Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

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

net/publication/227270692

The recent development of analysis methodology for rapid crack propagation


and arrest in gas pipelines

Article  in  International Journal of Fracture · February 2000


DOI: 10.1023/A:1007614308834

CITATIONS READS

32 309

2 authors, including:

O'Donoghue Padraic
National University of Ireland, Galway
76 PUBLICATIONS   843 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

High Temperature creep modelling of power plant steels View project

All content following this page was uploaded by O'Donoghue Padraic on 11 December 2015.

The user has requested enhancement of the downloaded file.


International Journal of Fracture 101: 269–290, 2000.
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.

The recent development of analysis methodology for rapid crack


propagation and arrest in gas pipelines

ZHUO ZHUANG1 and P. E. O’DONOGHUE2


1 Department of Engineering Mechanics, Tsinghua University, Beijing 100084, China
2 Department of Civil Engineering, National University of Ireland, Galway Galway, Ireland

Received 28 July 1998; accepted in revised form 6 December 1999

Abstract. Several dynamic analysis issues relating to rapid crack propagation (RCP) and arrest in gas piplines
were developed recently, and are presented in this paper. This is based on a fluid/structure/fracture interaction
package, PFRAC. Some developments have been implemented into this finite element code to simulate the be-
havior of the fractured pipes. The criteria for crack initiation, propagation and arrest have been discussed. As the
crack propagates along the pipeline, the gas pressure decompression ahead of the crack tip and an efficiency of
a linear decay behind the crack have been used in the computation. For the calculation of crack driving force G,
the numerical approaches using the nodal force release and energy balance methods are described. This paper also
presents a novel analysis methodology that has been developed to investigate the suitability of crack arrestors.
Several numerical results for the cracked steel pipes with arrestors are presented along with comparisons with
pipes that do not have arrestors.

Key words: Dynamic fracture mechanics, finite element, pipeline, rapid crack propagation, arrest

1. Introduction

There have been a number of catastrophic instances where cracks have propagated in a rapid
manner for a few kilometers in steel gas transmission pipelines. Some fast running cracks
also occurred in polyethylene (PE) gas distribution pipelines. Clearly it is important that these
rupture events must be prevented. This problem is recognized as one of the most challenge
issues of dynamic fracture mechanics [1–6]. For rapid crack propagation (RCP) to occur in a
gas pipeline, it is necessary that a through-wall crack exists in the axial direction. This crack is
generally a result of third party damage or possibly due to slow crack growth (SCG) through
the pipe wall in PE materials or stress corrosion in steel pipes. Unlike other failures, which in
principle could have been avoided by preventing initiation (e.g., through a conventional frac-
ture mechanics approach), third party damage cannot be anticipated. Following initiation, the
crack will either propagate or else will quickly arrest. The focus of this paper will examine the
conditions under which long running cracks will take place rather than investigating initiation
conditions. Once these conditions are identified, the design is such that long running steady
state cracks will be prevented.
During axial propagation, the crack travels at a relatively constant speed and steady state
conditions prevail. Arrest finally occurs when the crack encounters an arrestor device or enters
a tougher section of the pipe. Thus, it is desirable that the conditions are such that steady state
propagation is prevented and, in this case, the crack will arrest after a short distance. There will
then be an increasing need for reliable pipeline design, inspection and maintenance procedures
that will preclude service failures. A new analysis methodology is presented in this paper to
270 Z. Zhuang and P.E. Donoghue

Figure 1. Prediction of crack propagation and arrest from computed dynamic crack driving force and measured
dynamic crack resistance data (schematic).

investigate the suitability of crack arrestors. Since crack initiation could not be absolutely
prevented, then a crack arrest strategy is required as a line of defence against catastrophic
rupture. It must be pointed out that in most cases, the pipe material is sufficiently tough to
prevent continued crack propagation. Thus, the damage resulting from this propagation event
will be limited and contained to a small region.
A finite element simulation model for gas, structure and fracture dynamics interaction
has been developed at early 1990’s [7], which is named Pipeline FRacture Analysis Code
(PFRAC). Some analysis capabilities of this code have been extended by the later works [8].
The complication for this model lies principally in their ability to correctly treat the combined
effects of (1) the large-scale inelastic deformation of the pipelines; (2) the pressure from
unsteady fluid flow; (3) the dynamic crack propagation process. This significant interaction
behaviour is at the heart of the development of dynamic fracture mechanics model. This paper
deals primarily with the recent development of PFRAC for the analysis of RCP and arrest in
the gas pipelines.

2. The conditions of crack propagation and arrest

The conditions under which a crack propagates in a gas pipeline are such that the motive force
for crack propagation, here expressed in terms of the energy release rate, G, or crack driving
force, must continue to balance the dynamic fracture toughness of material, Gd , to sustain
propagation. This is given by [8]

G(a, p, D0 , SDR, E) = Gd (T , ȧ, h), (1)

where G is a function of the crack length, a, the gas pressure, p, the outer diameter of the pipe,
D0 , the standard dimension ratio of the pipe size, SDR = D0 / h, and the stiffness of material,
E. In turn, Gd is a material property dependent on the temperature, T , the crack speed, ȧ and
the thickness of the pipe wall, h, due to constraint effects. Dynamic fracture only takes place
when this balance is maintained and if G drops below Gd , the crack will arrest.
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 271

Schematic G data for gas pipeline fracture are shown in Fig. 1 with decompression at
low speeds and inertial effects at high speeds resulting in a peak value of driving force at
intermediate speeds. This peak value is referred to as maximum crack driving force Gmax . It
also illustrated the fracture resistance data Gd . To complete the relations such as given in the
equation (1), the minimum value of Gd is obviously a key parameter. Various experiments
have been devised in order to determine values of Gd . More specifically, when the Gmax
available in the pipe is such that Gmax = Gd , there is sufficient energy for the crack to
propagate, thus enabling it to extend over a long distance. The relative values of Gmax and
Gd therefore provide the basis for developing practical guide-lines to prevent RCP in service.
The attention in dynamic fracture mechanics research focuses on the G computations and Gd
determinations.

