See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/280624082

Meshless computation of water hammer

Conference Paper · October 2007

CITATIONS READS

6 409

2 authors:

Arris S Tijsseling Anton Bergant

Eindhoven University of Technology The Academic and Research Network of Slovenia
121 PUBLICATIONS   1,817 CITATIONS    115 PUBLICATIONS   2,229 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Arris S Tijsseling on 03 August 2015.

The user has requested enhancement of the downloaded file.

 Scientific Bulletin of the 2nd IAHR International Meeting of the Workgroup
“Politehnica” University of Timisoara on Cavitation and Dynamic Problems
Transactions on Mechanics in Hydraulic Machinery and Systems
Tom 52(66), Fascicola 6, 2007 Timisoara, Romania
October 24 - 26, 2007

MESHLESS COMPUTATION OF WATER HAMMER

Arris S. TIJSSELING * Anton BERGANT
Department of Mathematics and Computer Science Litostroj E.I. d.o.o.
Eindhoven University of Technology Ljubljana
*Corresponding author: Den Dolech 2, P.O. Box 513, 5600 MB Eindhoven, The Netherlands
Tel.: 31402472755, Fax: +31402442489, E-mail: a.s.tijsseling@tue.nl

ABSTRACT is easy to implement and its exact solutions can be

used to asses the numerical error in more conventional
Water hammer concerns the generation, propa-
methods for the test cases of frictionless water-hammer
gation, reflection and damping of pressure waves in
and lumped friction.
liquid-filled pipe systems. The numerical simulation
of water-hammer events is usually based on one-
KEYWORDS
dimensional mathematical models. Two equations
representing the conservation of mass and momentum Unsteady pipe flow; Hydraulic transients; Pressure
govern the unsteady flow in the pipes. Boundary surges; Water hammer; Wave propagation; Wave
and/or intermediate conditions model the behaviour of reflection; Method of characteristics; Mesh-free
hydraulic machinery and the additional complication method; Exact solution.
of column separation. The method of characteristics is
the preferred method of solution and the conventional 1. INTRODUCTION
approach is to define a fixed grid in the distance-time
plane. On this grid, the unknown pressures and ve- 1.1. Water-hammer equations
locities (or heads and discharges) are numerically Classical water-hammer theory [1-7] adequately
computed in a time-marching procedure that starts describes the low-frequency elastic vibration of liquid
from a given initial condition. Rectangular or diamond columns in fully-filled pipes. The pipes are straight,
grids are used that in general do not exactly match the prismatic, thin-walled, linearly elastic and of circular
given pipe lengths or the chosen time step. Therefore cross-section. The two equations, governing velocity,
numerical interpolations and/or adjustments of wave V, and pressure, P, are
speeds or pipe lengths are necessary, and these ne-
∂V 1 ∂P VV
cessities introduce error. + = − f , (1)
This paper presents a different way of water-hammer ∂t ρ ∂z 4R
computation. The procedure is based on the method
∂V 1 ∂P
of characteristics, but a numerical grid is not required. + *
=0 , (2)
Any point in the distance-time plane can be selected ∂z K ∂t
to compute the local solution without explicitly using with
stored previous solutions. The computation is based 1 1 2R
on back-tracking waves by means of a very simple *
= + . (3)
recursion, so that the programming effort is small. K K Ee
Exact solutions are thus obtained for frictionless Notation: E = Young modulus, e = wall thickness,
water-hammer. Approximate solutions are obtained f = Darcy-Weisbach friction factor, K = bulk modulus,
when the distributed friction in individual pipes is K* = effective bulk modulus, R = inner pipe radius,
concentrated at the pipe boundaries. t = time, z = distance along pipe, and ρ = mass den-
The new algorithm is fully described and the new
sity. The pressure wave speed is
method is evaluated and compared with conventional
water-hammer calculations. The following boundary K
*
conditions are tested: reservoir, instantaneous and c = . (4)
gradual valve closure, and pipe junction. The algorithm ρ
66 Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems

1.2. Conventional methods of solution 2. THEORY

The method of characteristics (MOC) is the pre- 2.1. General hyperbolic equations
ferred technique to solve the water-hammer equations
The general linear equations
[8-9], because the wave speed c is constant and, unlike
∂ ∂
finite-difference [10-11], finite-volume [12-13], finite- A φ( z,t ) + B φ ( z , t ) + C φ ( z , t ) = 0 (5)
element [14-15] and lattice-Boltzmann [16] methods, ∂t ∂z
steep wave fronts can be properly dealt with. The describe wave propagation in one spatial dimension.
distance-time (z-t) plane is covered with either a The constant matrices A and B are invertible and
rectangular (collocated) or a diamond (staggered) A-1 B is diagonalizable. The constant matrix C, which
computational grid in accordance with the Courant may be non-invertible, causes frequency dispersion
condition Δ x = c Δ t . Horizontal-vertical [1-7] or (if C ≠ O). The N dependent and coupled variables
diagonal [17] time-marching from a known initial φi , constituting the state vector φ , are functions of
condition gives approximate and exact solutions in the independent variables z (space: 0 ≤ z ≤ L) and t
systems with and without friction, respectively. In (time: t ≥ 0). Herein N = 2 and C = O.
multi-pipe systems the Courant condition cannot be
satisfied in all pipes if a common time step Δt is 2.2. Method of characteristics (MOC)
adopted, and wave-speed or pipe-length adjustments The MOC introduces a new set of dependent vari-
and/or interpolations have to be applied. These intro- ables through
duce numerical error. In single pipes, interpolations
are necessary when numerical data is required in η ( z , t ) = S −1 φ ( z , t ) or φ ( z , t ) = S η ( z , t ) , (6)
between grid points, for example at the exact location so that each ηi is a linear combination of the original
of measuring equipment. Wood et al. [18, 19] devel-
oped and commercialised an alternative MOC-based variables φi . Substitution of (6) into (5), with C = O,
solution procedure. Closed-form analytical solutions gives
can be obtained for water hammer in a single pipe with ∂ ∂
lumped [20] or linearised [21] friction. AS η( z ,t ) + BS η( z ,t ) = 0 (7)
∂t ∂z
1.3. New algorithm Multiplication by S-1A-1 yields
The new approach presented in this paper has no ∂ ∂
adjustments (of wave speeds or pipe lengths), no η( z ,t ) + Λ η( z ,t ) = 0 , (8)
numerical interpolations and no approximations (ex- ∂t ∂z
cept for lumped friction). The method automatically in which
tracks waves backward in time by means of a simple −1 −1
Λ = S A B S. (9)
recursion. The exact solutions obtained for frictionless
water-hammer in multi-pipe systems can be used: to A set of decoupled equations is obtained when L
check numerical results and schemes, to serve as is diagonal:
reference solutions in benchmark problems, and to
⎛ λ1 0 ⎞
perform parameter variation studies without parameter Λ =⎜ ⎟. (10)
changes generated by the numerical method itself ⎝0 λ2 ⎠
(e.g. changed wave speeds or pipe lengths). The method
is mesh-free in the sense that a conventional compu- Substitution of (10) into (9) and solving for S re-
tational grid is not required. The theory is presented veals that a non-trivial solution exists only when the
in general form so that it can be applied to analogous diagonal elements of L are eigenvalues satisfying
two-equations systems in, for example, open chan- the characteristic equation
nel flow, structural vibration, and electronics. It can det ( B − λ A ) = 0 , (11)
be extended to four-equation systems encountered
in which case S consists of the eigenvectors ξ i be-
in, for example, the theory of linear(ised) waves in
two-phase flows, two-layer density currents, liquid- longing to λ i :
saturated porous media, and water hammer with
S = ( ξ1 ξ 2 ) . (12)
fluid-structure interaction [22]. The method even
works for higher-order systems like those developed The decoupled equations (8),
for water hammer by Bürmann [23], noting that then
∂ η i ( z ,t ) ∂ ηi ( z ,t )
the necessary eigenvalues and eigenvectors cannot + λi =0 , i = 1 , 2 , (13)
be found in closed form. ∂t ∂z
Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems 67

transform to dTz (t ) φ( zb , t ) = qz (t ) or
b b

d ηi ( z , t ) d Tz (t ) S η( zb , t ) = qz (t ) , (19)
=0 , i = 1, 2 , (14) b b
dt
where d Tzb is a row vector with 2 coefficients and the
when they are considered along characteristic lines
in the z-t plane defined by scalar qzb is the (boundary) excitation. For conven-

dz ience, the two row vectors d T0 and d TL are stacked in

= λi , i = 1, 2 . (15) the 2 by 2 matrix D, and the vector q consists of the
dt
two scalars q1 = q0 and q2 = qL. Section 2.6 gives an
The solution of the ordinary differential equa- example of matrix D.
tions (14) and (15) is
η i ( z , t ) = η i ( z − λ i Δt , t − Δt ) , i = 1 , 2 , (16)

when a numerical time step Δt is used, or, more

general and with reference to Fig. 1,
η i ( P ) = η i ( Ai ) , i = 1 , 2 ,

⎛ η 1 ( A1 ) ⎞
or η ( P ) = ⎜ ⎟. (17)
⎝η 2 ( A 2 ) ⎠
The value of the unknown variable ηi does not
change along the line Ai P.
The original unknowns in φ are obtained from η
through (6). This gives
2
Figure 2. Boundary point P and "feeding"
φ(P) = ∑ S R i S −1 φ ( Ai ) , (18)
characteristic line in the distance-time plane.
i =1

where the i-th diagonal element of the matrix R i is

2.4. Junction conditions
1 and all other elements are 0.
At junctions at z = zj the unknowns ηi (P) may be
discontinuous. The relations (16) or (17) provide two
equations for ηi (P ± ), i = 1, 2 (see Fig. 3). Two addi-
tional equations are required to find the unknowns at
both sides of the junction, z = z j − and z = z j +. These
are given by the linear junction conditions
J1(t ) φ ( z j− , t ) = J2(t ) φ ( z j+ , t ) + r (t ) or
J1(t ) S1 η1 ( z j , t ) = J2(t ) S2 η2 ( z j , t ) + r (t ) , (20)

Figure 1. Interior point P and "feeding"

characteristic lines in the distance-time plane.

