Вы находитесь на странице: 1из 17

COMMUNICATIONS IN NUMERICAL METHODS IN ENGINEERING Commun. Numer. Meth.

Engng 2005; 21:183199 Published online 24 December 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/cnm.741

Analysis of unsteady pipe ow using the modied nite element method


Romuald Szymkiewicz1; ; and Marek Mitosek2
1 Faculty 2 Faculty

of Civil and Environmental Engineering; Gdansk University of Technology; 80-952 Gdansk; Poland of Environmental Engineering; Warsaw University of Technology; 00- 653 Warsaw; Poland

SUMMARY A modied nite element method is proposed to solve the unsteady pipe ow equations. This approach yields a six-point implicit scheme with two weighting parameters. An accuracy analysis carried out using the modied equation approach showed that the proposed scheme has higher accuracy compared to other methods. A comparison of experimental data and the results of numerical solution showed that the required damping and smoothing of a pressure wave can be obtained when numerical di usion is produced by the applied method. It suggests that the physical dissipation process observed in the water hammer phenomenon is not represented properly in its mathematical model. Therefore, the classical system of water hammer equations seems to be incomplete. Copyright ? 2004 John Wiley & Sons, Ltd.
KEY WORDS:

pipe ow; water hammer; nite element method; numerical di usion

1. INTRODUCTION Taking into account the elasticity of pipe walls, the compressibility of liquid, the hydraulic losses due to pipe friction, assuming uniform distribution of pressure and velocity over the cross-section area and that the pipeline is always completely lled with water, Parmakian [1] proposed the following mathematical model of the water hammer phenomenon: @V @H f @V +V +g + V |V | = 0 @t @x @x 2D @H c2 @V @H +V + =0 @t @x g @x (1) (2)

to: R. Szymkiewicz, Faculty of Civil and Environmental Engineering, Gdansk University of Technology, 80-952 Gdansk, Poland. E-mail: rszym@pg.gda.pl

Correspondence

Copyright ? 2004 John Wiley & Sons, Ltd.

Received 1 December 2003 Accepted 2 November 2004

184

R. SZYMKIEWICZ AND M. MITOSEK

where x is the space co-ordinate, t the time, V the velocity in the pipe, H the piezometric head, f the friction factor, D the inside diameter of the pipe, g the acceleration of gravity, and c the velocity of pressure wave. The pressure wave velocity is expressed as follows: c= 1 D 1 + K Ee (3)

where is the density of the liquid, K the bulk modulus of elasticity of the liquid, E the modulus of elasticity of pipe-wall material, and e the thickness of pipe wall. Equation (1) is the momentum equation and Equation (2) the equation of continuity. They form a rst-order hyperbolic system of partial di erential equations, for which an initialboundary problem is formulated. It is solved in domain: 06x6L, t 0 (where L is the length of the pipeline). Proper initial and boundary conditions have to be imposed. Steady ow assumed at t = 0 enables to determine the functions V (x; t = 0) and H (x; t = 0) over the entire pipeline (06x6L). Since from each boundary, only one characteristic enters the domain of solution, at x = 0 and x = L one function, V (t ) or H (t ), has to be imposed. For given initial and boundary conditions this system is solved numerically. The water hammer equations can be solved by all methods suitable for hyperbolic equations. The method of characteristics (MOC) is most commonly used. It was proposed by Streeter and Lai [2] as well as by Evangelisti [3]. Its description was given by Almeida and Koelle [4], Goldberg and Wylie [5], Wylie and Streeter [6] as well. Usually MOC is applied for a xed grid. Consequently, this approach needs an interpolation in space or in time. To obtain a stable solution, the CFL condition has to be satised. For the Courant number equal to unity, MOC ensures an exact solution, whereas for other values it produces numerical di usion. The dissipation error is introduced by linear interpolation between the nodes used to calculate the values of function V and H at the points of intersections of characteristics. The numerical di usion gives rise to smoothing and damping of the pressure wave. One can show that in this case the coe cient of numerical di usion is equal to the one corresponding to the well-known pure advection equation solved by the nite di erence method using the up-wind scheme [7]. Therefore one should remember that the numerical di usion generated by MOC increases with the increase in distance step and with the decrease in Courant number. To improve MOC, Sibetheros et al. [8] made an attempt to apply an interpolation by the spline functions. Another approach to nd a suitable method to solve the water hammer equations was proposed by Chaudhry and Hussaini [9]. It concerned the nite-di erence schemes of second order of accuracy. Both mentioned approaches do not represent remarkable advantages. Consequently, MOC with linear interpolation in space or in time is still most commonly used to solve the water hammer equations. Taking into account the experience resulting from the solution of the hyperbolic equations by various numerical schemes, Fletcher [10] inferred the general guideline that rst-order formulae for derivatives should be avoided. A rst-order representation for the advective term in the governing equation will generate spatial derivatives of second order and higher in the modied equation, which is the equivalent governing equation actually solved. The introduction of spurious second or third derivatives can change the character of the solution signicantly. Similar recommendation is given by Leonard and Drummond [11].
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