3. Numerical models in the simulation of cracked pipe

The computation of Gmax is one of the critical steps in the analysis process. Due to the
complexities of the RCP event, e.g., gas escaping from the opening breech, the calculation
of G is a non-trivial task. Fortunately this can be accomplished using a unique finite element
code PFRAC [7, 9]. There are three basic segments to this code: a structural mechanics unit;
a fluid mechanics unit; and a fracture mechanics unit. These three models are fully coupled
together and are used to analyse the crack propagation in a pipeline.
The primary requirements of the structural dynamics unit are that it must model the large
deformations of the cylindrical pipe wall. The program incorporates a Lagrangian finite ele-
ment description with four-node quadrilateral elements that allow for geometric non-linearity.
An explicit finite difference scheme is used to march forward in time. This code is ideally
suited to shell-like structures undergoing large deformations, e.g. the flap opening experienced
in ruptured pipes. A three-dimensional finite difference scheme is used to model the complex
highly transient flow that takes place when the pipeline is fractured and the gas is escaping.
Finally, a node release algorithm is implemented to simulate crack propagation in the finite
element model. The G is then calculated by the work done through the release of the nodal
forces. Validation of the PFRAC methodology was provided through comparison with full
scale test data [7, 9].
Using the variation of the elastic strain energy in the cracked pipe, an early analysis, de-
veloped by Irwin and Corten [10], identified the crack extension force as the rate of release of
circumferential strain energy stored ahead of the crack tip. Based on the fact that the pressure
behind the crack tip is zero. They developed the following closed-form expression for G:

πp02 (D0 − 2h)2 (D0 − h)


G0 = (2)
8Eh2
where p0 is the initial line pressure, D0 and h are the outside diameter and wall thickness of
the pipe and E is the Young’s modulus of the material. For the SDR11 PE pipe with a diameter
of 0.25 m and a crack velocity of 175 m s−1 , Fig. 2 shows the variation of G against pressures.
In this figure, both G and G0 increase quickly as the initial line pressure increases, while
the dimensionless values of G/G0 remain relatively constant. The value of G calculated
by PFRAC is much higher than the value of G0 (typically 10–30 G0 ) calculated by the
Irwin–Corten analysis equation. The latter prediction might be expected to work for a pipe
pressurised by an incompressible fluid. For the gas pipe, however, the pressure decays steadily
272 Z. Zhuang and P.E. Donoghue

Figure 2. Profiles of calculated crack driving force with pressure (G/G0 is the dimensionless value).

from the crack tip pressure, p1 , to atmospheric pressure, behind crack tip over a characteristic
length. Within this outflow region, the remaining gas can continue to develop a potent crack
driving force. Pressurised water in a pipeline can store as much internal energy as the pipe
wall, whilst pressurised gas can contain a thousand times as much. It is the internal energy
that effectively determines the driving force G. This is why the G for gas pressurisation is
much higher than the Irwin–Corten value. In turn, this is why the critical pressure for gas pipe
is much lower than that for water pipe. This has already been proved by full-scale tests.

4. Gas pressure simulation during steady state propagation

During crack propagation in the gas pipeline, the gas pressure provides the motive forces
for the moving walls, which are simultaneously being torn apart at the crack tip, while the
shape of the walls in turn affects the gas flow and the fracture processes. Evaluation of the gas
pressure on the pipe wall is a key step for quantifying this coupling phenomenon in order to
analyze the crack propagation process.

4.1. G AS PRESSURE DECOMPRESSION AHEAD OF THE CRACK TIP

As the crack propagates along the pipeline, the gas escapes behind it and a decompression
wave travels ahead of the crack tip, which reduces the pressure below the initial line pressure
p0 to some proportion of it which depends essentially on crack velocity. For a long running
crack with a constant velocity, the pressure attains a steady state value. A one dimensional
(1D) model, that has been established by comparison with previous full-scale pipe fracture
data [11] is given by:
  γ2γ−1
2 γ − 1 ȧ
p1 = p 0 + ȧ < C0 ,
γ + 1 γ + 1 C0
p1 = p0 ȧ ≥ C0 , (3)
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 273

Figure 3. Pressure transducer record for CSM (Centro Sviluppo Materiali, Italy) full scale steel pipe burst test
(from Ref. [1]).

where γ is the specific heat ratio of the gas, C0 is the speed of sound in the gas at line condi-
tions, p0 is the initial line pressure, and ȧ is a steady-state crack speed. For the present case
where the pipes are pressurized by natural gas, the values γ = 1.4, and C0 = 330 m s−1 are
used. The measured steady state crack pipe pressures from the full scale burst test experiments
were found to agree well with Equation (3). However, the remaining gas pressure behind the
crack tip does not collapse to zero instantly, but rather, it decays steadily over a characteristic
length because it can now escape through the opening breach. Within this outflow region, the
gas continues to exert pressure on the walls, resulting in the flaring of the pipe wall. Figure 3
illustrates pressure measurements in a pipe section during the propagation of a crack [1] (from
Centro Sviluppo Materiali (CSM), Italy).

