ISSN 0005−1144

ATKAAF 42(3−4), 177−188 (2001)

Drago Matko, Gerhard Geiger, Thomas Werner

Modelling of the Pipeline as a Lumped Parameter System

UDK 681.518.3
IFAC IA 5.3.1;2.8.1
Original scientific paper

The paper deals with the simplification of pipeline models. A nonlinear distributed parameters model is line-
arised and its transfer function given. The pipeline is represented as a two-port system. Two causal representations
– the hybrid ones which are used in practice – are studied further. They involve three different transcendent trans-
fer functions which are then approximated by rational transfer functions using a Taylor series expansion. The de-
rived models are valid for low frequencies and are used to discuss how to obtain better approximation. They
equalise the high frequency gain of the transcendent and rational transfer functions and employ a Padé approxi-
mation. Due to the approximation of the high frequency gain, the derived models are only valid for a class of
models – namely – well damped pipelines. The derived models which describe the pipeline as a lumped parame-
ter system were verified on a real pipeline using experiment data.
Key words: modelling, model approximation, distributed-parameter system, lumped-parameter system, pipelines

1 INTRODUCTION are given. They include three different transcendent

Models of pipelines are used for different pur-
poses, such as controller design, leakage monitoring
etc. Observer-based leak detection and localisation
schemes for example require a pipeline model to
compute the states of a pipeline without leak [1, 2].
The first industrial applications demonstrate the
performance of observer-based methods [3].
In these schemes, the observer is derived using a
mathematical model of the pipeline. The one-di-
mensional compressible fluid flow through pipelines
is governed by nonlinear partial differential equa-
tions [4]. Pipelines are therefore distributed para-
meter systems. To date, there is no general closed-
-form solution. Numerical approaches like the Met-
hod of Characteristics are used instead [5] yielding
the computational background for the observer algo-
rithms.
However, sometimes simple models of the
pipeline in the form of a lumped parameter system
are use-ful: the classical system theory for Multi-
Input Multi-Output (MIMO) systems can be used
e.g. for controller design and system identification.
The re-sulting algorithms are less time-consuming
and hen-ce better suited for critical real time appli-
cations. Additionally, the analysis of fluid transients
caused by leaks is much easier.
The paper is organised as follows: in Section 2
the nonlinear distributed parameter model of the
pipeline is linearised and represented as a two-port
system. Two representations of the pipeline's model
transfer functions which are approximated by ratio-
nal transfer functions in Section 3. In Section 5, the
verification of the simplified models by real experi-
ment data is presented.
2 MATHEMATICAL MODELS OF THE PIPELINE
The analytical solution for unsteady flow prob-
lems is obtained by using the equations for conti-
nuity, momentum and energy. These equations cor-
respond to the physical principles of mass conserva-
tion, Newton's second law F = ma and energy con-
servation. Applying these equations leads to a cou-
pled non-linear set of partial differential equations.
Further problems arise in the case of turbulent
flow, which introduces stochastic flow behaviour.
Therefore, the mathematical derivation for the flow
through a pipeline is a mixture of both theoretical
and empirical approaches.
The following assumptions for the derivation of a
mathematical model of the flow through pipelines
are made: Fluid is compressible, viscous, adiabatic
and homogenous. The resulting nonlinear distribu-
ted parameters model is described by the following
two partial differential equations:
A ¶p ¶q
⋅ =− , (1)
2
a ¶t ¶x
1 ¶q
⋅ + ρg sin α +
λq bg
q2 = −
¶p
, (2)
A ¶t 2 DA 2 ρ ¶x

AUTOMATIKA 42(2001) 3−4, 177−188 177

Modelling of the Pipeline ... D. Matko, G. Geiger, T. Werner

where A, D and α are the profile, the diameter and LM 1

the inclination of the pipeline respectively, p the
LM OP = M coshc nL h
PL
c hOP L P O
− Z K tanh nL p
pressure, q the mass flow rate, λ the friction coeffi- p
PP MQ P 0
(10)
cient, ρ the density of the fluid, g the gravity con-
stant and a the velocity of sound.
NQ Q MM Z1 tanhc nL h
0 p
1
cosh c nL h PQ
N Q L
N K p

Nonlinear equations (2, 1) are linearised and writ- 3. Impedance representation: Inputs Q0 , QL , outputs
ten in a form using notations common in the analy- P0 , PL:
sis of electrical transmission lines. Also, the gravity
L 1 O
LM P OP = MM Z cothc nL h −ZK
effect is neglected – included into the working
point. The corresponding system of linear partial 0
K p
c h PP LMQ OP (11)
sinh nL p L
differential equations is
N P Q MM− Z sinh 1nL
L K −ZK coth c nL h P N Q Q
p
PQ 0