185

As far as the MOC with linear interpolation in space is considered, for the Courant number less than unity, it represents an approximation of rst order of accuracy with regard to x as well as to t . The only way of achieving accurate solutions is to use small time step and a Courant number close to 1. This requirement can be fullled easily for a simple pipe only. When the pipes diameter varies, the wave celerity varies as well. In such a case, it is di cult to keep constant Courant number close to 1 and consequently this method will always produce numerical di usion. To avoid the aforementioned problems, the modied nite element method (FEM) can be applied. This approach proposed to solve the 1D unsteady ow and transport equations in open channel [12] can be successfully applied to solve the water hammer equations as well. The modication deals with the process of integration and leads to a more general form of algebraic equations approximating the governing equations. Its particular cases are the standard nite element method and the well-known nite di erence schemes. Usually the accuracy analysis is carried out in a classical way. It deals with the so-called coe cients of convergence, which characterize the amplitude and phase errors. However, these coe cients explain only general tendencies of the applied numerical method. As they are related to the wavelengths, information resulting from such an analysis cannot be used directly to assess the numerical errors generated while solving the di erential equations. For better assessment of the numerical errors the so-called modied equation approach seems to be very useful. This approach proposed by Warming and Hyett [13] is particularly suitable for the hyperbolic equations. Such analysis provides a large amount of information and suggests the way of increasing the solution accuracy. Proper choice of values of the two weighting parameters ensures third-order accuracy of the proposed method.

2. SOLUTION OF THE WATER HAMMER EQUATIONS BY THE MODIFIED FEM According to the Galerkin procedure, described in detail by Zienkiewicz [14], the solution of Equations (1) and (2) has to satisfy the following condition:
L 0

(fa ; : : :)N d x =

M 1 k =1

xk +1 xk

(fa ; : : :)N d x = 0

(4)

where is the symbolic representation of Equations (1) and (2), fa the approximation of any function f(x; t ) occurring in Equations (1) and (2), N(x) the vector of linear basis functions, L the length of pipeline, k the index of node, and M the number of grid points. In the standard approach the function f(x; t ) is approximated as follows: fa (x; t ) =
M k =1

fk (t )Nk (x)

(5)

where fk (t ) represents nodal value of function f(x; t ).


Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

186

R. SZYMKIEWICZ AND M. MITOSEK

When linear basis functions are applied in each element, only the two following integrals in Equation (4) exist I1 = I2 =
xk +1 xk xk +1 xk

fa (x; t )Nk (x) d x = fa (x; t )Nk +1 (x) d x =

1 2 fk (t ) + fk +1 (t ) 3 3 2 1 fk (t ) + fk +1 (t ) 3 3

xk 2 xk 2

(6a) (6b)

where xk is the distance between the nodes. In this case, the standard approach can be modied. Namely, the integral of the product of approximation of function and basis function in an element can be expressed as a product of certain average value of the function in the element and the integral of the basis function in this element. Therefore, I1 = I2 =
xk +1 xk xk +1 xk

fa (x; t )Nk (x) d x = fc (t )

xk +1 xk

Nk (x) d x = fc (t )

xk 2 xk 2

(7a) (7b)

fa (x; t )Nk +1 (x) d x = fc (t )

xk +1 xk

Nk +1 (x) d x = fc (t )

The weighted average fc (t ) in the considered element can be expressed using the following formula: for Equation (7a): for Equation (7b): fc (t ) = !fk (t ) + (1 !)fk +1 (t ) fc (t ) = (1 !)fk (t ) + !fk +1 (t ) (8a) (8b)

where ! is the weighting parameter ranging from 0 to 1. Consequently Equations (7a) and (7b) can be rewritten as follows: I1 = (!fk (t ) + (1 !)fk +1 (t )) I2 = ((1 !)fk (t ) + !fk +1 (t )) xk 2 xk 2 (9a) (9b)

