitude of artificial
approaches have been developed to deal with the quantification of displacement of the wave by the numerical scheme.
numerical dissipation and dispersion. The wide range of methods
in the literature is indicative of the dissatisfaction and distrust 5.3.4 Mass Balance Approach. The mass balance method
among researchers of more conventional, existing techniques. 关83,84兴 is a more general technique than the other existing meth-

58 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

ods since this approach can be applied to a nonlinear transient tionary helical vortices and that the breakdown of these vortices
problem with realistic boundary conditions. The basic idea is to into turbulence is very rapid. The breakdown of the helical vorti-
check how closely a particular numerical method conserves mass. ces into turbulence resulted in strong asymmetry in the flow with
Note that the mass balance approach can become ineffective in respect to the pipe axis. Brunone et al. 关31,88兴 carried out mea-
cases where a numerical scheme conserves mass but not energy surements of water hammer velocity profiles in turbulent flows.
and momentum. They also observed strong flow asymmetry with respect to the
5.3.5 EHDE Approach. Ghidaoui and Karney 关85兴 devel- pipe axis. In particular, they found that a short time after the wave
oped the concept of an equivalent hyperbolic differential equation passage, flow reversal no longer appears simultaneously in both
共EHDE兲 to study how discretization errors arise in pipeline appli- the top and the bottom sides of the pipe. Instead, flow reversal
cations for the most common interpolation techniques used to deal appears to alternate between the bottom and top sides of the pipe.
with the discretization problem in fixed-grid MOC. In particular, it This is consistent with the asymmetry observed by Das and Arak-
is shown that space-line interpolation and the Holly-Preissmann eri 关87兴. The impact of instabilities on wall shear stress in un-
scheme are equivalent to a wave-diffusion model with an adjusted steady pipe flows was measured by Lodahl et al. 关89兴. They found
wave speed, but that the latter method has additional source and that inflectional flow instabilities induce fluctuations in the wall
sink terms. Further, time-line interpolation is shown to be equiva- shear stress, where the root mean square of the wall shear stress
lent to a superposition of two waves with different wave speeds. fluctuation in the pipe was found to be as high as 45% of the
The EHDE concept evaluates the consistency of the numerical maximum wall shear stress.
scheme, provides a mathematical description of the numerical dis- Das and Arakeri 关87兴 applied linear stability analysis to un-
sipation and dispersion, gives an independent way of determining steady plane channel flow to explain the experimentally observed
the Courant condition, allows the comparison of alternative ap- instability in unsteady pipe flow. The linear stability and the ex-
proaches, finds the wave path, and explains why higher-order perimental results are in good qualitative agreement. Ghidaoui
methods should usually be avoided. This framework clearly points and Kolyshkin 关90兴 investigated the linear stability analysis of
out that numerical approximation of the water hammer equations
unsteady velocity profiles with reverse flow in a pipe subject to
fundamentally changes the physical problem and must be viewed
three-dimensional 共3D兲 perturbation. They used the stability re-
as a nontrivial transformation of the governing equations. For ex-
ample, implicit methods, while noted for their stability character- sults to reinterpret the experimental results of Das and Arakeri
istics, transform the water hammer problem into a superposition 关87兴 and assess their planar flow and quasi-steady assumptions.
of wave problems, each of which has a wave speed different from Comparison of the neutral stability curves computed with and
the physical wave speed and at least one of which has an infinite without the planar channel assumption shows that this assumption
wave speed. The infinite numerical wave speed associated with is accurate when the ratio of the boundary layer thickness to the
implicit schemes ensures that the numerical domain of depen- pipe radius is below 20%. Any point in the neutral stability curve
dence is larger than the physical domain of dependence, and ex- represents the parameters combination such that the perturbations
plains why these are highly stable. While good for stability, the neither grow nor decay. Critical values for any of these parameters
large discrepancy between the numerical and physical domains of can be obtained from the neutral stability curve. For unsteady pipe
dependence hinders the accuracy of these schemes. Another prob- flows, the parameters related are Re and t. Therefore, critical Re
lem with implicit schemes is that they are often computationally can be obtained.
inefficient because they require the inversion of large matrices. The removal of the planar assumption not only improves the
accuracy of stability calculations, but also allows for the flow
5.3.6 Energy Approach. Ghidaoui et al. 关86兴 developed an
stability of both axisymmetric and nonaxisymmetric modes to be
integrated energy approach for the fixed-grid MOC to study how
the discretization errors associated with common interpolation investigated, and for the experimental results to be reinterpreted.
schemes in pipeline applications arise and how these errors can be For example, both the work of Ghidaoui and Kolyshkin 关90兴 and
controlled. Specifically, energy expressions developed in this the experiments of Das and Arakeri 关87兴 show that the nonaxisym-
work demonstrate that both time-line and space-line interpolation metric mode is the least stable 共i.e., the helical type兲.
attenuate the total energy in the system. Wave speed adjustment, With the aim of providing a theoretical basis for the emergence
on the other hand, preserves the total energy while distorting the of helical instability in transient pipe flows, Ghidaoui and Koly-
partitioning of the energy between kinetic and internal forms. shkin 关42兴 performed linear stability analysis of base flow velocity
These analytic results are confirmed with numerical studies of profiles for laminar and turbulent water hammer flows. These base
several series pipe systems. Both the numerical experiments and flow velocity profiles are determined analytically, where the tran-
the analytical energy expression show that the discretization errors sient is generated by an instantaneous reduction in flow rate at the
are small and can be ignored as long as there is continuous work downstream end of a simple pipe system. The presence of inflec-
in the system. When the work is zero, however, a C r value close tion points in the base flow velocity profile and the large velocity
to one is required if numerical dissipation is to be minimized. The gradient near the pipe wall are the sources of flow instability. The
energy approach is general and can be used to analyze other water main parameters governing the stability behavior of transient
hammer numerical schemes. flows are Reynolds number and dimensionless time scale. The
stability of the base flow velocity profiles with respect to axisym-
6 Flow Stability and the Axisymmetric Assumption metric and asymmetric modes is studied and the results are plotted
Existing transient pipe flow models are derived under the in the Reynolds number/time scale parameter space. It is found
premise that no helical type vortices emerge 共i.e., the flow remains that the asymmetric mode with azimuthal wave number one is the
stable and axisymmetric during a transient event兲. Recent experi- least stable. In addition, it is found that the stability results of the
mental and theoretical works indicate that flow instabilities, in the laminar and the turbulent velocity profiles are consistent with pub-
form of helical vortices, can develop in transient flows. These lished experimental data. The consistency between the stability
instabilities lead to the breakdown of flow symmetry with respect analysis and the experiments provide further confirmation 共i兲 that
to the pipe axis. For example, Das and Arakeri 关87兴 performed water hammer flows can become unstable, 共ii兲 that the instability
unsteady pipe flow experiments where the initial flow was laminar is asymmetric, 共iii兲 that instabilities develop in a short 共water
and the transient event was generated by a piston. They found that hammer兲 time scale and, 共iv兲 that Reynolds number and the wave
when the Reynolds number and the transient time scale exceed a time scale are important in the characterization of the stability of
threshold value, the flow becomes unstable. In addition, they ob- water hammer flows. Physically, flow instabilities change the
served that the flow instability results in the formation of nonsta- structure and strength of the turbulence in a pipe, result in strong

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 59

flow asymmetry, and induce significant fluctuations in wall shear Jackson 关93兴 provided an estimate for the time delay from the
stress. These effects of flow instability are not represented in ex- moment the wall vortex ring was generated at the pipe wall to the
isting water hammer models. moment when significant changes in the structure and strength of
In an attempt to gain an appreciation of the importance of in- turbulence appeared near the pipe axis.
cluding the effects of helical vortices in transient models, Ghidaoui et al. 关46兴 proposed a dimensionless parameter P for
Ghidaoui et al. 关46兴 applied current transient models to flow cases assessing the accuracy of quasi-steady turbulence modeling in wa-
with and without helical vortices. In the case where stability re- ter hammer problems. This parameter is defined as the ratio of the
sults indicate that there are no helical vortices, Ghidaoui et al. time scale of radial diffusion of vorticity to the pipe core to the
关46兴 found that the difference between water hammer models and time scale of wave propagation from one end of the pipe to the
the data of Pezzinga and Scandura 关91兴 increases with time at a other. This parameter provides a measure for the number of times
mild rate. However, for the case where stability results and ex- a wave front travels from one end of the pipe to the other before
periments indicate the presence of helical vortices, it is found that the preexisting turbulence conditions start to respond to the tran-
the difference between water hammer models and the data of sient event. It follows that the frozen and quasi-steady assump-
Brunone et al. 关31兴 exhibits an exponential-like growth. In fact, tions are 共i兲 acceptable when PⰇ1, 共ii兲 questionable when P is of
the difference between models and the data of Brunone et al. 关31兴 order 1, and 共iii兲 applicable when PⰆ1. However, the last case
reaches 100% after only six wave cycles. This marked difference does not belong to the water hammer regime. These conclusions
between models and data suggests that the influence of helical are supported by the work of Ghidaoui et al. 关46兴, where they
vortices on the flow field is significant and cannot be neglected. compared the results of quasi-steady turbulence models with
available data and by the work of Ghidaoui and Mansour 关32兴
where they compared the results of frozen eddy viscosity models
7 Quasi-Steady and Frozen Turbulence Assumptions
with experimental data.
The convolution integral analytical models for wall shear in
unsteady turbulent flows derived in Vardy et al. 关59兴 and Vardy
and Brown 关60兴 assume that eddy viscosity remains ‘‘frozen’’ 共i.e., 8 Two-Dimensional Mass and Momentum Equations
time independent兲 during the transient. Turbulence closure equa- Quasi-two-dimensional water hammer simulation using turbu-
tions used by Vardy and Hwang 关25兴, Silva-Araya and Chaudhry lence models can 共i兲 enhance the current state of understanding of
关37兴, and Pezzinga 关38兴 assume that the turbulence changes in a energy dissipation in transient pipe flow, 共ii兲 provide detailed in-
quasi-steady manner and that the eddy viscosity expressions de- formation about transport and turbulent mixing 共important for
rived for steady-state pipe flows remain applicable for water ham- conducting transient-related water quality modeling兲, and 共iii兲 pro-
mer flows. An understanding of the response of the turbulence vide data needed to assess the validity of 1D water hammer mod-
field to water hammer waves is central to judging the accuracy of els. Examples of turbulence models for water hammer problems,
using either the frozen or the quasi-steady turbulence assump- their applicability, and their limitations can be found in Vardy and
tions. Hwang 关25兴, Silva-Araya and Chaudhry 关37,98兴, Pezzinga
There is a time lag between the passage of a wave front at a 关38,44兴, Eichinger and Lein 关45兴, Ghidaoui et al. 关46兴, and Ohmi
particular location along the pipe and the resulting change in tur- et al. 关99兴. The governing equations for quasi-two-dimensional
bulent conditions at this location 共e.g., 关46,92,93兴兲. In particular, at modeling are discussed in this section. Turbulence models and
the instant when a water hammer wave passes a position x along numerical solutions are presented in subsequent sections.
the pipe, the velocity field at x undergoes a uniform shift 共i.e., the The most widely used quasi-two-dimensional governing equa-
fluid exhibits a slug flowlike motion兲. The uniform shift in veloc- tions were developed by Vardy and Hwang 关25兴, Ohmi et al. 关99兴,
ity field implies that the velocity gradient and turbulent conditions Wood and Funk 关100兴, and Bratland 关101兴. Although these equa-
are unaltered at the instant of the wave passage. However, the tions were developed using different approaches and are written in
combination of the uniform shift in velocity with the no-slip con- different forms, they can be expressed as the following pair of
dition generates a vortex sheet at the pipe wall. The subsequent continuity and momentum equations:

冉 冊
diffusion of this vortex ring from the pipe wall to the pipe core is
the mechanism responsible for changing the turbulence conditions g ⳵H ⳵H ⳵u 1 ⳵rv
⫹u ⫹ ⫹ ⫽0 (42)
in the pipe. a2 ⳵t ⳵x ⳵x r ⳵r
A short time after the wave passage, the extent of vorticity
diffusion is limited to a narrow wall region and the turbulence ⳵u ⳵u ⳵u ⳵H 1 ⳵r␶
⫹u ⫹v ⫽⫺g ⫹ (43)
field is essentially frozen. In this case, both the quasi-steady tur- ⳵t ⳵x ⳵r ⳵x r␳ ⳵r
bulence and ‘‘frozen’’ turbulence assumptions are equally appli- where x,t,u,H,r are defined as before, v (x,r,t)⫽local radial ve-
cable. A similar conclusion was reached by Greenblatt and Moss locity, and ␶ ⫽shear stress. In this set of equations, compressibility
关92兴 for a temporally accelerating flow; by Tu and Ramparian is only considered in the continuity equation. Radial momentum is
关94兴, Brereton et al. 关95兴, and Akhavan et al. 关96,97兴 for oscilla- neglected by assuming that ⳵ H/ ⳵ r⫽0, and these equations are,
tory flow; He and Jackson 关93兴 for ramp-type transients; and therefore, only quasi-two-dimensional. The shear stress ␶ can be
Ghidaoui et al. 关46兴 for water hammer flows. As the time after the expressed as
wave passage increases, the extent of the radial diffusion of vor-
ticity becomes more significant and begins to influence the veloc- ⳵u
ity gradient and turbulence strength and structure in the buffer ␶ ⫽ ␳␯ ⫺ ␳ u ⬘v ⬘ (44)
⳵r
zone. The experiments of He and Jackson 关93兴 show that the axial
turbulent fluctuations are the first to respond to the changes in the where u ⬘ and v ⬘ ⫽turbulence perturbations corresponding to lon-
radial gradient of the velocity profile and that there is a time delay gitudinal velocity u and radial velocity v , respectively. Turbu-
between the changes in the axial turbulent fluctuations and its lence models are needed to describe the perturbation term
redistribution among the radial and azimuthal turbulent compo- ⫺ ␳ u ⬘ v ⬘ since most practical water hammer flows are turbulent.
nents. The production of axial turbulent kinetic energy and the The governing equations can be further simplified by neglect-
time lag between production and redistribution of axial turbulent ing nonlinear convective terms, as is done in the 1D case since the
kinetic energy within the buffer zone are not incorporated in wave speed a is usually much larger than the flow velocity u or v .
steady-state-based turbulence models. The characteristics of the Then the equations become the following:
flow in the core region will start to change only when the wave-
induced shear pulse emerges from the buffer zone into the core g ⳵H ⳵u 1 ⳵rv
⫹ ⫹ ⫽0 (45)
region. On the basis of their unsteady flow experiments, He and a2 ⳵t ⳵x r ⳵r

60 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

 ⳵u ⳵H 1 ⳵r␶ 9.1 Five-Region Turbulence Model. The model used by
⫹g ⫽ (46) Vardy and Hwang 关25兴 is a direct extension of the model devel-
⳵t ⳵x r␳ ⳵r
oped by Kita et al. 关103兴 for steady flow
These governing equations are usually solved by numerical
means. ⳵u ⳵u
␶⫽␳共 ␯⫹⑀ 兲 ⫽ ␳␯ T (50)
For an adequately anchored or restrained pipe, i.e., the pipe is ⳵r ⳵r
rigid and the radial velocity at the pipe wall is zero. From flow
where ⑀ ⫽eddy viscosity, ␯ T ⫽total viscosity, and the other terms
symmetry, the radial velocity at the centerline is also zero. Inte-
were previously defined. The total viscosity distribution is com-
grating Eq. 共45兲 across the pipe section, the radial velocity van-
partmentalized into five regions as follows:
ishes, leaving the following:
⳵H a2 ⳵Q ␯T 1
⫹ ⫽0 1. viscous layer ⫽1 0⭐y ⭐ (51)
⳵t gA ⳵ x
(47) ␯ * a