¶q ¶p N c h p
L + Rq = − (3)
¶t ¶x 4. Admittance representation: Inputs P0 , PL , outputs
¶p ¶q Q0 , QL:
C =− (4)
¶t ¶x
L 1 − 1 ⋅ 1 O
where L = 1 , R = 2
λ( q ) q −
( q is the flow at the wor-
LM Q OP = MM Z cothc nL h Z sinhc nL h PP LM P OP (12)
0 K
p
K p 0
A A ρD NQ Q MM Z1 ⋅ sinh 1nL − Z1 cothc nL h PP N P Q
L p L

king point) and C = A2 are the inertance (inductivi- N K c h p Q K

a
ty), resistance and capacitance per unit length, res- It should be noted that the forms 1 to 4 repre-
pectively. sent causal models, whilst Equations (5, 6) repre-
sent a noncausal model with inputs P0, Q0 and out-
The next step in the derivation of a simple mo- puts PL, QL. This completes the derivation of the
del of the pipeline is the analytical solution of linear transfer functions of the linearised pipeline. The re-
quations (3, 4). The Laplace transformation of them sulting transfer functions are transcendent. In the
yields the linear non-causal model next section, their approximations by rational trans-
fer functions will be given.
af af c h af c h
PL s = Z K Q0 s sinh nL p + P0 s cosh nL p (5)
3 LINEAR PIPELINE MODEL WITH LUMPED
Q a sf = Q a s f cosh c nL h − 1 P a s f sinh a nLpf
L 0 p 0 (6) PARAMETERS
Z K
In this section, the rational transfer functions of
where Lp is the length of the pipeline and
the pipeline will be derived. There are seven diffe-
n= a Ls + Rf ⋅ Cs rent transcendent transfer functions in the four pi-
(7)
peline forms of the previous section:
Z K = Ls + R = Ls + R .
(8) 1 , 1 tanh nL ,
c h c h
Z K tanh nL p ,
n Cs p
c h
cosh nL p ZK
Finally, the linearised model of the pipeline can Zk coth c nL h , ZK ⋅ 1 ,
p
be written in one of the following four forms which
differ from each other with respect to the model in-
c h
sinh nL p
1 ⋅ coth nL 1 ⋅ 1
puts (independent quantities) and outputs (depen-
dent quantities) ZK p c h and c h
Z K sinh nL p
.

1. Hybrid representation: Inputs Q0 , PL , outputs QL , First, simple models will be derived. These models
P0 : are only valid for low frequencies. On the basis of
a discussion of simple models, better models, which
LM 1
c h OP
Z K tanh nL p approximate the transcendent transfer functions in
LM OP M
P0
=
cosh nL p c h P LM P OP L
(9)
a broad frequency range, will be derived in subsec-
− 1 tanh nL p tion 4.
N Q MM
QL
ZK
c h coshc1nL h PP NQ Q 0
N Q p
3.1 Models for low frequencies
2. Hybrid representation: Inputs QL , P0 , outputs Q0 , Simple models are obtained by expanding the
PL: transcendent transfer function into a Taylor series.

178 AUTOMATIKA 42(2001) 3−4, 177−188

D. Matko, G. Geiger, T. Werner Modelling of the Pipeline ...

They are as follows: at the other end of the pipeline are kept constant)
1 the shock wave originating at one end of the pipe-
1. Hybrid representation: = line returns after reflection at the other end of the
c h
cosh nL p
pipeline with the opposite phase. The half-period of
Reverse pressure gain at upstream dead end: the oscillations is consequently equal to the time
P0 needed by the shock wave to travel along the pipe-
PL Q0 = 0 line and back. This gives the radial eigen-frequency
Flow gain at downstream reservoir: ω0 = 2p = p ⋅ 1 (16)
QL 4 LC L p 2 LC L p
PL = 0
Q0
which is only 10 % different from ω′0 . The static gain
Pressure gain at downstream dead end: of the treated transfer function is 1, meaning that the
PL step change of the flow at one end of the pipeline
QL = 0 results in the same steady state change of the flow
P0
at the other end (if the pressure at the first end re-
Reverse flow gain at upstream reservoir :
mains unchanged). Similarly, the step change of the
Q0 pressure at one end results in the same steady state
P0 = 0
QL change of the pressure at the other end (if the stea-
dy state flow remains unchanged).
1 = 1 ≈ 1
c h
cosh nL p cosh ea f
Ls + R ⋅ Cs L p j 2. Hybrid representation:
ZK
d i
tanh nL p =

