coupling conditions for gas networks governed by the isothermal euler equations

8/10/2019 COUPLING CONDITIONS FOR GAS NETWORKS GOVERNED BY THE ISOTHERMAL EULER EQUATIONS

1/20

Manuscript submitted to Website: http://AIMsciences.orgAIMS JournalsVolume X, Number 0X, XX 200X pp. XXX

COUPLING CONDITIONS FOR GAS NETWORKS GOVERNED

BY THE ISOTHERMAL EULER EQUATIONS

Mapundi K. Banda

University of KwaZulu-NatalSchool of Mathematical Sciences

Private Bag X013209 Pietermaritzburg, South-Africa

Michael Herty

Technische Universitat KaiserslauternFachbereich Mathematik

Postfach 3049D-67653 Kaiserslautern, Germany

Axel Klar

Technische Universitat KaiserslauternFachbereich Mathematik

Postfach 3049D-67653 Kaiserslautern, Germany

(Communicated by Aim Sciences)

Abstract. We investigate coupling conditions for gas transport in networkswhere the governing equations are the isothermal Euler equations. We discuss

intersections of pipes by considering solutions to Riemann problems. We in-troduce additional assumptions to obtain a solution near the intersection and

we present numerical results for sample networks.

1. Introduction. There has been intense research on gas networks in the lastdecade. Different models for transient flow have been proposed and refer to [21, 4]and the publications of Pipeline Simulation Interest Group [20] for an overview ofseveral approaches. Numerical methods have been proposed [26, 12] for simulatinggas networks and optimization problems have been investigated [8, 18, 22, 25].Today a variety of models and optimization problems related to gas transport innetworks exist.

Our starting point for describing transient gas flow are the isothermal Euler equa-tions. Using the properties of solutions to Riemann problems for those equations,we derive coupling conditions for an intersection of pipes with variable tempera-tures. The coupling of the pipes is such that the arising waves can change theirdirections. This is a major difference to most of the models proposed in the above

2000 Mathematics Subject Classification. Primary: 76N15; Secondary: , 35Lxx.

Key words and phrases. Isothermal Euler equation, Networks, Gas Flow.This work was supported by the University of Kaiserslautern Excellence Cluster Dependable

Adaptive Systems and Mathematical Modelling and the University of KwaZulu-Natal competitiveresearch grant (2005) Flow Models with Source Terms, Scientific Computing, Optimization inApplications.

1


2/20

2 MAPUNDI K. BANDA, MICHAEL HERTY, AXEL KLAR

references. The Ansatz is similar to the work done in [9, 14] for deriving couplingconditions of a hyperbolic system for traffic flow.

The present paper is a companion paper to [3]. In [3] a network of pipelines withconstant temperatures (or to be more precise, equal sound speeds on all consideredpipes) and coupling conditions based on the equality of pressure at the junctionshas been investigated. In Section 4 of the present paper several theorems used in[3] are proven using a demand and supply analysis of the situation at the junctions.In Section 5 different coupling conditions are introduced. Imposing additional as-sumptions on the distribution of the gas flow and the structure of the arising waves(equality of pressure at the junction or subsonic flow, respectively) we point out theconstruction of a solution near the intersection for pipes having different tempera-tures. Restricting ourselves to the case of a simple intersection of pipes with oneingoing and one outgoing pipe we prove the existence of solutions. For the case ofgeneral junctions, i.e. tee fittings, with multiple ingoing and outgoing pipelines butconstant temperature for all pipes we refer to [3]. Finally, we describe in Section

6 a numerical method used to solve the isothermal Euler equations on the networkbefore we present examples on sample problems.

2. The Isothermal Euler equations. The isothermal Euler equations are a sim-plification of the Euler equations. They are obtained from the isentropic equationswhich in turn are derived from the Euler equations under the assumption that theenergy equation is redundant. A further assumption for pipe networks is that thereis negligible wall expansion or contraction under pressure loads; i.e. pipes haveconstant cross-sectional area. In the isothermal Euler equations the temperature isconstant. The equation of state applied here is

p=

Z

RT

Mg , (1)

wherep is the pressure, Zis the natural gas compressibility factor,R the universalgas constant, T the absolute gas temperature, Mg is the gas molecular weight and the density. For a simplified presentation we finally assume that all pipes havethe same diameter D.

We consider the isothermal equations in conservative form with a constant a2 =ZRT /Mg, i.e.,

t+x(u) = 0 (2a)

t(u) +x(u2 +a2) = 0 (2b)

Here, u denotes the velocity of a gas and u is the flux. The first equation is theconservation of mass and the second states the conservation of momentum.

We briefly recollect the main properties of Riemann problems for the system (2).For a general reference on the solution of hyperbolic systems we refer to [7, 10] andfor an application to the isothermal Euler equations to [17]. The understandingof Riemann problems is of importance for the coupling conditions of several pipeslater on. We will use the following notation throughout the remaining sections,

q:= u, U :=

q

, F(U) :=

q

q2/+a2

. (3)


3/20

COUPLING CONDITIONS FOR GAS NETWORKS 3

A Riemann problem for (2) is an initial value problem for (2) with Heavi-side initialdata, i.e.,

tU+ xF(U) = 0, U(x, 0) =

Ul x 0Ur x > 0

(4)

where Ul and Ur are given constants. It is wellknown that the isothermal Eulerequations enjoy the following properties: The eigenvalues are

1(U) = q/ a and 2(U) = q/ +a. (5)Both fields are genuinely nonlinear, see [17]. A point Ucan be connected to Ul bya 1-(Lax-)shock, if and only if there exists such that

U=Ul+l

1

s1(; Ul)

and 0. (6)

It can be connected to Ul by a 2-(Lax-)shock, if and only if there existssuch that

U=Ul+l

1

s2(; Ul)

and [1, 0]. (7)

The shock speeds s1 and s2 are given by

s1,2(; Ul) = ql/l a

1 +. (8)

The set of all points U that can be connected to Ul by a 1-(Lax-)shock (resp. 2-(Lax-)shock) is said to be the 1-(Lax-)shock curve (resp. 2-(Lax-)shock curve) andit is parameterized by .

