Você está na página 1de 7

Analytical solution for upheaval buckling in buried pipeline

Fausto Andreuzzi
a,
*
, Arcangelo Perrone
a
a
Via XXIV Maggio 80/A, F-0046 Grottaferrata Rome, Italy
Received 25 January 2000
Abstract
This work deals with the evaluation of the conditions that generate the local buckling in a buried pipeline due to a temperature
gradient across its wall. The transport pipelines are often subjected to thermal gradients since the uid temperature inside the pipe is
greater than the external one. The minimal value of internal wall temperature depends on the physical properties of the inner uid and
thermal exchanges toward the surrounding environment.
The present analysis has been performed for applications to submarine pipelines end supported by S.I.L.MAR. s.r.l. in a basic
design study for pipelines in the North Sea. The problem here presented concerns the calculation of the limit elastic instability axial
load in a rectilinear pipe whose ends are restrained and on which acts a constant frictional force exerted by the surrounding soil. The
condition for the local buckling of the beam leads then to the comparison between the critical eulerian load for elastic instability with
friction and the compressive forces acting at the extremities contrasting the axial deformation of the pipe. The mathematical model
here presented allows the analytical evaluation of this critical load and then is particularly interesting in cases of rapidly varying axial
load for high frictional forces where other methods (nite elements, nite dierences) can give rise to signicant errors due to dis-
cretization related to the modelling of the varying axial compressive force inside the elements. Finally, we hope that students and
researches do not forget the objective of doing good to humanity by using human mind's conquests such as the analytical
methods. 2001 Elsevier Science B.V. All rights reserved.
Notations
E Young's modulus of pipe material
I Moment of inertia with respect to the neutral axis
A Area of steel in pipe cross-section
a Coecient of linear expansion
Dt Temperature dierence across the pipe wall
w Total submerged weight (pipe + soil cover) per unit length
f Friction coecient (pipe soil)
N Axial compressive load acting at the pipe restrained rst end
l Pipe length within the restrained ends
F = wf Frictional force per unit length
L = N=F Anchoring length
x Longitudinal coordinate along the pipe
y(x); z(x) Pipe deection at x
s
0
Adimensional threshold parameter
s Adimensional characteristics variable
www.elsevier.com/locate/cma
Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087
*
Corresponding author. Tel.: +39-06-94315876; fax: +39-06-94315877.
E-mail address: silmar@hurricane.it (F. Andreuzzi).
0045-7825/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 0 4 5 - 7 8 2 5 ( 0 0 ) 0 0 3 6 6 - 2
c Adimensional free boundary parameter
s
e
Adimensional eulerian parameter
s
c
Adimensional critical parameter
J
m
(t) Bessel function of rst kind order m
s
l;m
(t) Lommel's s function order m
Ji
c
(t) Primitive of J
m
(t) integral
si
l;m
(t) Primitive of s
l;m
(t) integral
1. Simplied mathematical model
In [1] it is discussed and solved the problem concerning the evaluation of the limit elastic instability axial
load in a vertical beam subjected to its own weight and simply restrained at the lower end. In this case, the
longitudinal tension N increases linearly along the beam axis starting from the value N = 0 at the free upper
extremity. In our present problem the axial tension decreases linearly from an unknown value (critical load)
at the rst restrained end of the beam through the frictional force exerted from the soil which contrasts the
axial deformation due to temperature difference. The compressive load acting at one of the pipe extremities
is
P = EAaDt: (1:1)
The dierential equation determinating the statical equilibrium conguration of the beam is
EIy
IV