⳵u ⳵H 1 ⳵r␶ ␯T 1 a
⫹g ⫽ 2. buffer I layer ⫽ay ⭐y ⭐ (52)
⳵t ⳵x r␳ ⳵r
(48) ␯ * a * CB

Q 共 x,t 兲 ⫽ 冕
A
udA (49) 3. buffer II layer
␯T
␯
⫽C B y 2
*

where Q⫽discharge. These equations are the same as those pre- a ␬

⭐y ⭐ (53)
sented by Pezzinga 关38兴. CB * C B ⫹ ␬ /4C m R
2
In cases where the radial velocity component 共mass flux兲 is *
negligible, Eqs. 共47兲–共49兲 can be usefully applied. However, the ␯T
inclusion of radial fluxes in Eqs. 共45兲 and 共46兲 remove the incon- 4. logarithmic region ⫽ ␬ y 关 1⫺ 共 ␬ /4C m 兲共 y /R 兲兴
␯ * * *
sistency that occurs near boundary elements due to the simulta-
neous imposition of the no-slip condition and the plane wave as- ␬
sumption 关24兴. Since numerical integration of Eq. 共49兲 is needed ⫻
C B ⫹ ␬ 2 /4C m R
to relate velocity distribution to discharge, even very small errors *
from neglecting radial fluxes can produce spurious oscillations in 1
⭐y ⭐ 2C m 共 1⫹ 冑1⫺C c /C m 兲 R
pressure head calculations. * ␬ *
Ghidaoui 关26兴 derived quasi-two-dimensional equations from
the complete 3D continuity equation and Navier-Stokes equations (54)
using an ensemble averaging process in which the assumptions ␯T 1
inherent in the quasi-two-dimensional equations 共such as flow axi- 5. core region ⫽C c R 2C m 共 1⫹ 冑1⫺C c /C m 兲 R ⭐y
symmetry and the plane wave assumption兲 are made explicit. The ␯ * ␬ * *
scaling analysis 关26兴 shows that the viscous terms associated with ⭐R (55)
the compressibility of the fluid are significantly smaller than the *
viscous term associated with angular deformation. Therefore, the where y⫽R⫺r, y ⫽u y/ ␯ , R ⫽u R/ ␯ , u ⫽ 冑␶ w / ␳ , and the
* * * * *
compressibility is neglected in the momentum equations of both coefficients are a⫽0.19, C B ⫽0.011, ␬ ⫽0.41, C m ⫽0.077, and
1D and 2D models. C c ⫽a function of Reynolds number 共usually a value of 0.06 is
In Silva-Araya and Chaudhry 关37,98兴 and Eichinger and Lein used兲. The total viscosity distribution depends on friction velocity
关45兴, an integration of the momentum equation is also carried out. u and position y only. This is true for steady flow since all
*
In each case, the system reduces to a 1D formulation. The quasi- information at interior points will ultimately propagate to the wall
two-dimensional momentum Eq. 共46兲 is only used to provide an boundary. Given sufficient time, the velocity profile adjusts and
unsteady friction correction for 1D governing equations. These finally depends on wall shear stress only. However, this model
corrections include: 共i兲 an energy dissipation factor, which is the may be problematic for unsteady flow since the interior conditions
ratio of the energy dissipation calculated from the cross-sectional cannot solely be represented by wall shear stress.
velocity distribution to that calculated from the Darcy-Weisbach
9.2 Two-Layer Turbulence Model. In the two-layer turbu-
formula 关37,98兴 or 共ii兲 direct calculation of wall shear stress, ei-
lence model, flow is divided into two layers: 共i兲 a smooth pipe,
ther by velocity gradient at the pipe wall or through energy dissi-
viscous sublayer is assumed to exist near the wall; and 共ii兲 outside
pation 关45兴.
the viscous sublayer, the Prandtl mixing length hypothesis is used
1. viscous sublayer ⑀ ⫽0 y ⭐11.63 (56)
9 Turbulence Models *
Turbulence models are needed to estimate the turbulent pertur-
bation term for ⫺ ␳ u ⬘ v ⬘ . In the water hammer literature, the
2. turbulent region ⑀ ⫽l 2冏 冏
⳵
⳵r
u
y ⭓11.63
*
(57)
widely used turbulence models are algebraic models where
关25,37,38,98,99兴 in which eddy viscosity is expressed as some
algebraic function of the mean flow field. Other sophisticated l y
models 共such as the k⫺ ⑀ model, which require additional differ- ⫽ ␬ e ⫺(y/R) (58)
R R
ential equations for eddy viscosity兲 have also been tried 关45兴.
Similar results for the pressure head traces have been obtained.
The algebraic turbulence models used by Vardy and Hwang
关25兴 and Pezzinga 关38兴 are discussed further to illustrate some
␬ ⫽0.374⫹0.0132 ln 1⫹ 冉 83100
Re 冊 (59)

features of algebraic turbulence models. These models were com- and in which l⫽mixing length and Re⫽Reynolds number for ini-
paratively studied by Ghidaoui et al. 关46兴. The comparison shows tial flow. The thickness of the viscous sublayer is determined by
that very similar dissipation is produced by the two models. Other the wall shear stress. The eddy viscosity in the turbulent region
different variations of algebraic turbulence models are available in includes some information about the velocity profile. The two-
Rodi 关102兴. layer model appears to be more suitable for unsteady flow simu-

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 61

lation, but one should note that the expression for mixing length Vardy and Hwang 关25兴 solves the hyperbolic part of governing
and the empirical coefficient ␬ are based on steady flow equiva- equations by MOC and the parabolic part using finite difference.
lents. This hybrid solution approach has several merits. First, the solu-
Ghidaoui et al. 关46兴 compared both models 共i.e., the five- tion method is consistent with the physics of the flow since it uses
region and the two-layer model兲 and obtained very similar results MOC for the wave part and central differencing for the diffusion
for pressure head estimates. The comparative study suggests that part. Second, the use of MOC allows modelers to take advantage
the pressure head is not sensitive to eddy viscosity distribution in of the wealth of strategies, methods, and analysis developed in
the pipe core region. As these models are based on steady flow conjunction with 1D MOC water hammer models. For example,
principles, the application of these models to unsteady flow prob- schemes for handling complex boundary elements and strategies
lems implicitly includes the quasi-steady assumptions discussed in developed for dealing with the 1D MOC discretization problem
Section 7. 共e.g., wave speed adjustment and interpolation techniques兲 can be
These algebraic turbulence models are widely used, mainly adapted to quasi-two-dimensional MOC models. Third, although
because of their simplicity and robustness. As more powerful the radial mass flux is often small, its inclusion in the continuity
computers become available and improvements are made to nu- equation by Vardy and Hwang 关25兴 is more physically correct and
merical solution techniques, detailed turbulence structures may be accurate. A major drawback of the numerical model of Vardy and
obtained using more sophisticated turbulence models, such as the Hwang 关25兴, however, is that it is computationally demanding. In
two-equation k⫺ ⑀ models, or perhaps even Reynolds stress mod- fact, the CPU time required by the scheme is of the order N r3
els, for which no eddy viscosity hypothesis is needed. where N r ⫽number of computational reaches in the radial direc-
All of the models mentioned above are based on the Reynolds- tion. Vardy and Hwang’s scheme was modified by Zhao and
averaged Navier-Stokes 共RANS兲 equation. The averaging process Ghidaoui 关105兴 to a much more efficient form. The CPU time
is clearly a time average and valid for steady flows. For unsteady required is reduced to order N r , making the scheme more ame-
flows, the use of the time average is highly questionable unless the nable to application to the quasi-two-dimensional modeling of
unsteadiness has a much larger time scale than the time scale of pipe networks and for coupling with sophisticated turbulence
turbulence. Obviously, this is not the case for fast transients. models. Several numerical schemes for quasi-two-dimensional
As an alternative, large eddy simulation 共LES兲 has been devel- modeling are summarized in the following material.
oped recently. In LES, the Navier-Stokes equation is filtered,
large-scale motion is resolved while the small-scale motion is 10.1 Vardy-Hwang Scheme. The characteristic form of
modeled. If results from LES were available, then some of the Eqs. 共45兲 and 共46兲 is as follows 关25兴:
assumptions mentioned previously could, in principle, be more
rigorously evaluated. Unfortunately, in carrying out LES, a full dH a du a2 1 ⳵q a 1 ⳵共 r␶ 兲
⫾ ⫽⫺ ⫾
3D system of equations must be solved using very fine grids dt g dt g r ⳵r g r␳ ⳵r
关104兴. For steady flow simulations, when the turbulence statistics
reach steady, the ensemble average can be obtained over a time dx
interval from a single run 关104兴. However, the ensemble average along ⫽⫾a (60)
dt
cannot be obtained from a single run for transient flow. The re-
quirement of many runs makes the resulting computational pro- where q⫽r v .
cess prohibitively time consuming. As yet, such analyses have not The pipe is divided into N r cylinders of varying thickness. At
been performed in pipe transients. a given time t and location x along the pipe, two equations apply
to each cylinder. Since there are N r cylinders in total, the total
10 Numerical Solution for 2D Problems number of equations is 2N r . Therefore, the governing equations
The 2D governing equations are a system of hyperbolic- for all cylinders can be written in matrix form as follows: Az
parabolic partial differential equations. The numerical solution of ⫽b, where A is a 2Nr⫻2Nr matrix whose form is as follows:

a
1 ⫹ ⑀ C u2 共 1 兲 ␪ C q2 共 1 兲 ⫺ ⑀ C u3 共 1 兲
g

冋 册

¨ ©
a
1 ⫺ ⫹ ⑀ C u2 共 1 兲 ␪ C q2 共 1 兲 ⑀ C u3 共 1 兲
g
⯗ ⯗ ⯗
a
1 ¯ ⫺ ⑀ C u1 共 j 兲 ⫺ ␪ C q1 共 j 兲 ⫹ ⑀ C u2 共 j 兲 ␪ C q2 共 j 兲 ⫺ ⑀ C u3 共 j 兲 ¯
g

1 ¯ ⑀ C u1 共 j 兲 ⫺ ␪ C q1 共 j 兲 ⫺ 冋 a
g
⫹ ⑀ C u2 共 j 兲 册 ␪ C q2 共 j 兲 ⑀ C u3 共 j 兲 ¯

⯗ ⯗ ⯗
a
1 ¯ ⫺ ⑀ C u1 共 Nr 兲 ⫺ ␪ C q1 共 Nr 兲 ⫹ ⑀ C u2 共 Nr 兲
g

1 ¯ ⑀ C u1 共 Nr 兲 ⫺ ␪ C q1 共 Nr 兲 ⫺ 冋 a
g
⫹ ⑀ C u2 共 Nr 兲 册
where j⫽index along radial direction; C u1 ,C u2 ,C u3 ⫽coefficients associated with axial velocity u; C q1 ,C q2 ⫽coefficients
associated with radial flux q; and ⑀ and ␪ are weighting coefficients. The unknown vector z
⫽H in⫹1 ,u i,1
n⫹1 n⫹1
,q i,1 , ¯ ,u i,n⫹1
j ,q n⫹1
i, j , ¯ ,u i,Nr⫺1 ,q i,Nr⫺1 ,u i,Nr 其 in which i⫽index along axial direction and the superscript T de-
n⫹1 n⫹1 n⫹1 T

62 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

notes the transpose operator and b⫽a known vector that depends on head and velocity at time level n. Therefore, the solution for head,
and longitudinal and radial velocities, involves the inversion of a 2N r ⫻2N r matrix. The sparse nature of A is the reason the scheme is
inefficient.
Improving the efficiency of the Vardy-Hwang scheme is essential if quasi-two-dimensional models are to become widely accepted
as tools for analyzing practical pipe systems or for conducting numerical experiments. Algebraic manipulation of the coefficient matrix
leads to a highly efficient scheme in which the original system becomes two subsystems with tridiagonal coefficient matrices expressed
as the following: Bu⫽bu and Cv⫽bv , where B is a tridiagonal N r ⫻N r matrix given by

冢 冣
a
⫹ ⑀ C u2 共 1 兲 ⫺ ⑀ C u3 共 1 兲
g
⯗
a
¯ ⫺ ⑀ C u1 共 j 兲 ⫹ ⑀ C u2 共 j 兲 ⫺ ⑀ C u3 共 j 兲 ¯
g
⯗
a
⫺ ⑀ C u1 共 Nr 兲 ⫹ ⑀ C u2 共 Nr 兲
g

冉 冊
The unknown vector u⫽ 兵 u i,1
n⫹1
, ¯ ,u i,n⫹1
j , ¯ ,u i,Nr 其 represents longitudinal flow velocity; bu is a known vector whose elements
n⫹1 T

depend on H, u, and q at time level n; and C is a tridiagonal N r ⫻N r matrix given by

1 ␪ C q2 共 1 兲
0 ⫺ 关 ␪ C q1 共 2 兲 ⫹ ␪ C q2 共 1 兲兴 ␪ C q2 共 2 兲
⯗
¯ ␪ C q1 共 j⫺1 兲 ⫺ 关 ␪ C q1 共 j 兲 ⫹ ␪ C q2 共 j⫺1 兲兴 ␪ C q2 共 j 兲 ¯
⯗
¯ ␪ C q1 共 Nr⫺1 兲 ⫺ 关 ␪ C q1 共 Nr 兲 ⫹ ␪ C q1 共 Nr⫺1 兲兴

