Professional Documents
Culture Documents
Finite Volume Method For Water Hammer Simulation: June 2002
Finite Volume Method For Water Hammer Simulation: June 2002
Finite Volume Method For Water Hammer Simulation: June 2002
net/publication/277938439
CITATIONS READS
2 652
1 author:
Michał Szydłowski
Gdansk University of Technology
44 PUBLICATIONS 132 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Michał Szydłowski on 09 June 2015.
Michał Szydłowski
160
TiASWiK’02 161
The original problem (2) neglecting the source term Applying the Roe idea of approximate Riemann
can be transformed to the following form problem solution the numerical flux at i+1/2 cell–
∂U ∂U interface can be expressed as
+A =0 (10) 1 1
∂t ∂x Fi +1 / 2 = (FL + FR ) − A ∆U (18)
where U is the same as in equation (2) and jacobian 2 2
A=∂F/∂U can be written as where subscripts L and R denote left and right side of
i+1/2 cell–interface, ∆U is an increment of pressure
a2 and mass discharge between cells i and i+1. Values of
0
A= A (11) pressure p and mass discharge q needed to calculate
A 2q
the elements of matrix A from equation (12) can be
ρA
evaluated as follows
In the approximate Roe solution of the Riemann p + pR q + qR
problem the equation system (10) is linearized by p= L , q= L (19a,b)
2 2
replacing the jacobian A by averaged matrix A . The If the conserved variables UR and UL necessary to
both forms of jacobian matrix must be diagonalizable estimate flux (18) are equal to cell–centre parameters
in the Roe approach. The above condition means that (Ui and Ui+1) then the numerical scheme is of first–
following equation must be satisfied order accuracy in space. In order to ensure the second–
A=RΛL (12) order accuracy approximation a function extrapolation
where Λ is a diagonal matrix containing the technique can be imposed. In the scheme presented
eigenvalues of matrix A, whereas R and L contain here function values are extrapolated from the cell
associated right and left eigenvectors. The eigenvalues centre–point to the cell–interface by MUSCL
λi of matrix A can be evaluated by solution of reconstruction technique (van Leer, 1979).
characteristic equation (Coulson and Jeffrey, 1982)
A − λI = 0 (13)
4. TIME INTEGRATION SCHEME AND
where I is the identity matrix. Considering jacobian
BOUNDARY CONDITIONS
matrix (11) the roots of (13) are equal to
λ1 = u − u 2 + a 2 (14a) In order to complete the solution of equation (9) a
numerical scheme of time integration must be
λ2 = u + u 2 + a 2 (14b)
implemented. For given computational cell i equation
where u=q/(ρA) is the flow velocity. The (9) can be written in another vector form
corresponding right (r) and left (l) eigenvectors can be ∂ Ui
calculated solving following equations (Coulson and = Xi (20)
Jeffrey, 1982) ∂t
Ari = λi ri (i = 1,2) (15a) where vector X containing all the terms of equation (9)
except for the time derivatives can be expressed as
l i A = λi l i (i = 1,2) (15b)
(F − Fi −1 / 2 ) (
Solution of equations (15a,b) for water–hammer X i = − i +1 / 2 − S i +1 / 2 + S i −1 / 2 ) (21)
problem yields following right and left eigenvectors, ∆xi
respectively where source terms at the both cell–interfaces can
1 be set equal to averaged values Si+1/2=(Si+Si+1/2)/2 and
Si–1/2=(Si–1+Si)/2, respectively. The solution of equation
r1 = A 2 2
(16a)
2 u − u + a (20) can be obtained by integration in the time
a increment 〈t, t+∆t〉. The general time integration
1 scheme, defining a new function values can be written
r2 = A 2 2
(16b) as follows
2 u + u + a t + ∆t
a Ui
n +1 n
= U i + ∫ X i dt (22)
A t
a2 2 u + u + a
2 2
where superscripts n and n+1 denote previous and next
l1 = a (17a)
2A u 2 + a 2 time level on the numerical grid. Scheme (22) can be
−1 realized in explicit or implicit way. In the present work
A 2
the two–step explicit scheme is applied
− 2 u − u + a
2
a2
l2 = a (17b) p n
U i = U i + 0.5∆tX i
n
(23a)
2A u2 + a2 1
n +1 n p
Ui = U i + ∆tX i (23b)
162 M. Szydłowski, Finite volume method for...
The above scheme is of second–order accuracy in time This kind of verification is possible only for simplified
and its stability is restricted by the local value of the water flow case when wave celerity can be assumed
Courant number (Potter, 1977) constant. In the second test the numerical solution is
examined against experimental data.
u i + u i2 + ai2
Cr = ≤1 (24)
∆ xi / ∆t Test–case no 1
where subscript i denotes concerning cell. The water hammer phenomenon is analysed in single,
horizontal, frictionless pipe of length L=500m and
A consistent set of initial and boundary condition is diameter D=0,1 m. The downstream valve closure
required to complement the water hammer equations experiment is simulated (Fig. 3).
solution. All conditions must be imposed in accordance
with characteristics theory (Godunov, 1975). To start
the computation the initial values of pressure and mass
p
discharge must be known inside every computational h0 =ρ . 0g
0
cell at time t=0. Analysing the characteristic curves of
equations (2) (Fig. 2) it is seen that the boundary D
0.0
0 1 2 3 4 5 6 7 8 9 10
5. NUMERICAL RESULTS t [s]
In order to verify the numerical solution method Fig. 4. Test–case no 1 – analytical () and numerical
presented in this paper two test–cases are considered. solution obtained using dissipative (+) and non–
In the first one computed results are compared to dissipative (•) scheme
analytical solution of unsteady water flow equations.
162
TiASWiK’02 163
The pressure variation in time at the valve section is steady state friction term produces results absolutely
shown in this picture. In the calculated results obtained different from physical ones (Fig. 5a,c).
using the scheme of first–order accuracy the numerical
diffusion is observed. The wave fronts are strongly a)
smeared and the maximum pressure value is decreasing p [MPa]
in time. If scheme of second–order accuracy is applied 1.0
a quite good agreement with analytical solution is
0.8
observed. The maximum values of pressure and wave
period are properly reproduced. The steep wave fronts 0.6
are slightly smoothed due to replacing in solution the 0.4
function discontinuities by narrow regions where
0.2
parameters are continuous. The better agreement could
be ensured by numerical mesh refining. 0.0
0 1 2 3 4 5 6 7 8 9 10 11 12
t [s]
Test–case no 2
b)
In this test case the computed results are compared to p [MPa]
measured data. The physical experiment was performed
at hydraulic laboratory of Environmental Engineering 1.0
164
TiASWiK’02 165