4.2. D ECAY FUNCTIONS FOR GAS PRESSURE BEHIND CRACK TIP

It has been mentioned that PFRAC is a fully coupled code that integrates the three-way inte-
gration of the gas, structure and fracture dynamics processes involved in pressure boundary
crack propagation into a single computational tool. Because of the high level of interaction
and the large number of elements and nodes, the interaction algorithm is possibly more time
consuming than the actual finite element or finite difference calculations themselves. Due to
the computational expense of the fully coupled simulations, a simplification is to replace the
gas dynamic analysis by a curve fit pressure distribution on the pipe walls obtained from ex-
perimented data. To demonstrate this, a curve fit is generated for the transient pressure profiles
measured in the full-scale burst test data of Fig. 3. This set of relations allows an approxima-
tion of the axial and circumferential pressure profiles on the pipe walls behind the crack tip as
a function of crack position and initial line pressure. These approximations have been included
in PFRAC to carry out the simulations of the gas pressure decay behind the crack; e.g. expo-
nentially decaying Fourier series or parabolic decay function. Thus, the shell fracture model
can be uncoupled from the fluid dynamic model. The relatively close agreement between the
coupled and uncoupled computations [2] demonstrates that the fluid dynamics computation
274 Z. Zhuang and P.E. Donoghue

Figure 4. Comparison of computed crack driving force with crack propagation distance between the linear decay
function and the exponentially decaying Fourier series.

can be replaced by an equivalent moving boundary condition specification to model the shell
fracture in many cases.
A very simple and easy to implement function is a linear decay model. When PFRAC is
applied to the analysis of RCP in a pressurised pipe, linear decay behind the crack is assumed
from the crack tip pressure, p1 , to atmospheric pressure over a decay length L. Although
there are the circumferential pressure variations on the pipe walls behind the crack in the
experiments as shown in Fig. 3, sufficiently accurate results are obtained using the linear
decay model with the assumption of no circumferential pressure variation. The linear decay
function is
 z
p(z) = p1 1 − z < L, (4)
L
where p1 is the crack tip pressure from Equation (3) and z is the position of the cross
section to be calculated. When the crack speed is 175 m s−1 , in a 0.25 m diameter, SDR11
pipe and 2.0 D decay length, comparison of computed G is made between the linear decay
function and the exponentially decaying circumferentially varying Fourier series, as shown in
Fig. 4. According to this figure, there is very good agreement between both numerical pressure
decay models. Specifically, the steady state values are virtually the same in both cases and this
is generally the quantity of greatest interest. In addition, results from linear decay represent
the more steady values than the results from exponentially decaying Fourier series. Because of
the simplicity, accuracy and efficiency of the linear decay, it has been used in the subsequent
calculations.

5. Evaluation of crack driving force

Since the steel pipe fracture events are accompanied by extensive yielding of the material, an
elastic-plastic measure of energy release rate had to be devised. The more elementary crack
tip parameters such as the stress intensity factor and the path independent J integral are not
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 275

strictly valid under these conditions. A more advanced form of the J integral, referred to as the
T ∗ integral [12] is more appropriate here because it is valid for dynamic crack propagation in
elastic-plastic materials. Another option is based on the energy outflow from a finite element
model through the work done by the nodal force release. This method is very simple and is
easy to implement. Thus, it is adopted in PFRAC. Since the nodal force release method is
suitable for a wide range of materials, it can also be used in PE pipes.
In a typical generation phase computation, the PFRAC input includes the crack position as
a function of time. Thus the amount of crack advance for each time step is known. During the
time stepping procedure that is used, several increments are required for the crack to traverse
a single element. Since the finite element is only restrained at the corners (four node element),
the propagation of the crack along an element is simulated by the incremental release of a
single nodal force. The energy flow to the crack tip per unit area of crack extension, G, is
calculated by the code. This is approximated by calculating the work done by the restraining
force as follows:
Z t¯+1t
2
G= F v̇n dt, (5)
hL̄ t¯
where h is the wall thickness, L̄ is the length of the element in the direction of propagation
and 1t is the time taken for the crack to propagate through the element and v̇n is the nodal
velocity normal to the direction of crack extension. In Equation (5), F is the force at the node
being released which is taken to vary linearly with the position of the crack on the element
boundary. It is expressed as follows
 
a(t)
F (t) = F0 1 − , (6)