Lastly, v⫽ 兵 H in⫹1 ,q i,1

n⫹1
, ¯q i,Nr⫺1
n⫹1
其 T is an unknown vector of equations and that obtained from the velocity profile integration, it
head and radial velocities and bv ⫽a known vector whose ele- is neglected, since the calculation of the velocity profile is only
ments depend on H,u,q at time level n. Inversion of tridiagonal used to estimate the friction term. This latter difference has only a
systems can be performed efficiently by using the Thomas algo- minor influence on the calculation of the unsteady friction term, as
rithm. argued by Eichinger and Lein 关45兴.
Silva-Araya and Chaudhry’s 关37,98兴 procedure is similar to the
10.2 Pezzinga Scheme. The numerical solution by Pezz- foregoing methods. Once the velocity is obtained, energy dissipa-
inga 关38兴 solves for pressure head using explicit FD from the tion and discharge can be calculated. The dissipation is used to
continuity Eq. 共47兲. Once the pressure head has been obtained, the estimate an energy dissipation ratio, which provides a correction
momentum Eq. 共48兲 is solved by implicit FD for velocity profiles. factor for the friction term in the 1D equations. The adjusted 1D
This velocity distribution is then integrated across the pipe section equations are then solved to give a new discharge, which is com-
to calculate the total discharge. The scheme is fast due to decou- pared to that calculated from velocity profile integration. If the
pling of the continuity and momentum equations and the adoption difference is small 共say, less than 5%兲, the calculation proceeds to
of the tridiagonal coefficient matrix for the momentum equation. the next time step. Otherwise, the pressure gradient is adjusted,
It has been applied to network simulations. and the procedure is repeated. A mixing length algebraic turbu-
While the scheme is efficient, the authors have found that there lence model 共smooth pipe, 关37兴, rough pipe 关98兴兲 is used in the
is a difficulty in the numerical integration step. Since the integra- calculation of the velocity profile.
tion can only be approximated, some error is introduced in this
step that leads to spurious oscillations in the solution for pressure. 11 Boundary Conditions
To get rid of these oscillations, a large number of reaches in the
radial direction may be required or an iterative procedure may The notion of boundary conditions as applied to the analysis of
need to be used 关37兴. fluid transient problems is analogous to, but slightly different
from, the conventional use of the terminology in solving differen-
10.3 Other Schemes. In Ohmi et al. 关99兴, the averaged 1D tial equations. Just as a ‘‘boundary value’’ problem in the math-
equations are solved to produce pressure and mean velocity. The ematical sense implies conditions that must be satisfied at the
pressure gradient is then used to calculate a velocity profile using edges of the physical domain of the problem, boundary conditions
the quasi-two-dimensional momentum equation, from which wall in fluid transients implies the need for additional head-discharge
shear stress is determined. relations to describe physical system components such as pumps,
A similar procedure is used in Eichinger and Lein 关45兴. One- reservoirs and valves. Thus, one or more simplified auxiliary re-
dimensional equations are first solved to obtain the pressure gra- lations can be specified to solve for piezometric head, flow veloc-
dient. This pressure gradient is used to solve Eq. 共46兲 using a ity, or other variables associated with the physical devices them-
finite difference method. The eddy viscosity is obtained from a selves. Examples of boundary conditions include, but are not
k⫺ ⑀ model. Once the velocity profile is known, the friction term limited to, valves, nozzles, pumps, turbines, surge tanks, air
can be calculated from the velocity gradient at the wall, which is valves, tanks and reservoirs, heat exchangers, condensers and
then used in the 1D equations. An iterative procedure is employed many other application-specific devices.
in this calculation to obtain eddy viscosity. Although there might This section of the paper discusses a generalized approach to
be some difference between the discharge calculated from 1D incorporating boundary conditions within the method of charac-

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 63

teristics framework that preserves the complex physical and topo-
logical character of the compressible fluid system. The approach
utilizes unambiguous definitions of the nodes, links, and boundary
conditions that represent the components of a physical pipe sys-
tem or network. Attention is restricted to the method of character-
istics solution because it is the most powerful and physically con-
sistent method for dealing with physically and behaviorally
complex devices without imposing unrealistic or difficult modifi-
cations to the numerical scheme. The discussion begins by re-
viewing the governing equations and the form of the method of
characteristics solution that has been developed for this purpose.
11.1 Governing Equations and Their Solution. Two
equations—a relation of mass conservation and a momentum
equation—are generally used to model transient flow in closed
conduits 共e.g., 关20–23兴兲, which can be written from Eqs. 共20兲,
共21兲, and 共28兲 as
Fig. 6 Method of characteristics grid
⳵V ⳵H f V兩V兩
⫹g ⫹ ⫽0 (61)
⳵t ⳵x 2D

⳵H a2 ⳵V
⳵t
⫹
g ⳵x
⫽0 (62) 冕A
P
Q 兩 Q 兩 dx⫽ 关 Q A ⫹ ⑀ 共 Q P ⫺Q A 兲兴 兩 Q A 兩 ⌬x (66)

To be compatible, x and V must be positive in the same direction. in which 兩 ⑀ 兩 ⭐1.

Equations 共61兲 and 共62兲 are valid as long as the flow is 1D, the This linearization of the friction term includes the ‘‘classical’’
conduit properties 共diameter, wave speed, temperature, etc兲 are Q A 兩 Q A 兩 ⌬x approximation ( ⑀ ⫽0.0) and the ‘‘modified’’
constant, the ‘‘convective’’ and slope terms are small, and the Q P 兩 Q A 兩 ⌬x linearization ( ⑀ ⫽1.0) as special cases. The approxi-
friction force can be approximated by the Darcy-Weisbach for- mation associated with ⑀ ⫽0.0 has been traditionally employed,
mula for steady flow. In addition, it is usually assumed that the but is troublesome for high friction cases; the modified lineariza-
friction factor f is either constant or weakly dependent on the tion is often more accurate and has improved stability properties
Reynolds number. Note that, for simplicity, the shear model in the 关107兴, but has not yet been universally adopted. Not only does Eq.
momentum Eq. 共61兲 above is equivalent to Eq. 共41兲 without the 共66兲 allow a single program to be used for both approximations,
convolution term. Other shear models can be readily adapted for but intermediate values of ⑀ can be used to optimize accuracy for
use in the boundary condition framework described herein. a given ⌬t. Preliminary results indicate values near 0.81 are well
Because the equations governing transient fluid flow can sel- suited to most applications. Higher-order approximations of the
dom be solved analytically, numerical solutions are used to ap- energy loss term can also be incorporated, but generally require
proximate the solution. The most widely used procedure is the iterative solution procedures. The linearized first-order approaches
fixed grid method of characteristics, which has the desirable at- result in explicit formulations and provide acceptable results over
tributes of accuracy, simplicity and numerical efficiency. The the initial wave cycle for systems of low to moderate friction.
method is described in many standard references including If Eq. 共63兲 is integrated as illustrated above, two equations can
Chaudhry 关20兴 and Wylie et al. 关23兴. Again, the procedures de- be written for the unknowns at P
scribed here can be easily adapted for use with any of the inter-
polation, reach-back, reach-out, and wave speed or pipe length H P ⫽C P ⫺B P Q P (67)
adjustment schemes mentioned previously. and
In essence, the method of characteristics combines the momen-
tum and continuity equations to form a compatibility expression H P ⫽C M ⫹B M Q P (68)
in terms of discharge Q and head H, that is in which
R C P ⫽H A ⫹Q A 关 B⫺R 兩 Q A 兩 共 1⫺ ⑀ 兲兴 (69)
dH⫾BdQ⫾ Q 兩 Q 兩 dx⫽0 (63)
⌬x
B P ⫽B⫹ ⑀ R 兩 Q A 兩 (70)
where B⫽a/gA and
C M ⫽H B ⫺Q B 关 B⫺R 兩 Q B 兩 共 1⫺ ⑀ 兲兴 (71)
f ⌬x B M ⫽B⫹ ⑀ R 兩 Q B 兩 (72)
R⫽ (64)
2gDA 2
In more complex systems, a subscript to indicate the pipe number
This equation is valid only along the so-called C ⫹ and C ⫺ char- is often added to these equations. At points P internal to a pipe-
acteristic curves defined by line, H P can be eliminated from Eqs. 共67兲 and 共68兲 to obtain

dx C P ⫺C M
⫽⫾a Q P⫽ (73)
dt
(65) B P ⫹B M
At the ends of a conduit, however, the solution of the character-
For this reason, the x-t grid in Fig. 6 is chosen to ensure ⌬x istic equations is algebraically complicated by one or more
⫽⫾a⌬t. Then, if the dependent variables are known at A and B, ‘‘boundary conditions.’’
Eq. 共63兲 can be integrated along both A P and B P. Integration of
the first two terms is straightforward, while the third requires the 11.2 Boundary Conditions. The subject of what constitutes
variation of Q with x to be known. Although this function is a boundary condition can be treated generally. Karney 关108兴 pre-
generally unknown, the term can usually be approximated 关23兴. A sents concise terminology for describing pipe networks and
convenient linearization of the A to P integration is given by boundary conditions. His nomenclature is followed throughout
Karney and McInnis 关106兴 as follows: this paper and is briefly reviewed here. Once the time domain is

64 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

 proposed by Karney 关108兴. It differs from the latter only by the
inclusion of the variable friction term linearization.
For all pipes belonging to the set N1 , Eq. 共67兲 holds while Eq.
共68兲 applies for members of N2 . These equations can be rear-
ranged to obtain
H P C Pi
Q P i ⫽⫺ ⫹ , i苸N1 (74)
B Pi B Pi
and
HP CMj
⫺Q P j ⫽⫺ ⫹ , j苸N2 (75)
BMj BMj
in which the second subscript represents the variable at the bound-
ary section of a particular pipe in the set.
The continuity equation for the junction requires the sum of
the flows entering the node to equal the sum of the flows leaving
the node
Fig. 7 Generalized node with one external flow
兺
i苸N1
Q Pi⫺ 兺
j苸N2
Q P j ⫺Q ext⫽0 (76)

discretized into ⌬t segments, most conduits in the network are Equations 共74兲 and 共75兲 can be substituted directly into Eq. 共76兲 to
divided into one or more reaches of length ⌬x. For clarity, the produce the following expression for H P :
term ‘‘pipe’’ is henceforth restricted to conduits containing at least H P ⫽C C ⫺B C Q ext (77)
one characteristic reach. The end of each reach, where head and
flow values must be determined, is called a section. At sections in which

冉兺 冊
internal to a pipe, the discharge can be obtained from Eq. 共73兲. ⫺1
1 1
However, at each end of the pipe, an auxiliary relation between
head and discharge must be specified. Such a head-discharge re-
B C⫽
i苸N1
⫹ 兺
B P i j苸N2 B M j
(78)
lation is called a boundary condition.
The term ‘‘node’’ is used herein to indicate a location where and
boundary sections meet. The degree of a node is the number of
pipes 共i.e., characteristic sections兲 connected to it. However, in a
general network, not only pipes may be connected to a node, but
C C ⫽B C 冉兺
i苸N1
C Pi
B Pi
⫹ 兺
j苸N2
CMj
BMj 冊 (79)

various other elements as well. For example, a node may represent Equation 共77兲 represents a single relationship between junction
a suction or discharge flange of a pump, the location of a valve head H P and external flow Q ext in a multipipe frictionless junc-
discharging from the network, or a connection for a pressure relief
tion. The form of this equation is equivalent to the single C ⫹
valve. All such nonpipe junctions are labeled external and the
compatibility Eq. 共67兲 and shows that any one-node boundary
number of such connections is called the complexity of the node.
condition located in a network can be evaluated in exactly the
A node of complexity zero is called simple, a node of complexity
same manner as if the boundary condition occurred at the down-
one ordinary, and a node of complexity greater than one complex.
stream end of a single pipe.
In this paper, boundary conditions associated with complex nodes
Once a functional relationship representing a particular hy-
are referred to as boundary systems. Generally, the difficulty of
draulic device is substituted into Eq. 共77兲, a single equation and
solving a network increases as the complexity of the nodes in the
unknown results. If this relationship is either linear or quadratic,
network increases but, as the following section shows, is indepen-
an explicit formula for the unknown can be obtained.
dent of the degree of any node in the network. The terminology
For example, the simplest boundary condition occurs when
related to nodes can be extended in a natural way to networks as
Q ext is either constant or a known function of time 共e.g., constant
well and has been used by Karney 关109兴 to develop a general
displacement pumps or fixed demands兲. In this case, the value of
approach for analyzing complex networks.
Q ext can be substituted into Eq. 共77兲 to obtain the junction head. In
11.2.1 Simple and Ordinary One-Node Boundary Conditions. particular, this equation becomes H P ⫽C C when Q ext is zero. This
Junctions of several pipes are usually modeled as frictionless in solution for a simple node is algebraically equivalent to Eq. 共73兲 if
transient flow applications 共e.g., 关109,110兴兲. Complications arising the node has only two pipes.
by attempting to calculate junction losses at a general node are Comprehensive treatment of various boundary conditions such
considerable and are not discussed in this paper. Generally, energy as valves, pumps, turbines, accumulators, air valves and many
losses at junctions are relatively small and neglecting them does others can be found in Wylie et al. 关23兴, Karney 关108兴, Chaudhry
not appear to significantly impair the accuracy of the method of 关20兴, McInnis 关111兴, Karney and McInnis 关82兴, and McInnis et al.
characteristics solution for a simple pipe junction. 关112兴. Formulations for many system-specific devices abound in
The assumption that local losses are negligible is equivalent to the literature.
representing the hydraulic grade line elevation at the node by a
single number, designated H P .
Consider now Fig. 7, which depicts a junction of any number of
pipes at a node. Let N1 be the set of all pipes whose assumed flow 12 Water Hammer Software
direction is toward the node in question and N2 be the set of pipes With the advent of the Windows operating system, computer
whose assumed flow direction is away from the node. Let one languages such as Visual Basic and Visual C, geographic informa-
flow be identified as external and governed by an auxiliary rela- tion systems 共GIS兲, and the World Wide Web, many water ham-
tion. Positive flows are assumed to be from the junction. The mer models, previously only suited to academics and expert engi-
following derivation is similar to that appearing in Chapter 11 of neering practitioners, are now accessible to even the most novice
Fluid Transients by Wylie and Streeter 关22兴, but uses the notation analyst.

