Sunteți pe pagina 1din 14

Chapter 28

351













CHAPTER 28



INTEGRATED RELIABILITY-BASED OPTIMAL DESIGN OF
STRUCTURES
1


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



1. INTRODUCTION
In conventional optimal design of structural systems the weight or the initial cost of the
structure is usually used as objective function. Further, the constraints require that the
stresses and/or strains at some critical points have to be less than some given values.
Finally, all variables and parameters are assumed to be deterministic quantities. In this
paper a probabilistic formulation is used. Some of the quantities specifying the load and
the strength of the structure are modelled as random variables, and the constraints
specify that the reliability of the structure has to exceed some given value. The
reliability can be measured from an element and/or a systems point of view. A number
of methods to solve reliability-based optimization problems have been suggested, see
e.g. Frangopol [1], Murotsu et al. [2], Thoft-Christensen & Srensen [3] and Srensen
[4].
For structures where the reliability decreases with time it is often necessary to
design an inspection and repair programme. For example the reliability of offshore steel
structures decreases with time due to corrosion and development of fatigue cracks.
Until now most inspection and repair strategies are based on experience rather than on
rational investigations, see e.g. Jubb [5] and Dunn [6]. As a result it can be expected
that inspection and repair of the structure on the above-mentioned bases are not only
uneconomic, but perhaps also unsatisfactory from a safety point of view.
In chapter 2 of this paper reliability-based optimal design is discussed. Next, an
optimal inspection and repair strategy for existing structural systems is presented. An
optimization problem is formulated, where the ob- jective is to minimize the expected
total future cost of inspection and repair subject to the constraint that the reliability at
any time is acceptable (see Thoft-Christensen & Srensen [7]). The reliability is
estimated using first-order reliability methods, Thoft-Christensen & Murotsu [8] and

1
Proceedings IFIP WG 7.5 Conference on Reliability and Optimization of Structural Systems,
Aalborg, Denmark, May 1987, Springer-Verlag, 1988, pp. 385-398.
Chapter 28


352
Madsen et al. [9]. Finally, integration of the optimal inspection/repair strategy and the
reliability-based optimal design problem is considered. A practically usable procedure
to solve the described integrated optimization problem is presented and de- monstrated
on an offshore structure.


2. OPTIMAL DESIGN
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 an
an objective function and where the con- straints ensure that stresses, displacements,
etc. do not exceed given critical values. The optimization variables are denoted z =
(z
1
,...,z
n
) and in most cases they are geometrical quantities, such as cross-sectional
areas. However, instead of using the structural weight as an objective function the total
or the initial cost of the structure can be used.
In reliability-based structural optimization some of the quantities describing the
load and/or the strength of the system are modelled as random variables. The random
variables are denoted X = (X
1
,...,X
n
). A reliability model of the structural system is
then formulated. The elements in this model are failure elements modelling potential
failure modes of the elements of the structural system, e.g. fatigue failure of a tubular
joint. Each failure element is described by a failure function
( , ) 0 g x z = (1)
Realizations x of X resulting in ( , ) 0 g x z s correspond to failure states, while
( , ) 0 g x z > correspond to safe states. In first-order reliability methods (FORM) a
transformation T of the generally correlated and nonnormally distribued variables X
into standardized and normally distributed variables U = (U
1
,...,U
n
) is defined. Let X
= T (U ). In the u-space the reliability index is defined as

( ( ), ) 0
min ( )
T
g T u z
u u
=
= (2)
The solution point
*
u to the optimization problem in (2) is called the design
point. If the safety margin M =g(T (U )) is linearized in the design point we get
T
M U ~ + (3 )
where

*
1
i
i
u i
u g
g u

c
= =
V c
(4)
u
g V is the gradient of g with respect to u in the design point
*
u .
If the failure function is not severely non-linear then the probability of failure P
f
can with good approximation be determined from
( )
f
P ~ u (5)
where ( ) u is the standard normal distribution function.
Let the structural system be modelled by s failure elements and let failure of the
system be defined as failure of one of these elements. Then a generalized reliability
index
S
of this series system can be determined from
Chapter 28