2.3. Boundary conditions

At the boundaries at z = 0 and z = L, the relations
(16) or (17) provide just one equation (see Fig. 2).
To find the two unknowns ηi (P) , i = 1, 2, at z = zb
(zb = 0 or zb = L), one additional equation per bound-
ary is required. These are given by the linear boundary Figure 3. Junction point P and "feeding"
conditions characteristic lines in the distance-time plane.
68 Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems

where J1 and J2 are 2 by 2 coefficient matrices and dTL = (1 0 ) and qL = 0 , (27)

the vector r is the (junction) excitation. Section 2.6
gives an example of J1 and J2. where Pres is the reservoir pressure. Now Eqs. (26)
Thus the state vector φ can be discontinuous at and (27) are combined to give
z = z j . In general, the Riemann invariants η1 = S1 −1 φ ⎛0 1⎞ ⎛P ⎞
D=⎜ ⎟ and q = ⎜ res ⎟ . (28)
and η2 = S2 −1 φ in the domains left and right of the ⎝1 0⎠ ⎝ 0 ⎠
junction are different. Only if J1(t ) S1 = J2(t ) S2 the The junction conditions are defined through coeffi-
junction will not cause wave reflection. cient matrices J1 and J2, and excitation vector r [see
Eq. (20)]. For example, continuity of flow rate and
2.5. Initial conditions pressure at the junction of two pipes requires that
The initial condition at t = 0 can be the steady- ⎛ A1 0 ⎞ ⎛ A2 0⎞ ⎛0⎞
state solution φ ( z,0) = φ0 , where the constant state J1 = ⎜ ⎟ , J2 = ⎜ ⎟ and r = ⎜ ⎟ , (29)
⎝ 0 1⎠ ⎝ 0 1⎠ ⎝0⎠
φ0 is consistent with the boundary and junction
conditions, or it can be any non-equilibrium state where A1 and A2 are the cross-sectional pipe areas
φ ( z,0) = φ0 ( z ) exciting the system (e.g. sudden left and right of the junction.
release of pressure). 2.7. Nonlinear non-instantaneous valve closure
2.6. Water-hammer two-equation model In steady turbulent pipe flow the pressure loss,
Δ P0 , across a fully open valve is given by the orifice
In terms of the general equation (5), frictionless
water-hammer [Eqs. (1-2) with f = 0] is represented equation
by the state vector Δ P0 = ξ 0 12 ρ V0 V0 , (30)
⎛V ⎞ where ξ0 is an empirical loss coefficient [4, pp. 44-45].
φ=⎜ ⎟ (21)
⎝ P⎠ The same relation is assumed to hold for a closing
and the matrices of coefficients valve,
⎛1 0 ⎞ ΔP = ξ 1
2
ρV V , (31)
A = ⎜⎜ ⎟ ,
* −1 ⎟
(22)
⎝0 K ⎠ where x depends on valve position and hence of time.
Division of (31) by (30) and introduction of the dimen-
⎛ 0 ρ −1 ⎞
B = ⎜⎜ ⎟⎟ . (23) sionless valve closure coefficient τ = ξ0 / ξ gives
⎝1 0 ⎠
the nonlinear boundary condition
The characteristic (dispersion) equation (11),
P0 V V = τ 2 (t ) V0 V0 P , (32)
corresponding to the matrices (22) and (23), yields
K* where the pressure downstream of the valve has been
λ 1,2 2 = = c2 , (24) taken zero, so that Δ P0 = P0 and Δ P = P .
ρ
The specific function τ (t ) used herein is
where λ1 is positive and λ2 is negative.
The transformation matrix used [Eq. (12)] is ⎧ (1 − t / Tc ) 3.53 for 0 ≤ t ≤ 0.4 Tc
⎪⎪
, (33)
S = ( TA ) with T defined by
–1 τ ( t ) = ⎨ 0.394 (1 − t / Tc )1.70 for 0.4 Tc ≤ t ≤ Tc
⎪ 0 for t ≥ T
⎪⎩ c
⎛ 1 λ1 ⎞ ⎛ 1 ( ρ c ) −1 ⎞
T=⎜ ⎟ , so that S −1 = ⎜ ⎟ . (25) in which Tc is the valve closure time and ξ0 = 0.2 (see
⎜1 λ 2 ⎟ ⎜ 1 − ( ρ c ) −1 ⎟
⎝ ⎠ ⎝ ⎠
Fig. 4). This function, which is based on measured
It is noted that transformation matrices are not unique. ball-valve discharge coefficients, was provided by
For example, S = ( TB ) is an equally valid trans-
–1
Professor DC Wiggert (Michigan State University,
formation matrix. USA) to Delft Hydraulics (The Netherlands) in 1987.
The boundary conditions are defined through More information on ball-valve discharge character-
coefficient matrix D and excitation vector q [see Eq. istics can be found in [24-25].
(19)]. For example, a reservoir at z = 0 and a valve The quadratic equation (32) is solved simultane-
at z = L, as in the test problems in Section 4, are rep- ously with one linear equation (one compatibility
resented by equation). This is done exactly in Appendix A. In
using the quadratic formula for (32), care should be
dT0 = ( 0 1) and q 0 = Pres , (26) taken of possible cancellation [26].
Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems
1 in terms of the Riemann invariants ηi , i = 1, 2, and it
tau (-)
0.5
0 A2, on the right boundary at z = L. The component η 1
0 0.01 0.02
time (s)
Figure 4. Valve-closure function t(t) for Tc = 0.03 s.
The resistance of the valve in steady state leads
to a small initial pressure, so that
⎛ V0 ⎞
φ0 = ⎜ ⎟ ,
⎝ P0 ⎠
with P0 equal to Pres to ensure initial equilibrium.
2.8. Modelling of friction
Distributed friction (f > 0) is modelled by lumping
it to one of the pipe ends, which is the common pro-
cedure in graphical [27] and wave-plan [18] methods.
For example, the pressure loss at a reservoir-pipe
entrance is then represented by the boundary condi-
tion [4, p. 103]
L ρV V
Pres − P = (1 + k + f ) , (35)
2R 2 tion of some basic results for frictionless water-hammer.
where k is the (minor) entrance loss and L is the pipe
length. The term 1+k is neglected herein.
2.9. Modelling of column separation
Column separation [28] occurs when the absolute
pressure drops to the vapour pressure Pv . This can be
modelled at closed pipe ends by adopting the boundary
condition P = Pv as long as the cavity volume ∀ is
positive. The calculation of ∀ through
∀ (t ) =∀ (t − Δ t ) ± A V (t ) Δ t (36)
can only be done by explicitly storing (and interpo-
lating) previous cavity volumes.
3. ALGORITHM
This section with its two appendices is the heart of
the paper. It describes an algorithm for finding the
exact value of the state vector φ in any point (z, t) in
the distance-time plane. The algorithm is formulated
R E S E R V O IR
69
is based on a "coast-to-coast" approach, which is
explained in words now. Figure 2 is essential. The
vector η in point P on the left boundary at z = 0 at
time t consists of the two components η 1 and η 2.
The component η 2 is assumed to be known because,
according to equation (17), it is equal to η 2 in the point
0.03
follows from the boundary condition (19) and it de-
pends on d T0 , S, q0 and η 2. Naturally, the same story
holds for any point P on the right boundary. For the
calculation of η in point P on the left boundary, one
needs information from the "earlier" point A2 on the
right boundary; and for the calculation of η in the point
(34) A2 on the right boundary, one needs information from
the "earlier" point A1 on the left boundary. This whole
process can nicely be captured in a simple recursion
that stops when characteristic lines intersect the t = 0
line, at which η has a given initial value. The recursion
and the treatment of internal points (Fig. 1) are pre-
sented for a single pipe in Appendix B. The treatment
of junctions (Fig. 3) and double pipes is similar and
dealt with in Appendix C.
4. RESULTS
The new method is validated through the calcula-
The obtained exact solutions give details as fine as
the chosen resolution in time or distance. Conven-
tional MOC results obtained on very fine computa-
tional grids are given as a control.
4.1. Single pipe
The single-pipe test problem [22] sketched in
Fig. 5 is a water-filled reservoir-pipe-valve system
defined by: L = 20 m, R = 398.5 mm, e = 8 mm,
E = 210 GPa, K = 2.1 GPa, ρ = 1000 kg/m3, f = 0,
and Q0 = 0.5 m3/s, so that the initial flow velocity
V0 = 1.002 m/s and the wave speed c = λ1 = −λ2 =
1025.7 m/s [Eq. (24)]. A pressure transducer (PT) is
assumed to be located at distance z = 11.15 m from
the reservoir. Rapid closure of the valve generates the
Joukowsky pressure P = ρ cV0 = 1.028 MPa only if
the closure time Tc is smaller than the wave reflection
time 2L/c. This is the case for the pressure histories
(taken relative to a constant static pressure high enough.
P IP E VALVE
Figure 5. Single-pipe test problem.
70 Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems

to prevent the occurrence of cavitation) portrayed in 6

1.5 .10
Fig. 6. The classical case of instantaneous valve clo-
sure is shown in Fig. 6a. Figure 6b, which is for non-
6

pressure (Pa)
instantaneous valve closure with Tc = 0.03 s, reveals 1 .10
that the pressure effectively starts to rise after time
0.7Tc . Figure 7 displays the corresponding pressure 5 .10
5

distribution along the pipe at time t = 0.24 seconds.

The results in Figs. 6 and 7 were obtained with 0
0 2 4 6 8 10 12 14 16 18 20
the new method and with the conventional MOC distance (m)
method utilising a very fine computational grid (924
reaches and a time step of 21 microseconds) and a (7a)
marginally changed wave speed c = 1025.3 m/s. Both 0
solutions are displayed in the Figs. 6 and 7, but the
differences are hardly visible, thus confirming the 5

pressure (Pa)
5 .10
validity of the new method. The new method calcu-
lated solutions in 240 points in time (1 millisecond 6
1 .10
step) to produce Fig. 6 and in 240 reaches to produce
Fig. 7. These values were chosen to obtain high resolu-
6
tion, not high accuracy, because exact solutions are 1.5 .10
0 2 4 6 8 10 12 14 16 18 20
always obtained. It is noted that finding exact solu- distance (m)
tions for frictionless water-hammer in a single pipe
(7b)
6
1.5 .10 Figure 7. Single-pipe test problem.
1 .10
6 Pressure distribution along the pipe calculated at
5 t = 0.24 s: (a) Tc = 0 s, (b) Tc = 0.03 s. Solid line:
pressure (Pa)

5 .10
new method. Dotted line: conventional method.
0
5 .10
5 is nothing special, except that the exact solutions are
6 not necessarily in equidistantly spaced grid points.
1 .10
1.5 .10
6 4.2. Double pipe
0 0.05 0.1 0.15 0.2
time (s)
In the double-pipe test problem sketched in Fig. 8,
the first 3.85 m of the pipeline has a twice thicker wall
(6a) (e1 = 16 mm, e2 = 8mm). The wave speed in the thick
1.5 .10
6 section is c1 = 1184.0 m/s, which is larger than c2 =
1 .10
6 1025.7 m/s. The calculated pressure histories are shown
5 in Fig. 9. The instantaneous valve closure in Fig. 9a
pressure (Pa)