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 65

 In this section, we describe several commercially available wa- 12.2 HAMMER. With HAMMER, a schematic of the piping
ter hammer software packages. The information presented here is layout for both pipelines and networks can be drawn on-screen
intended to aid readers in locating software appropriate to their and groups of hydraulic elements duplicated to save time during
water hammer analysis needs. the input process. As an initial condition, steady-state data can be
It is important to note that two of the authors of this paper are imported from EPANET and WATERCAD. Some of the available
also authors of a commercially available water hammer software boundary devices include pumps, air vessels, open surge tanks,
package. To avoid any conflicts of interest and to be fair to all reservoirs, surge control valves, vacuum relief and air release
water hammer software developers, a critique of each water ham- valves, and bypass lines with check valves.
mer software package is not presented here. Commercially available for over 15 years, this method of
Instead, the intention of this section is to summarize the perti- characteristics-based model can be used to simulate pump power
nent features of each computer model. These features include, but failure, valve closure, pipe breaks, and pump startup. Time history
are not limited to, the available hydraulic devices, selectable surge animations and plots of transient pressure, flow rate, and air or
protection measures, input facilities, and output graphical visual- vapor volume at nodes and along pipes are available for both
ization options. Also listed for each software package is the nu- pipelines and networks. The model also produces profile views of
merical method used by the water hammer model to solve the network pipeline paths, showing the initial steady-state pressure
unsteady flow problem. The reader is directed to Sec. 5 for back- as well as the maximum and minimum pressure envelopes 共Con-
ground on these numerical methods. tact: www.ehg.dns2go.com兲.
Each computer model has special features that distinguish it 12.3 HYTRAN. Drag-and-drop facilities enable on-screen
from the other reviewed models. These differences are most often construction or deletion of a pipeline or network in either plan or
the result of a desire to serve a specific commercial market. For profile views. Alternatively, node, pipe, and boundary device data
example, some packages are best suited to fire protection sprinkler can be directly imported from EPANET. Some of the selectable
systems, fueling systems, or oil pipelines, while others are clearly boundary devices include pumps, turbines, air chambers, vacuum
tailored to large municipal water distribution systems. Still other relief valves, check valves, tanks, reservoirs, pressure relief
models specialize in the analysis of hydroelectric systems, sewage valves, pressure regulating valves, and demands. On-line help,
force mains, or industrial applications such as cooling water sys- including a database of valve coefficients and pipe material prop-
tems. However, despite their obvious market focus, it is often erties, is available to the user.
possible to analyze just about any piping system with each of A method of characteristics-based solver generates pressure and
these models. flow rate history traces at nodes and along the pipeline following
The software packages described herein are in no particular pump power failure or startup. In addition, the computed transient
order and more information on a product 共e.g., up-to-date pricing, hydraulic grade line at any instant in simulation time can be plot-
new features, computer system requirements, etc兲 can be obtained ted in combination with the pipeline profile. Pressure traces, hy-
upon browsing the appropriate Internet homepage, which is listed draulic grade line plots, and pipe flow direction can be animated
at the end of each review. Unless otherwise noted, the software for real-time viewing. A column separation indicator warns the
packages reviewed below operate within a Windows-based envi- user when cavitation is detected 共contact: www.hytran.net兲.
ronment. Please also note that the information summarized below
12.4 HYPRESS. This model has an object-oriented interface
is largely derived from each water hammer modeler’s Internet
that allows for flexible input of pipe, node, and boundary device
homepage and is current at the writing of this paper 共2003兲.
data for pipelines and networks. Some of the boundary devices
Due to space limitations, all of the water hammer software
that can be represented by the model include pumps, turbines,
packages now readily available could not be included in this sum-
valves, reservoirs, surge chambers, and air vessels.
mary. The reader is encouraged to search out alternatives on the
Using a fourth-order implicit finite difference based numerical
Internet prior to selecting one of the models described herein.
solver, HYPRESS calculates the maximum, minimum, and instan-
12.1 Pipenet. This fluid flow program predicts pressure taneous transient hydraulic grade line for a pipeline following
surges, calculates hydraulic transient forces, models control sys- pump power failure. The hydraulic grade lines are plotted in com-
tems, and has been commercially available for over 20 years. bination with the pipeline elevation profile and the instantaneous
The interface drag-and-drop facilities are used to build a sche- transient hydraulic grade line, which can be animated in real time
matic of the pipeline or network and the associated boundary de- 共Contact: www.hif.cz兲.
vices. Pipe schedules as well as fitting, lining, pump and valve 12.5 IMPULSE. Liquids such as water, petroleum, chemical
data are provided on-line for the user’s convenience. The user can products, cryogens, and refrigerants can all be modeled using the
specify the units of both the input and output data. Fluid proper- IMPULSE water hammer model. A piping schematic is created in
ties such as viscosity and specific gravity can also be input by the the workspace using drag-and-drop facilities and data can be input
user. Boundary devices include pumps, air chambers, reservoirs, directly by the user or obtained from a built-in database contain-
tanks, caissons, vacuum relief valves, check valves, flow control ing properties for nine fluids and eight pipe materials. Some of the
valves, surge relief valves, and air release valves. hydraulic devices that can be incorporated into the pipe network
PIPENET performs a surge analysis using the method of charac- include pumps, reservoirs, liquid accumulators, gas accumulators,
teristics and calculates pressures and flow rates at nodes, pipes, vacuum breaker valves, demands, relief valves, and pressure con-
and boundary devices, as well as transient pressure forces on trol valves.
pipes and bends. As an option, the program calculates the forma- IMPULSE will calculate a system steady state and transfer it to
tion, growth, and collapse of a vapor cavity if the pressure in the the method of characteristics solver. Pipe length adjustment, as
pipe system drops to vapor pressure. PIPENET also has facilities for opposed to wave speed adjustment, is used in combination with
incorporating control theory 共e.g., proportional, integral, deriva- the time step to spatially discretize the piping network. In some
tive loops兲 in the operation of pumps and valves. Note that a cases, this means that the modeled pipe length can approximate
special module is available for analyzing sprinkler systems. the true length of the pipe. Transient events 共e.g., pump power
Output data can be plotted as time history plots, each with user failures, pump starts, valve closures, etc兲 can be initiated based on
defined titles. Examples include pressure and flow rate time his- time or a device setpoint. Liquid column separation, vapor cavi-
tory plots at nodes, pipe sections, or boundary devices. In addi- tation, and cavity collapse can be modeled. This model will iden-
tion, graphs of fluid level in an air chamber versus simulation time tify when and where maximum pressures occur and plot flow rate,
may be plotted 共Contact: www.sunrise-sys.com兲. pressure, and velocity time histories, which can be formatted by

66 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

the user. At each time interval, output, such as maximum and 12.10 WHAMO. This model uses a four-point implicit finite
minimum pressures, are tabulated for each node, pipe, and bound- difference method to calculate time-varying flow and head in
ary device 共Contact: www.aft.com兲. pipelines and networks. The user can select boundary devices
such as pumps, turbines, valves, tanks, reservoirs, vented or un-
12.6 WANDA. Both pressurized and nonpressurized vented air chambers, pressure control valves, electric governors,
branched and looped pipe systems can be simulated with this wa- and constant or time varying demands. A schematic of the piping
ter hammer model. Free-surface flow 共i.e., a partly filled pipe兲 is system can be drawn on-screen with the help of a palette of
modeled using a conjugate gradient method with an upwind ad- boundary device symbols.
vection approximation. The effects of draining and filling a pipe- Both steady-state and transient conditions are generated and
line can be simulated with this component. simulations of pump power failure, valve closure, turbine load
A schematic of the piping system can be created on-screen us- rejection, turbine startup, and governor controlled turbine opera-
ing a palette of boundary devices. A user defined image 共e.g., a tion are possible 共Contact: www.cecer.army.mil/usmt/whamo/
street map兲 can be imported as a background to the schematic and whamo.htm兲.
properties of nodes, pipes, and boundary devices are input using
dialog boxes. Some of the available boundary devices include 12.11 TRANSAM. Using this model, real-time, 3D 共i.e., dis-
pumps, control valves, check valves, taps, air vessels, air inlet tance, time, and pressure兲 animations of the transient pressure
valves, surge towers, pressure relief valves, weirs, and condensers. surface can be viewed along user-defined network and pipeline
The method of characteristics-based solver can be interrupted paths. A piping layout map can be created in a designated work-
and resumed during a simulation. Cavitation and control theory space using point-and-click options and a combination of pull-
共e.g., proportional integral derivative loops, sensors, etc兲 modules down menus and dialog boxes are available for node, pipe, and
are optional. Pressure versus time histories can be plotted at user- boundary device data input. An EPANET to TRANSAM conversion
defined locations within the pipe system. In addition, it is possible utility is supplied. Some of the boundary devices that can be rep-
to view an animation of pressure wave propagation and reflection resented by this model include pumps, turbines, air chambers,
along pipeline routes of a network 共Contact: www.wldelft.nl/soft/ reservoirs, tanks, flow control valves, air and vacuum relief
wanda/兲. valves, check valves, pressure relief valves, surge anticipating
valves, pressure reducing/sustaining valves, constant and time
12.7 FLOWMASTER. This model calculates transient pres- varying demands, and bypass lines with check valves.
sures and flow rates in piping networks. In addition, calculation of Pump power failure and startup, variable speed pump and valve
heat transfer and simulation of partly empty pipe segments 共e.g., operations 共e.g., full and partial openings or closures兲, turbine
sprinkler systems兲 is possible. load rejection, and pipe breaks are just some of the event or time
Pipe networks can be drawn on-screen using a list of piping initiated unsteady flow conditions that can be simulated using this
components and some of the boundary devices that can be repre- method of characteristics-based model. Simulation of the forma-
sented include pumps, reservoirs, weirs, orifices, valves, accumu- tion, growth, and collapse of vapor cavities is optional. Time his-
lators, diaphragms, diffusers, heat-exchangers and pipe fittings. tory plots of pressure 共and flow rate at nodes兲 can be produced at
User-defined boundary devices can be programmed in either FOR- nodes and along pipes. Real-time animations of the instantaneous
TRAN or C. Operational issues can be studied using predefined transient, maximum, and minimum hydraulic grade lines can be
controllers or user-defined controllers programmed in Visual Ba- viewed for pipe paths 共Contact: www.hydratek.com兲.
sic or Java.
The method of characteristics solver generates results that can
be viewed graphically or in tabular formats. Note that in addition 13 Emerging Applications in Water Hammer
to liquids, gas flow dynamics can be simulated 共Contact: www- By now, the reader is likely aware that the principal use of
.flowmaster.com兲. transient analysis, both historically and present day, is the predic-
tion of peak positive and negative pressures in pipe systems to aid
12.8 SURGE2000. With this model, a schematic of the pip-
in the selection of appropriate strength pipe materials and appur-
ing layout can be drawn on-screen and over it can be placed an
tenances and to design effective transient pressure control sys-
imported background image, such as a street or elevation contour
tems.
map. Boundary devices include pumps, valves, reservoirs, tanks,
Two important areas in which transient modeling is now taking
air vessels, air and vacuum valves, pressure relief valves, surge
a key role are parameter estimation for leakage detection and wa-
anticipating valves, and heat exchangers.
ter quality predictions in potable water systems. Brief discussions
This model uses the wave-plan method as opposed to the
of these two important areas of application are provided in this
method of characteristics or finite difference methods employed
section.
by the other models reviewed in this paper. Pump power failure,
pump startup, and valve operations 共e.g., closure兲 are just some of
the unsteady fluid flow events that can be simulated with 13.1 Parameter Estimation for Leakage Detection and In-
SURGE2000. Output, such as pressures, can be tabulated, plotted verse Models. In many pipeline related industries, such as a
as contours over the system map, and displayed in time history potable water supply or in oil or gas transmission, owners know
plots at nodes. In addition, for each pipeline path, the maximum, that information is the key to successful management of their
minimum, and instantaneous transient hydraulic grade lines can pipeline operation. For example, in the case of a water supply,
be plotted on an elevation versus distance graphic 共Contact: ww- physical system characteristics, customer data, production rates,
w.kypipe. com兲. maintenance records, quality assays, and so on, each provide man-
agement, engineering, operations, and maintenance staff with in-
12.9 LIQT. First introduced in 1972, LIQT can model fluid formation they need to keep the system running efficiently and
transients in pipelines and networks subject to pump power failure safely, and at a reasonable cost to the consumer. A large body of
and startup, turbine load loss, and valve closure. Some of the literature on the subject of information requirements and data
boundary devices that can be selected by the user include pumps, management already exists, and all private and public pipeline
turbines, check valves, air and vacuum valves, surge tanks, stand- utilities are aware of the importance of collecting, archiving, and
pipes, accumulators, and pressure relief valves. LIQT operates analyzing data. Perhaps the most costly and time consuming as-
within a DOS environment window and uses the method of char- pect of information management, however, is the collection of
acteristics to compute pressures and flow rates that can be ex- data. This section outlines how inverse transient analysis can be
ported to spreadsheets, databases, and graphic software for pre- applied to gather some types of physical system data. The tech-
sentation 共Contact: www.advanticastoner.com兲. nology has the potential to be both cost efficient and accurate.

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 67

 Using high-frequency pressure transducers, it is now possible to more measured data sets using some assumed set of system pa-
safely measure induced, or naturally occurring, pressure surge rameters. System parameters could be pipe wave speed, friction
events. Coupling water hammer models to inverse models offers factor, water consumption rates and locations, leakage rates and
the possibility of inexpensive data collection with a wide coverage locations, and so on. A ‘‘merit’’ function is used to compare the
of the system. System demands, leakage, pipe condition 共rough- goodness of fit between the observed data and the model output.
ness兲, closed or partially closed valves, even pockets of trapped Common merit functions are the error sum of squares, sum of the
gas or air can 共in theory at least兲 all be detected using recorded absolute values of errors, etc. Some sort of search or optimization
high-frequency pressure data. In addition to pressure measure- procedure is employed to find the set of parameter values that
ments, transient flow data can also be used in the inverse analysis minimizes the discrepancy between observed data values and
procedures. However, flow meters capable of accurately resolving those predicted by the forward model. It is the nature of the search
the variation in flow rates that occurs during water hammer events technique employed in the optimization step that characterizes the
are quite expensive and more troublesome to install. High- inverse modeling approach.
frequency pressure transducers, on the other hand, are relatively
13.1.1.1 Adjoint models. Adjoint models use a form of La-
inexpensive and easy to mount at common access points, such as
grangian optimization coupled with a gradient search to minimize
valve and meter chambers, or even at fire hydrants. To date, hy-
the errors between the observed data and the forward model pre-
draulic model parameters such as pipe roughness and wave speeds
diction. In transient flow applications, the problem statement
have been successfully calibrated using these techniques. would take the following general form 共see also Liggett and Chen
13.1.1 Inverse Analysis of Transient Data. Whether a tran- 关121兴兲:
sient is small or large, accidental or planned, pressure waves
propagate from their respective points of origin to other parts of
the system. They travel at speeds ranging from about 250 m/s to
Minimize E⫽ 兺 关共 h m
⫺h c 兲 2 ⫹ 共 q m ⫺q c 兲 2 兴 (80)

nearly 1500 m/s, depending on pipe material, soil and anchoring subject to the following physical constraints:
conditions. The shock fronts interact with any part of the system
that either dissipates energy or does work in a thermodynamic ⳵h a2 ⳵q
⫹ ⫽0 (81)
sense. Thus, the energy content of the wave is diminished by ⳵ t gA ⳵ x
virtue of its interaction with the physical system, and its frequency
components, amplitude, phasing, and attenuation characteristics ⳵q ⳵h q兩q兩
⫹gA ⫹f ⫽0 (82)
become modified through successive interaction with the system. ⳵t ⳵x 2AD
In effect, a pressure signal at a given location constitutes a record
where E is the error sum of squares and Eqs. 共81兲 and 共82兲 are the
of conditions in the system during the course of a given transient
continuity and momentum equations rewritten in terms of dis-
event.
charge and assuming steady Darcy-Weisbach friction. The super-
Deciphering this record of interaction and extracting its infor-
script m denotes measured data values and the superscript c de-
mation content is precisely what an inverse transient model does.
notes the values computed by the forward model, h is piezometric
The inverse model evaluates the recorded pressure 共or flow兲 signal
pressure head, q is the flow rate, a is the pipeline celerity, f is the
and determines which set 共or sets兲 of system parameters, i.e., pipe
Darcy-Weisbach friction factor, t is time, x is a spatial coordinate,
roughness, water consumption 共leakage兲, wave speed, etc, best
g is gravitational acceleration, and A and D are the pipeline cross-
matches the measured data. In this way, information 共data兲 of
sectional area and diameter, respectively.
several types can be gathered from those areas of the pipe system
Equation 共80兲 can be combined with Eqs. 共81兲 and 共82兲 by
that the transient waves have traversed. For example, pressure
using Lagrangian multipliers ␭ 1 and ␭ 2 as follows:
traces from two pump trip tests can be sufficient to estimate pipe
roughness values for every major pipe and consumption values at
each node in a small city. Of course, the accuracy of the estimates
can be improved by increasing the number of tests performed or
E *⫽ 冕 冕冋
x t
共 h m ⫺h c 兲 2 ␦ 共 x m ⫺x c 兲 ␦ 共 t m ⫺t c 兲 ⫹ 共 q m ⫺q c 兲 2 ␦ 共 x m

by monitoring pressures 共flows兲 at more than one location.