where F0 is the force at that node just before the node release commences and a(t) is the
distance the crack has traveled along the element.
A typical computational result for the G variation with propagation distance (or time) is
given in Fig. 5. In this computation, for SDR11 PE pipe with pressure 0.311 MPa, diameter
0.25 m and decay length 2.0 D, the crack is assumed to propagate from a zero length crack at
a constant velocity of 175 m s−1 . After crack initiation, G rises until steady state conditions
are achieved. After this, G remains relatively constant. For the transmission and distribution
pipelines, this value is referred to as the steady-state driving force. The small perturbations
are related to the interaction of element mesh and the release of nodal force.
The crack velocities corresponding to the steady state values of G with the decay lengths
1.5 D, 2.0 D and 3.0 D are plotted in Fig. 6. The values of G, plotted as a function of steady
state crack velocity, exhibit the peaks at an intermediate speed. These peaks are referred to
as Gmax corresponding to the different decay lengths. For slower crack speeds, G is assumed
to have been reduced by decompression and rapid outflow behind the crack. However for
faster crack speeds, G has been depleted by inertial effects which divert pressure work into
kinetic energy. This provides justification for the presence of a plateau. From this plot, it is
also illustrated significant effect of the decay length to the values of G.
The pipe wall deformation will now be investigated. During crack propagation in the pipe,
the wall opens up as the flaps move out. This is illustrated more clearly in Fig. 7, where
computational results are presented for several pipe locations at a given instant of time. This
illustrates the wall deforming from its initial circular configuration to an ovalised shape as
276 Z. Zhuang and P.E. Donoghue

Figure 5. Crack driving force variation with crack propagation distance during RCP in a PE pipe (p = 0.311 MPa,
D = 0.25 m, SDR11, L = 2.0 D, ȧ = 175 m s−1 ).

Figure 6. Profiles of steady state crack driving force with constant crack velocities (p = 0.311 MPa, D = 0.25 m,
SDR11, PE pipe).

the crack passes. Behind the crack plane, considerable pipe wall deformation is evident as the
flaps move out.
From PFRAC parameter studies, variations of G with crack velocity are calculated for
the pipe configuration to find the value of crack velocity corresponding to Gmax . It has been
demonstrated that when the internal pressures are changed the value of crack velocity corre-
sponding to Gmax is not changed. This point is illustrated in Fig. 8 for SDR11 and 0.25 m
diameter pipe. It indicates that when the crack velocity is 175 m s−1 , the peak values of crack
driving force will be obtained for different pressures for a decompression length of 2 D. At
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 277

Figure 7. Deformed cross sections of cracked PE pipe, computed using PFRAC.

Figure 8. Computed results showing that when the pressure is changed the peak value location of the maximum
G corresponding to crack velocity is not changed.

this velocity, the steady state values of G are repeatedly calculated for various pressures. The
value of pressure which gives Gmax = Gd is the critical pressure pc [13].
For analysis of the size effect on dynamic fracture event in a pipeline, it is interesting to
determine the location of the peak value of G for different pipe sizes. Figure 9 illustrates
the velocity at which Gmax occurs for different wall thicknesses (SDR). This result is for PE
pipe with a diameter of 250 mm and a pressure of 0.311 MPa. The crack velocity at which
Gmax occurs decreases with increasing SDR (diameter/thickness). If the diameter of the pipe
278 Z. Zhuang and P.E. Donoghue

Figure 9. The locations of Gmax are calculated during crack propagation in the PE pipe according to crack velocity
and SDR.

remains constant, the crack velocity at the Gmax location decreases with decreasing the wall
thickness of the pipe. This is a very useful solution for the Gmax computation.

6. Energy balance approach

Another approach for the calculation of G in fracture mechanics is the energy balance method.
This has also been incorporated into PFRAC and used in the calculation of the pipe fracture
events [7].

6.1. E NERGY VARIATION IN CRACK PROPAGATION PIPELINES

The important issue is to investigate the energy variation in the pipe during crack propagation.
By way of example, for the SDR11 PE pipe with 10 m length, 0.25 m D (diameter), 0.311 MPa
pressure and 2.0 D decay length, four separate analyses are considered each with a different
steady state crack velocity 100, 150, 200 and 300 m s−1 . The results of energy variation with
crack propagation time are given in the Figs. 10–13 respectively.
The crack is assumed to start at time = 0.007 s and propagates at constant velocity. During
this time, the external work and internal energy increase quickly. The approximate constant
slopes of the external work variation corresponding to the different velocities are measured
from each figure. The internal energy consisting of strain and kinetic energy, also increase
in a similar fashion. For the intermediate crack velocity 150 m s−1 , as shown in Fig. 11,
the differences between external work and internal energy increase during the propagation
process. The longer the crack propagates, the greater the difference becomes. For the faster
crack velocity 200 m s−1 , as shown in Fig. 12, both external work and internal energy are
close to each other and are in very good agreement at a crack velocity of 300 m s−1 , as shown
in Fig. 13. For the slow crack velocity, as shown in Fig. 10, the energy quantity is much lower
than those at intermediate and faster crack velocities. This reduction is because of the internal
pressure decompression and the rapid outflow during the relative slow crack propagation in the
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 279

Figure 10. Energy variation during crack propagation in the pipe (SDR11 PE pipe, 0.25 m diameter, 0.311 MPa
pressure, 2.0 D decay length and 100 m s−1 crack velocity).

Figure 11. Energy variation during crack propagation in the pipe (SDR11 PE pipe, 0.25 m diameter, 0.311 MPa
pressure, 2.0 D decay length and 150 m s−1 crack velocity).

pipe. Note that the magnitudes of the internal energy and external work increase dramatically
as the velocity increases. However, the slopes of the external work and the internal energy are
approximative constant in each case. The kinetic energy can be ignored before crack initiation.