5 .10
clearly brings out all the reflections from the thicker
0 pipe at the inlet. Joukowsky’s value is exceeded and a
5
5 .10 beat is starting to develop. The mathematical challenge
1 .10
6 is to capture all the tiny peaks in the entire period of
1.5 .10
6 simulation. The new method is able to do so, because
0 0.05 0.1 0.15 0.2 it is exact. The traditional water-hammer simulation
time (s) can only achieve this by using again a very fine
(6b) computational grid of 900 reaches with a time step
Figure 6. Single-pipe test problem. of 21 microseconds, and marginally changed wave
speeds c1 = 1184.3 m/s and c2 = 1025.5 m/s. The
Pressure history calculated at z = 11.15 m:
new method used a time step of 500 microseconds
(a) Tc = 0 s, (b) Tc = 0.03 s. Solid line:
to produce Fig. 9 and 480 reaches to produce Fig. 10.
new method. Dotted line: conventional method.
Figure 9b is the more realistic case of gradual valve

3 .8 5 1 1 .1 5
20
0 .8
0 zj PT L
Figure 8. Double-pipe test problem. Distances in metres.
Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems 71

closure. All tiny peaks are smeared out now, but (10a)
Joukowsky is still exceeded and a beat still devel- 0
ops. Figure 10 is the pressure distribution along the
5
pipeline at time t = 0.24 seconds. 2 .10
Both new and conventional solutions are displayed 4 .10
5
in the Figs. 9 and 10, but the differences are hardly

pressure (Pa)
5
6 .10
visible, thereby confirming again the validity of the
new method. Finding exact solutions for frictionless 8 .10
5

water-hammer in a double pipe is − to the author’s 6

1 .10
knowledge − new.
6
6
1.2 .10
1.5 .10
6
1.4 .10
6 0 2 4 6 8 10 12 14 16 18 20
1 .10
distance (m)

5 .10
5
(10b)
pressure (Pa)

0
Figure 10. Double-pipe test problem.
Pressure distribution along the pipeline calculated
5
5 .10 at t = 0.24 s: (a) Tc = 0 s, (b) Tc = 0.03 s. Solid line:
6
new method. Dotted line: conventional method.
1 .10

6
1.5 .10 5. CONCLUSION
0 0.05 0.1 0.15 0.2
time (s)
The equations for water hammer without friction
(9a) have been solved exactly for time-dependent bound-
1.5 .10
6 ary and constant (steady state) initial conditions with
a new method based on the MOC. The strength of
6
1 .10 the method is the simplicity of the algorithm (recur-
5 .10
5 sion) and the fast and accurate (exact) calculation of
pressure (Pa)

transient events. Its weakness is that the calculation

0 time strongly increases for events of longer duration.
5
The solutions are exact for the selected points of
5 .10
calculation in the distance-time plane, but they do not
1 .10
6 give information on what exists in between these points
(no guaranteed constant values as in methods that
6
1.5 .10 track discontinuities). The efficiency of the algorithm
0 0.05 0.1 0.15 0.2
time (s) can be much improved because many repeat calcula-
(9b) tions occur, but this will be at the expense of simplicity
Figure 9. Double-pipe test problem. and clarity. The method is mesh-free and ideal for
Pressure history calculated at z = 11.15 m: (a) Tc = 0 s, parallellisation and adaptivity. It is highly suitable to
(b) Tc = 0.03 s. Solid line: new method. Dotted line: quickly calculate a limited number of exact transient
conventional method. pressure histories and distributions in simple frictionless
pipe systems.
6
1 .10 The new method has been successfully validated
5
8 .10 against conventional water-hammer software in two
6 .10
5 simple test problems. In the second test problem, for
4 .10
5 the first time, an exact solution for frictionless water-
pressure (Pa)

5 hammer in a two-pipe system is presented.

2 .10

0 NOMENCLATURE
5
2 .10
Scalars
5
4 .10
A cross-sectional area, m2
5
6 .10 c pressure wave speed, m/s
0 2 4 6 8 10 12 14 16 18 20
distance (m) det determinant
E Young modulus of pipe wall material, Pa
72 Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems

e pipe wall thickness, m APPENDIX A. SOLUTION OF NONLINEAR

f Darcy-Weisbach friction factor EQUATION (32)
K bulk modulus, Pa
The nonlinear boundary condition (32) is solved
K* effective bulk modulus, Pa
simultaneously with one linear equation for the two
k minor (entrance) loss
variables V and P. The system of equations can be
L pipe length, m
written as
P pressure, Pa
Q flow rate, m3/s a1 V + a2 P = a3 , (37a)
qz
b
boundary excitation b1 V 2 + b2 P = b3 , (37b)
R inner radius of pipe, m
where the coefficients ai and bi (i = 1, 2, 3) can be time
Tc valve closure time, s
dependent and a2 ≠ 0 and b1 ≠ 0 .
t time, s
Elimination of P gives
V velocity, m/s
∀ cavity volume, m3 a2 b1 V 2 − a1 b2 V + a3 b2 − a2 b3 = 0 , (38)
z axial coordinate, m so that
Δ numerical step size; change in magnitude 1