This concept applied for open channel unsteady ow and transport equations [12] can be developed for the water hammer equations as well. Calculation of one integral in expression (4) over an element of length xk = xk +1 xk is made in the following way:
Xk +1 Xk

@Ha @Vc 1 @Va2 + +g @t 2 @x @x ! d Vk dV k +1 + (1 !) dt dt

f V |V | 2D

Nk d x
c

= +

1 xk 2 + (Vk2 + V k +1 ) 2 4 (10)

xk g (!fk Vk |Vk | + (1 !)fk +1 V (Hk + Hk +1 ) + k +1 |V k +1 |) 4D 2

Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM


Xk +1 Xk

187

@Hc @Ha c2 @Va + Vc + @t @x g @x ! d Hk d Hk +1 + (1 !) dt dt

Nk d x xk 2 (11)

c2 1 (!Vk + (1 !)V (Vk + V k +1 )(Hk + Hk +1 ) + k +1 ) 2 2g f V |V | 2D

Xk +1 Xk

@Ha @Vc 1 @Va2 + +g + @t 2 @x @x (1 !) dV d Vk k +1 +! dt dt

Nk +1 d x
c

1 xk 2 + (Vk2 + V k +1 ) 2 4 (12)

g xk (Hk + Hk +1 ) + ((1 !)fk Vk |Vk | + !fk +1 V k +1 |V k +1 |) 2 4D

Xk +1 Xk

@Hc @Ha c2 @Va + Vc + @t @x g @x (1 !) d Hk +1 d Hk +! dt dt

Nk +1 d x 1 xk + ((1 !)Vk + !V k +1 )(Hk + Hk +1 ) 2 2 (13)

= +

c2 (Vk + V k +1 ) 2g

In these expressions, the subscript a denotes approximation according to formula (5) while the subscript c denotes approximation by Equations (8a) and (8b). It means that the proposed modication concerns an alternative way of approximation of equations except for the terms containing the spatial derivatives. The derivatives of rst order with regard to x are approximated using the standard approach. When all integrals in each element are summarized according to Equation (4), the global system of ordinary di erential equations over time is obtained. It has the following form: S dX + CX = 0 dt (14)

where S is the constant matrix, symmetrical and banded, C the variable matrix, asymmetrical and banded, X = (V1 ; H1 ; V2 ; H2 ; : : : ; VM ; HM )T the vector of unknowns set up from nodal values T d V1 d H1 dV M d HM the vector of time derivatives, and T the symbol of of V and H , ddX t = dt ; dt ; : : : ; dt ; dt transposition.
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

188

R. SZYMKIEWICZ AND M. MITOSEK

The matrices S and C have dimensions of (2M ) (2M ) with bandwidth equal to 7. The following two-level method is applied to time integration: Xj+1 = Xj + t d Xj+1 d Xj + (1 ) dt dt (15)

where is the weighting parameter ranging from 0 to 1, of time level. It leads to the system of non-linear algebraic equations (S + t Cj+1 )Xj+1 = (S

t the time step, and j the index

t (1 )Cj )Xj

(16)

which has to be completed by imposed boundary conditions. They are as follows:


at the beginning of the pipe (k = 1), a constant piezometric head H1 = const is imposed, at the end of the pipe (k = M ), the velocity of out owing water is calculated depending

on the gate opening [2]; in this paper, it is assumed that the gate is closed immediately so VM (t ) = 0 for t0. The presented method involves two weighting parameters ! and . Taking ! = 2 3 , the FEM in its standard form is obtained, whereas for ! = 1, this method coincides with the nite di erence one. When = 1 2 , the method becomes the CrankNicolson scheme (equivalent to the implicit trapezoidal rule), whereas with = 1, it is the implicit Euler scheme (Figure 1). To solve the non-linear system, an iterative method, for example Newtons one, has to be used.

1 2

1 2

t j+1 (1 ) t t

tj x x

x k1

xk

x k+1

Figure 1. Numerical grid for the modied nite element method.


Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

189

3. STABILITY ANALYSIS AND NUMERICAL ERRORS GENERATED BY THE FEM Stability analysis is carried out for the simplied linear form of the governing equations which is as follows: @H @V +g =0 @t @x @H c2 @V + =0 @t g @x Approximation of this system on the grid points presented in Figure 1 with the following algebraic equations: (1 !) +
j +1 j j V V j+1 V V j+1 Vkj k 1 V k 1 k +1 + 2! k + (1 !) k +1 t t t

(17) (18) x = const yields

g g +1 j +1 Hkj Hkj1 + Hkj+1 = 0 1 + Hk +1 + (1 ) x x


+1 j Hkj H j+1 Hkj+1 H j+1 Hkj 1 Hk 1 + 2! k + (1 !) k +1 t t t

(19)

(1 !) +

c2 c2 j +1 j +1 j j V V k 1 + V k +1 + (1 ) k 1 + V k +1 = 0 g x g x

(20)

The stability analysis was carried out by the Neumann method [15]. The so-called amplication matrix, being a result of this approach, has the following eigenvalue: =1 with Cr tan R= m x 2 m x 2 (22) 2R 4R2 + i 4R2 2 + 1 4R2 2 + 1 (21)

1 + (2! 1)tan2

where is the eigenvalue of amplication matrix, i = (1)1=2 the imaginary unit, Cr = c t= x the Courant number and m the wave number. The condition of numerical stability [15]:
| |61

(23)
1= 2

in this case takes the following form:


| |=

2 1 2 + 1= (4R2 )

61

(24)

Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

190

R. SZYMKIEWICZ AND M. MITOSEK

1.5 = 0.4

I I

1.0

= 0.5 = 0.67 =1 = 0.5; C r = 1

0.5 0 2 4 6 8 10 N 12 14 16 18 20

Figure 2. Amplitude portrait for di erent .

It has to be satised for any wave number m. This takes place when
1 2

(25)

and ! 1 2 (26)

As the relation (24) is satised for any Courant number, the above relations ensure unconditional stability. The conclusions presented above can be illustrated by the graphs of the eigenvalue modulus of amplication matrix (24) called the amplitude portrait [7]. Since m x = 2 =N (N = number of computational intervals x per wavelength corresponding to examined component of Fourier series), the modulus of eigenvalue is a function of N : | | = f(N ). In Figures 2 and 3, one can notice that for Cr = 1 and = 1 2 , | | = 1. It means that the scheme does not change the , one obtains | |1. In this case the wave amplitude is amplied wave amplitude. For 1 2 and consequently the scheme becomes unstable, whereas for 1 2 , the amplitude is damped. Errors generated by any numerical method can be examined more exactly by the modied equation approach [10]. The solution of the hyperbolic equation has a form of waves that are dened by their amplitudes and phase celerities. The numerical methods applied to obtain this solution should not change the wave parameters. A method that changes the amplitude is called dissipative whereas a method changing the phase celerity is called dispersive. Both mentioned phenomena are caused by truncation error while approximating derivatives. To carry out an accuracy analysis using the modied equation approach all nodal values of functions V and H in the algebraic equations that approximate Equations (17) and (18), are replaced by Taylor series expansion around the node (k; j + 1) (Figure 1). Next, the obtained
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

191

1.0 C r = 0.5 II

Cr = 1 Cr = 2 = 0.5; = 0.67

0.5 0 2 4 6 8 10 N 12 14 16 18 20

Figure 3. Amplitude portrait for di erent Courant numbers.

relations are rearranged so that they contain only spatial derivatives. Finally the following system of equations is derived: @H @V +g = @t @x @H c2 @V + = @t g @x
n

@2 V + @x2 @2 H + @x2

@3 H + @x3 @3 V + @x3

(27) (28)

The approximation of Equations (17) and (18) by the FEM modies them to the system of di erential equations with an innite number of terms. It is well known that for smooth function the total value of the right-hand side is dominated by its rst term. It is known as well that in the modied equation, the even-order derivatives are associated with dissipation whereas the odd-order oneswith dispersion. Consequently, the type of error that dominates the numerical solution is determined by the rst term of the right-hand side of the modied equation. In Equations (27) and (28) vn is the coe cient of numerical di usion given by expression
n

1 2

c xCr

(29)

whereas

and

are the coe cients of numerical dispersion dened as follows:


n=

g x2 ((2 3!) (2 3 )Cr ) 6 c2 x 2 ((2 3!) (2 3 )Cr ) = 6g

(30) (31)

The modied equations can be used to demonstrate the consistency and the order of accuracy of the applied method and to deduce practically all information on its numerical properties. From relation (29) the following results:
FEM does not generate any numerical di usion for
Copyright ? 2004 John Wiley & Sons, Ltd.