353
1 1
( ) ( ( ; ))
S f S
P

= u = u u (6)
where
1
( ,..., )
s
= are the reliability indices of the failure elements and is a
correlation coefficient matrix determined from the linearized safety margins. ( )
S
u is
the s-dimensional standard normal distribution function. More sophisticated models of
systems failure are described by Thoft-Christensen & Murotsu [8].
Let the systems reliability index
S
be used as a measure of the reliability. Then
the reliability-based optimization problem can be fonnulated as, see Srensen & Thoft-
Christensen [10]
min ( )
z
C z (7)
min
. . ( )
S S
s t z > (8)
l u
z z z s s (9)
where C( z ) is the objective function (e.g. the initial cost of the structure),
min
S
is some
target reliability index and
l
z and
u
z are lower and upper bounds of the optimization
variables.
Alternatively, the reliability indices of the failure elements can be used as
constraints. Then, instead of (8) the constraints are
min
( ) , 1,...,
i i
z i s > = (10)
min
i
is the lower bound of the acceptable reliability index of element i.
The reliability-based optimization problems are generally non-linear and non-
convex. Most optimization algorithms require calculation of the gradients of the
objective function and of the constraints. From (2) and (4) it is seen that
*
1 ( , )
i
j u i j
g u z
z g z
c c
=
c V c
(11)
Equation (11) implies that gradients of (10) can be estimated effectively. Approximate
methods to estimate quasi-analytical gradients of (8) are described in Srensen [4]. A
procedure by which the systems reliability optimization problem (7) - (9) is solved by
using a sequence of element reliability constraint problems is also described in [4].


3. OPTIMAL INSPECTION AND REPAIR STRATEGIES
The purpose of an optimal inspection and repair strategy is to minimize the cost of
inspection and repair of a given structure so that the structure in its expected service life
has an acceptable reliability. The strategy is illustrated in figure 1, where T is the
lifetime of the structure and is the reliability index of the structure.
The reliability index is assumed to be a non-increasing function with time t if
no inspection and no repair are performed. T
i
, i = 1,2,,N are the inspection times and
min
is the minimum acceptable reliability of the structure in its lifetime. The quality of
inspection at the time T i is measured by a variable q
i
, i = 1,,N. Depending on the
magnitudes of the inspection qualities the reliability index function will increase at the
inspection times.
Chapter 28


354
It is further assumed
that the structure is modelled
by s failure elements and that
the damage of each failure
element can be modelled by
an increasing function of time
a(t) which is a realization of a
time-dependent stochastic
variable A(t). Failure of the
element occurs when
( )
cr
a t a > (12)
where a
cr
is a critical damage
measure, e.g. a critical crack
length.

Repair at the time T
i
is assumed to be performed if
m in
a a > (13)
where a
m
is the damage measured during the inspection and a
in
is a critical damage
measure. A repair is assumed to take place immediately and to be complete.
The inspection uncertainty which is assumed to decrease with inspection quality
q is divided into two groups: measurement uncertainty (e.g. measuring of crack length)
and non-detection of critical damage (e.g. ignoring a critical crack). The measurement
uncertainty is modelled as
A
m
= A
1
A(t) (14)
where A
l
is a random variable with parameters depending on q. Likewise, the non-
detection of a critical damage is modelled by a random variable A
2
with parameters
depending on the inspection quality q.
Generally, the cost of construction, inspection and repair has to be measured in
real prices. The cost due to failure is neglected in this paper.
The cost of inspection C
IN
is modelled as a function of the quality of inspections
C
IN
= C
IN
(q) (15)
The cost of a repair C
R
and of construction C
I
are assumed only to be dependent
on the design variables z
C
R
= C
R
( z ) (16)
C
I
= C
I
( z ) (17)
If the number of inspections is N and the real rate of interest is constantly equal
to r then the total real capitalized cost C is
, ,
1 1
( , , ) ( ) ( ) ( ) [ ( , )]
i i
N N
rT rT
I IN i i R i i
i i
C q t z C z C q e C z E R q t e

= =
= + +

(18)
where [ ( , )
i
E R q t is the expected number of repairs at the i
th
inspection.
The reliability of the failure elements or of the system has to fulfil the following
inequalities
min
( ) , 0 , 1,...,
i i
t t T i s > s s = (19)
min
( ) , 0
S S
t t T > s s (20)
Figure 1. Reliability index with inspection at the
times T
l
,..., T.