There is extensive literature about inverse analysis in both sci-
entific and engineering journals. The techniques have been applied
⫺x c 兲 ␦ 共 t m ⫺t c 兲 ⫹␭ 1 冉 ⳵h a2 ⳵q
⫹
⳵ t gA ⳵ x
⫹␭ 2
⳵q
⳵t
⫹gA
⳵h
⳵x冊 冉
for many years to structural engineering applications e.g., system
identification and damage detection 关113兴. Sykes 关114兴, Sun and
Yeh 关115兴, and Sun 关116兴 have used inverse methods to identify
⫹f
q兩q兩
2AD 冊册 dxdt (83)

parameters in 2D groundwater flow. Jarny et al. 关117兴 applied the The merit 共error兲 function has now been designated E * to
adjoint technique to heat conduction problems. Cacuci et al. 关118兴 indicate that it includes the Lagrangian terms for the continuity
and Hall 关119兴 applied the adjoint method to meteorology and and momentum equations, and has been expressed as an integral
climate modeling. Marchuk 关120兴 applied the adjoint technique to to be consistent with the continuum form of the momentum and
air pollution problems. continuity equations. The Dirac delta functions are included to
Most, though not all, inverse models utilize real measurements ensure that merit function terms are evaluated only at those loca-
in a ‘‘data-fitting’’ exercise that typically provides ‘‘best-fit’’ pa- tions and times for which observed data exist, i.e.,

再
rameters for the mathematical model postulated to fit the data.
Least-squares data-fitting is a simple example of an inverse 1 for x m ⫽x c
method that tries to fit the best mathematical model 共i.e., linear, ␦ 共 x m ⫺x c 兲 ⫽
0 for x m ⫽x c

再
exponential, polynomial, etc兲 to some observed data set. The
‘‘goodness of fit,’’ i.e., how well the particular assumed math- 1 for t m ⫽t c
ematical model represents the data, can be measured statistically and ␦ 共 t m ⫺t c 兲 ⫽ (84)
0 for t m ⫽t c
by an analysis of the errors between the observed data and those
predicted by the model. In fact, these errors are explicitly mini- The conventional approach to Lagrangian optimization is to
mized using Lagrangian optimization such that the optimal param- take partial derivatives of the merit function with respect to the
eter set is directly solved for. unknown system parameters (a or f in this simple formulation兲
The same concept can be applied to more complex physical and the Lagrangian multipliers ␭ 1 and ␭ 2 , and equate these slope
systems using sophisticated models. In an inverse problem, output functions to zero. This provides four equations from which the
from a ‘‘forward’’ model is used to generate an estimate of one or four unknown variables a, f , ␭ 1 , and ␭ 2 could be determined.

68 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

However, as these equations are quasi-linear hyperbolic partial detection. The transient response of a pipeline system to a given
differential equations, a more elaborate procedure must be used. flow disturbance with and without leakage points was measured as
Partial derivatives of the merit function are taken with respect to well as computed. The influence of the size and shape of small
␭ 1 and ␭ 2 and at critical points of the merit function must have a leaks, along with discharge conditions and initial flow regime, on
slope of zero. The two derivative functions given in Eq. 共85兲 the transient response of a pipeline system were analyzed. It is
below are known as the adjoint equations. found that the influence of the leak on the shape and amplitude of
⳵E* ⳵h a2 ⳵q the pressure signal is quite noticeable, even when the leak flow is
⫽ ⫹ ⫽0 only a few percent of the total flow in the pipe. Therefore,
⳵␭1 ⳵ t gA ⳵ x Brunone 关123兴 and Brunone and Ferrante 关124兴 formulated a
⳵E* ⳵q ⳵h q兩q兩 scheme for leakage detection on the basis of studying the differ-
and ⫽ ⫹gA ⫹f ⫽0 (85) ence in transient response of a pipeline system with and without a
⳵␭2 ⳵t ⳵x 2AD
leak. It is observed that the measured pressure head traces for the
It suffices to say that the adjoint model is solved iteratively. pipeline with a leak is different from that for an intact pipe. When
Values for physical system parameters a and f are assumed, and the transient wave encounters a leak, part of the wave is reflected
the forward model is run to determine the transient head and flow. back. The leak location is determined from the time when the
The adjoint equations are solved using the known heads and flows reflected wave arrives at the measurement station. The leak in-
in a backward pass to calculate the Lagrangian parameters ␭ 1 and duces additional drop in the pressure head traces, the amount of
␭ 2 . These values are used in a gradient search step 共the conjugate drop depending on the size of the leak. The size of the leak is
gradient technique is often used兲 to select new estimates of the using a formal inverse approach. The agreement between the ac-
optimal parameters. The search procedure terminates when the tual and the computed location and size of leak points is good.
value of E * cannot be reduced any further.
The advantage of the adjoint method is that it can be extremely 13.1.1.4 Frequency response method. The frequency re-
efficient for a well-conditioned problem. The model can be for- sponse method is used by Mpesha et al. 关125,126兴. A hydraulic
mulated to solve for other parameters of interest beside wave system is made up of several components. Each component can be
speed and friction. represented by a transfer matrix. Transient flow is caused by the
periodic opening and closing of a valve 关125兴 or by the sudden
13.1.1.2 Genetic algorithms. Genetic algorithms 共GAs兲 have
gained widespread popularity in recent years. There are many rea- opening or closing of a valve 关126兴. A frequency response dia-
sons for this success: 共i兲 GAs can be applied to a wide variety of gram at the valve is developed based on the transform matrix. For
problems; 共ii兲 GAs do not require the development of additional a system with leaks, this diagram has additional resonant pressure
code needed to solve the adjoint of the forward problem; 共iii兲 a amplitude peaks that are lower than the resonant pressure ampli-
single GA can be used with various models that solve the same tude peaks for the system with no leaks. From the frequency of
forward problem; 共iv兲 any model parameters can be specified as the peaks, the location of the leak can be detected. Very good
the unknown system parameters in a GA; 共v兲 GAs are quite suc- agreement have been obtained between the computed and the real
cessful in problems containing local extrema; and 共vi兲 GAs can leak condition.
find not only the global optimum, but can also describe other In Ferrante and Brunone 关127兴, the governing equations for
suboptimal solutions of interest, particularly for flat merit func- transient flow in pipes are solved directly in the frequency domain
tions. Genetic algorithms do not work for every problem, how- by means of the impulse response method. Therefore, the solution
ever, and one must be aware of their limitations. GAs work best of the response of the system to more attractive transient events is
for problems in which genotypes consist of a small number of available. Harmonic analysis of the transient pressure is used to
genes that can be expressed in short length strings, i.e., problems identify the location and the size of a leak.
having few decision variables 共parameters兲 that can be identified
by a small number of binary digits. Problems with large numbers 13.1.1.5 Mode damping method. Wang et al. 关128兴 investi-
of real-valued parameters over an extensive and continuous do- gated the damping characteristics of a transient pressure wave by
main are demanding of computer resources when solved by ge- wall friction and by system leakage. It is found that wall friction
netic algorithms. Despite these limitations, the method seems to damps all modes similarly, but leakage damps different Fourier
work well with pipeline problems, albeit solution procedures are modes differently. In addition, mode damping by leakage is found
slower than those of the adjoint method. to depend on leak location. The marked difference in mode damp-
In the simplest sense, genetic algorithms are an efficient form ing between wall friction and system leakage was successfully
of enumeration. A candidate set of parameters is assumed or ran- used to identify the location and size of leaks 关128兴. In particular,
domly generated to form individuals in a population. Subsequent Wang et al. 关128兴 were able to accurately identify system leaks by
iterations use evolutionary 共mutation兲 and reproductive 共cross- investigating mode damping characteristics of transient pressure
over兲 functions to generate further generations of solutions. The data obtained from numerical as well as laboratory studies. The
mathematical principle upon which genetic algorithms are based damping characteristic technique was successfully applied to
is intended for use with problems in which the decision variables single and multiple leaks.
are discrete, and in these situations the method can be extremely
efficient. Modifications to the method have been developed to 13.1.1.6 Wavelet transform method. Frequency analysis can
extend its application to continuous real-valued problems, al- only deal with a stationary signal 共i.e., the signal has to be either
though the procedures are less efficient in these cases. periodic or decomposable into a set of periodic signals兲. Wavelet
Karney and Tang 关122兴 have successfully applied the genetic transform can be used to detect local singularities in a measured
algorithm method to parameter estimation problems in water dis- signal. Whenever there is a singularity in a measured signal, a
tribution systems using transient pressure readings. Using data local maximum of the transform coefficient for the measured sig-
from only two pump trip tests 共one for model validation and the nal appears. The application of the transient wavelet transform to
other for the parameter estimation兲, Karney and Tang have suc- leakage detection in a pipeline was pioneered by Ferrante and
cessfully estimated pipe roughness factors and wave speed for Brunone 关129兴. The wavelet transform of pressure head history is
several large water distribution systems.
performed. According to the transform of the signal, the disconti-
13.1.1.3 Pressure wave method. Brunone 关123兴 and Brunone nuities in pressure head traces are detected. These discontinuities
and Ferrante 关124兴 conducted numerical and physical experiments correspond to wave reflections at boundary elements and at leak
to investigate the possibility of using transient data for leakage points. Using the time at which a discontinuity is observed, the

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 69

distance between the leak and the measurement station can be other sources of drinking water contamination as high as 1:1.
calculated. The location found by wavelet transform agrees with Studies by Payment 关130兴 suggest that one third of the 99 million
the real location very well. gastroenteritis cases in the U.S. each year might involve exposure
to waterborne pathogens in the distribution system. Conserva-
13.1.1.7 Identifiability and uniqueness requirements. In or- tively estimating that 20% of these cases result from pathogen
der for any inverse method to be successfully applied, two key intrusion into water pipes, then, in America alone, as many as 20
mathematical properties of the problem must be satisfied: identi- million cases of gastroenteritis annually might be directly caused
fiability and uniqueness. Identifiability refers to the notion that a by contamination of drinking water distribution systems.
single set 共or a finite number of distinct sets兲 of parameter values Recent research into the problem is now attempting to address
must reproduce, within an established level of error, the same four critical questions that naturally arise in response to these
response exhibited by the actual system. Uniqueness means that alarming statistics. 共i兲 What is the nature of the pathogen intrusion
the merit function exhibits a single, global minimum. For complex mechanism共s兲? 共ii兲 Why don’t routine water sampling and labora-
problems, there is no rigorous mathematical procedure that can tory testing detect intrusion events? 共iii兲 Is the health of water
assure us that identifiability and uniqueness requirements are sat- consumers in a particular system at risk? 共iv兲 Can the risk of
isfied in general. However, a simple example is described in the distribution system intrusions be reduced 共and by how much兲 or
following section that provides insight into the suitability of dif-
eliminated altogether? Answers to these four questions depend
ferent inverse modeling techniques for pipeline transient prob-
entirely on developing a clear understanding of the complex in-
lems.
teractions with hydraulic transients in pipe systems.
13.1.1.8 Identifiability. Let Hm ⫽ 兵 H 0 ,H 1 ,H 2 , . . . ,H n 其 de-
13.2.1 Distribution System Intrusion Pathways. There are
note a set of measured values, e.g., piezometric head, at time steps
several potential intrusion pathways whereby bacterial, protozoan,
0 to n corresponding to some sampling rate (t n ⫺t 0 )/(n⫺1). Let
and viral pathogens can enter a water supply, transmission, or
hc⫽ 兵 h 0 ,h 1 ,h 2 , . . . ,h n 其 similarly denote the set of computed
distribution pipeline: 共i兲 at the source; 共ii兲 during loss of pressure
heads at the same time steps but for a particular pair of unknown,
and subsequent exposure of the pipe interior to the external air,
but desired, parameter values ␴ 1 and ␴ 2 . Then, the following
soil or groundwater 共such as may happen during a main break
criterion for identifiability can be stated:
repair兲; 共iii兲 via cross-connections on a consumer’s property; and
Identifiability criterion: A set of parameter values ␴ 1 and ␴ 2 are
共iv兲 via cross-connections in the distribution system.
identifiable if and only if Hm ⬅hc⫾ ⑀ , where ⑀ represents the ab-
solute value of data, measurement, and model error. The identifi- The first two intrusion mechanisms are ‘‘controlled’’ situations
ability criterion can be visualized by plotting the difference be- insofar as the quality of finished water is carefully monitored and
tween Hm and hc for each pair of feasible values of parameters ␴ 1 treated to ensure compliance with drinking water standards, while
and ␴ 2 in the domain and selecting the zero contour of the differ- the latter two pathways are largely ‘‘uncontrolled.’’ Cross-
ences. Identifiable parameter pairs for which Hm ⬅hc would ap- connections can arise whenever a possible source of contaminated
pear as intersections 共loci of intersecting lines兲 of all such con- water or other liquid can be introduced into the potable water
tours. system by virtue of backpressure 共an excess of pressure causing
Uniqueness: The second condition that needs to be met if ad- flow to occur in a direction opposite to its normal intended flow
joint methods are to be used with a reasonable expectation of direction兲 or siphonage 共suction or ‘‘negative’’ pressure inducing
success is uniqueness, i.e., there should ideally be only a single flow from a contamination source into the distribution system兲.
critical point of the merit function within the feasible search do- While backpressure and negative pressures are usually eliminated
main. through proper hydraulic design, there is one major source of
The adjoint technique can still be useful if the feasible search negative pressures that is not normally accounted for in distribu-
domain can be restricted to a smaller region containing the global tion system design—hydraulic transients.
minimum. To this end, a more robust optimization scheme is often Water hammer occurs regularly in some systems and periodi-
employed to locate the probable region of the global minimum. cally in others whenever flow conditions are changed rapidly.
Following this initial screening, the adjoint scheme can then be Whether these changes in flow are the result of planned operations
applied to refine the solution. This two-phase optimization ap- like pump starts and stops, or are unplanned events initiated by
proach is only worthwhile if the time required to find a global power outages, accidental valve closures, or rupturing of a pipe,
minimum by other methods is too costly. Compared to the adjoint the ensuing episodes of negative pressure can introduce contami-
technique, genetic algorithms are better suited to solving problems nated fluids into the pipeline. Contamination can occur on a cus-
with multiple critical points and those that appear to give good tomer’s property or on the utility side of a service connection.
results for inverse modeling in pipeline transient applications. Contaminated fluids introduced at a cross-connection would be
largely transported in the prevailing direction of flow in the pipe
13.2 Pathogen Intrusion in Water Supply Systems. In the after entering the system.
first sentence of its proposed Ground Water Rule: Public Health Pressure dependent leakage is commonly known to occur from
Concerns document, the U.S. EPA Office of Water states that, the potable system to the surrounding 共soil兲 environment through
‘‘Assurance that the drinking water is not contaminated by human pipe joints, cracks, pinholes, and larger orifice-like openings.
or animal fecal waste is the key issue for any drinking water Funk et al. 关131兴 developed analytic hydraulic parameters to as-
system.’’ The proposed Ground Water Rule is designed to protect sess the potential for transient intrusion in a water distribution
against pathogenic bacteria and viruses in source water, against system. Their intrusion model was based on the percentage of
growth of opportunistic pathogenic bacteria in ground water dis- water lost through leakage lumped at system nodes and an
tribution systems, and to mitigate against any failure in the engi- ‘‘equivalent orifice’’ needed to pass the discrete leakage flow at
neered systems, such as cross-connections or sewage infiltration the prevailing system pressure.
into distribution systems. A paper by McInnis 共in progress兲 extends the work of Funk
There is considerable evidence in the literature that the number et al. to incorporate alternative intrusion flow models based on
of disease outbreaks 共including a large number that are unreported laminar flow, turbulent orifice flow, or a mixture of the two flow
关www.epa.gov/orgwdw000/standard/phs.html兴兲 due to fecal con- types. Work done by Germanopoulos and Jowitt 关132兴 on pressure
tamination of distribution systems is already large and might be dependent leakage suggests that most distributed leakage is prob-
growing. From 1971–1994, 50 of 356 reported waterborne dis- ably laminar in nature, occurring through larger numbers of small
ease outbreaks occurred as a result of pathogen entry into distri- openings. The 2D water hammer equations with turbulence mod-
bution systems. More recent statistics from the U.S. Center for els developed by Vardy and Hwang 关25兴, Pezzinga 关38兴, and
Disease Control put the ratio of distribution system intrusions to Silva-Araya and Chaudhry 关37,98兴 will be useful in generating