λ eigenvalue, wave speed, m/s a b ± {a12 b22 − 4 a2 b1 (a3 b2 − a2 b3 )}2

V = 1 2 . (39)
ξ valve loss coefficient 2 a2 b1
ρ mass density, kg/m3 This formula is susceptible to numerical error,
τ valve closure function - see Eq. (33) because of cancellation [26].
Matrices and Vectors In the present paper the coefficients in (37) for
the single pipe are
A coefficients - see Eqs. (5) and (22)
B coefficients - see Eqs. (5) and (23) a1 = S −11
1
, a2 = S −12
1
, a3 = η1 (A1 ) ,
C coefficients - see Eq. (5) and C = O herein
D boundary condition coefficients - see Eqs. b1 = ± 1 , b2 = −τ 2 (t ) V 0 V 0 / P0 , b3 = 0 ,
(19) and (28) and for the double pipe these are
d Tz row of matrix D
b a1 = S2 −11
1
, a2 = S2 −12
1
, a3 = η 21 ("A1 ") ,
J1, J2 junction condition coefficients - see Eqs.
(20) and (29) b1 = ± 1 , b2 = −τ 2 (t ) V 20 V 2 0 / P 20 , b3 = 0 .
O zero matrix
The matrices S and S2 are given by (25) and the
q boundary excitation vector - see Eqs. (19)
R matrix - see Eq. (18) function τ(t) by (33).
r junction excitation vector - see Eqs. (19) According to relation (32), which describes posi-
S transformation matrix - see Eqs. (12) and (25) tive flow (in z-direction) for positive pressure P, the
T transformation matrix - see Eq. (25) following condition must hold:
Λ diagonal matrix of eigenvalues - see Eq. (10) b1 V ≥ 0 . (40)
η Riemann invariants - see Eq. (6) In the examples presented herein, b1 = + 1 and the
ξ eigenvector - see Eq. (12) + sign (+ root) in (39) gave valid solutions satisfying
φ dependent variables - see Eqs. (5) and (21) condition (40).
Subscripts
APPENDIX B. RECURSION FOR A SINGLE
b boundary
PIPE
j junction
L boundary position z = L Input
res reservoir The coefficient matrices A, B, D(t), the excitation
v vapour vector q(t), the initial condition φ0 ( z ) , the length L,
0 boundary position z = 0; initial state t = 0
and the positive wave speed λ1 = −λ2 describe the
Extensions complete single-pipe system in general terms. It is
1 pipe 1 stressed once again that C = O in Eq. (5), so that the
2 pipe 2 algorithm does not apply to systems with distributed
friction, viscous damping, etc.
Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems 73

Output Implementation
The output is the state vector φ = Sη as a function The algorithms have been implemented in Math-
of distance and time. cad 11 [29]; the interested reader may find the Math-
cad worksheets in [30].
Boundary points
Table 1. Definition of the coefficients α and β
See Figure 2. The subroutine BOUNDARY cal-
in subroutine BOUNDARY. DS(t) is the matrix
culates η in the boundary points (finally) from con-
product of D(t) and S.
stant initial values. The pseudo-code reads:
z=0 z=L
BOUNDARY (input: z, t ; output: η)
if (t ≤ 0) then −DS ( t ) 1 , 2 −DS ( t ) 2 , 1
α12( t ) := α21( t ) :=
η := η 0 DS ( t) 1 , 1 DS ( t) 2 , 2
else
1 1
if (z = 0) then β11( t ) := β22( t ) :=
CALL BOUNDARY (L, t + L / λ2 ; η) DS ( t) 1 , 1 DS ( t) 2 , 2
h2 := η 2
η 1 := a12(t) η2 + b11(t) q1(t) APPENDIX B. RECURSION FOR A DOUBLE
η 2 := η2 PIPE
if (z = L) then Input
CALL BOUNDARY (0, t – L / λ1 ; η) The coefficient matrices A1, B1, A2, B2, D(t),
η1 := η 1 J1(t), J2(t), the excitation vectors q(t) and r(t), the
η 1 := η1 initial conditions φ1 0 ( z ) and φ2 0 ( z ) , the lengths z j
η 2 := a21(t) η1 + b22(t) q2(t) and L, and the positive wave speeds λ11 = −λ12 and
λ21 = −λ22, describe the complete double-pipe system
if (z ≠ 0 AND z ≠ L) then in general terms. The extensions 1 and 2 in the nota-
"z is not at a boundary" tion refer to the domains left and right of the junction
end at z = z j . See Figure 3.
Note that l2 is a negative number herein. The coef-
Output
ficients a and b are given in Table 1. They define the
unknowns η 1 and η 2 at the left and right boundaries, The output is the state vector φ = Sη as a function
respectively, as the exact solution of the linear equa- of distance and time, with φ = S1 η1 if 0 ≤ z ≤ z j− and
tion (19). For the case of nonlinear non-instantaneous
φ = S2 η2 if z j+ ≤ z ≤ L .
valve closure, as described in Section 2.7, in subroutine
BOUNDARY: q2 (t ) = V from Eq. (39), for z = L.
Boundary and junction points
Interior points See Figure 3. The subroutine BOUNDARY cal-
See Figure 1. The subroutine INTERIOR calcu- culates η1 and η2 in the boundary and junction points
lates η in the interior points from the boundary val- (finally) from constant initial values. The pseudo-code
ues. The pseudo-code reads: reads:

INTERIOR (input: z, t ; output: η) BOUNDARY (input: z, t ; output: η1, η2)

if (0 < z < L) then if (t ≤ 0) then
CALL BOUNDARY (0, t – z / λ1 ; η) η 1 := η1 0
η1 := η 1 η 2 := η2 0
CALL BOUNDARY (L, t – (z-L) / λ2 ; η) else
η2 := η 2 if (z = 0) then
CALL BOUNDARY (z j, t + z j / λ12 ; η1,
η 1 := η1
η2)
η 2 := η2 h2 := η1 2
else η11 := a12(t) η2 + b11(t) q1(t)
"z is not an interior point"
end η1 2 := η2
74 Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems

if (z = L) then Note that l12 and l22 are negative numbers herein.
CALL BOUNDARY (z j, t – (L–z j) / λ21 ; The coefficients a and b are given in Table 2. They
η1, η2) define the unknowns η 1 and η 2 at the left and right
η1 := η 2 1 boundaries, respectively, as the exact solution of the
system of two linear equations (19). For the case of
η 2 1 := η1
nonlinear non-instantaneous valve closure, as de-
η 2 2 := a21(t) η1 + b22(t) q2(t) scribed in Section 2.7, in subroutine BOUNDARY:
if (z = z j) then q2 (t ) = V from Eq. (39), for z = L. The coefficients γ
CALL BOUNDARY (0, t – z j / λ11 ; η1, and δ are given in Table 3. They define the unknowns
η2) η 2 and η 1 at the left and right sides of the junction,
η1 := η11 respectively, as the exact solution of the system of
CALL BOUNDARY (L, t – (z j-L) / λ22 ; two linear equations (20).
η1, η2) Table 2. Definition of the coefficients α and β
η2 := η 2 2 in subroutine BOUNDARY. DS1(t) is the matrix
η11 := η1 product of D(t) and S1, and DS2(t) is the matrix
product of D(t) and S2.
η 2 1 := γ11(t) η1 + γ12(t) η2 + δ11(t) r1(t) +
δ12(t) r2(t) z=0 z=L
η1 2 := γ21(t) η1 + γ22(t) η2 + δ21(t) r1(t) + −DS1 ( t ) 1 , 2 −DS2 ( t ) 2 , 1
α12( t ) := α21( t ) :=
δ22(t) r2(t) DS1 ( t) 1 , 1 DS2 ( t) 2 , 2
η 2 2 := η2
1 1
if (z ≠ 0 AND z ≠ L AND z ≠ z j) then β11( t ) := β22( t ) :=
"z is not at a boundary or junction" DS1 ( t) 1 , 1 DS2 ( t) 2 , 2
end

Table 3. Definition of the coefficients γ and δ in subroutine BOUNDARY. JS1(t) is the matrix
product of J1(t) and S1, and JS2(t) is the matrix product of J2(t) and S2.

junction side 1, z = zj− junction side 2, z = zj+

JS2( t) 2 , 1⋅ JS1( t) 1 , 1 − JS2( t) 1 , 1⋅ JS1( t) 2 , 1 JS1( t) 2 , 2⋅ JS1( t) 1 , 1 − JS1( t) 1 , 2⋅ JS1( t) 2 , 1

γ21( t ) := γ11( t) :=
det ( t ) det ( t)

JS2( t) 1 , 1⋅ JS2( t) 2 , 2 − JS2( t) 2 , 1⋅ JS2( t) 1 , 2 JS1( t) 1 , 2⋅ JS2( t) 2 , 2 − JS1( t) 2 , 2⋅ JS2( t) 1 , 2

γ22( t ) := γ12( t ) :=
det ( t ) det ( t )

−JS2( t ) 2 , 1 −JS1( t) 2 , 2
δ21( t ) := δ11( t ) :=
det ( t ) det ( t )

JS2( t) 1 , 1 JS1( t) 1 , 2
δ22( t) := δ12( t) :=
det ( t ) det ( t )