≈ 1 . (13) Negative output admittance at upstream

FH 21 L LC + 241 L R C IK s
2
p
4
p
2 2 2
+ 1 L2p RCs + 1
2
QL
dead end: Q =0
PL 0
This transfer function describes a change of one Input admittance at downstream dead end:
quantity (flow, pressure) at one end of the pipeline, Q
if the same quantity is changing at the other end, − 0 QL = 0
whilst the other quantity remains constant, e.g. the P0
change of the flow at the pipeline's output, if the
1 tanh nL = Cs ⋅ tanh
flow changes at the input of the pipeline whilst the
pressure at the output of the pipeline remains con- ZK p c h Ls + R e a Ls + RfCs L j ≈
p

stant. The transfer function (13) can be interpreted 1 L3 RC 2 s 2 + L Cs

as a second order system with the eigen-frequency 6 p p
≈ .
1 2 ⋅ 1
FH
2 p
24 p
IK
1 L2 LC + 1 L4 R 2 C 2 s 2 + 1 L2 RCs + 1
2 p
ω 0′ = ≈ (14)
1 L2 LC + 1 L4 R 2 C 2 LC L p (17)
2 p 24 p This transfer function describes a change of the
and damping flow at one end of the pipeline, if the pressure is
changing at the same end whilst the flow at the
ζ≈ C ⋅ R⋅ L (15) other end of the pipeline remains constant. It has a
p
2L zero in the origin of the s plane which causes its
static gain to be 0. This is obvious since the flow at
where the second term in the square root of equa- both ends of the pipeline must be the same in the
tion (14) was neglected for small R. The eigen-fre- steady state.
quency can be interpreted as follows: with hybrid
boundary conditions (pressure at one end and flow A higher order Taylor extension of the transcen-
dent transfer function would result in:

1 tanh nL = Cs ⋅ tanh
ZK
c h
p
Ls + R e a Ls + RfCs L j ≈
p

60 6
FH 120
IK
1 C L p 5 RLs 4 + 1 C 2 Lp 3 L + 1 C 3 Lp 5 R 2 s 3 + 1 L3 RC 2 s 2 + L Cs
6 p p
≈ . (18 )
24 12
FH 24 p 2 p
IK
1 L2 C 2 Lp 4 s 4 + 1 RLC 2 Lp 4 s 3 + 1 L2 LC + 1 L4 R 2 C 2 s 2 + 1 L2 RCs + 1
2 p

AUTOMATIKA 42(2001) 3−4, 177−188 179

Modelling of the Pipeline ... D. Matko, G. Geiger, T. Werner

For R→0 this transfer function has two complex Negative output impedance at upstream
conjugate zeros at the radial frequency PL
dead end: − Q0 = 0
6 ⋅ 1 QL
ω= (19)
LC L p
Ls + R coth
which is close to the twice eigen-frequency of the
Z K coth nL p = c h Cs e a Ls + RfCs L j ≈ p

pipeline with hybrid boundary conditions. FH 21 L LC + 241 L R C IK s + 21 L RCs + 1

2
p
4
p
2 2 2 2
p
≈
3. Hybrid representation: Z K tanh nL p = c h FH 16 C Lp L + 1201 C Lp R IK s + 16 L RC s + L Cs .
2 3 3 5 2 3 3
p
2 2
p
Input impedance at downstream reservoir : (23)
P0
Q0 PL = 0 This transfer function describes the change of
the pressure at one end of the pipeline if the flow
Negative output impedance at upstream is changing at the same end while the flow at the
PL other end remains constant. It has a pole in the
reservoir : − P0 = 0 origin of the s plane and has on infinite static gain.
QL
This is obvious since different steady state flows on
Ls + R tanh both ends of the pipeline cause a constant increase
c h
Z K tanh nL p =
Cs e a Ls + RfCs L j ≈ p of the pressure. The remaining dynamics of the
transfer function for small R is a second order sys-
1 L PLCs + F L L + 1 L R CI s + L R
3 2 3 2 tem with the eigen-frequency
≈
3 p H 6 p K p p

FH 21 L LC + 241 L R C IK s + 21 L RCs + 1 . (20)

2
p
4
p
2 2 2 2
p ω= 6 ⋅ 1 (24)
LC L p
This transfer function describes a change of the
pressure at one end of the pipeline, if the flow is which is close to the twice eigen-frequency of the
changing at the same end whilst the pressure at the pipeline with hybrid boundary conditions and the
other end remains constant. Its static gain is LpR damping
which corresponds to the static change of the pres- 3C ⋅ R ⋅ L .
sure due to changing flow. ζ≈ p (25)
2L
A higher order Taylor extension of the transcen- This is the inverse of the case 2.
dent transfer function would result in:

Ls + R ⋅ tanh
e a Ls + RfCs L j ≈
Z K tanh nL p = c h Cs p

1 RL Lp C s + F 1 L Lp C + 1 R LLp C I s + 1 L RLCs + F L L + 1 L R CI s + L Cs
2 5 2 4 2 3 2 5 2 3 3 2 3 2

≈
40 H6 40 K 3 H 6 K .
p p p p
(21)
1 L C Lp s + 1 RLC Lp s + F 1 L LC + 1 L R C I s + 1 L RCs + 1
2 2 4 4 2 4 3 2 4 2 2 2 2
24 12 H2 24 K 2 p p p