Similarly, the set of all points that can be connected to Ul by 1-rarefaction (resp.2-rarefaction) wave is said to be the 1-rarefaction (resp. 2-rarefaction) curve. Aparametrization of the 1-rarefaction (resp. 2-rarefaction) curve is given by

U() = lexp(( 1)/a)

1+a

, 1 := 1(Ul) (9)

and

U() = lexp(( 2)/a)

1 a

, 2 := 2(Ul), (10)

respectively. We refer to Figure 1 for an example of the shape of all curves in theq plane and to [17] for more details.

For the discussion later on, we also state a parametrization of the 1,2-(Lax-)shockcurves and 1,2-rarefaction curves for a given right state Ur,i.e., a description of thepossible points Ul that can be connected to Ur.

1-s. U() = Ur+r 1

s1(; Ur)

,1 0. (11a)2-s. U() = Ur+r

1

s2(; Ur)

, 0. (11b)

1-r. U() = rexp(( 1)/a)

1+a

, 1 := 1(Ur). (11c)

2-r. U() = rexp(( 2)/a)

1 a

, 2 := 2(Ur). (11d)


4/20


0 1 2 3 4 5 6 7 8 9 102

0

2

4

6

8

10

12

14

ul

q

1S

1R

2S2R

Figure 1. 1- and 2-curves in the q plane for given Ul anda = 1.

Remark 2.1. The shock speedsi for two given statesUl, Ur on the same i-(Lax)-shock curve is given by the slope of the line connecting Ul and Ur. Further, theslope of the 1- and 2-rarefaction curves at a pointUis equal to the first and second

eigenvalue at this pointU, respectively. For the isothermal Euler equations, the 1-(Lax-) shock and 1-rarefaction curves do not coincide. They are connected smoothlyup to the second derivative.

3. Pipeline Networks. Before we discuss a network of pipes, we introduce twomain assumptions, that simplify the discussion following.

A0 There are no vacuum states present, i.e., >0. (12a)

A1 The direction of flow does not change, i.e., u 0. (12b)

At the appropriate places we will point out additional difficulties that arise, if theassumptions (12b) is relaxed.

Due to assumption (12b) we model a network of pipes as a directed, finite graph(J, V) and in addition we may connect edges tending to infinity. Each edge j Jcorresponds to a pipe. Each pipe j is modelled by an interval [xaj , x

bj ]. For edges

ingoing or outgoing to the network we have xaj = or xbj = +, respectively.Each vertex v Vcorrespond to an intersection of pipes. For a fixed vertex v Vwe denote byv (

+v) the set of all indices of edges j J ingoing (outgoing) to the

vertex v .


5/20


On each edge j Jwe assume that the dynamics is governed by the isothermalEuler equations: for all x [xaj , xbj ] and t 0,

t

jqj

+x

qj

q2j /j + a2jj

= 0, (13a)

Uj(x, 0) :=

jqj

(x, 0) =

0jq0j

(x), (13b)

whereaj is a constant depending on the pipe and U0j = (

0j , q

0j )

T is the initial data.At each vertex v V we have to couple systems of the type (13) by suitable

coupling conditions. For given constant initial data we construct a solution (Uj)j ata single vertex by solving (half-)Riemann problems, see below. The main idea is toconstruct wave fans consisting of waves of negative (positive) speeds on the ingoing(outgoing) pipes of the intersection. This is done by introducing intermediatestates at the vertex, similar to the constructions given in [9, 13, 14]. We have one

intermediate state for each connecting pipe and those states have to satisfy thecoupling conditions at the vertex.To be more precise: Consider a single vertexv with ingoing pipes j v and

outgoing pipes j +v. Assume constant initial data U0j given on each pipe. Afamily of function (Uj)jv +v is called a solution at the vertex, provided that Ujis a weak entropic solution on the pipe j and (for Uj sufficiently regular)

jv

qj(xbj, t) =

j+v

qj(xaj+, t) t > 0. (14)

This condition resembles Kirchoffs law and is referred to as Rankine-Hugoniotcondition at the vertex. It states the conservation of mass at the intersection:

A2 At each vertex the total mass is conserved. (15)

Remark 3.1. It turns out that (A0)-(A2) is not sufficient to obtain a unique solu-tion(Uj)j at the vertex. Further conditions need to be imposed, compare also withthe discussion in [9, 13] for coupling conditions of different hyperbolic system.

4. The (half-)Riemann problems. Before introducing additional conditions toobtain solutions at a vertex, we discuss special Riemann problems at the junctions:At a fixed junction v Vwe consider the (half-)Riemann problems for a pipe j ,

tUj+ xFj(Uj) = 0, (16a)

Uj(x, 0) =

U0j x < x

bj

Uj x xbjj v , (16b)

Uj(x, 0) = Uj x > xajU0j x x

aj

j

+

v

, (16c)

with given constant initial data Uj,0 and Fj(Uj) is defined as in (3).We determine all possible states Uj , j v , (resp. j +v) such that the solution

to the Riemann problem (16a,16b) (resp. (16a,16c)) have negative (resp. positive)speed. By the same reasoning as in [14] we neglect solutions of zero speed (stationaryshocks).

Proposition 4.1. Given constant initial dataU0j =:Ul withl > 0 andql 0 onan ingoing pipej v.