d
dx
(T(x)y
/
) = 0; (1:2)
y
/
= dy=dx and so for other derivatives with the boundary conditions
y(0) = y
/
(0) = y(l) = y
/
(l) = 0; (1:3)
and being T(x) = N Fx.
(1.2) and (1.3) constitute a linear fourth-order dierential homogeneous problem with variable coe-
cients and determine an eigenvalue problem, i.e., nd the non-trivial solutions y(x) (eigenfunctions) asso-
ciated to values of N (eigenvalues) which satisfy the differential equation and the boundary conditions. It is
known that the critical load is the lowest positive eigenvalue. The simplied mathematical model here
analyzed evaluates the frictional force per unit length as F = wf .
2. Analytical solution of the proposed mathematical model
We observe the (1.2) holds only if T(x) is a not-negative function and then if l < L. In the interval
L < x < l we must assume T(x) = 0 (no compressive force, no contrasting frictional force). Then the correct
problem to be solved can be stated as: nd the lower eigenvalue N and associated L(N) such that
EIy
IV
(T(x)y
/
)
/
= 0; 0 < x < L(N);
y(0) = y
/
(0) = 0:
(2:1)
EIz
IV
= 0; L(N) < x < l;
z(l) = z
/
(l) = 0;
(2:2)
and with the equilibrium and congruency condition at x = L(N)
y
///
= z
///
; y
//
= z
//
; y
/
= z
/
; y = z: (2:3)
5082 F. Andreuzzi, A. Perrone / Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087
It is suitable to specify that the fourth derivative is discontinuous at x = L, where we have y
IV
z
IV
= Fy
/
.
The discontinuity is originated from the discontinuity at the same point of the rst derivative of the
function min[T(x); 0[ which represents the axial force within the integration interval (0; l).
In the following exposition we will indicate (2.1) as (I), (2.2) as (II), (2.3) as Interface.
For any given N and L(N)
(I) Assume s = 1 x=L.
Writing (2.1) as y
IV
N=EI
+
(sy
/
)
/
= 0; y(0) y
/
(0) = 0.
By integrating once (2.1) with respect to the x variable, changing the independent variable x in s and
putting u(s) = y
/
we obtain
u
//
a
2
su = a (note that y
///
= a=L)
with a
2
= N
3
=(F
2
EI):
(2:4)
Then introducing the new variable g = ss
1=3
; s = 2a=3 we have
g
2=3
u
//
_

1
3
g
1
u
/
u
_
= 4 aL
3
9s
4=3
_ _ _
;
where s = (2=3)(N
3
=(F
2
EI))
1=2
is the adimensional characteristic variable.
By substituting the new dependent so dened function w
u g ( ) = g
1=3
w(g);
the previous dierential equation becomes
g
1=3
w
//
g
2=3
w
/
g
1=3
_

1
9
g
5=3
_
w = ag
2=3
; a =
4 a
9s
4=3
;
by multiplying all the members by g
5=3
we nally obtain
g
2
w
//
gw
/
g
2
_

1
9
_
w = ag; (2:5)
which is an inhomogeneous Bessel equation.
A general solution of (2.5) is given by
w g ( ) = AJ
1=3
g ( ) BJ
1=3
g ( ) as
0;1=3
g ( ); (2:6)
where J
m
(g) is the rst kind Bessel function of m order and s
l;m
(g) is the Lommel's s function, a particular
solution of the inhomogeneous Bessel equation
g
2
w
//
gw
/
g
2
_
m
2
_
w = g
l1
:
The use of the Lommel's s
l;m
(g) indeed S
lm
(g) is obligatory in our problem because S
lm
(g) diverges at
g = 0, where the general solution must satisfy the interface conditions (2.3). For Bessel and Lommel's
functions, series representations, small and large arguments, asymptotic expansions for large order, re-
currence formulas, numerical computations refer to [24]. Briey, for clarity of exposition, we recall that for
small arguments
J
m
(g)
g
2
_ _
m
_
C m ( 1); s
g;m
(g) g
l1
_
l (
_
1)
2
m
2
_
; (2:7)
where C(t) is the Euler's Gamma function of argument t.
In the following we introduce the adimensional threshold parameter
s
0
=
2
3
Fl
3
EI
_ _
1=2
F. Andreuzzi, A. Perrone / Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087 5083
and the adimensional free boundary parameter
c = s
2=3
_
s
2=3
0
_
3=2
:
The boundary condition at the rst end (x = 0 ==g = s) leads to
w s ( ) = AJ
1=3
s ( ) BJ
1=3
s ( ) as
0;1=3
s ( ); y s ( ) = 0; (2:8)
moreover
y(c) = d
4
J
1=3
c; s ( )A
_
J
1=3
c; s ( )B R
1=3
c; s ( )a
_
;
d
4
= 2 3s
2=3
_ _ _
;
where
J
m
c; s ( ) = Ji
m
s ( ) Ji
m
c ( ); R
m
c; s ( ) = si
o;m
s ( ) si
o;m
c ( );
Ji
m
t ( ) =