For R→0 this transfer function has two complex

conjugate zeros at the radial frequency

ω= 6 ⋅ 1 (22)
LC L p 5. Impedance representation: Z K 1 =
which is close to the twice eigen-frequency of the
sinh nL p c h
pipeline with hybrid boundary conditions. Negative reverse transimpedance at upstream
P0
dead end: − Q0 = 0
4. Impedance representation: Z K coth nL p = c h QL
Input impedance at downstream dead end: Negative transimpedance at downstream
P0 PL
dead end: − QL = 0
QL = 0 Q0
Q0

180 AUTOMATIKA 42(2001) 3−4, 177−188

D. Matko, G. Geiger, T. Werner Modelling of the Pipeline ...

1 Ls + R ◊ 1 This transfer function describes the change of

ZK = ª the flow at one end of the pipeline if the pressure
sinh nL p c h Cs sinh ea f
Ls + R Cs L p j is changing at the same end while the pressure at
1 the other end remains constant. This is an inverse
ª
FH 16 C Lp L + 1201 C Lp R IK s + 16 L RC s + L Cs .
2 3 3 5 2 3 3
p
2 2
p
of the case 3, so a higher order Taylor extension of
the transcendent transfer function would result in:
(26)

1 coth nL = Cs coth
e a Ls + RfCs L j ª
ZK p c h Ls + R p

1 L C Lp s + 1 RLC Lp s + F 1 L LC + 1 L R C I s + 1 L RCs + 1
2 2 4 4 2 4 3 2 4 2 2 2 2

ª
24 12 H2 24 K 2 .
p p p
(28)
1 RL Lp C s + F 1 L Lp C + 1 R LLp C I s + 1 L RLCs + F L L + 1 L R CI s + L R
2 5 2 4 2 3 2 5 2 3 3 2 3 2
40 H6 40 K 3 H 6 K p p p p

This transfer function describes the change of

the pressure at one end of the pipeline, if the flow
is changing at the other end while the flow at the For R ª 0 this transfer function has a pole at the
same end remains constant. It has a pole at the ori- origin of the s plane, i.e. the integral character. The
gin of the s plane, causing its integral character. remaining dynamics of the transfer function is a
The remaining dynamics of the transfer function for second order system with the eigen-frequency
small R is again a second order system with the
same eigen-frequency and damping as in the previ- 6 ◊ 1
ous case. ω= (29)
LC L p
1 coth nL =
6. Admittance representation:
ZK p c h which is close to the twice eigen-frequency of the
pipeline with hybrid boundary conditions and the
Input admittance at downstream reservoir : damping
Q0
P0 PL = 0
ζª 3C ◊ R ◊ L . (30)
p
2L
Negative output admittance at upstream
QL
reservoir : - P0 = 0 7. Admittance representation: 1 =
PL
Z K sinh nL p c h
1 coth nL = Cs coth Negative reverse transadmittance at upstream
ZK p c h Ls + R e a Ls + RfCs L j ª p
Q0
reservoir : - P0 = 0
FH
1 L2 LC + 1 L4 R 2 C 2 s 2 + 1 L2 RCs + 1 IK PL
2 p 24 p 2 p Transadmittance at downstream reservoir:
ª .
3 p
6
FH
1 L3 RLCs 2 + L L + 1 L3 R 2 C s + L R
p p p
IK QL
(27) PL = 0
P0

1 = Cs ◊ 1 ª
Z K sinh nL p c h Ls + R sinh
e a Ls + Rf Cs L j p

ª 1 . (31)
40 6
FH
40 3 6 p
IK
1 RL2 Lp 5 C 2 s 4 + 1 L2 Lp 3 C + 1 R 2 LLp 5 C 2 s 3 + 1 L3 RLCs 2 + L L + 1 L3 R 2 C s + L R
P p p
FH IK

This transfer function describes the change of

the flow at one end of the pipeline if the pressure
is changing at the other end while the pressure at

AUTOMATIKA 42(2001) 3-4, 177-188 181

Modelling of the Pipeline ... D. Matko, G. Geiger, T. Werner

the same end remains constant. It has a static gain The eigen frequency ω0 (Equation 16) determi-
1 which corresponds to the static change of the nes the coefficient
Lp R
flow due to changing pressure. For R ª 0 this trans- a2 = 12 = 42 L2p LC . (36)
fer function has a pole at the origin of the s plane, ω0 p
i.e. the integral character. The remaining dynamics
of the transfer function is a second order system The high frequency gain is determined by
with the eigen-frequency and damping as in the
previous case. 1
a f
lim G0 jω = lim (37)
ωÆ∞ ωÆ∞
cosh e a Ljω + Rf Cjω L j p
4 MODELS FOR WELL-DAMPED PIPELINES
which does not exist. However,
In this section, the pipeline will be presented as
a second order lumped parameter system in the C RL p
± e ( Ljω + R ) C j ω L p j ± ◊
form lim e =e L 2 (38)
ωÆ∞
2
b2 s + b1 s + b0 - sTd
af
G s =
a2 s 2 + a1 s + a0
e (32) does exist and in the case