6.2. E NERGY BALANCE APPROACH

The methodology using energy balance to calculate G in the fracture component has been
established. The global energy balance together with an associated crack extension criteria
280 Z. Zhuang and P.E. Donoghue

Figure 12. Energy variation during crack propagation in the pipe (SDR11 PE pipe, 0.25 m diameter, 0.311 MPa
pressure, 2.0 D decay length and 200 m s−1 crack velocity).

Figure 13. Energy variation during crack propagation in the pipe (SDR11 PE pipe, 0.25 m diameter, 0.311 MPa
pressure, 2.0 D decay length and 300 m s−1 crack velocity).

has been used to evaluate G


 
1 dUe dUi
G= − , (7)
h da da
where the internal energy Ui = Us + Uk . Thus, G can be computed by the external work
Ue done on the pipe, the internal strain energy Us and the kinetic energy Uk during a crack
length increase of da in the pipe. Since each term in the bracket on the right-hand side of
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 281

Figure 14. Crack driving force calculated using energy balance method at different crack velocities (for decay
length=2.0 D, pipe length = 10 m, p = 0.311 MPa, D = 0.25 m, SDR11 PE pipe).

Equation (7) is an energy differential, it can be approximated numerically as follows


1
G= (1Ue − 1Ui ), (8)
h 1a
where 1U represents the incremental value of the energy in the pipe during a crack length
increase of 1a.
A series of computations has been carried out using the energy balance method to obtain
the driving force G. For the SDR11 PE pipe with 0.25 m diameter, 0.311 MPa internal
pressure, 2.0 D decay length and 10 m pipe length, results for steady state G at different
steady state crack velocities are calculated by the energy balance equation (8) and illustrated
in Fig. 14.
Figure 14 exhibits the peak values of G at an intermediate speed. This can also be illus-
trated by energy variation at these crack speeds with the differences between external work
and internal energy increasing during the crack propagation as profiled in Fig. 11. Results are
not surprising since the biggest difference in slopes occurs at crack velocities 150–180 m s−1 ,
while slopes are relatively similar at very low and very high speeds. For the high crack speed,
as shown in Fig. 12, the differences between external work and internal energy are very small
and are virtually identical at a crack speed of 300 m s−1 (Fig. 13). There is also no surprise
that very small differences in slope are obtained at crack velocities greater than 200 m s−1 .
As shown in Fig. 10, a similar trend is evident at low crack speeds. Thus, low values of G are
achieved for the high and the low velocities, which also can be seen in Fig. 14.
The comparison of G calculated by nodal force release and energy balance is also exhibited
in Fig. 14, and reasonable agreement is observed. Better results are achieved with more refined
meshes for 6400 elements (25 × 25 mm) than 4000 elements (40 × 25 mm). In particular,
the values of G are very close to each other along with at crack speeds in the region 100–
220 m s−1 . These speeds occur usually in the PE gas pipelines during the RCP experiments
[13].
282 Z. Zhuang and P.E. Donoghue

In the computation, the energy differential in one element is quite sensitive if the element
length is very short in the crack propagation direction. Because G is determined from the
subtraction of large numbers, any small variations from large numbers will lead to a significant
influence on the computations. These small perturbations related to the interaction of element
mesh and the release of nodal force have been demonstrated in Fig. 5. If 1a is taken as a few
elements in length, the sensitivity is less critical and better results are obtained. Given that Ue
and Ui increase monotonically, a relatively long length 1a is permitted (e.g. 1a = 0.1 m is
used in the computation).
Calculation of G using the energy balance method in Equation (8) and nodal force release
method in Equation (5) should be in agreement if the solution is stable and energy is con-
served. A number of disparities have been observed as evident by comparison of the results
between these methods in Fig. 14, particularly at very low and very high velocities. Although
the results from both methods are not in great agreement, the peak values of G are very close
to each other along with the corresponding crack velocities.

7. A simulation model for crack arrestor in gas pipeline

Mechanical arrestors are frequently used in the gas transmission pipelines to prevent the
possibility of long running crack. These arrestors have the effect of reducing the pipe from
opening as the crack passes. This decreases the available crack driving force and, as a result
crack arrest may take place. A novel analysis methodology has been developed to simulate
these fracture events and to determine the suitability of crack arrestor configurations.

7.1. C URVED BEAM ELEMENT FOR PFRAC

The basic idea to simulate pipe arrestor in PFRAC is necessary to have a suitable curved
beam element. While the PFRAC library contained a number of element types, there was
no curved beam element available. Accordingly, a two-node curved beam element, based
on the approach of Ref. [14] was selected for insertion into PFRAC. This beam element
formulation that accounts for initial curvature and moderate variations of rigid body rotation is
outlined. However, as this way only a two-dimensional formulation, it was not directly suited
to simulate an arrestor, as out-of-plane deformation can occur during pipe fracture.
Accordingly, it is necessary to modify the curved beam approach [14] using the technique
in Ref. [15]. The latter involves a large displacement, small strain transient analysis for space
frames. Thus, the two-dimensional procedure is extended to the curved beam element with
initial in-plane curvature and six degrees of freedom at each node of the element. Contact
conditions are also implemented to simulate pipe/arrestor interaction. The expressions for the
dynamic three-dimensional curved beam element are now described in detail. These have been
successfully incorporated into PFRAC.