6/20


Then the (half-)Riemann problem (16a,16b) withUj =:Ur such thatqr 0, r >0, admits either the constant solutionUr Ul or solutions with negative wave speed,if and only if

Ur belongs either to the 1-(Lax)-shock curve (6) or (17a)

to the 1-rarefaction curve (9) throughUl (17b)

and eitherlmax{1, q2l

2l a2j

} r (17c)

orlmin{1, exp( qlla

1)} r l. (17d)Remark 4.2. Due to assumptions (A0) and (A1), the assertion of the previousproposition exclude states Ur which consist either of the vacuum state = 0 orstates Ur with negative speed u = q/ < 0. (Half-)Riemann problems with suchstates are not solved and in particular the previous proposition does not give anyinformation on those problems.

Moreover, there exists points Ur satisfying (17d), only if Ul is subsonic, i.e.,ql/l aj . If Ur satisfies (17d) it is connected to Ul by a 1-rarefaction wave. Ifequation (17c) is satisfied then Ur is connected to Ul by a 1-(Lax-)shock. Theequations (17) allow an interpretation in terms of demand functions, see below andin [13, 16].

Proof. Given a state Ul with ql 0, l > 0. We discuss the possible states Urwithr >0, qr 0 that can be connected by a 1-wave. AssumeUr is connected bya 1-(Lax-)shock toUl.Then there exists a 0 such that the shock speed is givenbys1(; Ul) = ql/l aj

1 +. Hence, Ur can be connected by a 1-(Lax-)shock of

negative speed, if and only if

1la2

j

q2ll

la2j and 0.Rephrasing this conditions in terms ofr() := ql + l we obtain (17c). AssumeUris connected by a 1-rarefaction wave to Ul.Then the wave speed is 1(U) = q/ ajand 1 is increasing along the 1-rarefaction curve in the q plane. Hence, ifql/l aj ,then all pointsUr are connected toUlby a 1-rarefaction wave of positivespeed. Ifql/l < aj, then there exists a point U such that1(U) = 0 and U andUl can be connected by a 1-rarefaction wave. We have = lexp(ql/(ajl) 1)and this yields (17d).

Now assume, Ul is connected to a state Um by a 1-wave of negative speed. Weconsider states Ur that can be reached from Um by a 2-wave of negative speed. A2-rarefaction wave connecting any point Um= Ur to a point Ur with qr 0 andr > 0 has always positive speed. Hence, there are no 2-rarefaction waves in thesolution to (16a, 16b). Further, if any pointUm

= Ur is connected to Ur by a 2-

(Lax-)shock, then there exists [1, 0], such that Ur = qm/m+ m

1s2

. The

shock speed iss2(, Um) = qm/m+ aj

1 +. Hence, ifqm 0,thens2 0. IfUmis such thatqm< 0, then the 2-(Lax-)shock curve lies in the fourth quadrant of theq plane: Indeed, the curve qm+ ms2(; Um) has no zero for 1< 0.Therefore,Um cannot be connected to a point Ur (qr 0, r >0) by a 2-curve ofnegative speed.

For the subsequent discussion we introduce the demand function d(; Ul, aj)depending on a given (left) initial datum Ul and the constant of the pipe aj : for a


7/20


8/20


1-(Lax-)shock curve at = l. This implies that there exists Ur such that qr =q

for q d(l; Ul, aj). If qr q, then Ur and Ul are connected by a 1-rarefactionwave and else by a 1-shock wave.

If we skip assumption (12b), we obtain further states Ur in the fourth quadrantof the q plane, which can be connected to Ul by waves of negative speed. Wegive an example of the area of the possible states in Figure 3: For the given Ul thestates Ur lie either on the 1-(Lax-)shock curve through Ul (1 S) or in either ofthe following areas. The area A1 is bounded from above by the curve{(, q)}and to the right by the curve 1 S. States Ur in A1 can be connected to Ul by a2-(Lax-)shock and a 1-(Lax-)shock both of negative speed. The areaA2 is boundedto the left by 1 Sand to the right by the line q= aj; points in this area can beconnected toUl by a 2-rarefaction and a 1-(Lax-)shock. Note that the intersection

point U of 1 S and q = aj satisfies 2(U) = 0. Further, an expression of theareasAi in terms of a demand function (c.f. Proposition 4.3) is impossible.

0 5 10 15 20 25 30 35 4040

35

30

25

20

15

10

5

0

5

q

xUl

1S

(,q)

q = aj

2S

2R

Figure 3. Plot of the boundaries of the areas for states Ur thatcan be connected to Ul by waves of negative speed.

Next, we consider an outgoing pipe j

+v and the Riemann problem (16a, 16c)

with given right initial data U0j =:Ur. We are looking for all states Uj =:Ul thatcan be connected to Ur by waves of positive speed. A first result uses the notionof the supply function and describes all statesUl that can be connected by 1-wavesonly.

Proposition 4.4. Let j +v. Given a constant initial dataU0j =: Ur with qr0, r > 0. Denote bys(; Ur, aj) the supply function through the stateUr.

Then,

0 q s(r; Ur, aj), (20)


9/20


there exists a unique stateUj := Ul such that ql = q, l > 0 and such that the

solution to (16a, 16c) is either constant or consists of a 1-(Lax-)shock or a 1-

rarefaction wave of positive speed.Proof. Assume q = qr. By the discussion in the previous section, the states

Ul that can be connected to Ur by a 1-(Lax-)shock belong to the curve (11a).