C ◊ RL p >> 1
where Td is the dead time. The transcendent trans- (39)
L 2
fer functions will be approximated by a rational
transfer function with dead time however only for a the high frequency gain (37) can be approxima-
class of pipelines, as shown later. The procedure of ted by
the approximation is as follows: as the simple mo- C ◊ RL p
b2 -
dels are valid for low frequencies, their static gain is
exact and will be applied to the derived models.
lim G s =
ωÆ∞
af a2
ª 2e L 2 (40)
Also, the eigen-frequency as discussed in the earlier yielding
subsection (Equation 16) will be used. Next, the high
C ◊ RL p
frequency gain will be approximated from transcen- -
dent function and known dead time will be applied. b2 = 82 LCL2p e L 2 . (41)
p
In this way, four coefficients of the transfer func-
tion (32) are determined. The remaining two are The condition (39) can also be written as
obtained using a Padé approximation of the trans-
cendent transfer function. Since the high frequency ζ >> 1 ,
2
gain approximation is valid only under certain con-
ditions, the derived models are only valid for one so the derived models are valid for well damped
class of pipelines. The derived models are as fol- pipelines only.
lows The remaining coefficients b1 and a1 are de-
termined using a Padé approximation of (33) i.e.
1. Hybrid representation: Reverse pressure gain at up- by comparing the corresponding terms of the
stream dead end, Flow gain at downstream reser- following two series expansions:
voir, Pressure gain at downstream dead end and
Reverse flow gain at upstream reservoir a2 s 2 + a1 s + 1
=
1 1 b2 s 2 + b1 s + 1
af
G0 s =
c h =
c h
cosh nL p e - sTd
e - sTd . (33)
cosh nL p a f b
= 1 + a1 - b1 s + a2 - b2 - b a1 - b1 s 2 +K a fg (42)
Since this transfer function connects quantities at and
different ends of the pipeline, the dead time for
- LCs L p
(33) is known – it is the time needed for the
shock wave to travel along the pipeline.
cosh e b Ls + Rg Cs L j e = p

F1
= 1 + G L RC - L LC J s +
2 I
Td = LC L p . (34) H2 p
K p

F
+ G L LC +
2 1 1
L R C - L RC 4 2 2 3 IJ
LC s 2 + K
The static gain of (33) is 1, so the coefficients H p
24 2
p p
K
a0 = 1 The result is (taking into account Equations 36
(35)
b0 = 1 and 41)

182 AUTOMATIKA 42(2001) 3-4, 177-188

D. Matko, G. Geiger, T. Werner Modelling of the Pipeline ...

F - C◊
RL p
- 24 p2 L - p2 L2p R 2 C + 12 p2 L p R LC
I
GH
L p C 96 L - 192 Le L 2
JK
b1 = (43)
d
12 p2 L p RC - 2 LC i
F - C◊
RL p
+ 5p2 L2p R 2 C - 12 p2 L p R LC
I
GH
L p C 96 L - 192 Le L 2
JK
a1 = . (44)
d
12 p2 L p RC - 2 LC i
2. Hybrid representation: Negative output admittance Cs tanh
at upstream dead end and Input admittance at Ls + R e a Ls + Rf Cs L j = p

downstream dead end

= L p Cs - 1 L3p RC 2 + K (52)
3
G0 s = 1 tanh nL p .
af c h (45)
yields
ZK
b1 = L p ◊ C (53)
The dead time of the treated transfer function is
zero since it connects a change of the flow at
one end of the pipeline if the pressure changes a1 = 1 2 L p 12 LC + p2 L p RC .
d i (54)
3p
at the same end:
Td = 0. (46) 3. Hybrid representation: Input impedance at down-
stream reservoir and Output impedance at up-
The static gain of G0(s) is zero, so the coeffici- stream reservoir
ents
a0 = 1 (47) af c h
G0 s = Z k tanh nL p . (55)
b0 = 0.
The dead time of the treated transfer function is
The eigen frequency ω0 is the same as in the zero, since it connects a change of the pressure
previous case: at one end of the pipeline if the flow changes at
the same end, so
a2 = 12 = 42 L2p LC . (48) Td = 0. (56)
ω0 p
The static gain of G0(s) is LpR, yielding
The high frequency gain is determined by
a0 = 1 (57)
b2
= lim 1 tanh nL p = c h b0 = Lp R .
a2 ωÆ∞ Z K
The eigen-frequency is the same as in the previ-
Cjω
= lim tanh e - LCω 2 + RCjω L p = j ous two cases:
ωÆ∞ Ljω + R
C a2 = 12 = 42 L2p LC . (58)
= (49) ω0 p
L
yielding The high frequency gain is determined by
b2
b2 = 42 L2p C LC .
p
(50) = lim Z tanh nL p =
a2 ωÆ∞ k
c h
Ljω + R
The remaining coefficients b1 and a1 are deter- = lim tanh e - LCω 2 + RCjω L p = j
ωÆ∞ Cjω
mined by the same procedure as in the previous
case. A comparison of L
= (59)
C
b2 s 2 + b1 s
a2 s 2 + a1 s + 1
a f
= b1 s + b2 - b1 a1 s 2 + K (51) yielding
b2 = 42 L2p L LC . (60)
and p