=1 2 , i.e. when it represents an approximation of the second order with regard to t . It can be shown, that the terms of
Commun. Numer. Meth. Engng 2005; 21:183199

192

R. SZYMKIEWICZ AND M. MITOSEK

1 0.8 0.6 R 0.4 0.2 0 0 2 4 6 8 10 N 12 14 16 18 20 from (32) = 2/3 =1

Figure 4. Phase portrait for di erent !.

higher order than 2, which have been omitted here, disappear for Cr = 1. In this case the method ensures an exact solution, whereas for Cr = 1 unphysical oscillations can appear in the solution because of dispersivity ( n = 0 and n ; n = 0). FEM is unstable for 1 2 . In this case vn 0. It means that the initial-value problem for the system of Equations (27) and (28) is ill-posed and consequently its solution does not exist. FEM ensures a stable solution for 1 2 , because with vn 0 the initial-boundary problem for Equations (27) and (28) is well posed and its solution exists always. However it contains a dissipation error, because in this case the method produces a numerical di usion. It increases with the increase in x and Cr giving rise to smooth solution and reduction of the pressure gradients. The FEM is of second order of accuracy with regard to x as well as with regard to time it is of rst order of accuracy with regard to t . t for = 1 2 only. For another value of However, in this case, the order of accuracy can be increased. Namely, for = 1 2 and with != 2 Cr2 3 6 (32)

the terms of second and third order in the system of modied Equations (27) and (28) are cancelled. It means that in this case the proposed version of FEM ensures an accuracy of third order with regard to x as well as to t . This holds for Cr 61 only because of the condition of stability (26). The advantages of the proposed approach are illustrated in Figures 4 and 5 where the socalled phase portraits are presented. They are represented by the convergence coe cient R(N ) being the ratio of numerical and exact wave celerities. The diagrams presented in Figure 4 for various ! indicate that the presented modication of FEM signicantly improves its numerical properties. Namely, one can notice that for ! dened by Equation (32), the wave speed is a ected only for the shortest waves. There is remarkable di erence on comparing with ! = 2 3 (standard FEM) and ! = 1 (the nite di erence method). Consequently, the proposed method is able to ensure better accuracy of the solution especially when the strong gradients occur.
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

193

Cr = 1 1 0.8 0.6
R

C r = 0.75 C r = 0.25

0.4 0.2 0 0 2 4 6 8 10 N 12 14 16 18 20

Figure 5. Phase portrait for di erent Courant numbers.

90 80 70 60 H [m] 50 40 30 20 10 0 0 1 2 3 4 5 t [s] 6 7

x = 5 m, Cr = 1, = 0.5 x = 5 m, Cr = 2, = 1.0 x = 10 m, C r = 2, = 1.0

Figure 6. Computed head oscillations by the modied FEM for simple hypothetical pipeline at downstream end.

The remarks presented above were conrmed by numerical tests carried out for the hypothetical pipeline. In the pipeline of length L = 500m, the initial velocity was V0 = 0:4m= s. The piezometric head at reservoirs was Hr = 50 m = const, whereas the wave speed was 1000 m= s. The valve was closed immediately. The calculated pressures at the end of the pipeline with x are presented in Figure 6. For = 1 != 1 2 for di erent values of , Cr and 2 and Cr = 1, an exact solution of the linear water hammer equations was obtained. For increasing value of , Cr and x the function H (t ) becomes more and more smooth and simultaneously its amplitude is strongly damped. The only reason for this process is the numerical di usion generated by the applied method.
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

