Documente Academic
Documente Profesional
Documente Cultură
When a structure is under certain loadings, it will experience stress/strain and deform
in some way. To ensure structural integrity, designers have to assess all the possible
failure modes of the structure. A 'limit state' is a condition beyond which the structure
is viewed as failure. In another word, the 'violation' of a limit state can be defined as
an undesirable condition for the structure. Limit states for structures can be divided
into three categories: (1) ultimate limit state, which defines the collapse of structures.
(2) damage limit state, which defines the damage of structures. (3) serviceability limit
state, which defines disruption of normal use of structures.
This is the conventional design methodology. It is also called 'working stress' design.
In this design, a safety factor is introduced to ensure certain safety margin in the
design. The format of strength assessment is
σy
σ ≤ [σ ] = (1)
n
γ 1 L1 + γ 2 L2 + ... ≤ φR (2)
The same safety format as in deterministic limit state design (Eq. 2) is used in
reliability-based limit state design. The partial factors (also called partial safety
factors) are calculated by reliability analysis. In this way, target reliability can be
achieved by using this method.
This is why some people thought limit state design was reliability-based limit state
design. This method will be explained in the next section in details.
1
RELIABILITY THEORY AND ITS APPILICATION
Structural reliability is concerned with lifetime structural safety, how much is required
and how this can be achieved in design. Structural reliability analysis will take
uncertainties, which are related to the process of structural design, into consideration.
Physical uncertainties
2
Uncertainties of this type can be reduced by some techniques and judgement. They
are sometimes referred to as subjective uncertainties.
Statistical uncertainties
Modelling uncertainties
actual response
Xm =
predicted (modelled) response
The so-called actual response is not known, but in practice the experimental
(measured) data is used as actual response. But actually experiments have also
uncertainties, which are not counted in the practice.
Phenomenological uncertainty
Human errors
Most of recorded structural collapse or losses are attributed to human errors rather
than insufficient prescribed safety in design. About 50% to 90% (from different
statistical resources) of accidents were caused by human errors. 85% has been quoted
as a typical value for marine structures.
Notional reliability
3
Because the failure probability caused by human errors is not normally considered in
structural reliability analysis, there is large difference between the actual (real) risk
and the predicted failure probability. At present this gap is typically 1 to 3 orders of
magnitude. So the predicted failure probability is referred to as being notional or
nominal.
Level 1: In this method reliability based partial safety factor (PSF) are applied to
characteristic value of load components and resistance factors in the safety check
equations used in design. This is a deterministic format most commonly advocated for
limit states design codes at present.
Level 2: In this method the information of mean and variance of random variables are
used in the analysis. It is called first order and second moment (FOSM) method, [or
advanced first order and second moment (AFOSM) method].
According to how the randomness of the structures is considered, the methods can be
divided into:
(1). random field method;
(2). random process method
(3). random variable method
(a) time-independent variable
(b) time-dependent variable
1). FOSM (First Order Second Moment) method, (or called Advanced First Order
Second Moment Method). In the development of reliability method, a method is
called Mean Value First Order Second Moment method (MVFOSM), which is not
4
correct. Actually it is only suitable for linear safety margin equation. When limit state
equation is not linear, this method will give wrong results (not accurate).
Special Case
A special case in reliability analysis is when the limit state equation can be expressed
as:
M = R−L (3)
0 − µM − µM
Pf = P[M ≤ 0] = Φ = Φ = Φ (− β) (4)
σM σM
[ ]
σ 2M = E (M − µ M ) = VAR[M ] = σ 2R + σ 2L
2
(6)
µM µR − µL
β= = (7)
σM (
σ 2R + σ 2L
1/ 2
)
A geometric explanation of the special case is shown in Figs. 1 and 2. A few
important points are worth noting:
5
• From Eq. 7 the reliability index can be calculated by the mean and standard
deviation of R and L. We don't need to know the probability distribution functions
of them.
• The failure probability can be easily calculated once reliability index is know by
Eq.(4). This is only true when R and L are normally distributed.
• Fig.2 shows that the reliability index is equal to the number of standard deviation
by which µ M exceeds zero.
Fig.1
Fig. 2
6
Based on Fig.1, another expression of failure probability can be derived as:
∞
Pf = P(M ≤ 0) = P[(R − L ) ≤ 0] = ∫ F (x )f (x )dx
R L (8)
−∞
It should be pointed out that this expression is valid when R and L have other kinds of
distribution functions.
S = 1 − Pf (9)
General Cases
If there are n random variables X = {x 1 , x 2 ,⋅ ⋅ ⋅, x n } , and the limit state equation is:
M = f {x 1 , x 2 ,⋅ ⋅ ⋅, x n } = 0 (10)
M<0 means failure, and M>0 means safe. In this general case the failure probability
is:
This method is widely used in engineering calculation. Even in this category there are
various algorithms. Among them, Fiessler and Rackwitz's algorithm, which show
good accuracy, efficiency and robustness, is the best. Therefore the procedure is
briefly described below.