AUTOMATIKA 42(2001) 3-4, 177-188 183

Modelling of the Pipeline ... D. Matko, G. Geiger, T. Werner

The remaining coefficients b1 and a1 are deter- The resulting transfer function has an integral
mined by the same procedure as in the two pre- character. Since the models are valid for well
vious cases. A comparison of damped pipelines only, a second order transfer
function is sufficient and the eigen-frequency of
b2 s 2 + b1 s + b0 the pipeline with impedance boundary condi-
= tions which is close to the twice eigen-frequency
a2 s 2 + a1 s + 1
of the pipeline with hybrid boundary conditions
a f b a
= b0 + b1 - b0 a1 s + b2 - b0 a2 + - b1 + b0 a1 a1 s 2 + K f g can not be recognised.
(61)
and

Cs e a Ls + RfCs L j = L R + FH L L - 13 L R CIK s +
Ls + R tanh
p p p
3
p
2

+ L 1 L RLC - L RFH 1 L LC + 1 L R C IK + 1 L RC FH 1 L R C - L LIK O s

3 2 4 2 2 2 3 2 2
+K (62)
NM 3 p
2 p
24 p
2 3 p PQ p p p

yields

b1 =
d
L p 288 L2p R 2 LC + p2 L4p R 4 C 2 - 288 L p RL LC - 72 p2 L2 i (63)
24 p 2
d L2p R 2 C - 3 L i
and

a1 =
d
L p 96 L p RLC - 96 L LC - 16 p2 L p RLC + 3p2 L3p R 3 C 2 i. (64)
8p 2
d L2p R 2 C - 3 L i

5. Impedance representation: Negative reverse trans-

4. Impedance representation: Input impedance at impedance at upstream dead end and negative
downstream dead end and negative output impe- transimpedance at downstream dead end
dance at upstream dead end
af 1 (68)
af
G0 s = Z K coth nL p c h (65) G0 s = Z K
c h
sinh nL p
.

The dead time of the treated transfer function is Since this transfer function connects change of
zero, since it connects a change of the pressure the pressure at one end of the pipeline if the
at one end of the pipeline if the flow changes at flow changes at the other end of the pipeline,
the same end, so the dead time for (68) is known – it is the time
Td = 0 (66) needed for the shock wave to travel along the
pipeline.
Transfer function (65) is the inverse of the
transfer function (45) so the coefficients are: Td = LC L p . (69)

a0 = 0 If the pipeline is approximated with the second

a1 = L p ◊ C order system, what is sufficient for well damped
pipelines, the known eigen-frequency can not be
a2 = 42 L2p C LC used. A determination of all coefficients of the
p second order transfer function by the Padé ap-
b0 = 1 (67) proximation results in a unsolvable system of equ-
ations. So an other procedure should be used.
b1 = 1 2 L p 12 LC + p2 L p RC
d i By this procedure it is supposed that the dyna-
3p mics (i.e. the denominator of the corresponding
b2 = 42 L2p LC . transfer function) of all hybrid representations
p of the pipeline is the same.

184 AUTOMATIKA 42(2001) 3-4, 177-188

D. Matko, G. Geiger, T. Werner Modelling of the Pipeline ...

Equation (68) can be written as: 7. Admittance representation: Reverse transadmittance

at upstream reservoir and admittance at down-
1 1
ZK
sinh nL p c h c h
= Z K coth nL p
cosh nL p c h (70) stream reservoir

af
G0 s = 1 = 1 e - sTd .
and is the product of transfer functions (65) and c h
Z K sinh nL p c h
Z K sinh nL p e - sTd
(33). The dead time is a part of the transfer
function (33). The denominator of the Padé ap- (74)
proximation of this transfer function cancels out Since this transfer function connects change of
the numerator of the Padé approximation of the the flow at one end of the pipeline if the pres-
transfer function (65), so the coefficients of the sure changes at the other end of the pipeline,
resulting transfer functions are: the dead time for (74) is known – it is the time

a0 = 0
a1 = L p ◊ C

a2 = 42 L2p C LC
p
b0 = 1
F RL p
I (71)
- C◊ 2 2
L C G 96 L - 192 Le
p
L 2 - 24p L - p L2p R 2 C + 12 p2 L p R LC J
b1 =
H K
d
12 p2 L p RC - 2 LC i
C RL p
b2 = 8 LCL2 e - L ◊ 2
2 p
p