194

R. SZYMKIEWICZ AND M. MITOSEK

4. PHYSICAL MODEL AND EXPERIMENTAL DATA In order to obtain reliable experimental data, an installation for physical experiment was used. To eliminate the interaction of various factors in uencing the water hammer phenomenon, the simplest cases were studied. An experimental apparatus for investigating the water hammer events in the pipelines is shown in Figure 7. It is composed of a straight pipeline (1), pressurized tank (2) and a ball valve (4) mounted at the end of the pipe. The valve was closed manually. The valve closure time Tc , measured to a thousandth of a second, was from 0.018 to 0:025 s in all tests. The measurements and further analysis of pressure head characteristics H (t ) referred to a simple water hammer in which the wave re ection time (water hammer period) T was always greater than the value of closure time. The pressure was recorded by means of a measuring system consisting of strain gauges (5), extensometer amplier (6) and a computer (7) with AD= DA (20 MHz) card. The signal from the transducer was sampled with a frequency of 2000 Hz. The gauges, with measuring ranges up to 1.2 and 2 MPa (with measurement uncertainty 0:5%), had linear performance characteristics in the whole range of the measured pressure with a correlation coe cient not less than 0.999. The pressure transducers are located at two points along the pipeline (Figure 7). The pressure p(t ) was recorded during the process of the water hammer for di erent velocities V0 of steady water ow (varied from 0.3 to 2 m= s in the experiments). The velocity V0 in the pipeline was measured by the volumetric method (with measurement uncertainty 1%). The tested pipes (1) were fed from a reservoir (2), which itself was lled from the pipe network. The water pressure ps at the installation was constant (ps = 0:65 0:01 MPa). The pressure was reduced by means of a pressure-reducing valve (3). The stream pressure value p0 at the pipe outlet and the water ow velocity V0 were adjusted in order to avoid column separation events at the water hammer [1618]. After an initial steady state had been established, a transient event was initiated by a rapid valve closure. The experiments were carried out for various pipelines having various parameters as length, diameter, initial velocity, etc. In this paper, we would focus attention on the simplest case of the water hammer, namely in steel pipeline having constant diameter only. The characteristics

8 5 4

Vo

1 D, e L

Ps

Po
9

Figure 7. Experimental installation.


Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

195

Table I. Steel pipe characteristics.


Parameter Length L (m) Inside diameter D (m) Thickness of pipe wall e (m) Roughness height k (m) Waves velocity c (m= s) Time of closure Tc (s) Initial pressure head H0 (m) Initial velocity V0 (m= s) Value 72.0 0.042 0.003 0.00008 1245.0 0.021 51.0 0.408

120 100 80 H [m] 60 40 20 0 0 1 2 3 4 t [s] 5 6 7 8

Figure 8. Observed head oscillations for single steel pipeline at its downstream end.

of this pipeline are described in detail in Table I. The experiments carried out for numerous cases allowed us to record 10 sets of data representing the head oscillations H (t ) at the downstream end and at the mid-length of pipe. To check the obtained results and to eliminate the accidental errors, the experiments were repeated many times. The typical graph of H (t ) is presented in Figure 8 as an example of acquired data. For each experiment, a strong damping of pressure wave was observed. Very quickly, after several seconds, the amplitudes became insignicant and the hydrostatic state was reached. Simultaneously the wave front became smoother and consequently the function H (t ) lost its initially sharp form. 5. NUMERICAL TESTS AND DISCUSSION OF OBTAINED RESULTS The calculations of the water hammer phenomenon were carried out by the proposed method, with the friction factor given by ColebrookWhite formula. The results obtained for the set of data that minimized the numerical di usion ( x = 0:5 m, t = 0:00025 s, Cr = 0:62) are
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