m=0
J
m2m1
t ( ) = m ( 1)tS
1;m1
t ( ) tJ
m1
t ( )S
o;m
t ( );
si
l;m
(t) by the denition of the s function and using the recurrence formulas for Bessel functions.
(II) For a given N > 0 and L(N) = N=F > 1 we have
(II.1) At g = c the boundary conditions y(c) = w(c) = 0 must be fullled. So that we obtain
J
1=3
c; s ( )A J
1=3
c; s ( )B R
1=3
c; s ( )a = 0;
J
1=3
c ( )A J
1=3
c ( )B s
0;1=3
c ( )a = 0;
J
1=3
s ( )A J
1=3
s ( )B s
0;1=3
s ( )a = 0:
The previous equations lead to the eigenvalue characteristics equation in the unknown adimensional
characteristic variable.
(II.1.1)
d
A
J
1=3
(c; s) d
B
J
1=3
(c; s) dR
1=3
(c; s) = 0
with
d
A
= s
0;1=3
s ( )J
1=3
c ( ) s
0;1=3
c ( )J
1=3
s ( );
d
B
= J
1=3
s ( )s
0;1=3
c ( ) J
1=3
c ( )s
0;1=3
s ( );
d = J
1=3
c ( )J
1=3
s ( ) J
1=3
s ( )J
1=3
c ( ):
Note that if c = 0 it is necessary to assume B = 0 (the Bessel function J
1=3
(t) diverges in t = 0, see also
(2.7), then in this case the eigenvalue characteristics equation reduces to
(II.1.2)
J
1=3
(0; s)s
0;1=3
s ( ) R
1=3
(0; s)J
1=3
(s) = 0:
(II.2) If s
0
> s the integration interval (0; l) must be subdivided in the subintervals (0; L) and (L; l).
(II.2.1) Subinterval (0; L) ==0 < g < s. The Eqs. (2.1) and (2.8) hold with c = 0 and then
y
/
x
(s) = AJ
1=3
(s) BJ
1=3
(s) as
0;1=3
(s) = 0;
moreover
y(0) = d
4
AJ
1=3
(0; s)
_
BJ
1=3
(0; s) aR
0;1=3
(0; s)
_
:
It is necessary to evaluate y
///
(0); y
//
(0); y
/
(0), where the derivatives are performed to respect the x
variable in g = 0.
5084 F. Andreuzzi, A. Perrone / Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087
Taking the limits (note that J
1=3
(t) diverges in t = 0 but the limit values for the previous derivatives
exist) by using the J
m
and s
l;m
approximations given in (2.7), one nds
(II.2.1.1)
y
/
x
(0) = d
3
B=L; y
//
x
(0) = d
2
A=L
2
; y
///
x
(0) = d
1
a=L
3
being d
1
=
9
4
s
4=3
; d
2
= s
2=3
=(2
1=3
C(4=3)); d
3
= 2
1=3
=C(2=3):
(II.2.2)
Subinterval (L; l) ==0 < n < 1; x L = kn; k = l 1
_

s
s
0
_ _
2=3
_
;
z n ( ) =
~
A
2
1
3
n
3
_
n
2
3
_

~
B
2
n
2
_
2n 1
_
:
This function satises the dierential equation and the boundary condition (2.2). In particular we have
z 0 ( ) =
~
A
3

