You are on page 1of 12

Chapter 26

CHAPTER 26

RELIABILITY ANALYSIS OF TUBULAR JOINTS IN OFFSHORE


STRUCTURES 1

P. Thoft-Christensen & J.D. Srensen, Aalborg University, Denmark

ABSTRACT
Reliability analysis of single tubular joints and offshore platforms with tubular joints is
presented. The failure modes considered are yielding, punching, buckling, and fatigue
failure. Element reliability as well as systems reliability approach are used and
illustrated by several examples. Finally, optimal design of tubular joints with reliability
constraints is discussed and illustrated by an example.

1. INTRODUCTION
It is well known that the majority of structural failures of jacket type platforms are
failure in the welded tubular joints. Therefore, in estimates of the reliability of such an
offshore platform it is important to include the different failure modes of the tubular
joints. In this paper four different failure modes are considered, namely
Yielding
Punching
Buckling
Fatigue.
Although fatigue failure is included in this reliability analysis time dependence as
such is not treated. The analysis is performed at a given time. For the sake of simplicity
only K-joints are considered. T-joints, Y-joints, etc. can be analysed in an analogous
way. The four failure modes are considered separately, i.e. no mechanical interaction
between them is included in the reliability analysis.
In section 2 fundamental reliability concepts such as safety margin, reliability
index, model uncertainty, etc. are briefly presented. Then, in section 3, modelling of the
four failure modes is shown. The corresponding safety margins are introduced.
Relatively simple safety margins are used for this purpose. This fact necessitates that
1
Reliability Engineering, Vol. 19, 1987, pp. 171-184.

71
Chapter 26

model uncertainties are included. On this basis the reliability of the tubular K-joints can
be estimated with regard to each of the failure modes considered. To estimate the
reliability of a tubular joint with regard to all the failure modes mentioned a systems
approach is needed. This is also necessary if the reliability of the tubular joints is
included in the estimate of the total reliability of the platform. Therefore, in section 4,
systems reliability concepts and methods are presented with special emphasis on
modelling of the residual strength of a tubular joint after failure has taken place.

2. STRUCTURAL RELIABILITY CONCEPTS


The fundamental assumption is that all relevant uncertainty variables can be modelled
as random variables X = (Xl,, Xn), called basic variables. A realization of X is
denoted x . x w , where w is the basic variable space. Each failure mode is described
by a safety margin M s = f( X ), where f: w R is called the failure function defined in
such a way that f( x ) 0 corresponds to unsafe situations and f( x )>0 to safe situations.
By a suitable transformation the correlated and non-normally distributed variables X
are transformed into uncorrelated and normally distributed variables Z . By this
transformation the failure surface f( x )=0 in the x-coordinate system is transformed into
a failure surface w given by f( z )=0 in the z-space. In the n-dimensional z-space the
reliability index b is defined by
1/ 2
n
b = min zi2 (1)
z w
i =1
It can be shown that the probability of failure Pf for this failure mode can be
determined with good approximation from
Pf F (- b ) (2)
where F ( ) is the standard normal distribution function. The basic variables X
included in this reliability analysis are resistance variables as well as load variables.
However, since these variables are not perfectly modelled and the failure functions do
not perfectly describe the limit states it is required that model uncertainty variables are
included among the basic variables. A more detailed presentation of elementary
structural reliability concepts is given by Thoft-Christensen & Baker [1].

3. FAILURE MODES
This section deals with the four failure modes mentioned in the introduction. The main
purpose is to formulate safety margins. The K-joint used for illustration is shown in
figure 1. Four critical sections are considered. They are indicated in figure 1 by I, II, III
and IV. The load effects in each critical section are supposed to be an axial force N and
a bending moment M. The loading is assumed to be l concentrated forces Pi, i = 1,,l,
and the structure is assumed to have linear elastic behaviour so that M and N can be
written
l l
M = ai Pi , N = bi Pi (3)
i =1 i =1