Therefore, the states Ul = U()

()q()

, which can be connected to Ur by a

1-(Lax-)shock of positive speed have to satisfy

() [0, rmin{1, q2r

2ra2}], resp. q() [0, qr]. (21)

Further, the states Ul that can be connected to Ur by a 1-rarefaction belong to the

curve (11c) or equivalently by{(, q())} with q() = qrr

ajlog

r

. Hence,

a state U = (, q()) can be connected to Ur by a 1-rarefaction wave of positivespeed, if and only ifqr < raj . Then the flux q() satisfies

qr < q() q(), (22)where =rexp

qrraj

1

,i.e., dd

q() = 0.Note that in this cases(r; Ur, aj) =

q(). Combining (21) and (22) we obtain: Ifs(r; Ur, aj) qr, then there exists aunique state U() =: Ul such that q = ql connected to Ur by a 1-(Lax-)shock. Ifs(r; Ur, aj)> qr, then there exists a unique state Ul with q =ql connected to Urby a 1-rarefaction wave.

The next result completes the discussion of the possible states Ul that can beconnected to Ur such that the solution to (16a, 16c) is a juxtaposition of 1-wavesand/or 2-waves of positive speed.

Proposition 4.5. Given constant initial dataU0j =:Ur withr > 0 andqr

0 on

an outgoing pipej +v.Then the (half-)Riemann problem (16a, 16c) withUj =: Ul andl > 0, ql 0

admits either the constant solution Ul Ur or the solution is a juxtaposition ofwaves of positive speed provided that

0 ql s(m; Um, aj) (23)for an arbitrary stateUm (m, qm) with the properties

m < and (24a)Um is either on the 2-rarefaction or (24b)

on the 2-(Lax-)shock curve throughUr, (24c)

where is the zero of the function

q() := qr

r+aj log

r

(25)

in the interval[rexp

qr

raj 1

, r].

Proof. According to (11b), a left state Um can be connected toUr by a 2-(Lax-)shock of positive speed, ifm > r. Further, the 2-rarefaction curve (11d) in theqplane can be equivalently rewritten as {(, q) : q= q()} whereq() is definedin (25). Since the speed of the 2-rarefaction waves generated is equal to the slope


10/20


d

dq(), then Um can be connected to Ur by a 2-rarefaction wave of positive wave

speed, if and only ifr > m rexp

qr

raj 1

.

Note that a stateUmwithqm < 0 cannot be connected to a state Ulwithql 0 bya 1-(Lax-)shock wave of positive speed. Indeed, assume the contrary, then the shock

speed is given by ql qml m < 0, sincel< m. Further, a 1-rarefaction connecting

Um, qm < 0 andUl, ql 0 consists of waves with speed 1(U) = q/ a < 0.Hence,we obtain the assertion on the possible states Um. For a fixed state Um

we can connect any left state Ul by a 1-wave of positive speed, provided that qls(m; Um; aj) (see Proposition 4.4).

Remark 4.6. As in Remark 4.2 Riemann problems are not solved, if there is vac-

uum present or if intermediate states with negative velocityu = q/ appear.To obtain a suitable solution to a network problem we need to reduce the rangeof possible statesUl. We already imposed the assumptions (A0)-(A2), but we see,that further conditions have to be imposed. The exact formulation of additionalcoupling conditions is subject to modelling and we present a possible choice in thenext section.

5. Coupling Conditions for Pipe Intersections. We present different couplingconditions for pipe to pipe intersections. We consider an intersection of two pipes.The pipes might have different sound speeds, i.e. aj=aj , but as a simplificationthey are assumed to have the same diameter D. Note that this situation can alsobe seen as a system of conservation laws having a discontinuous flux function. For adiscussion of pipe intersections like tees in the more simpler situation of pipes with

the same sound speed, we refer to the forthcoming paper [3].We introduce a further condition to obtain a unique solution, namely (A3),

A3 qj/j ajt > 0, j +v, (26)or (A3),

A3 The pressure at the vertex is a constant, i.e.,a2i i = a2jj , (27)

respectively. Furthermore, either of the above conditions need to be complementedby condition (A4) to obtain a unique solution. This assumption is as in [14, 13, 9]and resembles an entropy condition.

A4 The flux is maximal at the intersection. (28)

We discuss the modelling and theoretical properties of solutions at an intersection

satisfying (A0)-(A2),(A3) and (A4) in Section 5.1 and solutions satisfying (A0)-(A2),(A3) and (A4) in Section 5.2, respectively.

Some remarks are in order. In realworld applications the modelling of intersec-tion is more complex and additional effects are taken into account. In the engineer-ing literature various tables can be found which depending on the geometry andmaterial properties of the intersection describe an approximation of the behaviorfor coupled pipes. Condition (A3) is a rather simple model which is widely usedin the engineering community to model pipetopipe intersections. We call solu-tions satisfying (A3) therefore physical solutions. Moreover, solutions satisfying


11/20


(A0)-(A2),(A3) and (A4) have further interesting mathematical properties, see Sec-tion 5.2. But as seen later, condition (A3) cannotbe evaluated analytically for any

initial data and any sound speeds. Therefore, we propose assumption (A3) replac-ing (A3). Then, we can analytically compute solutions satisfying (A0)-(A2),(A3)and (A4). But condition (A3) can yield non-physical solutions and especially, inthe case of only two connected pipes having the same sound speed, (A3) yields thephysical solution only for particular choices of initial data. We present numericalresults showing the difference in the solutions to the different coupling conditionsin Section 6.

5.1. Coupling conditions I. The first assumption which simplifies the theoreticaldiscussion below is as follows. We restrict ourselves to subsonic solutions on theoutgoing pipe, c.f. [24]. This implies to require a solution to satisfy in particular(A3):

qj/j

aj

t > 0, j

+v (29)

Using the assumptions (A0)-(A2), (A3) and (A4), we can construct a solutionto the pipetopipe fitting using the notion of (half-)Riemann problems and sup-ply/demand functions introduced in the previous sections: Consider a single in-tersection v with an ingoing pipe j = 1 v and an outgoing pipe j = 2 +vconnected at xb1 = x

a2 . Assume constant initial data U

0j. We determine states Uj ,

such that the solution to (16a,16b) and (16a,16c), respectively, consists of waves ofnonpositive and nonnegative speed, only.