196

R. SZYMKIEWICZ AND M. MITOSEK

presented in Figure 9(a). Taking into account the applied values of x and Cr , one can evaluate the strength of the introduced numerical di usion. To obtain the results presented in Figure 9(a), an insignicant numerical di usion n = 3:86 m2 = s was introduced. The goal of this introduction was to eliminate the unphysical oscillations resulting from non-linearity of the Equations (1) and (2). One can notice an essential disagreement with the observed pressure oscillations. From experimental results it is shown that the oscillations are relatively quickly damped, and that after 8 s they practically disappear. Conversely, in the results of computation, such strong damping is not observed. It is well known that for the friction factor given by the ColebrookWhite formula, the coincidence is observed at the beginning of the water hammer phenomenon, only for the rst cycle of pressure wave. Then the experimental data and calculations di er more and more with time. Generally, the observed damping of pressure wave amplitude is more intensive compared with the calculated one. This disagreement is considered as a result of an inadequate estimation of the shear stress for unsteady ow using the equation for the steady state [1922]. It is accepted commonly that the steady-state equation for friction is not able to reproduce accurately the shear stress during transient ow [23]. For this reason, the e ort of many authors is oriented to improve the evaluation of the shear stress for unsteady pipe ow. For example, an improvement of equation for friction has been proposed by Zielke [22], Vardy and Hwang [24], Brunone et al. [25], Pezzinga [21], Axworthy et al. [19], Pezzinga [21] and others as well. Usually, the proposed improvement deals with modication of the equation for friction. However, the algebraic term representing the friction process in Equation (1) is not able to smooth the oscillations. Therefore, it seems that an improvement of the water hammer equations cannot be focused on the modication of the friction term only. As long as the water hammer equations preserve the hyperbolic character, they will be unable to smooth the head oscillations. The numerical experiment conrms the remarks presented above. For the set of parameters that minimized the numerical errors, the friction factor f resulting from ColebrookWhite formula was multiplied by constant value K . Therefore, in Equation (1), a modied friction factor equal to K f was used. This approach being the simplest way of improvement of the friction factor is considered here only to show the results of such idea. Usually, the proposed improvements have more sophisticated form. Obviously, one can expect that the damping of head oscillations increases with increase in K . The results achieved with K = 10 are presented in Figure 9(b). As one can notice, the head oscillations are damped better but without any smoothing. Consequently, the calculated function H (t ) keeping its sharp form di ers from the observed one. The observed and computed pressure oscillations can be well matched by introducing numerical di usion. As it was shown earlier, the numerical di usion produced by FEM increases with the increase in x, Cr and . Very good agreement was obtained for ! = 0:525, = 0:850, x = 3:6 m and Cr = 0:865. For these data, the coe cient of numerical di usion is equal to vn = 1357 m2 = s. In Figure 9(c), the results of computations are compared with the ones provided by physical experiment. Note, that the numerical di usion introduced into the solution ensures the damping of amplitude of pressure wave and its smoothing simultaneously. One can add that the numerical di usion is able to ensure an excellent agreement even for the friction factor f = 0. These results allow us to deduce the nature of desired physical mechanism in the water hammer equations.
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

197

120 100 80

H [m]

60 40 20 0 0 1 2 3 4 5 6 7 8

(a)

t [s]

120 100 80

H [m]

60 40 20 0 0 1 2 3 4 5 6 7 8

(b)

t [s]

120 100 80

H [m]

60 40 20 0 0 1 2 3 4 5 6 7 8

(c)

t [s]

Figure 9. Computed and observed head oscillations for single steel pipeline at downstream end: (a) for friction factor f by ColebrookWhite formula; (b) for friction factor f multiplied by K = 10; (c) for friction factor f by ColebrookWhite formula with numerical di usion.

Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

198

R. SZYMKIEWICZ AND M. MITOSEK

It seems that, apart from friction expressed in Equation (1) by algebraic terms another factor takes part in the process of wave damping as well. As it was shown the friction term is able to reduce the wave amplitude only. An excellent adjustment ensured by numerical dissipation suggests that the lacking process should be expressed by the di usive term that is able to ensure such smoothing. As far as an improvement of the friction factor f is considered it should be remembered that the required damping can be achieved by an interaction of the numerical di usion and the shear stress. Therefore, it is possible to obtain a satisfying agreement between the experimental data and the calculations for their various combinations. Since the quality of obtained solution is determined by the physical process of friction as well as by the numerical error, it is of essential importance to separate both e ects. For this reason, to evaluate the real meaning of any introduced improvement of the equation for the shear stress, the numerical errors should be evaluated. A satisfying agreement of the experimental data and calculations cannot be considered as a su cient proof as long as the numerical and physical e ects in the improved solution are not separated. In the other case, the conclusions may be misleading. This seems to be a very important problem while analysing unsteady pipe ow.

6. CONCLUSIONS The nite element method in the version obtained by modifying the standard approach applied in the Galerkin procedure was proposed to solve the unsteady pipe ow equations. Finally a six-point implicit scheme for xed mesh with two weighting parameters was obtained. It ensures an approximation of third order with regard to x and to t for Cr 61. Consequently the numerical solution is more accurate compared with the one given by standard version of the nite element and di erence method. The results of solution of the water hammer equations by the proposed method compared with the experimental data showed an essential in uence of the numerical factors such as the time step t , the space interval x and the weighting parameters on the quality of the results. The nature of this in uence was explained by the accuracy analysis carried out using the modied equations approach. From the point of view of the theory of the numerical methods it seems to be a paradox that the highly accurate methods give the results that disagree with measurements. Conversely, large numerical di usion produced by the methods, being a poor approximation, ensures satisfying agreement. Since in the water hammer process, an important role is played by the mechanism of physical dissipation, which is not represented in Equations (1) and (2), therefore, in practice, to obtain a satisfying adjustment, the lacking e ect of this mechanism is reproduced by numerical dissipation. It seems the water hammer equations should contain another additional mechanism of physical dissipation, which would be able to ensure both e ects simultaneouslydamping and smoothing of the pressure wave.
REFERENCES 1. Parmakian J. Water Hammer Analysis. Prentice-Hall, Inc: New York, 1955. 2. Streeter VL, Lai Ch. Water hammer analysis including uid friction. Journal of Hydraulics Division (ASCE) 1962; 88(HY3):79 111. 3. Evangelisti G. Water hammer analysis by the method of characteristics. LEnergia Elettrica 1969; 46(10): 673 692.
Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