det ( t ) := JS1 ( t ) 2 , 2⋅ JS2 ( t ) 1 , 1 − JS1 ( t ) 1 , 2⋅ JS2 ( t ) 2 , 1

Interior points η1 := η11

The subroutine INTERIOR calculates in the inte- CALL BOUNDARY (z j, t – (z- z j) / λ12 ; η1,
rior points from the boundary and junction values. η2)
The pseudo-code reads: η2 := η1 2
INTERIOR (input: z, t ; output: η1, η2) η 1 := η1
if (0 < z < z j) then
η 2 := η2
CALL BOUNDARY (0, t – z / λ11 ; η1, η2)
Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems
if (z j < z < L) then
CALL BOUNDARY (z j, t – (z-z j) / λ21 ; η1,
η2)
η1 := η 2 1
CALL BOUNDARY (L, t – (z-L) / λ22 ; η1,
η2)
η2 := η 2 2
η 1 := η1
η 2 := η2
else
"z is not an interior point"
end
Implementation
The algorithms have been implemented in Math-
cad 11 [29]; the interested reader may find the Math-
cad worksheets in [30].
REFERENCES
[1] Aronovich, G. V., Kartvelishvili, N. A., and Lyubim-
tsev, Ya. K., 1968, Gidravlicheskii Udar i Uravnitel’
nye Rezervuary, (Water Hammer and Surge Tanks,)
Izdatel’stvo “Nauka”, Moscow, Russia (in Russian).
(English translation by D. Lederman, 1970, Israel
Program for Scientific Translations, Jerusalem, Israel.)
[2] Fox, J. A., 1977, Hydraulic Analysis of Unsteady Flow
in Pipe Networks, Macmillan Press Ltd, London, UK;
Wiley, New York, USA.
[3] Chaudhry, M. H., 1987, Applied Hydraulic Tran-
sients (2nd edition), Van Nostrand Reinhold, New
York, USA.
[4] Wylie, E. B., and Streeter, V. L., 1993, Fluid Tran-
sients in Systems, Prentice Hall, Englewood Cliffs,
New Jersey, USA.
[5] Záruba, J., 1993, Water Hammer in Pipe-Line Systems
(2nd edition), Czechoslovak Academy of Sciences,
Prague, Czech Republic; Developments in Water
Science, 43, Elsevier Science Publishers, Amsterdam,
The Netherlands.
[6] Popescu, M., Arsenie, D., and Vlase, P., 2003, Applied
Hydraulic Transients for Hydropower Plants and
Pumping Stations, A.A. Balkema Publishers, Lisse,
The Netherlands.
[7] Thorley, A. R. D., 2004, Fluid Transients in Pipeline
Systems (2nd edition), Professional Engineering Pub-
lishing, London, UK.
[8] Ghidaoui, M. S., Zhao, M., McInnis, D. A., and Ax-
worthy, D. H., 2005, “A Review of Water Hammer
Theory and Practice,” ASME Applied Mechanics
Reviews, 58, 49-76.
[9] Bergant, A., Tijsseling, A.S., Vítkovský J. P., Covas D.,
Simpson A. R., and Lambert M. F., 2007, “Parame-
ters Affecting Water-Hammer Wave Attenuation,
Shape and Timing,” Accepted for Publication in
IAHR Journal of Hydraulic Research.
75
[10] Chaudhry, M. H., and Hussaini, M. Y., 1985, “Second-
Order Accurate Explicit Finite-Difference Schemes
for Waterhammer Analysis”, ASME Journal of Fluids
Engineering, 107, pp. 523-529.
[11] Arfaie, M., and Anderson, A., 1991, “Implicit Finite-
Differences for Unsteady Pipe Flow,” Mathematical
Engineering for Industry, 3, pp 133-151.
[12] Guinot, V., 2000, “Riemann Solvers for Water Ham-
mer Simulations by Godunov Method,” International
Journal for Numerical Methods in Engineering, 49,
pp. 851-870.
[13] Hwang Yao-Hsin, and Chung Nien-Mien, 2002, “A
Fast Godunov Method for the Water-Hammer Prob-
lem,” International Journal for Numerical Methods
in Fluids, 40, pp. 799-819.
[14] Jović, V., 1995, “Finite Elements and the Method of
Characteristics Applied to Water Hammer Modelling,”
Engineering Modelling, 8(3-4), pp. 51-58.
[15] Shu, J.-J., 2003, “A Finite Element Model and Elec-
tronic Analogue of Pipeline Pressure Transients with
Frequency-Dependent Friction,” ASME Journal of
Fluids Engineering, 125, pp. 194-199.
[16] Cheng Yong-guang, Zhang Shi-hua, and Chen Jian-
zhi, 1998, “Water Hammer Simulations by the Lat-
tice Boltzmann Method,” Transactions of the Chi-
nese Hydraulic Engineering Society, Journal of Hy-
draulic Engineering, 1998(6), pp. 25-31 (in Chinese).
[17] Thanapandi, P., 1992, “An Efficient Marching Algo-
rithm for Waterhammer Analysis by the Method of
Characteristics,” Acta Mechanica, 94, pp. 105-112.
[18] Wood, D. J., Dorsch, R. G., and Lightner, C., 1966,
“Wave-Plan Analysis of Unsteady Flow in Closed
Conduits,” ASCE Journal of the Hydraulics Division,
92, pp. 83-110.
[19] Wood, D.J., Lingireddy, S., Boulos, P.F., Karney, B.W.,
and McPherson, D. L., 2005, “Numerical Methods
for Modeling Transient Flow in Distribution Systems,”
Journal AWWA, 97(7), pp. 104-115.
[20] Jones, S. E., and Wood, D. J., 1988, “An Exact Solu-
tion of the Waterhammer Problem in a Single Pipeline
with Simulated Line Friction,” ASME - PVP, 140,
pp. 21-26.
[21] Sobey, R. J., 2004, “Analytical Solutions for Unsteady
Pipe Flow,” Journal of Hydroinformatics, 6, pp. 187-
207.
[22] Tijsseling A. S., 2003, “Exact Solution of Linear
Hyperbolic Four-Equation Systems in Axial Liquid-
Pipe Vibration,” Journal of Fluids and Structures, 18,
pp. 179-196.
[23] Bürmann, W., 1975, “Water Hammer in Coaxial Pipe
Systems,” ASCE Journal of the Hydraulics Division,
101, pp. 699-715. (Errata in 101, p. 1554.)
[24] Rij, A. A. van, 1970, “Standaard Afsluiterkarakteris-
tieken (Standard Valve Characteristics),” Delft Hydrau-
lics Laboratory, Report V 195, Delft, The Netherlands
(in Dutch).
[25] Schedelberger, J., 1975, “Closing Characteristics of
Spherical Valves,” 3R international, 14, pp. 333-339
(in German).
 76 Proceedings of the 2nd IAHR International Meeting of the Workgroup on Cavitation and Dynamic Problems in Hydraulic Machinery and Systems
[26] Bergant, A., and Simpson, A. R., 1991, “Quadratic-
Equation Inaccuracy for Water Hammer,” ASCE
Journal of Hydraulic Engineering, 117, pp. 1572-
1574.
[27] Bergeron, L., 1961, Waterhammer in Hydraulics and
Wave Surges in Electricity, John Wiley & Sons, New
York, USA.
[28] Bergant, A., Simpson, A.R., and Tijsseling, A.S.,
2006, “Water Hammer with Column Separation: A
View publication stats
Historical Review,” Journal of Fluids and Structures,
22, pp. 135-171.
[29] Mathcad 11, 2002, User's Guide, MathSoft Engineering
& Education, Inc., Cambridge, USA.
[30] Tijsseling, A. S, and Bergant, A., 2007, “Meshless
Computation of Water Hammer,” CASA Report 07-28,
ISSN 0926-4507, Department of Mathematics and
Computer Science, Eindhoven University of Tech-
nology, The Netherlands.