M = g (x 1 , x 2 ,..., x n ) (12)
When M < 0, the structure fails, and M > 0 the structure is safe. The failure surface is
given by M = 0.
7
M = g (x , x ) + ∑ (x ) ( )
n
*
1
*
2 ,..., x *
n i − x *i g i' X * (13)
i =1
where
g X =
' ∂g X *
( )
* ( )
i
∂x i
{
X * = x 1* , x *2 ,..., x *n } is the unknown design point.
If µ i and σ i represent the mean and standard deviation of the variable x i , the mean
value of M is:
( ) ( )
n
µ M = ∑ µ i − x *i g i' X * (14)
i =1
( )
n
σ M = ∑ α i g i' X * σ i (16)
i =1
where
αi =
( )
g i' X * σ i
(17)
1/ 2
2
{ ( ) }
n
∑ g j X σ j
' *
j=1
are referred as sensitivity factors since they reflect the relative influence of each
design variable on the reliability index. The larger the sensitivity factor, the more
influential the variable is.
∑ (µ ) ( )
n
− x *i g i' X *
µ i
β= M = i =1
(18)
∑ α g (X )σ
σM n
' *
i i i
i =1
8
∑ g (X )(µ )
n
'
i
*
i − x *i − α i βσ i = 0 (19)
i =1
Because the design point is not known, iteration is needed to get the solution.
Pf = Φ (− β) (21)
Suppose that the variable x i has density function f x i (x i ) and distribution function
Fx i (x i ) . The basic idea of the transformation is to let the original density function and
distribution function of the variable x i be equal to that of a normal variable at the
design point. That is:
x *i − µ iN
( )
Fx i x = Φ *
i
(22)
σi
N
1 x *i − µ iN
( )
f x i x = N ϕ *
σ i σ iN
i
(23)
From Eqs.(22) and (23), the equivalent mean and standard deviation are expressed as:
( ( ))
µ iN = x *i − Φ −1 Fx i x *i σ iN (24)
( { ( )})
ϕ Φ −1 Fx i x *i
σ = N
(25)
f (x )
i *
xi i
Step 1
9
give an initial approximation of design point. Mean values are normally used as the
initial design point, eg. X *(0) = {µ1, µ 2, ...,µ n }
Step 2
Calculate the following:
( )
g 0 = g x 1* , x *2 ,..., x *n
∂g (X )
= g (X ) =
*
g i' '
i
*
∂x i
Step 3
Transform the non-normal variables to equivalent normal variables
µ iN = x *i − Φ −1 Fx i x *i σ iN( ( ))
( { ( )})
ϕ Φ −1 Fx i x *i
σ =N
f (x )
i *
xi i
Step 4
Calculate the following:
( )
n
x = ∑ g i' X * x *i
i =1
( )
n
µ x = ∑ g i' X * µ i
i =1
1/ 2
n 2
σ x = ∑ {g i' (X * )σ i }
i =1
αi = i
g X σi
' *
( )
σx
x − g0 − µx
β=−
σx
Step 5
Calculate the design point for the next iteration
( m +1)
x *i = µ i − α i βσ i
Step 6
Check if the iteration converges. The criterion for this is
10
x *(i m +1) − x *(i m )
<ε (i = 1,…,n)
x *(i m +1)
β *( m +1) − β *( m )
<ε
β *( m +1)
Step 7
calculate
Pf = Φ (− β)
and print out the reliability index, probability of failure, design point and sensitivity
factors.
Fig. 3
Example 1
The limit state function of a ship’s mid-ship section for initial yielding is expressed
as: M = x 1 × x 2 − x 3
where x 1 , x 2 , x 3 are random variables, whose probabilistic properties are shown in
Table 1. Please calculate the reliability index using the Advanced First Order Second
11
Moment method (Rackwitz and Fiessler’s algorithm). Which variable is the most
influential one? Why?
(Iteration stop criteria ε is set as 0.01.)