6. Admittance representation: Input admittance at needed for the shock wave to travel along the
downstream reservoir and output admittance at pipeline.
upstream reservoir Td = LC L p . (75)
G0 s = 1 coth nL p
af c h (72) If the pipeline is approximated with the second
ZK
order system, what is sufficient for well damped
The dead time of the treated transfer function is pipelines, the known eigen-frequency can not be
zero, since it connects a change of the flow at used. A determination of all coefficients of the
one end of the pipeline if the pressure changes second order transfer function by the Padé ap-
at the same end, so proximation results in a unsolvable system of
equations. So again the procedure will be used
Td = 0 (73)
which supposes that the dynamics of all hybrid
Transfer function (72) is the inverse of the trans- representations of the pipeline is the same.
fer function (55) so the coefficients are: Equation (74) can be written as:

a0 = L p R

a1 =
d
L p 288 L2p R 2 LC + p2 L4p R 4 C 2 - 288 L p RL LC - 72 p2 L2 i
24p 2
d L2p R 2 C - 3 L i
a2 = 42 L2p L LC
p
b0 = 1

b1 =
d
L p 96 L p RLC - 96 L LC - 16p2 L p RLC + 3p2 L3p R 3 C 2 i
8p 2
d L2p R 2 C - 3 L i
b2 = 12 = 42 L2p LC .
ω0 p

AUTOMATIKA 42(2001) 3-4, 177-188 185

Modelling of the Pipeline ... D. Matko, G. Geiger, T. Werner

der lag with the time constant of 15 s as the flow

1 transducer dynamics.
= 1 coth nL p 1
c h
Z K sinh nL pc h
ZK cosh nL p
and is the product of transfer functions (72) and
(33). The dead time is a part of the transfer
function (33). The denominator of the Padé ap-
proximation of this transfer function cancels out
the numerator of the Padé approximation of the
transfer function (72), so the coefficients of the
resulting transfer functions are:
(76)
c h Accordingly these dynamics were taken into ac-
count in the verification of the lumped parameter
model. If the measured flow was used as the input
to a model, the applied input was obtained by
an approximate deconvolution (filtering by the
Gs-1 ª 15 s + 1 ) where ε should be small, but was ta-
εs + 1
ken to be 3 so as not to magnify the measurement
noise. The simulated output flow was first filtered

a0 = L p R

a1 =
d
L p 288 L2p R 2 LC + p2 L4p R 4 C 2 - 288 L p RL LC - 72 p2 L2 i
24p 2
d i
L2p R 2 C - 3 L

a2 = 42 L2p L LC
p
b0 = 1 (77)
F - C◊
RL p
- 24p2 L - p2 L2p R 2 C + 12 p2 L p R LC
I
GH
L p C 96 L - 192 Le L 2
JK
b1 =
d
12 p2 L p RC - 2 LC i
- C ◊ RL p
b2 = 82 LCL2p e L 2 .
p

5 VERIFICATION OF THE MODELS

by Gs = 1 and then compared with the mea-
15 s + 1
The models were verified on a real pipeline with sured flow.
the following data: Length of the pipeline Lp =
= 31 146 m, diameter D = 0.143 m, relative rough- Model forms 1 and 2 as described in section 2
ness kc = 0.0291 mm, inclination α = 0 and the fluid were evaluated with the following results:
data: Density ρ = 752 kg/m3, kinematic viscosity v = 1. Hybrid representation: Inputs Q0, PL, outputs QL,
= 7.18 ¥ 10-7 m2/s and velocity of sound a = 1110 m/s. P0
Fluid transients were generated for experimental
verification by closing a shunt valve at the begin-
ning of the pipeline. This leads to a quick drop in
pressure causing fluid transients. There were no
controllers for flow rate or pump pressure.
The excitation was quite high (more than 50 % of
the steady state values of the signals). For this rea-
son, the model was linearised at 75 % of the steady
state flow (18 kg/s) yielding the parameters L = 62.3,
R = 10.6 and C = 1.3 ¥ 10-8. The damping factor for
this case is ζ = 3.37. The experimental data were
first evaluated by the solving the nonlinear partial
differential equations. The evaluation gave unsatis-
factory results due to the dynamics of the transdu-
cers. Since all the data – both input and output –
are measured, only the relative dynamics of the
pressure and flow transducers is significant. A good Fig. 1 The simulated (solid line) and the measured (dashed line)
coincidence was obtained by including the first or- flows at the end of the pipeline