7.2. G OVERNING FORMULATIONS FOR CURVED BEAM

In addition to the global coordinate system, each element is described in terms of a set of
local element coordinates (x̂, ŷ, ẑ) as shown in Fig. 15. In this figure, the thick curved line
from 1 to 2 represents a single beam element. In relation to Fig. 16, the ring arrestor that goes
around the outer circumference of the pipe is split up into several curved beam elements. A
distinct coordinate system is associated with each element as indicated by a superposed hat.
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 283

Figure 15. The 3-dimension curved beam element.

Figure 16. Schematic diagram of a ring arrestor.

The element coordinate system is defined such that the x̂ axis always remains coincident with
a line joining the end points of the element. In addition, each node has a nodal coordinate
system and this is indicated by a superposed bar.
The translational displacement ûI J and rotational displacement θ̂I J of the element (I
representing x̂, ŷ and ẑ components and J representing the original position of node J )
are additively decomposed into rigid body displacements and deformation displacements.
Given any displaced configuration of the beam, the orientation of the element coordinate is
specified because x̂ axis always connects the end points of the beam. The translation and
rotation which are needed to bring the coordinate system into this orientation are the rigid
body displacements. Remaining displacements are the deformation displacements.
Velocities and accelerations of the beam denoted by u̇I J and üI J at the nodes, respectively,
where superposed dots denote time derivatives. The angular velocities and angular accelera-
tions are treated in nodal coordinate components, ω̄I J and ᾱI J , respectively. The forces and
moments at the nodes are similarly denoted by FI J and MI J , respectively.
The equations of the motion for the system are written separately for the translational and
rotational degrees of freedom. Let FIint ext
J and FI J be the sum of all internal and external nodal
forces, respectively, associated with node J . The three translational equations are:

mJ üI J = FIext
J − FI J
int
(no sum on J, I = x, y, z), (9)
284 Z. Zhuang and P.E. Donoghue

where mJ is the translational mass of the node J . Let M̄Iint ext


J and M̄I J be the sum of all internal
and external moments, respectively, expressed in terms of node coordinate components of
node J . The three rotational equations are the standard Euler equations

IxJ ᾱxJ + (IzJ − IyJ )ω̄zJ ω̄yJ = M̄xJ


ext
− M̄xJ
int

IyJ ᾱyJ + (IxJ − IzJ )ω̄xJ ω̄zJ = M̄yJ


ext
− M̄yJ
int
(no sum on J ) (10)
IzJ ᾱzJ + (IyJ − IxJ )ω̄yJ ω̄xJ = ext
M̄zJ − int
M̄zJ

where II J are the inertia moments. The nodal displacement matrix is expressed as

 
{d̂}T = 1̂12 θ̂z1 θ̂z2 θ̂y1 θ̂y2 θ̂x12 , (11)

where 1̂12 = elongation of element, θ̂z1 , θ̂z2 , θ̂y1 , θ̂y2 = bending deformation rotations, θ̂x12 =
torsional deformation.
The analysis is now specialised to the curved beam with the displacements in the axial
direction, ûx , which is described by linear shape functions. In contrast, the transverse and
anti-plane displacements, ûy and ûz , respectively, are described by cubic shape functions.
Thus, the deformation displacement field due to the nodal displacement {d̂} in Equation (11)
is given by
 m

 û 
 ûxm 

y
=

 û 
m
 z  
θ̂x
 
ξ 0 0 0 0 0
 0 L(ξ 3 − 2ξ 2 + ξ ) L(ξ 3 − ξ 2 ) n o
0 0 0
= 0
 d̂ , (12)
0 0 L(−ξ 3 + 2ξ 2 − ξ ) L(−ξ 3 + ξ 2 ) 0 
0 0 0 0 0 ξ
where the superscript ‘m’ is used to indicate that these are the displacements of the mid-
surface, the non-dimensional variable ξ is given by

ξ= . (13)
L
The axial length between nodes 1 and 2 of the beam element is L.
Warping of the cross-section due to torsion is neglected. If the usual Euler–Bernoulli beam
assumptions are imposed such that normals to the mid-line remain straight and normal, the
deformation at any point of the cross section of the beam is given in terms of the mid-line
displacements as
∂ ûm
y ∂ ûm
ûx = ûm
x − ŷ − ẑ z
,
∂ x̂ ∂ x̂
ûy = ûm
y − ẑ θ̂ x,

ûz = ûm
z + ŷ θ̂x . (14)
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 285

After some manipulation, the strain displacement equation now can be written in matrix
form as
 

 1̂12  
 
   1 
 θ̂ 


 ˆx  η1 η2 η 3 η4 0 

z1 

L θ̂
2ˆxy =  0 η5 η6 0 0 −ẑ  z2
, (15)
  
 θ̂y1 
2ˆxz 0 0 0 −η5 −η6 ŷ 
 


 θ̂y2 


 
θ̂x12

where the functions ηk (k = 1–6) are given by



η1 = − (6ξ − 4) + ω̂0 (3ξ 2 − 4ξ + 1),
L

η2 = − (6ξ − 2) + ω̂0 (3ξ 2 − 2ξ ),
L

η3 = (6ξ − 4),
L

η4 = (6ξ − 2),
L
η5 = (3ξ 2 − 4ξ + 1),
η6 = (3ξ 2 − 2ξ ) (16)

and the initial relative rotation is defined by

∂ û0y
ω̂0 = , (17)
∂ x̂
where û0y is the initial distance between the curved mid-line of the beam and the x̂ axis.
Once the strains of the beam have been determined, the stresses can be computed by the
constitutive law. By using a kinematics variational principle, the expressions for the deforma-
tion nodal forces are obtained. The resulting nodal forces are then summed to give the total
internal forces and moments [16].