Solutions:
The starting point of the design point is: x 1*(1) =3, x *2(1) =245000, x *3(1) =375000
So,
g 0 = g(x 1* , x *2 , x *3 ) = 3 × 245000 − 375000 = 360000
∂g
= x *2 = 245000
∂x 1
∂g
= x 1* = 3
∂x 2
∂g
= −1
∂x 3
3
x= ∑g x
i =1
'
i
*
i = 245000 × 3 + 3 × 245000 − 375000 = 1095000
3
µ x = ∑ g i' µ i = (245000 × 3) + (3 × 245000 ) + (− 1 × 375000 ) = 1095000
i =1
σ x = 111255.6
g 1' σ1 245000 × 0.15
α1 = = = 0.33
σx 111255.6
g '2 σ 2 3 × 24500
α2 = = = 0.66
σx 111255.6
12
g 3' σ 3 − 1 × 75000
α3 = = = −0.674
σx 111255.6
x − g 0 − µ x 1095000 − 360000 − 1095000
β=− = = 3.236
σx 111255.6
( 2)
x 1* = µ1 − α 1βσ 1 = 3 − 0.33 × 3.236 × 0.15 = 2.84
( 2)
x *2 = µ 2 − α 2 βσ 2 = 245000 − 0.66 × 3.236 × 24500 = 192673.88
( 2)
x *3 = µ 3 − α 3βσ 3 = 375000 + 0.674 × 3.326 × 75000 = 538579.8
check stop criteria (if ε is set as 0.01)
For iteration 2:
( )
g 0 = g x 1* , x *2 , x *3 = 2.84 × 192673.88 − 538579.8 = 8614.019
∂g
= x *2 = 192673.88
∂x 1
∂g
= x 1* = 2.84
∂x 2
∂g
= −1
∂x 3
3
x= ∑g xi =1
'
i
*
i = 192673.88 × 2.84 + 2.84 × 192673.88 − 538579.8 = 555807.838
3
µ x = ∑ g i' µ i = (192673.88 × 3) + (2.84 × 245000 ) + (− 1 × 375000 ) = 898821.64
i =1
σ x = 106309.214
g 1' σ1 192673.88 × 0.15
α1 = = = 0.272
σx 106309.214
13
g 3' σ 3 − 1 × 75000
α3 = = = −0.705
σx 106309.214
x − g 0 − µ x 555807.838 − 8614.019 − 898821.64
β=− = = 3.308
σx 106309.214
( 3)
x 1* = µ1 − α1βσ 1 = 3 − 0.272 × 3.308 × 0.15 = 2.865
( 3)
x *2 = µ 2 − α 2 βσ 2 = 245000 − 0.655 × 3.308 × 24500 = 191914.87
( 3)
x *3 = µ 3 − α 3βσ 3 = 375000 + 0.705 × 3.308 × 75000 = 549910.5
Iteration 3
( )
g 0 = g x 1* , x *2 , x *3 = 2.865 × 191914.87 − 549910.5 = −74.397
∂g
= x *2 = 191914.87
∂x 1
∂g
= x 1* = 2.865
∂x 2
∂g
= −1
∂x 3
3
x= ∑g xi =1
'
i
*
i = 191914.87 × 2.865 + 2.865 × 191914.87 − 549910.5 = 549761.705
3
µ x = ∑ g i' µ i = (191914.87 × 3) + (2.865 × 245000 ) + (− 1 × 75000) = 902669.61
i =1
σ x = 106680.325
g 1' σ1 191914.87 × 0.15
α1 = = = 0.270
σx 106680.325
14
g '2 σ 2 2.865 × 24500
α2 = = = 0.658
σx 106680.325
g 3' σ 3 − 1 × 75000
α3 = = = −0.703
σx 106680.325
x − g 0 − µ x 549761.705 − (− 74.397 ) − 902669.61
β=− = = 3.307
σx 106680.325
( 4)
x 1* = µ1 − α1βσ 1 = 3 − 0.27 × 3.307 × 0.15 = 2.866
* ( 4)
x2 = µ 2 − α 2 βσ 2 = 245000 − 0.658 × 3.307 × 24500 = 191687.853
* ( 4)
x3 = µ 3 − α 3βσ 3 = 375000 + 0.703 × 3.307 × 75000 = 549361.575
So the iteration converges. The reliability index is 3.307. The design point is
x 1* =2.866, x *2 =191687.853, x *3 =549361.575.
x3 is the most influential variable because the corresponding sensitivity factor has the
largest absolute value.
The basic idea and procedure of reliability-based limit state design of pipelines will be
described.
Reliability-based limit state design has been used by some design codes of pipelines,
such as the recent DNV code (1996) for pipelines.
15
Reliability-based limit state design involves the following tasks:
To ensure the integrity of pipelines following limit states must be checked in design:
• out of roundness for serviceability
• bursting due to internal pressure, longitudinal force and bending
• buckling/collapse due to pressure, longitudinal force and bending
• fracture of welds due to bending / tension
• low-cycle fatigue due to shutdowns
• ratcheting due to reeling and shutdowns
• accumulated plastic strain
This is also called level 1 method. The combination of characteristic values and
partial safety factors is used to ensure certain level of safety of the structures.