We start with a discussion for the outgoing pipe j = 2 and construct the sonicpoint Um first. Let Ur := U

02 and for the case qr = a2r, we set Um := Ur.

Assume qr < a2r. Then, in the q plane, there is a unique intersection pointUm = (m, qm) of the line{(, q) : q= a2} and the 2-(Lax-)shock curve throughUr given by (11b). Um is given by m = r+mr, qm= a2m where

m =1

2

1 qr

ra2

1 qra2r

+

qra2r

2

+ 4

.In the case qr > a2r, there exists a unique intersection point Um of the line{(, q) : q = a2} and the 2-rarefaction curve through Ur given by (11d). Umis given by m = rexp(1 qr

a2r), qm = a2m. Um fulfills m > 0 and qm 0

and qm/m = a2. Next and in order to fulfill assumption (A4) we consider themaximization problem:

minqR+

0

q (30a)

subject to q d(01; U01 , a1) (30b)

and to q s(m; Um, a2), (30c)where d(; U01 , a1) is the demand function on pipe j = 1 for the given (left) stateU01 and s(; Um, a2) is the supply function on pipe j = 2 for the (right) state Umconstructed above. Note, that in this particular cases(m; Um, a2) = a2m. Thesolution to (30) is q := min{d(l; U01 , a1), s(m; Um, a2)}. Moreover, due to Propo-sition 4.3 there exists a unique state U1, such that q1 = q

and such that the so-lutionU1(x, t) to the Riemann problem (16a, 16b)with initial data U

01 , U1 satisfies

U1(xb1, t) = U1 for allt >0.Similarly, by Proposition 4.4, there exists a unique state

U2 such that q2 = q. But, the solution to (16a,16c) may violate (A3). However,


12/20


if we additionally assume that Ur = U02 is subsonic, then we conclude as follows:

For every q qm = a2m = s(m; Um, a2) there exist a unique point U2 on eitherthe 2-(Lax-)shock curve or the 2-rarefaction wave curve through the (right) stateU02 , such that q2 = q

. Then, the solution U2(x, t) to the Riemann problem (16a,16c) with initial data U2, U

02 satisfies U2(x

a2 , t) = U2 for all t >0 and consists of a

2-wave connecting the states U2 to Ur U02 . Furthermore, (A0)-(A4) are satisfied.This finishes the construction, refer to Figure 4 for a sketch of the states.

0 0.5 1 1.5 2 2.5 3

0

5

10

15

20

U2

0

U1

0

U1

U2

u

Um

Figure 4. Example with given data U01 = [2.8, 4.8], a1 = 6 andU0

2 = [2.6, 12], a2 = 8. The solid line is q = a2 and the dashed

line is the 2-wave curve through the right state U02 . The state Um(not in the picture) is the intersection of the dashed and the solidline. The dotted line is the 1wave curve through the left state U01 .In this particular case s(m; Um, a2) > d(01; U

01 , a1). The states

U1 U1 and U2 U2 are also depicted and have the same flux.In this example U1 and U01 are connected by a 1-rarefaction andU2 is connected by is connected to U

02 by a 2-rarefaction.

5.2. Coupling conditions II. We now discuss solutions satisfying (A0)- (A2),(A3) and (A4). We recall condition (A3): For all i v, j +v :

The pressure at the vertex is a constant, i.e., a2i i = a2jj . (31)

Before turning to the construction of a solution, we motivate the above choice.Common modelling of pipetopipe fittings in the engineering literature [19, 6]

is as follows: In realworld applications the momentum is not conserved at theintersection but reduced by a pressure drop depending on the physical constructionof the intersection. Usually the pressure drop fext at the intersection is approx-imated as in the incompressible case and modelled by a function fext dependingon the geometry of the intersection, a resistance coefficient (Kfactor), see [6]and the actual flows qi and pressures pi(i) at the intersection. The pressure dropfext is called minor loss and is not given analytically but in tables, see [6]. Those


13/20


minor losses are different from the friction inside the pipes. If we assume that thepressure loss at the intersection is given by the quantity fext, then (A3) has to be

replaced by the conditiona2jj =a

2i i fext (32)

for j +v, i v . For simplicity we assume in the following fext 0, i.e., (A3)holds.

Furthermore, in case of intersections with one(!) ingoing and one(!) outgoingpipe and a2 = a1 and mass conservation (A2), the condition (A3) implies theconservation of momentum, i.e.,

iv

q2i (xbi, t)

i(xbi, t)+a2i j(x

bi, t) =

j+v

q2j (xaj +, t)

j(xaj +, t)+ a2jj(x

aj +, t). (33)

Mathematically, the conservation of momentum and mass (14) implies that a solu-tion Uj , j = 1, 2 is also a weak solution in the sense of [14, 13]: The functionsUj

satisfy2

j=1

0

xbjxaj

Uj tj + Fj(Uj) xjdxdt xbjxaj

U0j j(x, 0)dx= 0,

for all smooth functions j : [0, ] [xaj , xbj ] IR2 having compact support in[xaj , x

bj ] and being a smooth across the vertex

i(xbi) = j(x

aj ) i v, j +v.

The equivalence of (A3) and (33) is not true for general pipe intersections or inter-section of pipes with different sound speeds. For a discussion of (A3) in the caseof tee fittings we refer to [3].

Now, we turn to the construction of a solution and proceed similar to Section 5.1.

Again, assume an intersection of two pipes j = 1 and j = 2 connected at xb1 =xa2 with possibly different sound speeds aj and constant initial data U

01 and U

02 .