Chapter 28


355
The minimum -values are assumed to be at least of the order 3 - 4 so that the
events of failure can be considered as rare events.
With the above assumptions the optimal inspection strategy for a single element
in a given structure can be determined from the following optimization problem where
the design variables are q , T and N
1
1
,...,
1 1
,...,
1,2,...
min ( , ) ( ) ( ) [ ( , )]
i i
N
N
N N
rT rT
IN i R i
q q
i i
t t
N
C q t C q e C z E R q t e

= =
=
= +

(21)
min
. . ( ) , 1,..., , 1
i
s t T i N N > = + (22)
max
1
0
N
i
i
T t t
=
s s

(23)
min max
, 1,...,
i
t t t i N s s = (24)
min max
, 1,...,
i
q q q i N s s = (25)
where t
min
, t
max
, q
min
and q
max
are minimum and maximum time intervals and inspection
qualities. ( )
i
T is the reliability index just before inspection at the time T
i
or at the time
T if i = N + 1.
The total number of different repair courses (branches) is 2N , see figure 2,
where 0 and 1 signify non-repair and
repair respectively. Let +R
i
and -R
.

signify repair and non-repair at the time
T
i
and let
i
j
B signify the event that branch
j occurs at the time Ti. Then
1
2
1
[ ] ( )
i
i
i i j
j
E R P R B

=
= +

(26)
If repair events are rare then
approximately
1
2
1
1
[ ] ( )
i
i i i
j
E R P R R

=
~ +

(27)

A simple upper bound of (26) is
1
1 1
[ ] 2 ( )
i
i i i
E R P R R R

s + (28)
Inequality (28) is expected to give reasonable estimates when the number of inspections
is not too large and when the repair events cannot be considered as rare.
The reliability index ( )
i
T at the time T
i
can be estimated using (6)
1
2
1
1
( ) [ ( ( ) )]
i
i
i i j
j
T P F T B

=
= u

(29)
where F(T
i
) is the event that failure occurs at the time T
i
. Corresponding to (27) and
(28) an approximation and a simple lower bound can be estimated from
1
1
( ) [ ( ( ) )]
i i i
T P F T R

~ u (30)
and
Figure 2. Repair realizations for a
single element.
Chapter 28


356
1 1
1 1
( ) [min{1, 2 ( ( ) )}]
i
i i i
T P F T R R

> u (31)
In order to solve the optimization problem (21) - (25) effectively it is very
important that the estimates and the gradients of E[R
i
] and ( )
i
T are not too
complicated to calculate numerically. But even if approximate estimates are used it can
be expected that the result of the optimization problem gives a reasonable distribution
of inspection intervals and qualities.
Let a given structure be modelled by s failure elements in a series system and let
all failure elements be inspected at the same times T
i
and with the same inspection
qualities q
i
. Then an optimal inspection strategy can be determined from the following
optimization problem
1
1
, ,
,...,
1 1 1 1
,...,
1,2,...
min ( , ) ( ) [ ( , )]
i i
N
N
N s N s
rT rT
IN j i R j ij
q q
i j i j
t t
N
C q t C q e C E R q t e

= = = =
=
= +

(32)
min
. . ( ) , 1,..., , 1
i
s t T i N N > = + (33)
max
1
0
N
i
i
T t t
=
s s

(34)
min max
, 1,...,
i
t t t i N s s = (35)
min max
, 1,...,
i
q q q i N s s = (36)
C
IN,j
and C
r,j
are the inspection and repair cost of element j. E[R
ij
] is the expected
number of repairs at the time T
i
in element j:
( 1)
2
1
[ ] ( )
s i
i
ij ij k
k
E R P R B

=
= +

(37)
The systems reliability index ( )
S i
T is given by
( 1)
3 2
1
1 1
( ) ( )
s i
i
S i j i k
k j
T P F T B

= =
( | |
= u
( |
( \ .

(38)
F
j
(T
i
) is the event that element j fails at the time T
i
. If repair events are rare then simple
approximations to (37) and (38) which include the essential features are
( )
1 ij ij i j
E R P R R

( ~ +

(39)
| | ( ) { }
1
1
1
s
S i j i i j
j
T P F T R

=
( | |
~ u
( |
( \ .

(40)
The two optimization problems (21) - (25) and (32) - (36) are a mixture of
integer, non-linear and non-convex optimization problems, which can be solved
sequentially for a fixed number of inspections N.
The probabilities in (26) - (40) can be estimated using first-order reliability
methods, see example 2.