γ c γ f Lk ≤ φm R k (26)
where γ c and γ f are load effect partial safety factors > 1.0, and φ m is a resistance
PSF < 1.0. They are illustrated in Fig. 4. Fig. 5 shows the design point in design
variable space.
16
Fig.4
Fig.5
17
The number of PSFs is decided by code writers. In principle, every design variable
can have a PSF, but this is not necessary because some design variables only have
marginal effect on reliability of the structure (shown by sensitivity factors). Hence
PSFs are only introduced to important design variables. In practice, if too few PSFs
(say 2 or 3), are used in a code, the resulting spread in reliability for a variety of
structural components will be unnecessarily large and wasteful. Five PSFs are the
common number nowadays.
Characteristic Values
R k = µ R (1 − k R VR ) = µ R − k R σ R (27)
0.05 = Φ (− k R ) (28)
µ − k R σR − µR
[Q 0.05= P(R < R k ) = P(R < (µ R − k R σ R )) = Φ R = Φ(− k R ) ]
σR
Hence k R = 1.645.
For ‘Loading’ variables, characteristic values are calculated on the high side of their
mean. The characteristic load effect is:
L k = µ L (1 + k L VL ) = µ L + k L σ L (29)
For convenience, typical k values for percentage probability are presented in Table 1.
Table 1
18
Lower bound 5% values are often assumed for yield stress, and (incorrectly) for
strength curves used in offshore design. 2 standard deviation values (2.32%
probability) are generally used for fatigue strength design.
Many methods for reliability analysis can be used for this purpose. The First Order
and Second Moment (FORM ) method is normally used because of its simplicity and
reasonable accuracy.
Faulkner [1984] has studied the target reliability for various steel structures, which are
shown in Fig. 5. It is observed that the reliability for merchant ships and British
frigates vary in a very large range. A value of β = 3.0 for frigates and 3.0 to 4.0 for
merchant ships was recommended.
19
Fig. 6 Target reliability
Based on a recent study [Mansour et al, 1996], the suggested target reliability for
ships is shown in Table 2. It is seen that the target reliability indices for collapse of
the entire structure (primary failure mode) is greater than that of a non-critical welded
detail relative to fatigue.
20
Table 2 Recommended target safety indices
[Mansour and Wirsching, 1996]
where SLS is Serviceability Limit States, ULS is Ultimate Limit States, FLS is
Fatigue Limit State, ALS is accidental Limit State.
x *i
γ i , φi = (30)
x ik
x *i 1 − β c α i Vi
γ i , φi = = (31)
x ik 1 ± k i Vi
where γ i are loading PSFs usually ≥ 1.0 and the second denominator term is + ve,
φ i are resistance PSFs usually ≤ 1.0 and the second term in denominator is – ve. For
each variable x i , x *i is the design point, x ik is characteristic value, β c is the target
reliability index for the code. α i is sensitivity factor.
Example 2
Follow example 1, calculate the partial safety factors of the design variables.
21
Recall the results in example 1,
The reliability index is 3.307. The design point and sensitivity factors are:
x 1* =2.866, x *2 =191687.853, x *3 =549361.575.
α 1 = 0.270 α 2 = 0.658 α 3 = −0.703
For x 1 :
For x 2 :
For x 3 :
This is a loading variable, so
The above is the procedure to calculate the PSFs. The example means that if a safety
check format
(PSF1 × x 1k )(PSF2 × x 2k ) ≥ (PSF3 × x 3k )
is used, the designed structure will have a safety index β =3.307. This is the idea of a
probability-based design code. In the design, as long as the characteristic values and
PSFs are used, target reliability is ensured implicitly (e.g. reliability analysis is not
carried out.).
22
The above mentioned reliability-based limit state design methodology can be applied
to new design and inspection and maintenance of pipelines. Hopkins and Jaswel
(1997) applied this method to uprating pipelines (i.e. their pressure increased beyond
their original design pressure).
REFERENCES
1. DNV, (1996): ‘Rules for Submarine Pipeline Systems’, Det Norske Veritas,
DNV'1996, Norway (Reported in Jiao, G., et al, ‘The Superb Project: Wall
Thickness Design Guideline for Pressure Containment of Offshore Pipelines,
Offshore Mechanics and Arctic Engineering Conference, OMAE 1996, Florence,
Italy.
2. Faulkner, D. (1984): ‘On Selecting a Target Reliability For Deep Water Tension
Leg Platforms’, Proceeding of 11th IFIP Conference on System Modelling and
Optimisation, Copenhagen, Denmark, July 25-29, pp. 490-513.
5. Sotberg, T., Moan, T., Bruschi, R., Jiao, G. and Mork, K.J., (1997) : 'The
SUPERB Project: Recommended Target Safety Levels for Limit State Based
Design of Offshore Pipelines', Proc. of OMAE;97.
23