Furthermore, we define c := a21/a22. Analogously to the the previous section, we

consider a maximization problem for the flux qat the vertex:

maxqIR+0

q subject to (34a)

0 q d(01; U01 , a1) and (34b)0 q s(m; Um, a2) and (34c)

a222 = a211. (34d)

But in contrast to (30), some quantities in (34) are defined implicitly: Given a fixedfluxq less than the demand, i.e., q < d(01; U

01 ). Then, 1 is uniquely defined, such

that the state U1 = (1, q) can be connected to U01 by waves of negative speed, c.f.

Proposition 4.1. Next, let U2 := (2, bq) := (c1, q). Finally, define Um = (m, qm)as the point intersection of the following two curves in the first quadrant of theqplane: First, the composition of the 1-(Lax-)shock curve and the 1rarefactionwave curve through the (left) state U2 and sound speeda2.Second, the compositionof the 2-(Lax-)shock curve and the 2rarefaction wave curve through the (right)stateU02 and sound speeda2.Assumem> 0, the ifUm exists, it is unique. Hence,we call a flux q is admissiblefor (34), if the above construction is possible or to beprecise, ifUm exists and ifm> 0 and ifq s(m; Um, a2).Note that it is possible,thatno flux q is admissible, see also discussion in Remark 5.1.


14/20


From now on, assume that the set of admissible fluxes q is non empty. Then,the maximization problem (34) is well-defined and has a unique solution q. As

described above we construct the states U1 and U2. Due to (34b) and Proposi-tion 4.1, the (half-)Riemann problem (16a,16b) with initial dataU01 and U1 consistsof waves of nonpositive speed and therefore U1(x

b1, t) = U1. Due to (34c) and

Proposition 4.5, the (half-)Riemann problem (16a,16c) with initial data U2 andU02 consists of waves of nonnegative speed. Its solutionU2(x, t) therefore satisfiesU2(x

a2+, t) =

U2.Due to (34d) and due to q2 = q1,the solutionUj , j = 1, 2 satisfies(A3) and (14). This finishes the construction. We give a few remarks on extensionsand related topics.

Remark 5.1. First, if we want to include the modelling of minor losses, we haveto replace (34d) by (32), which implies thatU2 =

c

1 a22fext

, q

.Second, in the simpler casec = 1, i.e., a2 = a1(!), there exists the following a

priori criterion for the existence of admissible fluxes c.f. [3]: If , then thereexist at least one admissible fluxq. Here, is as in Proposition 4.5 and( , 0)is thepoint of intersection in the first quadrant of the qplane of the 1-(Lax-)shockwave curve the (left) state U01 with the line{(, q) : > 0, q = 0}. This point isunique.

Third, again consider the simpler casec = 1.Then we might also define solutionsto the full problem (13) and (14) as restriction of the entropic solutionU = (, q)to a (usual) Riemann problem:

Ut +F(U)x = 0, U

(x, 0) =

U01 x < x

b1 = x

a2

U02 x > xb1

. (35)

I.e., U1(x, t) := U for x < xb1 and U2(x, t) := U

for x > xa2 . Both, the onegiven above and the solution constructed above by solving (34)coincideprovided that(x, t) > 0 and u(x, t) = q(x, t)/(x, t)

0. In the case u < 0 the assump-

tion (12a) is violated. However, the presented approach holds also true fordifferentsound speeds and can be extended to intersections of more than two pipes [3].

Finally, other mathematical models mentioned in the introduction propose thefollowing coupling conditions: As additional assumption the model of [8, 18] thetermstqand x(u

2) in (2) are neglected. Therefore, the authors finally deal witha scalar(!) conservation law. Coupling conditions for scalar conservation laws canbe found for example in [14].

The model introduced in [12] does neglect the term x(u2).Then it has to deal

with a linear(!) hyperbolic system and fixed eigenvalues.In [24] additional coupling conditions for hyperbolic and parabolic gas models

are defined: At the boundary the hyperbolic system is linearized to obtain fixedwave directions. This treatment is also common in engineering community when

numerically solving the pipe network problem.

6. Numerical Results. We present results for the isothermal Euler equations withfriction inside the pipes. The equation of state applied is (1). For actual operationsof pipelines we consider transient states of isothermal flow in pipes with a constantcross-sectional area with the steady state friction in all pipes [18, 26], i.e.,

tUj + xFj(Uj) = r(Uj), r(Uj) =

0

fg qj |qj |2Dj

, (36)


15/20


where D is the diameter of the pipe. The friction factor fg is calculated usingChens equation [5]:

1fg

:= 2log

/D

3.7065 5.0452

NRelog 1

2.8257

D

1.1098+

5.85060.8981NRe

whereNRe is the Reynolds numberNRe = uD/, the gas dynamic viscosity and the pipeline roughness.

For the numerical results we use a relaxed scheme [15, 2]: We apply a second orderMUSCL scheme together with a second order TVD time integration scheme [23, 11].Other approaches can be similarly applied [1]. A treatment of the source term(pipe friction) is undertaken as discussed in [26]. The integration of the sourceterm is done by solving an ordinary differential equation exactly after splitting(36). In all computations we use a space discretization of x= 1/800 and the timediscretization tis chosen according to the CFL condition with CFL = 3/4. In thefollowing the coupling conditions augmented in the relaxed scheme will be tested.

We present result for both coupling conditions (A3) and (A3), respectively. Themaximization problem (34) is solved numerically by direct search.

Example 1: The first example is similar to the one discussed in Figure 4: Giventwo connected pipesj = 1, 2 with constant initial dataU01 = [2, 8] andU

02 = [2, 12].

The sound speeds are a1 = 6 and a2 = 8. We prescribe inflow data at x = xa1 = 0and outflowx = xb2 = 1. The pipes are connected at x = x

b1 = x

a2 =

12

. We considerthe case of no friction inside the pipes. Furthermore, we consider the same initialdata but apply condition (A3). Snapshots of the solution are given in Figure 5.Next, we consider condition (A3) at the intersection and present snapshots of thenumerical solution for timest [0, 1] in Figure 6. In both cases the flux is continuousthrough the intersection, but in the first case (subsonic) the pressure is not due tothe different coupling (A3). In the second case the pressure is continuous throughthe intersection and hence fulfilling (A3). In all simulations the arising shock wavesare well captured by our second order scheme.