7.3. C ONTACT CONDITIONS

In the present approach, the arrestor simulation with curved beam elements can be considered
as a set of rings with a small clearance from the outer wall of pipe (see Fig. 16). The nodes
of the beam element are ordered by ring numbers and their positions in the ring. At each time
step, the initial PFRAC calculation assumes that the arrestor is not present. The procedure
then is to advance down the pipe in the axial direction checking contact with arrestor in turn.
When the crack propagates, the radius of the pipe wall, Rp , will change its magnitude owing
to the deformation of the wall. Contact takes place when the new location of Rp , is large than
the radius of the arrestor, Ra . Because of the arrestor restraint, the pipe cannot extend beyond
the arrestor location. Thus, the contact conditions between the pipe wall and the arrestor,
along with the various computational procedures are developed and illustrated in Table 1.
This process is repeated every time step that the interaction routine is implemented.
286 Z. Zhuang and P.E. Donoghue
Table 1. Computational procedure for the pipe wall deformation limited by crack
arrestor

When the contact occurs, the condition at the node is that the new accelerations are calcu-
lated by the equations of motion, which include the forces and masses contributed by the pipe
and the arrestor. The accelerations üa of the arrestor are given by

(fp + fa )ext − (fp + fa )int


üa = , (18)
mp + m a
where the subscripts p and a represent the pipe and arrestor, respectively. The velocities u̇a
and displacements ua of the arrestor can then easily be calculated by finite difference. In
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 287

Figure 17. Pipe cross-section deformations one diameter behind the crack tip for steel pipe: (a) with arrestor; (b)
without arrestor.

turn, the velocities and accelerations of the pipe have to be changed based on the modified
displacements.

7.4. A RRESTORS FOR STEEL PIPES

The significant effect of arrestors is to reduce the driving force during crack propagation.
The condition under which a crack arrests in a gas pipeline is that G for the pipe/arrestor
interaction is less than Gd of the pipe. The program developed for the simulation of the curved
beam element as an arrestor has been incorporated into PFRAC for the purposes of evaluating
G. Usually, an entire steel ring is used as the arrestor device.
An example of a ring arrestor application for a steel gas transmission pipeline is also
presented here. In this instance, a steady-state crack speed of 200 m s−1 is imposed on a
1.42 m diameter pipe, 18 mm wall thickness, under an internal pressure of 10.45 MPa. The
deformations of the pipe cross-section one diameter behind the crack tip are illustrated in
Fig. 17 for (a) the pipe with an arrestor; and (b) the pipe without an arrestor. For the steel
arrestor with cross area A = 12 cm2 and clearance 25 mm is used in this computation. The
trend is similar to the PE pipe example with flaring behind the crack being limited by the
arrestor as expected.
The crack driving force, G, is also significantly influenced by the arrestor. This is demon-
strated in Fig. 18 where the G is plotted as a function of the crack propagation distance.
The driving force decreases rapidly due to the presence of the arrestor. For example, if the
fracture toughness is 30 MPa − m, then crack arrest takes place. Figure 19 illustrates the
effects of different sized arrestors on the driving force. For an unstiffened pipe, the steady-
state G is 53 MPa − m. As expected, reductions in this value are more pronounced as the size
of the arrestor is increased. In the present implementation, the arrestor is treated as an elastic
material. When the yield stress is exceeded, the arrestor is allowed to break and clearly the
arrestor is no longer effective in this case.
288 Z. Zhuang and P.E. Donoghue

Figure 18. Profiles of crack driving force variation with propagation distance for the steel pipe (a) with arrestor,
(b) without arrestor.

Figure 19. Influence of different size arrestors on steady-state crack driving force.

8. Discussion and Conclusions

In these paper, several dynamic analysis issues are developed relating to crack propagation
and arrest in gas pipelines. Various aspects of the theory and the resulting applications of
the PFRAC code for the simulation of axial crack propagation events in gas pipelines are
discussed in greater detail.
Due to the computational expense of the fully coupled simulations, a simplification is to
replace the gas dynamics analysis by a curve fit pressure distribution, obtained from experi-
mental data, on the pipe walls behind the crack tip. Because of the simplicity, accuracy and
efficiency of the linear decay, it has been used in the PFRAC calculations.
Analysis Methodology for rapid crack propagation and arrest in gas pipelines 289