4. INTEGRATION OF OPTIMAL DESIGN AND OPTIMAL INSPECTION/
REPAIR STRATEGIES
Let the design variables z and the inspection variables N, T and q be design
variables in an integrated optimal design and optimal inspection strategy problem. Then
Chapter 28


357
the following optimization problem for a structural system modelled by s failure
elements can be formulated
( ) ( )
1
1
1
, ,
,...,
1 1 1 1
,...,
,...,
1,2,...
min ( , , ) ( ) [ ( , , )]
i i
m
N
N
N s N s
rT rT
I IN j i R j ij
z z
i j i j
q q
t t
N
C z q t C z C q e C z E R z q t e

= = = =
=
= + +

(41)
min
. . ( ) , 1,..., , 1
i
s t T i N N > = + (42)
max
1
0
N
i
i
T t t
=
s s

(43)
min max
, 1,...,
i
t t t i N s s = (44)
min max
, 1,...,
i
q q q i N s s = (45)
min max
, 1,...,
i
z z z i N s s = (46)
This second integer optimization problem is generally non-linear and non-
convex. The optimization problem can alternatively be formulated as the problem to
maximize the minimum systems reliability index during the lifetime of the structure
with the constraints (43) - (46) and with the further constraint that the total cost of
construction, inspection and repair must not exceed a maximum cost C
max
. Another
alternative is to formulate a multi-objective optimization problem where the solution
can be chosen from the Pareto optimum set.
In this paper the formulation (41) - (46) is used. The optimization problem is
solved sequentially for varying N using the NLPQL algorithm implemented by
Schittkowski [11]. The algorithm is based on the method by Han [12], Powell [13] and
Wilson [14]. Generally it is a very effective method where each iteration consists of
two steps. The first step is determination of a search direction by solving a quadratic
optimization problem formed by a quadratic approximation of the Lagrangian function
of the non-linear problem and a linearization of the constraints at the current design
point. The second step is a line search with an augmented Lagrangian merit function.


5. EXAMPLE 1. MODELLING OF FATIGUE CRACK FAILURE ELEMENTS
A simple fatigue crack failure element is considered in this example. In example 2 the
same failure element is used to model the failure modes in a plane model of an offshore
steel structure. Based on a linear elastic fracture mechanics approach, Paris' law and
constant amplitude loading the following safety margin can be formulated
corresponding to the failure event that the crack exceeds a critical crack length a , see
e.g. Madsen, Krenk & Lind [9]
( )
2 2
2 2
/ 2
0
2
2
M M
M
M M
F cr
M B
M t A a Y t
K


A | |
= +
|
\ .
(47)
where A
0
is the initial crack length, Y the stress intensity factor, B a load model
uncertainty parameter, M and K material parameters, v the stress cycle rate and t the
time. A is the standard deviation of stress variations which is generally dependent on
the design variables z .
Following the assumptions stated above a safety margin corresponding to repair can be
formulated as
Chapter 28


358
( )
2
2
2
2
/ 2
0
1 2
2
2
M
M
M
M M in
r
a M B
M t A Y t
A A K

| | A | |
= +
| |
\ .
\ .
(48)
where a
in
is the critical repair crack length and Ai and A2 the stochastic variables
modelling inspection un- certainty, see (13) - (14). In (47) and (48) Ao' y, B, M and K
are modelled as random variables.
From (48) it is seen that both A
1
and A
2
can be considered as inspection model
uncertainty variables of the multiplication type. Therefore, it seems reasonable to
model A
1
and A
2
as log-normally distributed variables. The following distribution
parameters are used in this example
A
1
: LN(0.l (q), 0.02 (q)) , 0 < q < 1 (49)
A
2
: LN( (q), 0.2 (q)) , 0 < q < 1 (50)
where LN( , ) is a log-normal distribution with the expected value and the
standard deviation . In this example ( ) q is
( )
3
1
0.14 , 0 1
1 (1 )
q
q q
q