ANALYSIS OF UNSTEADY PIPE FLOW USING MODIFIED FEM

199

4. Almeida AB, Koelle E. Fluid Transients in Pipe Networks. Elsevier Applied Science: London, 1992. 5. Goldberg DE, Wylie EB. Characteristics method using time-line interpolations. Journal of Hydraulic Engineering (ASCE) 1983; 109(5):670 683. 6. Wylie EB, Streeter VL. Fluid Transient in Systems. Prentice-Hall, Inc.: Englewood Cli s, New York, 1993. 7. Abbott MB, Basco DR. Computational Fluid Dynamics. Longman Scientic and Technical: Essex, UK, 1989. 8. Sibetheros IA, Holley ER, Branski JM. Spline interpolation for water hammer analysis. Journal of Hydraulic Engineering (ASCE) 1991; 117(10):1332 1351. 9. Chaudhry MH, Hussaini MY. Second order accurate explicit nite-di erence schemes for water hammer analysis. Journal of Fluid Engineering 1985; 107(4):523 529. 10. Fletcher CA. Computational Techniques for Fluid Mechanics, vol. 1. Springer: Berlin, Germany, 1991. 11. Leonard BP, Drummond JE. Why you should not use hybrid, power-law or related exponential schemes for convective modellingthere are much better alternatives. International Journal for Numerical Methods in Fluids 1995; 20:421 442. 12. Szymkiewicz R. Method to solve 1D unsteady transport and ow equations. Journal of Hydraulics Engineering (ASCE) 1995; 121(5):396 403. 13. Warming RF, Hyett BJ. The modied equation approach to the stability and accuracy analysis of nite-di erence methods. Journal of Computational Physics 1974; 14:159 179. 14. Zienkiewicz OC. The Finite Element Method in Engineering Science. McGraw-Hill: London, 1972. 15. Potter D. Computational Physics. Wiley: London, 1973. 16. Bergant A, Simpson AR. Pipeline column separation ow regime. Journal of Hydraulic Engineering (ASCE) 1999; 125(8):835 848. 17. Mitosek M. Study of cavitation due to water hammer in plastic pipes. Plastics, Rubber Composites Processing and Application 1997; 26(7):324 329. 18. Mitosek M. Study of transient vapour cavitation in series pipe systems. Journal of Hydraulic Engineering (ASCE) 2000; 126(12):904 911. 19. Axworthy DH, Ghidaoui MS, McInnis DA. Extended thermodynamics derivation of energy dissipation in unsteady pipe ow. Journal of Hydraulic Engineering (ASCE) 2000; 12(4):276 287. 20. Pezzinga G. Quasi-2D model for unsteady ow in pipe networks. Journal of Hydraulic Engineering (ASCE) 1999; 125(7):676 685. 21. Pezzinga G. Evaluation of unsteady ow resistances by quasi2D or 1D models. Journal of Hydraulic Engineering (ASCE) 2000; 126(10):778 785. 22. Zielke W. Frequency dependent friction in transient pipe ow. Journal of Basic Engineering (ASME) 1968; 90(1):109 115. 23. Elansary AS, Silva W, Chaudhry MH. Numerical and experimental investigation of transient pipe ow. Journal of Hydraulic Research 1994; 32(5):689 706. 24. Vardy AE, Hwang K. A characteristic model of transient friction. Journal of Hydraulic Research 1991; 229(5):669 684. 25. Brunone B, Golia UM, Greco M. Some remarks on the momentum equation for fast transients. Proceedings of the International Meeting on Hydraulic Transients and Water Column Separation. 9th Round Table, IAHR: Valencia, Spain, 1991; pp. 140148.

Copyright ? 2004 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2005; 21:183199

Вам также может понравиться