186 AUTOMATIKA 42(2001) 3-4, 177-188

D. Matko, G. Geiger, T. Werner
In Figure 1 comparison of the simulated (solid
line) and measured (dashed line) flows at the
end of the pipeline is shown.
Figure 2 depicts the simulated and the measured
pressure at the beginning of the pipeline.
Fig. 2 The simulated (solid line) and the measured (dashed line)
pressures at the beginning of the pipeline
In both figures an acceptable coincidence can be
observed. The error between the outputs of the
model and the real plant is caused by nonlineari-
ties due to high excitation and, of course, by the
approximation of the pipeline using a lumped pa-
rameter model. The »noise« in the simulated sig-
nals (especially in Figure 2) is due to the decon-
volution of the model input signal (flow at the
beginning of the pipeline) which increased the
measurement noise.
2. Hybrid representation: Inputs QL , P0 , outputs Q0 ,
PL
In Figure 3 the comparison of the simulated and
measured flows at the beginning of the pipeline
is shown.
Acceptable coincidence can again be observed.
The verification of the fourth model (output
pressure as the function of the output flow and
input pressure) failed. The reason for that failu-
re is as follows: according to Equation (10) the
resulting output pressure is the difference of two
terms (pressure changes due to the change of
the output flow and input pressure, respectively).
AUTOMATIKA 42(2001) 3-4, 177-188
Modelling of the Pipeline ...
Fig. 3 The simulated (solid line) and the measured (dashed line)
flows at the beginning of the pipeline
In the treated case, both terms have a magnitu-
de of about 25 bars, while their difference sho-
uld have a magnitude of 0.6 bars. The numerical
problems arising due to nonlinearities and mea-
surement noise magnified by the deconvolution
caused the failure of the validation.
6 CONCLUSION
Three transcendent transfer functions which are
parts of the linear model of the pipeline as a distri-
buted parameter system are approximated by ratio-
nal transfer functions with time delay. The approxi-
mations were verified on a real pipeline by using
real experiment data. The approximations are only
valid for one class of models – well damped pipe-
lines. Simple models for pipelines which are able to
oscillate are being investigated.
REFERENCES
[1] L. Billmann, Methoden zur Lecküberwachung und Regelung
von Gasfernleitungen. TH Darmstadt, 1985.
[2] H. Siebert, Untersuchung verschiedener Methoden zur Leck-
überwachung bei Pipelines. TH Darmstadt, 1981.
[3] G. Geiger, Application of a Model-Based Method for Leak
Detection and Localization. GMA-Kongress Meß- und Auto-
matisierungstechnik 18./19. Juni 1988, VDI Berichte Nr.
1397, 1998.
[4] V. L. Streeter, E. B. Wylie, Fluid Transients. McGraw-Hill,
New York, 1978.
[5] C. V. Rao, K. Eswaran, On the Analysis of Pressure Transi-
ents in Pipelines Carrying Compressible Fluids. Int. J. Pres.
Ves. and Piping, 56:107–129, 1993.
187
Modelling of the Pipeline ... D. Matko, G. Geiger, T. Werner

Modeliranje cjevovoda kao sustava s usredoto~enim parametrima. U ~lanku se opisuje postupak pojednostav-
ljenja matemati~kih modela cjevovoda. Lineariziran je nelinearni matemati~ki model s raspodijeljenim parametrima
iz ~ega je dobivena prijenosna funkcija. Cjevovod je prikazan kao dvoulazni sustav. Analizirana su dva me|usobno
povezana prikaza triju razli~itih transcedentnih prijenosnih funkcija koje se potom nadomje{taju racionalnim pri-
jenosnim funkcijama uporabom Taylorovog razvoja u red. Izvedeni modeli valjani su za niske frekvencije i koriste
se kako bi se do{lo do boljih aproksimacija. Uspore|uje se visokofrekvencijsko poja~anje transcedentnih i racional-
nih prijenosnih funkcija. Zbog aproksimacije visokofrekvencijskog poja~anja izvedeni su modeli valjani samo za
dobro prigu{ene cjevovode. Izvedeni modeli, koji opisuju cjevovod kao sustav s usredoto~enim parametrima, pro-
vjereni su na realnom cjevovodu.
Klju~ne rije~i: modeliranje, aproksimacija modela, sustav s raspodijeljenim parametrima, sustav s usredoto~enim
parametrima, cjevovod

AUTHORS’ ADDRESSES:
Drago Matko,
Faculty of Electrical Engineering
University of Ljubljana
Tr`a{ka 25, 1000 Ljubljana Slovenia
e-mail: drago.matko@@fe.uni-lj.si
Gerhard Geiger, Thomas Werner
Fachhochschule Gelsenkirchen, Germany
@fh-gelsenkirchen.de
e-mail: gerhard.geiger@
thomas.werner@@fh-gelsenkirchen.de

Received: 2001-07-12

188 AUTOMATIKA 42(2001) 3-4, 177-188