= < <

(51)
In a practical
situation the constants in
(49) - (51) should clearly
be calibrated carefully to
experimental data. The
proposed model assumes
that the inspection quality
q is in the open interval
from 0 to 1. The function
( ) q in (51) is shown in
figure 3. The slightly S-
shaped form of
( ) q implies that the
probability of discovering
a failure as a function of q
will also be S-shaped. A
consequence of this is that
even a small inspection quality gives a relatively high probability of discovering a
critical crack length. Contrary to this even a high inspection quality will not guarantee
that a critical crack is discovered.
The other stochastic variables are modelled as follows
A
0
: LN( 1, 0.25) [mm]
Y: LN( l, 0.05)
B: LN( 1, 0.25)
M: N( 3.8, 0.095)
K: LN( 6400, 1024) [10
6
N/m2]
N( , ) signifies that the variable is normally distributed with the expected value
and the standard deviation . M and lnK are assumed to be correlated with the
correlation coefficient p = - 0.44 (see Karadeniz et al. [15]).
Figure 3. Example of model of inspection quality
parameter ( ) q .

Chapter 28


359
Following Wirshing [16] M and K are assumed to be independent. Further, the
material variables A
0
, M and K before and after a repair are assumed to be independent.
Using the first-order reliability method for a single element model (27) can be
approximated by
| | ( )
2 1
( ( ), ; )
i R i R i
E R T T

~ u (52)
where ( )
R i
T is the reliability index corresponding to the safety margin (48) and is
the correlation coefficient between the safety margins at the time T
i
and T
i-l
. Only the
time parameter changes in the safety margin M
R
from the time T
i-1
to T
i
. Therefore,
can be expected to be close to -1 so that
| | ( ) ( )
1
N
i R N
i
E R T
=
~ u

(53)
( ) ( )
1
2 1
[ ( ( ), ; )]
i F i R i
T T T

~ u u (54)
where ( )
F i
T is the reliability index corresponding to the safety margin (47) and is
the correlation coefficient between the linearized safety margins M
F
(T
i
) and M
R
(T
i-1
).
Again,. due to no repair in the time interval between the two events, can be expected
to be close to -1. Therefore,
( ) ( )
1
1
[max{0, ( ( ) ( )}]
i R i F i
T T T

~ u u u (55)
For a series system model the above assumptions imply that the correlations
between repair safety margins of a given failure element between repair events can be
considered to be high. Further, the correlations between safety margins of different
elements and between safety margins before and after repair can be considered close to
0. This implies that
| | ( )
( )
1
j
s
i R N
j
E R T
=
~ u

(56)
and that ( )
S i
T given by (40) can be approximated by
( ) ( ) { } ( )
( )
( )
( )
( ) ( )
1 1
1 1
1 1
1
1
1
min{1, ( )}
min{1, }
j j
s s
S i j i i j i i j
j j
s
R i F i
j
T P F T R P F T R
T T




= =

=
( | | (
~ u > u
( | (
( \ .
(
= u u u
(


(57)
or if the correlations between the safety margins of different elements are taken into
account
( ) ( ) ( ) ( ) ( ) ( ) ( )
1
1 1
max{0, ; ; }
S i s R i R i s F i R i
T T T T T


(
~ u u u

(58)
where
R
and
F
are the correlation coefficient matrices for safety margins of
different elements for repair and failure events, respectively.





Chapter 28


360
6. EXAMPLE 2. OPTIMAL INTEGRATED DESIGN OF A PLANE OFFSHORE
STRUCTURE
In this example integration of optimal design and optimal inspection strategy for a
plane model of a steel jacket platform with fatigue crack failure elements is considered.
The structure shown in figure 4 is considered. The load modelling and the detailed
geometrical description are described in Thoft-Christensen [17]. Due to symmetry only
the 8 failure elements indicated by in figure 4 are considered in this example, i.e. s =
8. The structural design variables are the tubular thicknesses of the 6 groups of
elements indicated by in figure 4, i.e. m = 6. Further, the following values of the
constants in (47) and (48) are used: v = 6.31 10
6
cycles/year, time t measured in years,
a
cr
= 40 mm, a
in
= 10 mm. In this example a
cr
is chosen to be constant. If the critical
crack length is dependent on the tubular thickness then another possibility is to put a
cr

equal to the design variable modelling the thickness of the actual structural element.
The standard deviation of stress variations is modelled as
a = c (SCF
a
, SCF
b
, influence coefficients, sectional area and modulus) (59)
where is the hot spot stress depending on the stress concentration factors for axial
and bending load effects SCF
a
and SCF
b
(here the formulas by Kuang are used, see
Almar-Ns [18]), the influence coefficients for unit loads and the sectional area and
modulus. All these quantities are dependent on the design variables z . c is a constant
(in this paper c = 25 10
6
N/m
2
).




