70 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

models for predicting intrusion volumes, initial distribution of of applicability of existing models and to seek more appropri-
contaminant concentrations in the pipes, and the ultimate fate of ate models for problems where current models are known to
contaminants within the distribution system. fail
McInnis 关133兴 expands the consideration of transient intrusion
events from purely fluid mechanics aspects by developing a risk- The development of inverse water hammer techniques is an-
based framework for comparing the actual human health risks and other important future research area. A number of very promising
relative risk reduction achieved by alternative transient-intrusion inverse water hammer techniques have been developed in the last
mitigation strategies. McInnis 关133兴 applies transient modeling decade. Future work in this area needs to accomplish the follow-
with some assumed reference groundwater contamination level ing:
and computes hypothetical intrusion volumes for a given transient 1兲 further investigate the issues of efficiency, reliability, and iden-
event to predict the transitory impact of the event on system water tifiability of inverse water hammer techniques
quality. He has also proposed meaningful risk-based measures to 2兲 develop more realistic laboratory and field programs in order
provide quantitative comparisons of the relative reduction in the to further test existing inverse techniques as well as develop
risk of receptor infection achieved by alternative mitigation strat- new ones
egies. 3兲 develop systematic approaches 共e.g., using stochastic methods兲
that can incorporate the influence of modeling and measure-
ment errors on the reliability of inverse methods
4兲 develop identifiability-based methods to determine the quantity
14 Practical and Research Needs in Water Hammer
and quality of data necessary to carry out a successful inverse
Both theory and experiments confirm the existence of helical program
type vortices in transient pipe flows. The conditions under which
helical vortices emerge in transient flows and the influence of The practical significance of the research goals stated above is
these vortices on the velocity, pressure, and shear stress fields are considerable. An improved understanding of transient flow behav-
currently not well understood and, thus, are not incorporated in ior gained from such research would permit development of tran-
transient flow models. Future research is required to accomplish sient models able to accurately predict flows and pressures beyond
the following: the first wave cycle. One important consequence of this is that the
1兲 understand the physical mechanisms responsible for the emer- behavioral aspects of control devices activated 共or reactivated兲 by
gence of helical type vortices in transient pipe flows local flow or pressure is correctly modeled. Most importantly,
2兲 determine the range in the parameter space, defined by Rey- however, reducing the modeling errors beyond the first wave
nolds number and dimensionless transient time scale over cycle, along with better inverse techniques, will greatly improve
which helical vortices develop the accuracy and reliability of inverse transient models. This is
3兲 investigate flow structure together with pressure, velocity, and important because inverse models have the potential to utilize
shear stress fields at subcritical, critical, and supercritical val- field measurements of transient events to accurately and inexpen-
ues of Reynolds number and dimensionless time scale sively calibrate a wide range of hydraulic parameters, including
pipe friction factors, system demands, and leakage. At this time,
The accomplishment of the stated objectives would be sought such information can only be obtained through costly field mea-
through the use of linear and nonlinear analysis. Understanding surements of flows and pressures conducted on a few individually
the causes, emergent conditions, and behavior of helical vortices sampled pipes in the system. Transients, on the other hand,
in transient pipe flows as well as their influence on the velocity, traverse the entire system, interacting with each pipe or device in
pressure, and shear stress field are fundamental problems in fluid the system. Thus, they contain large amounts of information re-
mechanics and hydraulics. Understanding these phenomena would garding the physical characteristics of the system. Inverse tran-
constitute an essential step toward incorporating this new phe- sient analysis techniques are now being developed to decode this
nomena in practical unsteady flow models and reducing signifi- information for hydraulic model calibration as well as to identify
cant discrepancies in the observed and predicted behavior of en- and locate system leakage, closed or partially closed valves, and
ergy dissipation beyond the first wave cycle. damaged pipes. The potential annual savings in routine data col-
Current physically based 1D and 2D water hammer models lection costs for water supply utilities world wide is significant.
assume that 共i兲 turbulence in a pipe is either quasi-steady, frozen Equally important, an improved understanding of the true nature
or quasi-laminar; and 共ii兲 the turbulent relations that have been of turbulence in transient flows will be a groundbreaking step
derived and tested in steady flows remain valid in unsteady pipe toward modeling transient-induced water quality problems. Nega-
flows. These assumptions have not received much attention in the tive pressure waves can cause intrusion of contaminants from the
water hammer literature. Understanding the limitations and accu- pipe surroundings through cracks, pinholes, joints, and ruptures in
racy of assumptions 共i兲 and 共ii兲 is essential for establishing the the pipes. In addition, water hammer events cause biofilm slough-
domain of applicability of models that utilize these assumptions ing and resuspension of particulates within the pipe, potentially
and for seeking appropriate models to be used in problems where leading to unsafe or unpleasant drinking water. Without a better
existing models fail. Preliminary studies by Ghidaoui et al. 关46兴 understanding of transient flow behavior, the risk and degree of
show that agreement between physically based 1D and 2D water contamination of water supply systems during transient events
hammer models and experiments is highly dependent on the Rey- cannot be quantitatively assessed.
nolds number and on the ratio of the wave to turbulent diffusion
time scales. However, the lack of in-depth understanding of the
changes in turbulence during transient flow conditions is a signifi-
cant obstacle to achieving conclusive results regarding the limita- 15 Summary
tion of existing models and the derivation of more appropriate The scientific study of transient fluid flow has been undertaken
models. Therefore, a research program whose main objective is to since the middle of the nineteenth century. As is true of every
develop an understanding of the turbulence behavior and energy other area of engineering research, a great many advances have
dissipation in unsteady pipe flows is needed. This research pro- been made in the accuracy of analysis and the range of applica-
gram needs to accomplish the following: tions since then. Although only a few simple problems were ap-
proachable by earlier analytical methods and numerical tech-
1兲 improve understanding of and the ability to quantify changes niques, a much broader spectrum of transient problems could be
in turbulent strength and structure in transient events at differ- solved once graphical methods were developed. More recently,
ent Reynolds numbers and time scales the application of digital computing techniques has resulted in a
2兲 use the understanding gained in item 1 to determine the range rapid increase in the range and complexity of problems being

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 71

studied. This paper provides both a historical perspective and re- C M - quantity for negative characteristics used for
view of water hammer theory and an overview of recent develop- 1D MOC solution
ments in this field of fluid mechanics. C m - coefficient for five-region turbulence model
Specifically, advances in the last one or two decades dealing C P - quantity for positive characteristics used for
with some of the more complex and fundamental fluid mechanics 1D MOC solution
issues have been discussed: C q1 , C q2 - coefficients before q
1兲 The relation between state equations and wave speeds in single C r - courant number
as well as multiphase and multicomponent transient flows are C u1 , C u2 , C u3 - coefficients before u
illustrated and discussed. c - parameter associated with pipe anchor condi-
2兲 Various forms of 1D and 2D water hammer equations, such as tion
the Joukowsky model, classical 1D waterhammer equations, c - superscript denoting values predicted by for-
the 2D plane wave equations, and the quasi-two-dimensional ward model
plane wave equations are derived. c 2 - coefficient used in Daily et al. 关39兴
3兲 Governing equations of turbulent water hammer flows are ob- cs - control surface
tained by ensemble averaging of the quasi-two-dimensional cv - control volume
plane wave equations. D - diameter of pipe
4兲 Order of magnitude analysis is used throughout the paper to E - Young’s modulus of elasticity of pipe material
evaluate the accuracy of the assumptions in the various forms E - errors
of water hammer governing equations. E * - merit 共error兲 function
Water hammer models are becoming more widely used 共i兲 for e - thickness of pipe wall
the design, analysis, and safe operation of complex pipeline sys- F - wall resistance force
tems and their protective devices; 共ii兲 for the assessment and miti- F ext - external forces
gation of transient-induced water quality problems; and 共iii兲 for f - Darcy-Weisbach friction factor
the identification of system leakage, closed or partially closed g - gravitational acceleration
valves, and hydraulic parameters such as friction factors and wave Hm - set of measured piezometric head
speeds. In addition, turbulence models have been developed and H - piezometric head
used to perform numerical experiments in turbulent water hammer H A - piezometric head at point A
flows for a multitude of research purposes such as the computa- H B - piezometric head at point B
tion of instantaneous velocity profiles and shear stress fields, the H P - piezometric head at point P
calibration and verification of 1D water hammer models, the hc - set of computed piezometric head
evaluation of the parameters of 1D unsteady friction models, and i - index for pipes
the comparison of various 1D unsteady friction models. Under- J s - steady friction term
standing the governing equations that are in use in water hammer j - index for pipes
research and practice and their limitations is essential for inter- K - unsteady resistance coefficient
preting the results of the numerical models that are based on these K e - effective bulk modulus of elasticity
equations, for judging the reliability of the data obtained from K f - bulk modulus of elasticity of the fluid
these models, and for minimizing misuse of water hammer mod- K s - steady-state resistance coefficient
els. K u - unsteady resistance coefficient and momentum
flux of absolute local velocity
Acknowledgments k - unsteady friction factor
L - pipe length
The writers wish to thank the Research Grants Council of Hong l - mixing length
Kong for financial support under Project No. HKUST6179/02E. M - Mach number
m - superscript denoting measured data values
Nomenclature m - time level index
A - system matrix N1 - set of all pipes with flow toward conjuction
A - cross-sectional area of pipe node
a - water hammer wavespeed N2 - set of all pipes with flow away from conjuc-
a - coefficient for five-region turbulence model tion node
B - matrix for subsystem of longitudinal velocity N r - number of computational reaches in radial di-
component rection
B - coefficient for MOC formulation n - unit outward normal vector to control surface
BC - lumped quantity for characteristics solution for n - index of measured series
pipe network n - exponential for power law of velocity profile
BM - quantity for negative characteristics used for n c - number of complete water hammer wave
1D MOC solution cycles
BP - quantity for positive characteristics used for P - parameter for quasi-steady assumption
1D MOC solution P - piezometric pressure
b - known vector for system Q - discharge
bu - known vector for subsystem of longitudinal Q A - discharge at point A
velocity component Q B - discharge at point B
bv - known vector for subsystem of head and radial Q ext - discharge of external flow
component Q P - discharge at point P
C - matrix for subsystem of head and radial veloc- q - radial flux
ity component q - flow rate
CB - coefficient for five-region turbulence model R, R - radius of pipe, dimensionless distance from
*
CC - lumped quantity for characteristics solution for pipe wall
pipe network R - coefficient for MOC formulation
Cc - coefficient for five-region turbulence model Re - Reynolds number