0.2

0.4

0.6

0.8

1

0.2

0.4

0.6

0.8

1

60

70

80

90

100

110

120

xt

aj2

0.2

0.4

0.6

0.8

1

0.2

0.4

0.6

0.8

1

8

8.5

9

9.5

10

10.5

11

11.5

12

xt

q

Figure 5. Example 1: Snapshots of the solutionUj to the problemof two connected pipes at different times t. The intersection is atx= 1

2. The densities j are shown to the left and to the rightqj is

shown. The condition used at the intersection is (A3).

Example 2: The second example is similar to the closed valve example [26]. Weconsider an incoming wave on pipe j = 1, which cannot pass the intersection, dueto a given low flux profile on the outgoing pipe j = 2.We assume to have friction in


16/20


0.2

0.4

0.6

0.8

1

0.2

0.4

0.6

0.8

1

80

90

100

110

120

xt

aj2

0.2

0.4

0.6

0.8

1

0.2

0.4

0.6

0.8

1

6

7

8

9

10

11

12

xt

q

Figure 6. Example 1: Snapshots of the solutionUj to the problemof two connected pipes at different times t. The intersection is atx= 1

2. The densities j are shown to the left and to the rightqj is

shown. The condition used at the intersection is (A3).

the pipes present. We use the subsonic condition (A3) at the interface. The soundspeeds are ai = 360 m/s. On each pipe we assume a friction factor offg = 103

and a pipe diameter ofD = 0.1 m. We prescribe an inflow ofqin = 70kgm2s1

at x = xa1 and an outflow ofqout = 1 at x = xb2. Both pipes have the same initial

conditionU0j = [ 1360

, 1]. In Figure 7 we give a plot of the snapshots of the densityevolution. We observe that the inflow profile moves on pipe 1 until it reaches theintersection. Since the maximal flow on the outgoing pipe is q = 1, we obtain abackwards moving shock wave on pipe 1 and an increase in the density near theintersection.

0.2

0.4

0.6

0.81

0.2

0.4

0.6

0.8

1

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

xt

Figure 7. Example 2: Snapshots of the solution j to the problemof two connected pipes at different times t. The intersection is atx= 1

2. The condition used at the intersection is (A3).

.


17/20


Example 3: The next example deals again with two connected pipes and againwe use the subsonic condition (A3) for the coupling, which yields a simple bound

for maximum possible flow on the outgoing pipe. We start with an equilibriumsituation at the vertex and consider a rarefaction wave as initial data on the ingoingpipe. Parts of the wave exceed the maximal flow on the outgoing pipe and thereforethe flow passes the intersection until the outgoing pipe is filled. Then, we observea backwards moving shock wave on the ingoing pipe. We consider the case ofno friction. The pipes are parameterized as before. The rarefaction wave on theingoing pipe is given by the solution to the Riemann problem with the data Ul1 =[0.15, 70], Ur1 = [6 103, 10]. On the outgoing pipe we have U02 = [0.3, 10]. Thesound speeds are a1 = 360 and a2 = 75. The states and corresponding 1- and2-curves are depicted in Figure 8. Further, we present a plot of the flow q2(x =12

, t) = q1(x = 12 , t) together with the pressures a222(x = 12 , t) and a

211(x = 12 , t)

in Figure 8. We observe the sudden appearance of the shock wave on pipej = 1,once the maximal flow for the outgoing pipe (q = 10) is reached. Finally, we give

a contour plot ofj in the x tplane. One observes the ingoing rarefaction aswell as the shock wave.

0 0.1 0.2 0.3 0.4 0.5 0.6

0

10

20

30

40

50

60

70 U1,l

t=0

U

1,r

t=0U

2

t=0

U2

t=TU1

t=T

1S

1R

2S

2R

m=a2

0 0.01 0.02 0.03 0.04 0.05

20

40

60

80

100

120

140

t

q

a1

1

a2

2

x

t

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.01

0.02

0.03

0.04

0.05

Figure 8. Example 3: Example of an incoming rarefaction wavewhich partly passes the intersection and which is reflected as soonas the maximal flow on the outgoing pipe is reached. At the inter-section we use condition (A3).


18/20


Example 4: This example is uses the coupling condition (A3). We start with anequilibrium situation and then have a rarefaction wave incoming to the intersection

with the same flow as in the previous example. The initial data is as follows:Ul1, U

r1 as in Example 3 andU

02 = [1.25103, 10] witha1 = 360 anda2 = 800.The

intersection is at x = 12

.Plots of the contour lines of the pressurea2jj and the fluxqj are given in Figure 9. Here, the incoming flow passes the vertex and pressureand flux are continuous through the vertex.

x

t

aj

2

j

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

x

t

qj

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

Figure 9. Example 4: Example of an incoming rarefaction wavewhich passes through the vertex and where we use condition (A3).We give contour plots of the pressure a2jj (left) and fluxqj .

Example 5: The last example deals with a row of three connected pipes j =1, 2, 3. We give an example where an incoming flow partly passes the first intersec-tion and is thereafter reflected at the second intersection. The arising backwards

moving wave on the middle pipe finally reaches again the first intersection andcauses another backwards moving rarefaction wave on the first pipe. Each pipe isparameterized as before and we assume A3 as condition at the intersection. No fric-tion is assumed. The initial data on the pipes j = 1, 2 is the same as in the previousexample; they are depicted in Figure 10 where we also see the intermediate statesin the full solution. On pipej = 3 we assume U03 = [0.3, 3]. The sound speeds area1 = 360, a2 = 110 and a3 = 20 and we use condition (A3). A plot of the pressurea2jj in the x tplane is given in Figure 10. The intersections are at x = 1/3 andx= 2/3.