72
Chapter 26

where ai, i = 1,,l and bi, i = 1,,l are


coefficients of influence. P1,,Pl are random
variables. Therefore, M and N are also random
variables. In the failure criteria formulated in
this section no distinction is made between
random variables and realizations of random
variables.
The first failure mode to be considered
Figure 1. Tubular K-joint. I - IV is failure in yielding in the four critical sections
are the critical sections. indicates I, II, III and IV. The yield criterion for the
hot spots. tubular cross-sections is assumed to be
M p N (4)
- cos = 0
MF 2 N F
where the yield capacity in pure bending MF and in pure axial loading NF is
proportional to the yield stress Y and depends on the diameter and the wall thickness. Y
is modelled by a random variable. The corresponding safety margin is
M p N
M Y = ZY - - cos
(5)
M F 2 N F

where ZY is a random variable modelling the model uncertainty connected with the use
of equation (4).
The next failure mode to be considered is punching failure in the braces (cross-
sections I and II). Within some geometrical limits the punching criterion can be written
(see Yura et al. [2], [3])
1.2
N M
+ =1 (6)
NU M U
where NU and MU are ultimate punching capacities in pure axial loading and pure
bending, respectively. NU and MU are random variables proportional to the yield stress
Y and depending on the geometry of the joint.
To take into account the uncertainty in estimating NU and MU and in using (6)
three model uncertainty random variables ZP, ZN and ZM are introduced in the safety
margin MP corresponding to (6)
N M
1.2

M = Z - N
P P
+ (7)
Z NU Z M M U

The third failure mode is local stability failure (buckling) of the four tubular
members (cross-sections I, II, III and IV) forming the joint. According to the API-based
model for stability strength (see Miller [4]) the stability criterion can be written
N M
+ =1 (8)
NB M B
where NB and MB are functions of the yield stress Y. The corresponding safety margin
MB is
N M
M B = ZB - + (9)
N B M B

73
Chapter 26

where ZB is a model uncertainty random variable.


Finally, to obtain a safety margin for fatigue failure of the joint the fatigue load
is modelled as a stationary narrow-banded zero-mean Gaussian stochastic process with
standard deviation s P and zero-up crossing rate vP. Stress concentration factors at the
two hot spots (see figure 1) can be calculated from formulas suggested by Kuang [5].
The damage accumulation can then be determined by Miner's rule and the following
safety margin MF for each of the two hot spots is derived (see Wirshing [6]):
M F = Z F - ( Z L ) K -1 ( max ( t / 32, 1) )
m M1
g (10)
where m (= 3) and K are constants in the S - N relation used in Miner's rule. K is
modelled as a random variable. t is the wall thickness, Ml is a random variable and g is
a constant. ZF is a model uncertainty variable connected with using Miner's rule as a
measure of damage accumulation. ZL is a model uncertainty variable connected with the
estimate of the fatigue load, e.g. the estimate of hot spot concentration factors,
description of sea states and the use of the Morrison equation.
To summarize for the K -joint a total number of 12 safety margins is formulated
(4 related to yielding, 2 related to punching, 4 related to buckling and 2 related to
fatigue failure). The random variables involved in the analysis are the loads P1 ,..., Pl ,
the yield stress Y, the model uncertainty variables ZY, ZP, ZN , ZM, ZB, ZF, ZL, the constant
K in Miner's rule, and M1.

4. RELIABILITY OF STRUCTURAL SYSTEMS


For each of the 12 failure modes (failure elements) defined by their safety margin in
section 3 the reliability index can be estimated as shown in section 2. If the smallest of
these indices is used as a measure of the reliability of the joint then this is called
reliability modelling at level 0. A more satisfactory estimate of the reliability of the
joint or of the total structure is based on a systems approach (see e.g. Thoft-Christensen
& Murotsu [7]). A simple definition of systems failure for the K-joint concerned would
be failure in a single failure element, i.e. the structure is considered to be in a state of
failure when a single failure element fails. This is called reliability modelling at level 1.
The probability of failure Pf of the joint is then calculated as the probability of having
failure in failure element 1 and/or in failure element 2 . . . and/or in failure element 12,
and it can be shown that a good estimate of Pf is
Pf 1 - F12 ( b ; r ) ( 11)
where F12 is the 12-dimensional standardized normal distribution function,
b = ( b1 ,..., b12 ) is the reliability indices for the 12 failure elements (failure modes) and
r the correlation matrix for the safety margins. Reliability modelling at level 1 can
also be used for the total structure, but then all potential failure elements must be
included. The number of failure elements will often be very large, so from a practical
point of view it is important to be able to identify the most important (significant)
failure elements. This identification can e.g. be performed by the b -unzipping method
(see reference [7]). The significant failure elements are then elements in a series
system.
Failure in a single failure element of a joint or a structure will in general not be
considered a failure of the joint or the structure. It will in some situations be more