72 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

 r - radial coordinate 关4兴 Michaud, J., 1878, ‘‘Coups de bélier dans les conduites. Étude des
T - time scale moyens employés pour en atteneur les effects,’’ Bull. Soc. Vaudoise
Td - time scale for radial diffusion of vorticity Ing. Arch. 4共3,4兲, pp. 56 – 64, 65–77.
t - time 关5兴 Weston, E. B., 1885, ‘‘Description of Some Experiments Made on
the Providence, RI Water Works to Ascertain the Force of Water
t⬘ - time used for convolution integral Ram in Pipes,’’ Trans. Am. Soc. Civ. Eng. 14, p. 238.
U1 - longitudinal velocity scale 关6兴 Carpenter, R. C., 1893, ‘‘Experiments on Waterhammer,’’ Trans.
u - unknown vector for subsystem of longitudinal ASME, 15.
velocity component 关7兴 Frizell, J. P., 1898, ‘‘Pressures Resulting from Changes of Velocity
u - local longitudinal velocity of Water in Pipes,’’ Trans. Am. Soc. Civ. Eng. 39, pp. 1–18.
u - frictional velocity 关8兴 Joukowski, N. E., 1898, ‘‘Memoirs of the Imperial Academy Soci-
* ety of St. Petersburg,’’ 9共5兲 共Russian translated by O Simin 1904兲,
u⬘ - turbulence perturbation corresponding to u
Proc. Amer. Water Works Assoc. 24, pp. 341– 424.
V - cross-sectional average velocity 关9兴 Allievi, L., 1903, ‘‘Teoria generale del moto perturbato dell’acqu
v - velocity vector ani tubi in pressione,’’ Ann. Soc. Ing. Arch. Ithaliana 共French trans-
v - unknown vector for subsystem of head and lation by Allievi 共1904, Revue de mécanique兲.
radial velocity component 关10兴 Allievi, L., 1913, ‘‘Teoria del colpo d’ariete,’’ Atti Collegio Ing.
v - local radial velocity Arch. 共English translation by Halmos EE 1929兲, ‘‘The Theory of
v⬘ - turbulence perturbation corresponding to v Waterhammer,’’ Trans. ASME.
W - weighting function 关11兴 Courant, R. and Friedrichs, K. O., 1976, Supersonic Flow and
Shock Waves, Springer-Verlag, New York.
X - longitudinal length scale 关12兴 Jaeger, C., 1933, Theorie Generale du Coup de Belier, Dunod,
x - distance along the pipe Paris.
y, y - distance from pipe wall, dimensionless dis- 关13兴 Jaeger, C., 1956, Engineering Fluid Mechanics translated from Ger-
*
tance from pipe wall man by P.O. Wolf, Blackie, London.
Z - elevation of pipe centerline from a given da- 关14兴 Wood, F. M., 1937, ‘‘The Application of Heavisides Operational
tum Calculus to the Solution of Problems in Waterhammer,’’ Trans.
z - unknown vector for system ASME 59, pp. 707–713.
␣ - angle between pipe and horizontal direction 关15兴 Rich, G., 1944, ‘‘Waterhammer Analysis by the Laplace-Mellin
Transformations,’’ Trans. ASME, pp. 1944 – 45.
␣ - coefficient in weighting function 关16兴 Rich, G., 1951, Hydraulic Transients, 1st Edition, McGraw-Hill,
␤ - momentum correction coefficient New York, 1951 共Dover Reprint兲.
␤ - coefficient in weighting function 关17兴 Parmakian, J., 1955, Water-Hammer Analysis. Prentice-Hall Engle-
␥ - unit gravity force wood Cliffs, N.J., 1955 共Dover Reprint, 1963兲.
⑀ - distance from the water hammer front 关18兴 Streeter, V. L., and Lai, C., 1963, ‘‘Waterhammer Analysis Includ-
⑀ - eddy viscosity ing Fluid Friction,’’ Trans. Am. Soc. Civ. Eng. 128, pp. 1491–1524.
⑀ - implicit parameter for shear stress 关19兴 Streeter, V. L. and Wylie, E. B., 1967, Hydraulic Transients,
McGraw-Hill, New York.
⑀ - implicit parameter for friction
关20兴 Chaudhry, M. H., 1987, Applied Hydraulic Transients, Van Nos-
⑀ - measured and modeled data error trand Reinhold, New York.
␨ - a positive real parameter 关21兴 Watters, G. Z., 1984, Analysis and Control of Unsteady Flow in
␩ - difference from unity of Coriolis correction Pipelines, Butterworth, Stoneham, Ma.
␩ - constant for weighting function 关22兴 Wylie, E. B. and Streeter, V. .L 1984, Fluid Transients, FEB Press,
␪ - implicit parameter for radial flux Ann Arbor.
␬ - coefficient for weighting function 关23兴 Wylie, E. B., Streeter, V. L., and Suo, Lisheng, 1993, Fluid Tran-
␬ - coefficient for five-region turbulence model sient in Systems, Prentice-Hall, Englewood Cliffs,
关24兴 Mitra, A. K., and Rouleau W. T., 1985, ‘‘Radial and Axial Variations
␬ - coefficient for two-layer turbulence model in Transient Pressure Waves Transmitted Through Liquid Transmis-
␭1 - Lagrangian multiplier sion Lines,’’ ASME J. Fluids Eng. 107, pp. 105–111.
␭2 - Lagrangian multiplier 关25兴 Vardy, A. E., and Hwang, K. L., 1991, ‘‘A Characteristic Model of
␯ - kinematic viscosity Transient Friction in Pipes,’’ J. Hydraul. Res. 29共5兲, pp. 669– 685.
␯p - Poisson ratio 关26兴 Ghidaoui, M. S. 2001, ‘‘Fundamental Theory of Waterhammer,’’
␯T - total viscosity Special Issue of the Urban Water J. 共Special Issue on Transients,
␰ - strain Guest Editor: B. W. Karney兲, 1共2兲, pp. 71– 83.
关27兴 Walker, J. S., 1975, ‘‘Perturbation Solutions for Steady One-
␳ - fluid density Dimensional Waterhammer Waves,’’ ASME J. Fluids Eng. 6, pp.
␳0 - fluid density at undisturbed state 260–262.
␳e - effective density 关28兴 Hinze, J. O., 1975, Turbulence, McGraw-Hill Classic Textbook Re-
␴1 , ␴2 - unknown but desired parameters issue Series, New York.
␴x - axial stress 关29兴 Bergant, A. and Simpson, A. R., 1994, ‘‘Estimating Unsteady Fric-
␴␪ - hoop stress tion in Transient Cavitating Pipe Flow,’’ Proc. 2nd Int. Conf. on
␶ - shear stress Water Pipeline Systems, Edinburgh, UK, May 24 –26, BHRA Group
Conf. Series Publ. No. 110, pp. 3–15.
␶w - wall shear stress
关30兴 Axworthy, D. H., Ghidaoui, M. S., and McInnis, D. A., 2000, ‘‘Ex-
␶ ws - quasi-steady contribution of wall shear stress tended Thermodynamics Derivation of Energy Dissipation in Un-
␶ wu - discrepancy between unsteady and quasi- steady Pipe Flow,’’ J. Hydraul. Eng. 126共4兲, pp. 276 –287.
steady wall shear stress 关31兴 Brunone, B., Karney, B. W., Mecarelli, M., and Ferrante, M., 2000,
␾ - coefficient in unsteady friction formula. ‘‘Velocity Profiles and Unsteady Pipe Friction in Transient Flow,’’ J.
Water Resour. Plan. Manage. 126共4兲, pp. 236 –244.
关32兴 Ghidaoui, M. S., and Mansour, S., 2002, ‘‘Efficient Treatment of the
Vardy-Brown Unsteady Shear in Pipe Transients,’’ J. Hydraul. Eng.
References 128共1兲, pp. 102–112.
关1兴 Vanderburg, V. H., 1986, ‘‘Knowing Technology as if People Mat- 关33兴 Korteweg, D. J., 1878, ‘‘Über die fortpflanzungsgeschwindigkeit
tered,’’ Man-Env. Syst. 16, pp. 69–75. des schalles in elastischen rohren,’’ Ann. Phys. Chemie 5共12兲, pp.
关2兴 Kuhn, T., 1962, The Structure of Scientific Revolutions, University 525–542.
of Chicago Press, Chicago, IL. 关34兴 Lighthill, J., 1996, Waves in Fluids, Cambridge University Press,
关3兴 Menabrea, L. F., 1885, ‘‘Note sur les effects de choc de l’eau dans UK.
les conduites,’’ C. R. Hebd. Seances Acad. Sci. 47, July–Dec., pp. 关35兴 Tijsseling, A. S., 1995, ‘‘Fluid-Structure Interaction in Liquid-Filled
221–224. Pipe Systems: A Review,’’ J. Fluids Struct. 10, pp. 109–146.

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 73

关36兴 Streeter, V. L. and Wylie, E. B., 1985, Fluid Mechanics 共8th Edi- 关63兴 Wiggert, D. C., and Sundquist, M. J., 1977, ‘‘Fixed-Grid Character-
tion兲, McGraw Hill New York. istics for Pipeline Transients,’’ J. Hydraul. Div., Am. Soc. Civ. Eng.
关37兴 Silva-Araya, W. F., and Chaudhry, M. H., 1997, ‘‘Computation of 103共HY12兲, pp. 1403–1415.
Energy Dissipation in Transient Flow,’’ J. Hydraul. Eng. 123共2兲, pp. 关64兴 Goldberg, D. E., and Wylie, E. B., 1983, ‘‘Characteristics Method
108 –115. Using Time-Line Interpolations,’’ J. Hydraul. Eng. 109共5兲, pp. 670–
关38兴 Pezzinga, G., 1999, ‘‘Quasi-2D Model for Unsteady Flow in Pipe 683.
Networks,’’ J. Hydraul. Eng. 125共7兲, pp. 676 – 685. 关65兴 Lai, C., 1989, ‘‘Comprehensive Method of Characteristics Models
关39兴 Daily, J. W., Hankey, W. L., Olive, R. W., and Jordaan, J. M., 1956, for Flow Simulation,’’ J. Hydraul. Eng. 114共9兲, pp. 1074 –1095.
‘‘Resistance Coefficients for Accelerated and Decelerated Flows 关66兴 Yang, J. C., and Hsu, E. L., 1990, ‘‘Time-Line Interpolation for
Through Smooth Tubes and Orifices,’’ Trans. ASME 78共July兲, pp. Solution of the Dispersion Equation,’’ J. Hydraul. Res. 28共4兲, pp.
1071–1077. 503–523.
关40兴 Shuy, E. B., 1996, ‘‘Wall Shear Stress in Accelerating and Deceler- 关67兴 Yang, J. C., and Hsu, E. L., 1991, ‘‘On the Use of the Reach-Back
ating Turbulent Pipe Flows,’’ J. Hydraul. Res. 34共2兲, pp. 173–183. Characteristics Method of Calculation of Dispersion,’’ Int. J. Nu-
关41兴 Vardy, A. E., and Brown, J. M. B., 1997, ‘‘Discussion on Wall Shear mer. Methods Fluids 12, pp. 225–235.
Stress in Accelerating and Decelerating Pipe Flow,’’ J. Hydraul. 关68兴 Bentley, L. R., 1991, Discussion of ‘‘On the Use of the Reach-Back
Res. 35共1兲, pp. 137–139. Characteristics Method for Calculation of Dispersion,’’ by J. C.
关42兴 Ghidaoui, M. S., and Kolyshkin, A. A., 2001, ‘‘Stability Analysis of Yang, and EL Hsu, Int. J. Numer. Methods Fluids 13共5兲, pp. 1205–
Velocity Profiles in Water-Hammer Flows,’’ J. Hydraul. Eng. 1206.
127共6兲, pp. 499–512. 关69兴 Sibertheros, I. A., Holley, E. R., and Branski, J. M., 1991, ‘‘Spline
关43兴 Carstens, M. R., and Roller, J. E., 1959, ‘‘Boundary-Shear Stress in Interpolations for Water Hammer Analysis,’’ J. Hydraul. Eng.
Unsteady Turbulent Pipe Flow,’’ J. Hydraul. Div., Am. Soc. Civ. 117共10兲, pp. 1332–1349.
Eng. 85共HY2兲, pp. 67– 81. 关70兴 Karney, B. W., and Ghidaoui, M. S., 1997, ‘‘Flexible Discretization
关44兴 Pezzinga, G., 2000, ‘‘Evaluation of Unsteady Flow Resistances by Algorithm for Fixed Grid MOC in Pipeline Systems,’’ J. Hydraul.
Quasi-2d or 1d Models,’’ J. Hydraul. Eng. 126共10兲, pp. 778 –785. Eng. 123共11兲, pp. 1004 –1011.
关45兴 Eichinger, P. and Lein, G., 1992, The Influence of Friction on Un- 关71兴 Wood, D. J., Dorsch, R. G., and Lightnor, C., 1966, ‘‘Wave-Plan
steady Pipe Flow, Unsteady Flow and Fluid Transients, Bettess and Analysis of Unsteady Flow in Closed Conduits,’’ J. Hydraul. Div.,
Watts 共eds兲, Balkema, Rotterdam, The Netherlands, 41–50. Am. Soc. Civ. Eng. 92共HY12兲, pp. 83–110.
关46兴 Ghidaoui, M. S., Mansour, S. G. S., and Zhao, M., 2002, ‘‘Appli- 关72兴 Wylie, E. B. and Streeter, V. L., 1970, ‘‘Network System Transient
cability of Quasi Steady and Axisymmetric Turbulence Models in Calculations by Implicit Method,’’ 45th Annual Meeting of the So-
Water Hammer,’’ J. Hydraul. Eng. 128共10兲, pp. 917–924. ciety of Petroleum Engineers of AIME, Houston, Texas October
关47兴 Vardy, A. E. and Brown, J, M., 1996, ‘‘On Turbulent, Unsteady, 4 –7, paper No. 2963.
Smooth-Pipe Friction, Pressure Surges and Fluid Transient,’’ BHR 关73兴 Holly, F. M., and Preissmann, A., 1977, ‘‘Accurate Calculation of
Group, London, pp. 289–311. Transport in Two Dimensions,’’ J. Hydraul. Div., Am. Soc. Civ.
关48兴 Brunone, B. and Golia, U. M., 1991, ‘‘Some Considerations on Eng. 103共HY11兲, pp. 1259–1277.
Velocity Profiles in Unsteady Pipe Flows,’’ Proc. Int. Conf. on En- 关74兴 Chaudhry, M. H., and Hussaini, M. Y., 1985, ‘‘Second-Order Accu-
thropy and Energy Dissipation in Water Resources, Maratea, Italy, rate Explicitly Finite-Difference Schemes for Water Hammer
pp. 481– 487.
Analysis,’’ ASME J. Fluids Eng. 107, pp. 523–529.
关49兴 Greco, M., 1990, ‘‘Some Recent Findings On Column Separation
关75兴 Toro, E. F., 1997, Riemann Solvers and Numerical Methods for
During Water Hammer,’’ Excerpta, G.N.I., Padua, Italy, Libreria
Fluid Dynamics, Springer-Verlag, Berlin.
Progetto, ed., 5, 261–272.
关76兴 Toro, E. F., 2001, Shock-Capturing Methods for Free-Surface Shal-
关50兴 Brunone, B., Golia, U. M., and Greco, M., 1991, ‘‘Some Remarks
on the Momentum Equation for Fast Transients,’’ Proc. Int. Conf. low Flows, Wiley Ltd, Chichester, England.
on Hydr. Transients With Water Column Separation, IAHR, Valen- 关77兴 Guinot, V., 2002, ‘‘Riemann Solvers for Water Hammer Simulations
cia, Spain, 201–209. by Godunov Method,’’ Int. J. Numer. Methods Eng. 49, pp. 851–
关51兴 Brunone, B., Golia, U. M., and Greco, M., 1991, ‘‘Modelling of 870.
Fast Transients by Numerical Methods,’’ Proc. Int. Conf. on Hydr. 关78兴 Hwang, Y. H., and Chung, N. M., 2002, ‘‘A Fast Godunov Method
Transients With Water Column Separation, IAHR, Valencia, Spain, for the Water-Hammer Problem,’’ Int. J. Numer. Methods Fluids 40,
273–280. pp. 799– 819.
关52兴 Bergant, A., Simpson, A. R., and Vitkovsky, J., 2001, ‘‘Develop- 关79兴 O’Brian, G. G., Hyman, M. A., and Kaplan, S., 1951, ‘‘A Study of
ments in Unsteady Pipe Flow Friction Modelling,’’ J. Hydraul. Res. the Numerical Solution of Partial Differential Equations,’’ J. Math.
39共3兲, pp. 249–257. Phys. 29共4兲, pp. 223–251.
关53兴 Brunone, B., Golia, U. M., and Greco, M., 1995, ‘‘Effects of Two- 关80兴 Damuller, D. C., Bhallamudi, S. M., and Chaudhry, M. H., 1989,
Dimensionality on Pipe Transients Modeling,’’ J. Hydraul. Eng. ‘‘Modelling Unsteady Flow in Curved Channel,’’ J. Hydraul. Eng.
121共12兲, pp. 906 –912. 115共11兲, pp. 1471–1495.
关54兴 Wylie, E. B., 1997, ‘‘Frictional Effects in Unsteady Turbulent Pipe 关81兴 Samuels, G. P., and Skeel, P. C., 1990, ‘‘Stability Limits for Preiss-
Flows,’’ Appl. Mech. Rev. 50共11兲, Part 2, pp. S241–S244. mann’s Scheme,’’ J. Hydraul. Div., Am. Soc. Civ. Eng. 116共HY8兲,
关55兴 Vitkovsky, J. P., Lambert, M. F., Simpson, A. R., and Bergant, A., pp. 997–1011.
2000, ‘‘Advances in Unsteady Friction Modelling in Transient Pipe 关82兴 Karney, B. W., and Ghidaoui, M. S., 1992, ‘‘Discussion on Spline
Flow,’’ 8th Int. Conf. on Pressure Surges, The Hague, The Nether- Interpolations for Water Hammer Analysis,’’ J. Hydraul. Eng.
lands. 118共11兲, pp. 1597–1600.
关56兴 Zielke, W., 1968, ‘‘Frequency-Dependent Friction in Transient Pipe 关83兴 Sivaloganathan, K., 1978, ‘‘Flood Routing by Characteristic Meth-
Flow,’’ ASME J. Basic Eng. 90共1兲, pp. 109–115. ods,’’ J. Hydraul. Div., Am. Soc. Civ. Eng. 107共HY7兲, pp. 1075–
关57兴 Trikha, A. K., 1975, ‘‘An Efficient Method for Simulating 1091.
Frequency-Dependent Friction in Transient Liquid Flow,’’ ASME J. 关84兴 Wylie, E. B., 1980, ‘‘Inaccuracies in the Characteristics Method,’’
Fluids Eng. 97共1兲, pp. 97–105. Proc. Spec. Conf. on Comp. and Physical Modelling in Hydr. Eng.
关58兴 Suzuki, K., Taketomi, T., and Sato, S., 1991, ‘‘Improving Zielke’s ASCE, Chicago, 165–176.
Method of Simulating Frequency-Dependent Friction in Laminar 关85兴 Ghidaoui, M. S., and Karney, B. W., 1994, ‘‘Equivalent Differential
Liquid Pipe Flow,’’ ASME J. Fluids Eng. 113共4兲, pp. 569–573. Equations in Fixed-Grid Characteristics Method,’’ J. Hydraul. Eng.
关59兴 Vardy, A. E., Hwang, K. L., and Brown, J. M. B., 1993, ‘‘A Weight- 120共10兲, pp. 1159–1176.
ing Model of Transient Turbulent Pipe Friction,’’ J. Hydraul. Res. 关86兴 Ghidaoui, M. S., Karney, B. W., and McInnis, D. A., 1998, ‘‘Energy
31, pp. 533–548. Estimates for Discretization Errors in Waterhammer Problems,’’ J.
关60兴 Vardy, A. E., and Brown, J. M. B., 1995, ‘‘Transient, Turbulent, Hydraul. Eng. 123共11兲, pp. 384 –393.
Smooth Pipe Friction,’’ J. Hydraul. Res. 33, pp. 435– 456. 关87兴 Das, D., and Arakeri, J. H., 1998, ‘‘Transition of Unsteady Velocity
关61兴 Almeida, A. B. and Koelle, E., 1992, Fluid Transients in Pipe Net- Profiles with Reverse Flow,’’ J. Fluid Mech. 374, pp. 251–283.
works, Computational Mechanics Publications, Elsevier, New York. 关88兴 Brunone, B., Karney, B. W., and Ferrante, M., 1999, ‘‘Velocity Pro-
关62兴 Lister, M., 1960, The Numerical Solution of Hyperbolic Partial Dif- files Unsteady Friction Losses and Transient Modelling,’’ Proc. 26th
ferential Equations by the Method of Characteristics, A Ralston and Annu. Water Resour. Plng. and Mgmt. Conf. ASCE, Reston, VA 共on
HS Wilf 共eds兲, Numerical Methods for Digital Computers, Wiley CD-ROM兲.
New York, 165–179. 关89兴 Lodahl, C. R., Sumer, B. M., and Fredsoe, J., 1998, ‘‘Turbulent