7. Summary. We introduced different coupling conditions for the isothermal Eu-ler equations. We give general conditions that a solution near an intersection ofpipes has to satisfy and introduce the notion of demand and supply. We give addi-

tional assumptions and construct solutions to the network problem. We presented anumerical method using the coupling conditions as boundary values at the intersec-tions and presented results for sample intersections. The numerical scheme couldresolve well all arising shock and rarefaction waves. More complicated intersectionslike tees are considered in [3]. Additional elements (e.g. valves, compressors) andthe incorporation of realistic pressure loss models are considered in a forthcomingpublication. Since the presented approach holds for different sound speeds on eachroad, we will then also provide a comparison with the work on conservations lawswith discontinuous coefficients.


19/20


x

t

p = ai

2

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Figure 10. Example 5: Example for a double reflection on themiddle of three pipes. Condition (A3) is used.

Acknowledgements. This work was supported by the University of KaiserslauternExcellence Cluster Dependable Adaptive Systems and Mathematical Modellingand the University of KwaZulu-Natal competitive research grant (2005) FlowModels with Source Terms, Scientific Computing, Optimisation in Applications.

M. Herty would like to thank the University of KwaZuluNatal for their hospitalityduring a research visit there.

REFERENCES

[1] R. Aregba-Driollet and R. Natalini,Discrete kinetic schemes for multidimensional conserva-tion laws, SIAM J. Num. Anal., 6 (2000), pp. 19732004.

[2] M. K. Banda, Kinetic-based numerical schemes in fluid dynamics, Shaker Verlag, Aachen,2004.

[3] M. K. Banda and M. Herty and A. Klar,Gas flow in pipeline networks, to appear in NHM(2005).

[4] K. S. Chapmann, Virtual Pipeline System Testbed to Optimize the US Natural Gas Trans-mission Pipeline System, Technology Status Assessment Report, (2002).

[5] N.H. Chen, An explicit equation for friction factor in pipe, Ind. Eng. Chem. Fund. (1979),pp. 296297.

[6] Crane Valve Group, Flow of Fluids Through Valves, Fittings and Pipes, Crane TechnicalPaper No. 410 (1998).

[7] C. M. Dafermos, Hyperbolic Conservation Laws in Continuum Physics, Springer Verlag,

Berlin, (2000).[8] K. Ehrhardt and M. Steinbach, Nonlinear gas optimization in gas networks, Modeling, Sim-

ulation and Optimization of Complex Processes (eds. H. G. Bock, E. Kostina, H. X. Pu,R. Rannacher), Springer Verlag, Berlin, (2005).

[9] M. Garavello and B. Piccoli, Traffic flow on a road network using the Aw-Rascle model,

preprint, (2004).[10] E. Godlewski and P.-A. Raviart,Numerical Approximation of Hyperbolic Systems of Conser-

vation Laws, Springer Verlag, New York, (1996).


20/20


[11] S. Gottlieb and C-W. Shu and E. Tadmor, Strongly stability preserving high-order time dis-cretization methods, SIAM rev., 43(2001), pp. 89112.

[12] M. J. Heat and J. C. Blunt,Dynamic simulation applied to the design and control of a pipelinenetwork, J. Inst. Gas Eng., 9 (1969), pp. 261279.

[13] M. Herty and M. Rascle, Coupling conditions for the AwRascle equations for traffic flow,preprint, (2005).

[14] H. Holden and N. H. Risebro, A mathematical model of traffic flow on a network of unidi-

rectional roads, SIAM J. Math. Anal., 26 (1995), p. 9991017.

[15] S. Jin and Z. Xin,The relaxation schemes for systems of conservation laws in arbitrary space

dimensions, Comm. Pure Appl. Math., 48 (1995), pp. 235277.[16] J. P. Lebacque,Les modeles macroscopiques du trafic, Annales des Ponts., 67 (1993), pp. 24

45.[17] R. J. LeVeque,Numerical methods for conservation laws, Birkhauser Verlag, Basel, (1992).

[18] A. Martin and M. Moller and S. Moritz, Mixed Integer Models for the stationary case of gas

network optimization, TU Darmstadt Preprint, to appear in Math. Programming (2005).[19] F. M. White, Fluid Mechanics, McGrawHill, New York, (2002).

[20] Pipeline Simulation Interest Group, www.psig.org.[21] A. J. Osciadacz, Different Transient Models Limitations, advantages and disadvantages,

28th Annual Meeting of PSIG (Pipeline Simulation Interest Group), San Francisco, California(1996).

[22] E. Sekirnjak, Transiente Technische Optimierung, PSI Berlin, Technical Report, (2000).[23] C-W. Shu and S. Osher, Efficient implementation of essentially non-oscillatory shock-

capturing schemes, J. Comp. Phys., 77(1988), pp. 439471.

[24] J. D. Taylor, Numerical Analysis of fast and slow transients in gas transmission networks,PhD Thesis, University of Edinburgh (1997).

[25] Z. Vostry, Transient Optimization of gas transport and distribution, Proceedings of the 2ndInternational Workshop SIMONE on Innovative Approaches to Modelling and Optimal Con-trol of Large Scale Pipeline Networks, (1993).

[26] J. Zhou and M. A. Adewumi, Simulation of transients in natural gas pipelines using hybridTVD schemes, Int. J. Numer. Meth. Fluids, 32 (2000), pp. 407437.

Received November 11, 2004; revised January 2006.

E-mail address: [email protected]; [email protected]; [email protected]

coupling conditions for gas networks governed by the isothermal euler equations

Documents