74
Chapter 26

natural to define failure of the joint or of the structure as failure in two failure elements
(reliability modelling at level 2). To obtain these so-called significant pairs of failure
elements at level 2 the joint or the structure is modified by assuming in turn failure in
the significant failure elements and adding fictitious loads corresponding to the load-
carrying capacities (residual strength) of the elements in failure. For brittle failure (e.g.
fatigue) the residual strength is equal to zero. For other types of failure the reduced
strength can be taken into account by reducing the modulus of elasticity for the tubular
member in question. Thereby the flexibility of the structure is increased corresponding
to the formation of a failure mode.
The modified joint or structure is then reanalyzed and new reliability indices b
for the remaining failure elements are calculated. Failure elements with small b -values
are then combined with the failed element to form significant pairs of failure elements.
At level 2 the reliability of the joint or of the structure is modelled by a series system
where the elements are parallel systems with two elements (the significant pairs of
elements).

5. OPTIMAL DESIGN WITH RELIABILITY CONSTRAINTS


From a classical, deterministic point of view optimal design of a structural system is
usually formulated as an optimization problem where the structural weight is used as an
objective function and where the constraints ensure that the stresses, displacements, etc.
do not exceed given critical values. The optimization variables are denoted
t = (t1 ,..., tm ) and are usually geometrical quantities. Instead of using the structural
weight as an objective function the total and/or initial cost of the structure can be used.
In reliability-based structural optimization the same objective function as used
in deterministic optimal design is applied. However, the constraints state that the
reliability has to exceed a given critical value. If the systems reliability index
b S ( t ) = -F -1 ( Pf ) (see (11)) is used as a measure of the reliability then the reliability-
based optimization problem can be formulated as follows (see Srensen & Thoft-
Christensen [8]):
minimize w( t ) (12)
subject to b S ( t ) b S
min
(13)
l
t t t u
(14)
where w( t ) is the structural weight, b S some target reliability index, t l and t u
min

lower and upper bounds of the optimization variables. Alternatively the reliability
indices of the failure elements can be used as constraints. Then instead of (13) the
constraints are
bi ( t ) bimin , i = 1,..., N (15)
where bi is the reliability index of failure element i, bi is the lower bound of the
min

reliability index of element i and N is the number of failure elements.


The reliability-based optimization problem is generally a non-linear, non-
convex constrained optimization problem. In Srensen [9] some numerical procedures
to solve equations (12) - (15) are presented. The optimization problem with element
reliability constraints (15) is in example 4 solved by semi-analytical derivatives and a
general non-linear optimization algorithm. When the systems reliability index b S ,

75
Chapter 26

inequality (13), is used as a constraint, the optimization problem can be solved using a
procedure where a sequence of element reliability constraint problems formed by
linearization of (13) in terms of element reliability indices is formulated, see reference
[9].

6. EXAMPLE 1. RELIABILITY OF A SINGLE TUBULAR JOINT


Consider the K-joint B in the plane model of
the tubular steel-jacket offshore platform
shown in figure 2. The loads on the platform
are assumed to be proportional to two
independent random variables P1 and P2. P1
models the wave and wind loads, and P2
models the load from the top side, see
reference [10] for details. In reference [10] the
detailed geometrical modeling of the platform
is also described.
In figure 3 the failure elements of K-
joint B are indicated. There are 4 failure
elements modeling yielding failure, 2 failure
elements modeling punching failure, 4 failure
elements modeling buckling failure and 2
failure elements modeling fatigue failure. The
safety margins corresponding to the 4 types of
Figure 2. Offshore steel-jacket failure elements are described in section 3. In
structure, see ref. [10] for details. table 1 the statistical characteristics of the basic
variables are shown. The yield stress variables
Yl, Y2, Y3, and Y4 are assumed to be mutually correlated with correlation coefficient r
= 0.3. All other stochastic variables are assumed to be independent.