If the external work, strain and kinetic energy are calculated numerically for the crack
propagation in the pipeline, the energy balance equation can be used in the evaluation for the
crack driving force. The G calculated by the energy balance method and nodal force release
method are in reasonable agreement.
Arrest frequently occurs in the pipeline when the crack encounters an arrestor device on the
pipe, because the crack driving force, G, drops below the fracture toughness of material, Gd .
The analysis procedure allows for the consideration of steel rings as arrestors to limit crack
propagation in gas pipeline. It can also be conveniently used to establish guidelines on the use
of a ring arrestor, the space between the arrestor and the pipe wall, the size and material of the
arrestor.
The arrestors are placed at periodic intervals along the length of the pipe. These intervals
are determined by the industrial and safety requirements of limiting the extent of crack prop-
agation (say a few hundred meters). It should be clear that there is no interaction between the
arrestors. Thus, each must be capable of acting independently to arrest the crack. The model
of arrestors described here can allow for set of rings that are closely spaced. This has been
implemented in a model for a PE pipe fracture test that has a cage around the outside to prevent
excessive deformation [13]. A further application of the methodology involves the application
to circumferential stiffeners (tear straps) in aircraft fuselages which are also designed as crack
arrestors [17].
As the time frame of the test and analysis is relatively very short (less than 0.05 s) for
the PE pipe, a rigorous viscoelastic material model is not required as the compliance will not
change significantly. An example is a Small Scale test for the PE pipe with 2 or 3 m length.
If the crack speed is 100 m s−1 , it takes only 0.02 s or 0.03 s to propagate to the end of the
pipe [18]. However, strain rate effects are significant and a strain rate material model must be
included to capture the high rates that are exhibited near the crack tip. These will be involved
in the late research work.

References

Buzzichelli G. and Kanninen, M.F. (1987). The Development of a Ductile Pipe Fracture Model, Joint CSM and
SwRI report to the American Gas Association under AGA PRC contracts PR 182–527 and PR 15–526.
Kanninen, M.F., O’Donoghue, P.E., Green, S.T., Freitas, C.J. Morrow, T.B., Dexter, R.J. and Bowles, P.K. (1989).
The Development and Verification of Dynamic Fracture Mechanics Procedures for Flawed Fluid Containment
Boundaries, Final Report, 06-9452, Southwest Research Institute, San Antonio, Texas.
Greig, J.M. (1985). Fracture Propagation in 250 mm and 315 mm Polyethylene Gas Pipes, British Gas Engineering
Research Station Report No. E472.
Kanninen, M.F., O’Donoghue, P.E., Cardinal, J.W., Green, S.T. Curr, R.M. and Williains, J.G. (1989). A Fracture
Mechanics Analysis of Rapid Crack Propagation and Arrest in Polyethylene Gas Pipes, In: 11th Plastic Fuel
Gas Pipe Symposium, American Gas Association, pp. 334-343.
Vanspeybroeck, P. (1990). Evaluation of the Correlation between Instrumented Charpy V Tests and the Mod-
ified Robertson Test for Characterization of Their Resistance to Rapid Crack Propagation in Pressurized
Polyethylene Pipes, Proceedings of Pipeline Technology Conference, Belgium.
Vanspeybroeck, P. (1992). New Test Methods to Evaluate the Resistance to Rapid Crack Propagation in
Pressurized Polyethylene Pipes, Proceedings of Plastics Pipes Eight, D1/6, the Netherlands.
Zhuang, Z. (1995). The Development of Finite Element Methods for the Investigation of Dynamic Crack
Propagation in Gas Pipelines, Ph.D. thesis, University College Dublin, Ireland.
O’Donoghue, P.E., Green, S.T., Kanninen, M.F., Bowles, P.K. (1991). The development of fluid/structure inter-
action model for flawed fluid containment boundaries with applications to gas transmission and distribution
piping. Computers and Structures 38 (5/6), 501–513.
290 Z. Zhuang and P.E. Donoghue

O’Donoghue, P.E., Kanninen, M.F., Lueng, C.P., Demofonti, G. and Venzi, S. (1997). The development and
validation of a dynamic fracture propagation model for gas transmission pipelines. International Journal of
Pressure Vessel and Piping 70, 11–25.
Irwin, G.R. and Corten, H.T. (1968). Evaluating the Feasibility of Basing Pipeline Operating Pressure on In-Place
Hydrostatic Test Pressure, Report to Northern Natural Gas Company and EI Paso Natural Gas Company.
Maxey, W.A. (1974). Fracture initiation, propagation and arrest In: Proceedings of the 5th Symposium in Line Pipe
Research, American Gas Association, Houston, USA, J1–J31.
Kanninen, M.F. and Popelar, C.H. (1985). Advanced Fracture Mechanics, Oxford University Press, New York.
Zhuang, Z. and O’Donoghue, P.E. (1997) Material fracture toughness determination for polyethylene pipe
materials using small scale test results. ACTA Mechanics Sinica 13: 63–80.
Belytschko, T. and Glaum, L.W. (1979). Applications of higher order corotational stretch theories to nonlinear
finite element analysis. Computers and Structures 10, 175–182.
Belytschko, T. and Schwer, L. (1977). Large displacement, transient analysis of space frames. International
Journal for Numerical Methods in Engineering 11, 65–84.
O’Donoghue, P.E. and Zhuang, Z. (1999). A finite element model for crack arrestor design in gas pipelines. Fatigue
and Fracture of Engineering Materials and Structures 22 (1), 59–66.
Kanninen, M.F. and O’Donoghue, P.E. (1995). Research challenges arising from current and potential applications
of dynamic fracture mechanics to the integrity of engineering structures. International Journal for Solids
Structures 32 (17/18), 2423–2445.
Yayla, P. and Leevers, P.S. (1989). A New small scale test for rapid crack propagation. In: 11th Plastic Fuel Gas
Pipe Symposium, American Gas Association.

View publication stats

You might also like