The cost functions in the objective function in (41) are modelled as
C
I
( z ) = weight C
IO
, C
IO
= 35,000 kr/tons (60)
C
IN
(q
i
)=
2
i
q C
INO
, C
INO
= 200,000kr. (61)
C
R
,
j
( z ) = C
RO
, C
RO
= 25,000 kr. (62)
The values of the constants are chosen on the basis of information of actual cost
prices in the Danish part of the North Sea. The upper and lower bounds in (42) - (46)
are chosen as
Figure 4. Offshore steel jacket
structure.

Chapter 28


361
t
min
= 0.25 years t
max
= 2 years
q
min
= 0.1 q
max
= 0.95
l
i
z = 30 mm
u
i
z = l00 mm , i=1,...,6

min
= 3 T = 10 years
r = 0.
As mentioned above the
NLPQL algorithm [11] is used to
solve the optimization problem.
The probability estimates are
calculated using first-order
reliability methods and derivatives
are estimated using semi-
analytical derivatives based on
(11). Derivatives with respect to
the design variables z are
calculated using pseudo-load
vectors formed by numerical
differentiation of the stiffness
matrix of the structural system
with respect to the design
variables z .
For the number of
inspections N = 6 the optimal
values of the optimization
variables are shown in table 1.
The corresponding systems
reliability index
S
as a function of t is shown in figure 5. In this example it is seen
that the optimal time intervals between inspections decrease with time and that the
inspection qualities increase with time so that at the end of the expected lifetime of the
structure, when the reliability requirements become critical, the inspections should be
performed more often and be of higher quality. Note that the increase in reliability after
an inspection is growing with the time t.
In figure 6 the optimal value of the cost is shown for different values of the
number of inspections N. It appears that the total cost has a minimum of about 6 - 7
inspections.
In table 1 the optimization
results for some simplified models
of the general optimal strategy
problem are also shown. In column
2 the results are shown if no
inspection is performed. The
increase in the total cost is about
9%. This indicates, at least in this
example, that the total cost can be
reduced considerably when
inspection is included. In columns 3
and 4 the result of a sequential
optimization is shown.
Figure 6. Optimal total cost as a function of
the number of inspections.

Figure 5. Systems reliability index
S
as a
function of t for optimal values of design
variables and inspection variables for N = 6.

Chapter 28


362
















































Table 1. Optimal values of T , q and z for N = 6 and C
RO
= 25 10
6
kr.
Table 2. Optimal values of T , q and z for N = 6 and C
RO
= 2.5 10
6
kr.
Chapter 28


363
First, optimization with respect to the design variables z with constant time
intervals and inspection qualities is performed (column 3). Next, optimization with
respect to t and q with z equal to the values from column 3 is performed. The
optimal value of the cost by this sequential procedure is seen to be less than 1 % greater
than the result from column 1, although the optimal values of the variables differ
somewhat. Use of the sequential procedure reduces the computer time by about 20%
(full optimization on a VAX 8700 lasts about 1000 sec.).
In order to investigate the effect of the value of repair cost the same
optimization problems are solved when C
RO
is changed from 2510
6
kr. to 2.510
6
kr.
The results are shown in table 2. It is seen that the change of the repair cost has only
small effect on the optimal values of t , q , z , and the total cost. The reason for this is
that the probability that repair takes place is extremely small.


7. CONCLUSIONS
A model by which the integrated optimal design and optimal inspection times and
qualities can be determined has been formulated. The total cost of construction,
inspection and repair in the expected lifetime of the struc- ture is minimized in such a
way that the reliability at any time is acceptable. The design variables are the time
intervals between inspections, the quality of the inspections and some structural design
variables. The reliability measures are estimated using first-order reliability methods.
A numerical example, where an offshore steel jacket structure modelled by
fatigue crack failure elements is considered, indicates that the model works. The
example shows that the total cost is reduced significantly by including inspection in the
optimization. Further, the example indicates that the time intervals between in-
spections should decrease with time and the inspection qualities should increase with
time.
In order to model the reliability of a more complex structure a system consisting
of series and parallel systems of failure elements has to be used, see Thoft-Christensen
& Murotsu [8]. Using such models. the effects of redundancy can be more realistically
modelled.
As can be seen from the optimization problems and figure 2 the number of
possible branches becomes very large when the number of inspections increases and/or
when the number of failure elements increases. There- fore, it is very important to be
able to identify the most significant branches. The possibility of using expert systems
for this purpose should be investigated.
After an inspection where the actual damage (crack length) has been measured
the reliability estimates can be updated. Using the updated reliability measures new
optimal inspection time intervals and qualities can be determined by solving the
optimization problems (updated maintenance strategy).