~
B
2
; z
/
x
0 ( ) =
1
k
~
A
2
_

~
B
_
:
(II.2.2.1)
z
//
x
(0) =
~
B
k
2
; z
///
x
(0) =
~
A
k
3
:
Interface.
By imposing the equilibrium and congruency conditions and putting f = (s
0
=s)
2=3
1, it is obtained
d
5
=
d
2
2d
4
f
2
; d
6
=
d
1
3d
4
f
3
; b
a
=
d
1
2d
3
f
2
; b
A
=
d
1
d
3
f;
a
a
= R
1=3
0; s ( ) b
a
J
1=3
0; s ( ) d
6
; a
A
= J
1=3
0; s ( ) b
A
J
1=3
0; s ( ) d
5
;
and nally the eigenvalue characteristics equation.
(II.2.2.2)
a
a
J
1=3
(s) (b
a
a
A
b
A
a
a
)J
1=3
(s) a
A
s
0;1=3
(s) = 0:
It can be observed that if s = s
0
==f = 0, then d
5
= d
6
= b
a
= b
A
= 0.
a
a
= R
1=3
(0; s); a
A
= J
1=3
(0; s);
the equation in (II.2.2.2) then reduces to
J
1=3
(0; s)s
0;1=3
(s) R
1=3
(0; s)J
1=3
(s) = 0;
which is the same as in (II.1.2).
3. Numerical computations
We have previously proved that the so-dened function U
U s ( ) =
d
A
J
1=3
(c; s) d
B
J
1=3
(c; s) dR
1=3
(c; s); s Ps
0
;
a
A
J
1=3
s ( ) b
a
a
A
b
A
a
a
( )J
1=3
s ( ) a
a
s
0;1=3
s ( ); s < s
0
;
_
(3:1)
is continuous for all s > 0 including s = s
0
.
F. Andreuzzi, A. Perrone / Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087 5085
Then the evaluation of the critical load N requires the solution of the transcendental equation.
U(s) = 0; (3:2)
by means of numerical methods and choosing the smallest positive value of the adimensional characteristics
variable s which satisfy (3.2). We can observe that if the frictional force F = 0 the limit instability load is the
rst positive root of the equation
rsin r 2 cos r ( 1) = 0;
where r
2
= Nl
2
=EI. Being this root r = 2p => N
e
= (2p)
2
(EI=l
2
). We indicate the eulerian critical pa-
rameter as the value of s at the eulerian critical load N
e
and therefore s
e
= (16p
3
=3)(EI=Fl
3
)
If F > 0 the adimensional eulerian parameter s
e
can be assumed as the lower limit value for the actual
adimensional characteristic variable in nding the rst positive root of (3.2).
Then the algorithm needs the following steps:
Determine the interval where the root lies. This can be performed by increasing progressively the value of
s starting from s
e
until the sign of the function U at the current interval extremities are opposite.
Calculate the root within this interval by means of any numerical method not requiring derivatives (for
example dichotomous search, secant method) until the required accuracy has been reached.
We denote the lower positive root of (3.2) characteristic equation as adimensional critical parameter s
c
.
In the following Table 1 we summarize the results obtained from numerical computations, where suxed
(1) and (2), are reported the solutions from the present model and a nite element model, respectively. In
this example for f.e. solution the pipe has been discretized in 10 elements with the same length and by using
a constant stepwise function (averaged over any element) to describe the varying compressive load inside
the element. We can observe that the dierence (2) (1) increases by increasing the frictional coecient f
(here used as a parameter) exhibiting a diverging behaviour for f very large. The f.e. model determinates the
lowest eigenvalue of stiffness + geometric matrix of the quadratic form representing, in the f.e. model, the
total potential energy functional. The algorithm based on the f.e. model then calculates the critical load as
the lowest value of N for which this matrix becomes singular.
The increase of approximation errors is explained by taking into account the worst f.e. approximation
due to the discontinuous rst derivative of the function [min T(x); 0] within the element where this function
vanishes. More accurate f.e. model than the constant stepwise axial force, where the shape function is
continuous with all derivatives, must be employed to take into account accurately this property of the
solution. On the contrary, the analytical model treats this characteristic correctly.
The reported results are referred to the following parameters EI = 0:9 10
6
Kn m
2
; w =
130 Kn=m; l = 100 m .
Table 1
f s
e
s
0
s
(1)
c
s
(1)
c
=s
e
s
(1)
c
=s
0
s
(2)
c
s
(2)
c
=s
e
s
(2)
c
=s
0
Ds
c
=s
(1)
c
0.20 5.08 3.80 8.43 1.65 2.23 8.43 1.66 2.22 0.0
0.25 4.06 4.25 7.46 1.83 1.75 7.46 1.83 1.75 0.0
0.30 3.38 4.66 6.82 2.01 1.46 6.82 2.01 1.46 0.0
0.35 2.90 5.03 6.37 2.19 1.27 6.38 2.19 1.27 0.0
0.40 2.54 5.38 6.05 2.37 1.12 6.05 2.38 1.12 0.0
0.45 2.26 5.70 5.80 2.56 1.02 5.80 2.57 1.02 0.0
Threshold has been reached
0.50 2.03 6.01 5.60 2.75 0.93 5.74 2.83 0.96 0.025
0.55 1.85 6.31 5.44 2.94 0.86 5.67 3.07 0.90 0.042
0.60 1.69 6.59 5.31 3.13 0.81 5.54 3.27 0.84 0.043
0.75 1.36 7.36 4.91 3.67 0.67 5.31 4.43 0.72 0.081
1.00 1.02 8.50 4.54 4.46 0.53 4.91 4.82 0.53 0.082
1.25 0.81 9.51 4.15 5.11 0.44 4.53 5.57 0.48 0.092
5086 F. Andreuzzi, A. Perrone / Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087
4. Possible extension of the proposed model
The model we have here presented can be extended to take into account the soil resistance to the lateral
deections of the beam by introducing a linear spring resistance to the beam lateral displacements, rea-
sonably increasing the value of the critical axial load and then reducing the safety requirements.
References
[1] Landau, Lifsits, in: Riuniti (Ed.), Teoria dell'elasticita, 1979.
[2] Erdelyi, et al., Higher Transcendental Functions, Mc Graw-Hill, New York, 1953.
[3] Watson, A Treatise on the Theory of Bessel Functions, Cambridge University Press, Cambridge, 1922.
[4] Whittaker, Watson, A Course of Modern Analysis, Cambridge University Press, Cambridge, 1952.
F. Andreuzzi, A. Perrone / Comput. Methods Appl. Mech. Engrg. 190 (2001) 50815087 5087

Você também pode gostar