Table 1. Statistical characteristics (EX1: extreme type 1, N: normal, LN: log-normal).

76
Chapter 26

Figure 3. Failure modes of K-joint B.

The reliability of the joint is estimated using the first-order reliability method
presented in sections 2 and 4 with failure of the joint defined as failure modeling at
level 1. The series system modeling the significant failure modes is shown in figure 4.
It is seen that the most critical failure mode is fatigue failure in failure element 12
which has the reliability index b12 = 3.62.

Figure 4. Significant failure modes.

The correlation coefficient matrix of the linearized safety margins of the 4


significant failure elements is
1 0 0 0.86
0 1 0.82 0
r =
0 0.82 1 0

0.86
0 0 1
Using Hohenbichler's approximation, reference [11], an estimate of the reliability
index of the joint is b B = 3.58 .

7. EXAMPLE 2. RELIABILITY OF A PLATFORM (LEVEL 1)


Next we estimate the reliability of the total platform when failure is defined by
modeling at level 1. The 68 failure elements are numbered as shown in figure 5. The
basic variables are modeled as shown in table l, where the number of correlated yield
stress variables is increased to 6 and the number of independent K-variables is
increased from 2 to 8.
The reliability of the platform can now be estimated. The significant failure
modes are modeled as elements in a series system, see figure 6 ( Db1 = 1.1). It is seen
that the most critical failure element is failure element 51 with b51 = 3.62, see example
1. The most critical K-joint is joint B which has the reliability index b B = 3.58. Further,
we see in this example that the critical failure modes are those due to fatigue failure and
punching failure.

77
Chapter 26

Figure 5. Failure elements.


Using the linearized safety margins of the failure elements in figure 6 the
reliability index of the whole structure can be estimated at b S = 3.47 if Hohenbichler's
approximation is used.

Figure 6. Significant failure modes al level 1.

8. EXAMPLE 3. RELIABILITY OF A PLATFORM (LEVEL 2)


In continuation of example 2 the reliability of the structure is now estimated when
failure is defined by modeling at level 2. The same failure elements and basic variables
as in example 2 are used.
The two most critical failure elements at level 1 are the failure elements 51 and
67. Both of these failure elements correspond to fatigue failure. Fatigue failure is

78
Chapter 26

modeled by removing from the structure the brace where the fatigue failure element is
located.
Another critical failure element is element 61 which models punching failure in
K-joint E. Failure of this type of failure modes is modeled by reducing the stiffness of
the brace where the failure element is located. This is a simple model of the behavior
observed in experiments with punching failure 91 tubular joints, reference [3]. The
stiffness reduction is obtained by reducing fictitiously the modulus of elasticity of the
actual brace element by a factor g P (0 g P 1).
The model for punching failure can also be used when an element fails in
buckling failure mode. Yield failure can be modeled by introducing a yield hinge and
adding fictitious forces which correspond to the actual local forces producing yielding.
If g P = 0.5 is used the reliability model of the platform corresponding to failure
defined at level 2 can be established using the method described in section 4 ( Db1 =1.1
and Db 2 = 1). The significant failure modes are shown in figure 7. Also shown are the
equivalent failure elements with linear safety margins which model the parallel systems
in the reliability model.

Figure 7. Significant failure modes at level 2.


Figure 7 shows that the most significant failure mode is fatigue failure in failure
element 51 followed by fatigue failure in failure element 67. Using the linear safety
margins in the equivalent failure modes in figure 7, the reliability index of the whole
structure can be estimated at b S = 3.55 using Hohenbichler's approximation.

9. EXAMPLE 4. OPTIMAL DESIGN OF A TUBULAR JOINT


In order to optimize the parameters and thicknesses of the tubular members of the
offshore structure shown in figure 1 the structural elements are divided into 6 different
groups.
The number of optimization variables is 12, see table 2, where the initial values
of the variables are shown. As objective function a function proportional to the weight
of the structural model is used:
w = p (59.46 t1d1 + 97.76 t2 d 2 + 38.55 t3d3 + 43.18 t4 d 4 + 91.27 t5 d5 + 118.09 t6 d6 ) (16)
The lower bounds of the optimization variables are chosen as shown in table 2.