8. REFERENCES
[1] Frangopol, D. M.: Sensitivity of Reliability-Based Optimum Design. ASCE, Journal
of Structural Engineering, Vol. 111, No.8, 1985.
[2] Murotsu, Y., M. Kishi, H. Okada, M. Yonezawa & K. Taguchi:Probabilistic
Optimum Design of Frame Structure. In P. Thoft-Christensen (ed.): System
Modelling and Optimization, Springer-Verlag, Berlin, 1984, pp. 545-554.
Chapter 28


364
[3] Thoft-Christensen, P. & J. D. Srensen: Reliability Analysis of Tubular Joints in
Offshore Structures. To appear in Reliability Engineering, 1987.
[4] Srensen, J. D.: Reliability-Based Optimization of Structural Elements. Structural
Reliability Theory, Paper no. 18, The University of Aalborg, 1986.
[5] Jubb, J. E. M.: Strategies for Assessing Design and Inspection Requirements for
Redundant Structures. In Proc. Int. Symp., Nov. 1983, Washington (Faulkner et
al. (eds.)): The Role of Design, Inspection and Redundancy in Marine Structural
Reliability), 1984, pp. 118-137.
[6] Dunn, F. P.: Offshore Platform Inspection. In Proc. Int. Symp., Nov. 1983,
Washington (Faulkner et al. (eds.)): The Role of Design, Inspection and
Redundancy in Marine Structural Reliability), 1984, pp. 199-219.
[7] Thoft-Christensen, P. & J. D. Srensen: Optimal Strategy for Inspection and Repair
of Structural Systems. Civil Engineering Systems, 1987.
[8] Thoft-Christensen, P. & Y. Murotsu: Application of Structural Systems Reliability
Theory. Springer-Verlag, 1986.
[9] Madsen, H. 0., S. Krenk & N. C. Lind: Methods of Structural Safety. Prentice-Hall,
1986.
[10] Srensen, J. D. & P. Thoft-Christensen: Structural Optimization with Reliability
Constraints. Proc. 12th IFIP Conf. on System Modelling and Optimization,
Springer-Verlag, 1986, pp. 876-885.
[11] Schittkowski, K.: NLPQL: A FORTRAN Subroutine Solving Constrained Non-
Linear Programming Problems. Annals of Operations Research, 1986.
[12] Han, S. - P.: A Globally Convergent Method for Non-Linear Programming.
Journal of Optimization Theory and Applications. Vol. 22, 1977, pp. 297-309.
[13] Powell, M. J. D.: A Fast Algorithm for Non-Linearly Constrained Optimization
Calculations. In Numerical Analysis (ed. G. A. Watson): Lecture Notes in
Mathematics, Vol. 630, Springer-Verlag, 1978.
[14] Wilson, R. B.: A Simplicial Algorithm for Concave Programming. Ph.D. thesis,
Harvard University, Boston, 1963.
[15] Karadeniz, H., S. van Manen & A. Vrouwenvelder: Probabilistic Reliability
Analysis for the Fatigue Limit State of Off-Shore Structures. Bulletin Technique
du Bureau Veritas, July 1984, pp. 203-219.
[16] Wirshing, P. H.: Fatigue Reliability for Offshore Structures. ASCE, Journal of
Structural Engineering, Vol. 110, No. 10, 1984, pp. 2340-2356.
[ 17] Thoft-Christensen, P.: Structural Reliability Theory. Proc. ESRA Pre-Launching
Meeting, Ispra, Italy, 1985, pp. 82-99.
[18] Almar-Nres, A. (ed.): FatigueHandbook, Tapir, Trondheim, 1985.

S-ar putea să vă placă și