74 Õ Vol. 58, JANUARY 2005 Transactions of the ASME

 Combined Oscillatory Flow and Current in Pipe,’’ J. Fluid Mech. cient Valve Representation in Fixed-Grid Characteristics Method,’’
373, pp. 313–348. J. Hydraul. Eng. 123共8兲, pp. 709–718.
关90兴 Ghidaoui, M. S., and Kolyshkin, A. A., 2002, ‘‘A Quasi-Steady 关113兴 Beck, J. L. and Katafygiotis, L. S., 1992, ‘‘Updating Dynamic Mod-
Approach to the Instability of Time-Dependent Flows in Pipes,’’ J. els and Their Associated Uncertainties for Structural Systems,’’ Pro.
Fluid Mech. 465, pp. 301–330. the 9th Engineering Mechanics Conference, L. D. Lutes and J. M.
关91兴 Pezzinga, G., and Scandura, P., 1995, ‘‘Unsteady Flow in Installa- Niedzwecki, eds., ASCE, Reston, VA, pp. 681– 684.
tions with Polymeric Additional Pipe,’’ J. Hydraul. Eng. 121共11兲, 关114兴 Sykes, J. F., 1985, ‘‘Sensitivity Analysis for Steady State Ground
pp. 802– 811. Water Flow Using Adjoint Operators,’’ Water Resour. Res. 21共3兲,
关92兴 Greenblatt, D., and Moss, E. A., 1999, ‘‘Pipe-Flow Relaminariza- pp. 359–371.
tion by Temporal Acceleration,’’ Phys. Fluids 11共11兲, pp. 3478 – 关115兴 Sun, N. Z., and Yeh, W. G., 1990, ‘‘Coupled Inverse Problems in
3481. Groundwater Modeling–2: Identifiability and Experimental De-
关93兴 He, S., and Jackson, J. D, 2000, ‘‘A Study of Turbulence Under sign,’’ Water Resour. Res. 26共10兲, pp. 2527–2540.
Conditions of Transient Flow in a Pipe,’’ J. Fluid Mech. 408, pp. 关116兴 Sun, N. Z., 1994, Inverse Problems in Groundwater Modeling, Klu-
1–38. wer Academic Publishers.
关94兴 Tu, S. W., and Ramaprian, B. R., 1983, ‘‘Fully Developed Periodic 关117兴 Jarny, Y., Ozisik, M. N., and Bardon, J. P., 1991, ‘‘A General Opti-
Turbulent Pipe Flow–Part 1: Main Experimental Results and Com- mization Method Using Adjoint Equation for Solving Multidimen-
parison with Predictions,’’ J. Fluid Mech. 137, pp. 31–58. sional Inverse Heat Conduction,’’ Int. J. Heat Mass Transfer 34共11兲,
关95兴 Brereton, G. L., Reynolds, W. C., and Jayaraman, R., 1990, ‘‘Re- pp. 2911–2919.
sponse of a Turbulent Boundary Layer to Sinusoidal Free-Stream 关118兴 Cacuci, D. G., and Hall, M. C. G., 1984, ‘‘Efficient Estimation of
Unsteadiness,’’ J. Fluid Mech. 221, pp. 131–159. Feedback Effects with Application to Climate Models,’’ J. Atmos.
关96兴 Akhavan, R., Kamm, R. D., and Shapiro, A. H., 1991, ‘‘Investiga- Sci. 13共2兲, pp. 2063–2068.
tion of Transition to Turbulence in Bounded Oscillatory Stokes
关119兴 Hall, M. C. G. , 1986, ‘‘Application of Adjoint Sensitivity Theory to
Flows–Part 1: Experiments,’’ J. Fluid Mech. 225, pp. 395– 422.
an Atmospheric General Circulation Model,’’ J. Atmos. Sci. 43共22兲,
关97兴 Akhavan, R., Kamm, R. D., and Shapiro, A. H., 1991, ‘‘Investiga-
pp. 2644 –2651.
tion of Transition to Turbulence in Bounded Oscillatory Stokes
关120兴 Marchuk, G. I., 1995, Adjoint Equations and Analysis of Complex
Flows–Part 2: Numerical Simulations,’’ J. Fluid Mech. 225, pp.
423– 444. Systems, Kluwer, London.
关98兴 Silva-Araya, W. F., and Chaudhry, M. H., 2001, ‘‘Unsteady Friction 关121兴 Liggett, J. A., and Chen, L. C., 1994, ‘‘Inverse Transient Analysis in
in Rough Pipes,’’ J. Hydraul. Eng. 127共7兲, pp. 607– 618. Pipe Networks,’’ J. Hydraul. Eng. 120共8兲, pp. 934 –995.
关99兴 Ohmi, M., Kyomen, S., and Usui, T., 1985, ‘‘Numerical Analysis of 关122兴 Karney, B. W. and Tang, K., 2003, personal communication.
Transient Turbulent Flow in a Liquid Line,’’ Bull. JSME 28共239兲, 关123兴 Brunone, B., 1999, ‘‘Transient Test-Based Technique for Leak De-
pp. 799– 806. tection in Outfall Pipes,’’ J. Water Resour. Plan. Manage. 125共5兲,
关100兴 Wood, D. J., and Funk, J. E., 1970, ‘‘A Boundary-Layer Theory for pp. 302–306.
Transient Viscous Losses in Turbulent Flow,’’ ASME J. Basic Eng. 关124兴 Brunone, B., and Ferrante, M., 2001, ‘‘Detecting Leaks in Pres-
102, pp. 865– 873. surised Pipes by Means of Transient,’’ J. Hydraul. Res. 39共5兲, pp.
关101兴 Bratland, O., 1986, ‘‘Frequency-Dependent Friction and Radial Ki- 539–547.
netic Energy Variation in Transient Pipe Flow,’’ Proc. 5th Int. Conf. 关125兴 Mpesha, W., Gassman, S. L., and Chaudhry, M. H., 2001, ‘‘Leak
on Pressure Surges, BHRA, Hannover, Germany, 95–101. Detection in Pipes by Frequency Response Method,’’ J. Hydraul.
关102兴 Rodi, W., 1993, Turbulence Models and Their Application in Hy- Eng. 127共2兲, pp. 137–147.
draulics: A State-of-the-Art Review, 3rd Edition, Int. Association for 关126兴 Mpesha, W., Chaudhry, M. H., and Gassman SL 2002, ‘‘Leak De-
Hydraulic Research, Delft, Balkema. tection in Pipes by Frequency Response Method Using a Step Ex-
关103兴 Kita, Y., Adachi, Y., and Hirose, K., 1980, ‘‘Periodically Oscillating citation,’’ J. Hydraul. Res. 40共1兲, pp. 55– 62.
Turbulent Flow in a Pipe,’’ Bull. JSME 23共179兲, pp. 654 – 664. 关127兴 Ferrante, M., and Brunone, B., 2002, ‘‘Pipe System Diagnosis and
关104兴 Eggels, J. G. M., 1994, ‘‘Direct and Large Eddy Simulation of Leak Detection by Unsteady-State Tests–1: Harmonic Analysis,’’
Turbulent Flow in a Cylindrical Pipe Geometry,’’ PhD Dissertation, Adv. Water Resour. 26, pp. 95–105.
Delft University of Technology. 关128兴 Wang, Xiao-Jian, Lambert, M. F., Simpson, A. R., and Liggett, J.
关105兴 Zhao, M., and Ghidaoui, M. S., 2003, ‘‘An Efficient Solution for A., 2002, ‘‘Leak Detection in Pipelines Using the Damping of Fluid
Quasi-Two-Dimensional Water Hammer Problems,’’ J. Hydraul. Transients,’’ J. Hydraul. Eng. 128共7兲, pp. 697–711.
Eng., 129共12兲, pp. 1007–1013. 关129兴 Ferrante, M., and Brunone, B., 2002, ‘‘Pipe System Diagnosis and
关106兴 Karney, B. W., and McInnis, D., 1990, ‘‘Transient Analysis of Water Leak Detection by Unsteady-State Tests–2: Wavelent Analysis,’’
Distribution Systems,’’ J. AWWA 82共7兲, pp. 62–70. Adv. Water Resour. 26, pp. 107–116.
关107兴 Wylie, E. B., 1983, ‘‘The Microcomputer and Pipeline Transients,’’ 关130兴 Payment, P., 1999, ‘‘Poor Efficacy of Residual Chlorine Disinfec-
J. Hydraul. Div., Am. Soc. Civ. Eng. 109共HY12兲, pp. 539– 42. tant in Drinking Water to Inactivate Waterborne Pathogens in Dis-
关108兴 Karney, B. W., 1984, ‘‘Analysis of Fluid Transients in Large Distri- tribution system,’’ Can. J. Microbiol. 45共8兲, pp. 709–715.
bution Networks,’’ Ph.D. thesis, University of British Columbia, 关131兴 Funk, J. E., van Vuuren, S. J., Wood, D. J., and LeChevallier, M.,
Vancouver, Canada. 1999, ‘‘Pathogen Intrusion into Water Distribution Systems Due to
关109兴 Fox, J. A., 1977, Hydraulic Analysis and Unsteady Flow in Pipe Transients,’’ Proc. 3rd ASME/JSME Joint Fluids Engineering Conf.,
Networks, MacMillan Press, London. July 18 –23, San Francisco, California.
关110兴 Koelle, E., 1982, ‘‘Transient Analysis of Pressure Conduit Hydrau- 关132兴 Germanopoulos, G. and Jowitt, P. W., 1989, ‘‘Leakge Reduction by
lic Systems,’’ Proc the Int. Institute on Hydraulic Transients and Excessive Pressure Minimization in a Water Supply Network,’’
Cavitation, Sao Paulo, Brazil, B1.1–B1.38. Proc. Inst. of Civ. Eng. 共UK兲, 195–214.
关111兴 McInnis, D. A., 1992, ‘‘Comprehensive Hydraulic Analysis of 关133兴 McInnis, D. A., 2003, ‘‘A Relative-Risk Assessment Framework for
Complex Pipe Systems,’’ Ph.D. thesis, University of Toronto, Tor- Evaluating Pathogen Intrusion During Transient Events in Water
onto, Canada. Pipelines,’’ Urban Water J. 共Special Issue on Transients, Guest Edi-
关112兴 McInnis, D. A., Karney, B. W., and Axworthy, D. H., 1997, ‘‘Effi- tor: B. W. Karney兲, 1共2兲, pp. 113–127.

Applied Mechanics Reviews JANUARY 2005, Vol. 58 Õ 75

 Mohamed Ghidaoui received his B.A.Sc., M.A.Sc. and Ph.D. all in Civil Engineering from
the University of Toronto, Canada, in 1989, 1991, and 1993, respectively. Since July 1993,
he has been with the Department of Civil Engineering at the Hong Kong University of
Science and Technology (HKUST) where he is an Associate Professor. His research interests
include modeling of surface water flows and water hammer: unsteady friction in conduits,
turbulence modeling of fast transients, flow stability of time-dependent flows and turbulent
shallow shear flow, numerical modeling of surface and closed conduit flows, and application
of Bolzmann theory in hydraulics. He is a member of the International Association of
Hydraulic Research (IAHR) and the American Society of Civil Engineers (ASCE). He is a
founding member of IAHR-Hong Kong and currently serves as its president. He is an Asso-
ciate Editor of the Journal of Hydraulic Research and an advisory board member of the
Journal of Hydroinformatics. His awards include the Albert Berry Memorial Award, Ameri-
can Water Works Association; runner-up for the Hilgard Award for best paper, Journal of
Hydraulic Engineering; and Teaching Excellence Awards, School of Engineering, HKUST.

Ming Zhao obtained his Ph.D. in April 2004 from the Department of Civil Engineering, at
Hong Kong University of Science and Technolgy. He obtained both his B.A.Eng. in hydraulic
engineering and B.A.Sc. in eneterprise management in 1999 from Tsinghua University. His
research interests include numerical simulation of unsteady pipe flows, open channel flows,
turbulence modeling in hydraulics, and stability analysis for fluid flows.

David H. Axworthy is a registered professional engineer with a consulting engineering firm

in Los Angeles, California. He obtained his B.A.Sc. (1991), M.A.Sc. (1993) and Ph.D. (1997)
in civil engineering from the University of Toronto. Axworthy has analyzed pressure tran-
sients created by the operation of pump stations and valves and designed surge protection
for water supply, wastewater, fire protection, deicing, diesel, and jet fuel systems. A member
of the ASCE and AWWA, Axworthy is coauthor of a water hammer analysis model (Tran-
sAM), serves as a reviewer for the ASME Journal of Fluids Engineering, and has published
scientific papers in the area of pipe network transients.

Duncan A. McInnis (Ph.D., P.Eng., MHKIE) has degrees in environmental biology and civil
engineering. He has 20 years of scientific and professional engineering experience in com-
putational hydraulics, simulation, and computer modeling of surface water and pipeline
systems. McInnis has been a Lecturer of Civil Engineering and Senior Project Manager at
the Hong Kong University of Science and Technology. He is currently the Manager of Water
Resources with Komex International Ltd., an international environmental consulting firm.

76 Õ Vol. 58, JANUARY 2005 Transactions of the ASME