79
Chapter 26

Two

Table 2. Optimization variables.


optimization problems with reliability constraints are considered, see section 5. First,
the problem where all the reliability indices of the failure elements have to be greater
than a prescribed value b 0 = 4 is considered. This problem is solved using semi-
S

analytical derivatives and the NLPQL algorithm, reference [12]. The solution is shown
in table 3.
Next the problem where the systems reliability index corresponding to failure
modeled at level 1 is required to be greater than b 0 = 4 is considered. The problem is
S

solved using the NLPQL algorithm, semi-analytical derivatives and a sequence of


element optimization problems. The solution is shown in table 3. It is seen that the
weight is increased by 8% compared with the element optimization result.

Table 3. Results of optimizations. di and ti in [m], F in


[m3]. * the time is in sec. on a VAX/VMS 785.

80
Chapter 26

10. CONCLUSIONS
A method to estimate the reliability of tubular joints in offshore structures is presented.
Failure modes corresponding to yielding, buckling, punching, and fatigue failure are
used. The reliability is estimated using a first-order reliability method and failure of the
joint is defined as the event that one of the failure elements fails. In an example the
significant failure modes are due to punching and fatigue.
The reliability of the total structure can be estimated by the same methods
defining failure of the structure as simultaneous failure of a number of failure elements.
A new technique to model failure in e.g. punching is described. By this technique the
softening of the joint is modeled by reducing the modulus of elasticity of the damaged
structural members.
The problem of optimal design of a structure with respect to some reliability
constraints is formulated as an optimization problem. In an example the optimal tubular
diameters and wall thicknesses are determined such that the structural weight is
minimized.. The general non-linear optimization algorithm NLPQL is used.
The failure modes used in this paper do not take into account the interaction
between the different failure modes of a joint. This interaction should be further
investigated. The residual strength of a joint after failure has not yet been satisfactorily
modeled. For this purpose more experimental work is needed. The uncertainties related
to the interaction between different failure modes in a joint and the uncertainties related
to the modeling of the residual strength after failure may be treated by including further
model uncertainty variables.

11. REFERENCES
[1] Thoft-Christensen, P. & Baker, M.J. Structural Re1iability Theory and Its
Applications. Springer-Verlag, 1982.
[2] Hoadley, P. E. & Yura, J.A. UltimateStrength of Tubular Joints Subjected to
Combined Loads, University of Texas, Austin, 1983.
[3] Yura, J. A., Zettlemayer, N. & Edwards, I. P. Ultimate Capacity Equations for
Tubular Joints. OTC., Houston, Texas, OTC 3690, 1980, pp. 113-125.
[4] Miller, C.D. Buckling Design Methods for Steel Structures - A State-of-the-Art. In
Faulkner, D. et al.: Integrity of Offshore Structures. Applied Science Publishers,
London, 1981, pp. 397-418.
[5] Almar-Ns, A. (ed). Fatigue Handbook, Tapir, Trondheim, 1985.
[6] Wirshing, P. H. Fatigue Reliability for Offshore Structures. ASCE J. Structural
Eng. Vol. 110, 1984, No. 10,' pp. 2340-2356.
[7] Thoft-Christensen, P. & Murotsu, Y. Application of Structural Systems Reliability
Theory, Springer-Verlag, 1986 .
[8] Srensen, J. D. & Thoft-Christensen, P. Structural Optimization with Reliability
Constraints. Proc. 12th IFIP Conf. on System Modelling and Optimization,
Sprin-ger-Verlag, 1986 pp. 876-885.
[9] Srensen, J. D. Reliability-Based Optimization of Structural Elements. Structural
Reliability Theory, Paper No. 18, The University of Aalborg, 1986.
[10] Thoft-Christensen, P. Structural Reliability Theory. Proc. ESRA Pre-Launching
Meeting, Ispra, Italy, 1985, pp. 82-99.
[11] Hohenbichler, M. An Approximation to the Multivariate Normal Distribution.
LOG 6-82, Danish Academy of Engineers, Lyngby, 1982, pp. 79-100.

81
Chapter 26

[12] Schittkowski, K. NLPQL: A FORTRAN Subroutine Solving Constrained Non-


Linear Programming Problems. Annals of Operations Research, 1986.

82

You might also like