Professional Documents
Culture Documents
Introduction
- Facilities - Regulatory framework - Accident/failure experiences - Safety Management
T.Moan MARE WINT Sept.2013 1
Fatigue reliability
Background Fatigue life models (based on SN- and Fracture Mechanics) T.Moan MARE WINT Sept.2013 2 Reliability updating approaches
Introduction
Introduction
Experiences
Oil and gas platforms - significance of the oil and gas industry to the world econmy - need for technology development for deeper water, challenging natural and industrial environment, - ageing facilities Wind turbines Gathering of experiences development of procedures/methods/data
Escapeways/ lifeboats
Regulatory principles - Goal-setting viz. prescriptive - Probabilistic viz. deterministic - First principles viz. purely experiential
5 T.Moan MARE WINT Sept.2013 5
Failure - and accident data Safety management procedure - safety criteria, (limit states) including accidental limit state - risk and reliability analysis of design, inspection/monitoring Methods (hydrodynamics, structural analysis) Data (strength data for tubular joints)
CeSOS NTNU
Introduction
Introduction
Severe damage caused by hurricane Lilli in the Gulf of due to Mexico - inadequate state of art in offshore engineering or, - errors and omission during design or fabrication! T.Moan MARE WINT Sept.2013
a) Alexander L. Kielland fatigue failure, progressive failure and capsizing, North Sea, 1980
CeSOS NTNU
b) Ocean Ranger, flooding and capsizing, New Foundland, 1982. (Model during survival testing)
T.Moan MARE WINT Sept.2013 8 d) P - 36 explosion, flooding and capsizing, Brazil, 2001
Introduction
Introduction
In-service experiences with cracks in fixed offshore platforms (See Vrdal, Moan et al, 1997...)
Data basis - 30 North Sea platforms, with a service time of 5 to 25 years - 3411 inspections on jackets - 690 observations of cracks The predicted frequency of crack occurrence was found to be 3 times larger than the observed frequency; i.e. conservative prediction methods On the other hand: - Cracks which are not predicted, do occur. Hence, 13 % of observed fatigue cracks occurred in joints with characteristic fatigue life exceeding 800 years; due to - abnormal fabrication defects (initial crack size 0.1 mm !) T.Moan MARE WINT Sept.2013 - inadequate inspection 10
CeSOS NTNU
Mobile Fixed
Operational errors
CeSOS NTNU
Introduction
Introduction
Cause of failure
Inadequate design check to account for normal variability
Availability
- 96 - 98 % on land - 80 % for early wind farms offshore -Need for robust design, (reliable and few components) & smart maintenance, but also improved accessibility
(Courtesy: Fraunhofer)
- Larger turbine size? ( > 5 - 20 MW) - Predict, monitor and measure degradation
Introduction
Introduction
- QA/QC of engineering design process - QA/QC of the as-fabricated structure - QA/QC during operation (structural inspection )
Structural strength of the hull Strength of (possible) mooring system Escapeways and lifeboatstationes etc for evacuation
CeSOS NTNU
Introduction
10
Introduction
11
P,F
Component design check Collapsed cylinder Component design check depending on residual system strength and access for inspection
P,F
Check survival of the structure with damage under functional (F) and environmental loads (E) at an annual exceedance probability of 10-2. Load & resistance factors equal to 1.0
Fatigue crack
Plate thickness
Accidental collapse (ALS) - Ultimate capacity1) of damaged structure with credible damage
E
Jack-up collapsed
CeSOS NTNU
CeSOS NTNU
Introduction
12
Introduction
13
Sea loads
Ocean environment
Extreme moment (M) ULS: Collapse and resistance axial force (N) Local stress range history
FLS:
SN-curve/ fracture mechanics
Definition
Reliability: Probability of a component/system to perform a required function
Accidental loads
Analysis of damage
Risk: Expected loss (probability times consequences) - calibration of LFRD design approaches (1970s, 1980s) - RBI (Risk/Reliability Based Inspection) T.Moan (methods in 1980s-; industry adoption in 1990s-)
Piper Alpha
Design check
ALARP principle
MARE WINT Sept.2013 18
17
CeSOS NTNU
CeSOS NTNU
Pf = P [ R S 0] = P [ ln R ln S 0] =
where ( ) is the standard normal density function - R is the resistance - S is the loading
( u ) =
(t )dt
The main issues in the following will be to - describe the R and S as random variables - derive the expression for the Pf and generalise it for more complex problems
T.Moan MARE WINT Sept.2013 19
Pf = ( ) 101.2 1.4
T.Moan MARE WINT Sept.2013 20
Estimation of Failure Probability The probability of failure , Pf , for a time-invariant reliability problem, is Notional probaility, not true, actuarial
g ( x ) 0
f x ( x ) dx = ( )
Volume: fRS(r,s)drds is per definition the probability that s and r lies in the interval (s, s+ds) and (r, r+dr), respectively
- g(x) is the limit state function, i.e. g(x) = R - S - X is the set of n random variables - fx(x) in the joint probability density of the vector X. - Pf is determined by calculating the integral by MC simulation or FORM/SORM methods (Avoid FOSM methods etc)
Main issue:
Formulation of failure probability , which can be generalized NOTE: The probability density function for independent variables is
f R (r ) f S ( s )
T.Moan MARE WINT Sept.2013 22
FORM/ SORM
Instead of integrating the probability density in the domain of physical variable R and S, the integral may be transformed into a domain of independent standard normal variables, u1 and u2. This is done here for the simple problem where
P = P [ R S] = P [ R S 0] = P [ M 0 ] f
2 R = N ( R , R )
2 S = N ( S , S )
Transformation of variables
Distance d:
R,S space
U-space
R R R S S U2 = S U1 =
d=
R S
2 2 R + S
(u ) = Fx ( x)
x = Fx 1 ( (u ))
Failure probability
' Pf = P M 0 = ( M ' 0)
( M ' 0)
The same answer as before. The advantage of this method is when multiple variables are needed. T.Moan MARE WINT Sept.2013 24
{ }
p f = P [ M 0]
M = RS
with independent variables R and S given by probability density function fR(r) and fS(s), respectively. 1 , r s1 pf may be determined as follows:
=r i M si i
1)
( ) 2 , s2 ) (r 3 , s3 ) (r
n , (r
<0 M i
pf
k n
This estimate is accurate for large n. to determine a pf of (10-4-10-6) requires an n of the order (10-5-10-7)
sn )
(linearization)
P = ( ), f
T.Moan MARE WINT Sept.2013 27
Note that the conditional probability Pf = P[g( ) 0| I( )<0] = P[M1 0| M2 0] = P[(M1 0) (M2<0)] / P[M2 0]
System reliability
System failure may imply fatalities A system may - fail due to overload or fatigue failures (at multiple crack initiation sites), but may have - reserve capacity beyond first component failure N k PFSYS = P [ FSYS ] = P gi(...) ( )0 j i =1 j =1
What is the probability of failure in 20 years compared to the failure prob. in 1 year ?
Example: Wave loads (due to inertia forces) in North Sea: S max (1 year ) 0.8 S max (100 years) (ratio of loads prop. to ratio of wave heights) = 0.8
Smax(1 year )
max( 100 years )
Smax(100 years)
max( 1 year )
= 50
= 40
VS = 0.30 for both cases (incl. statistical + model uncertainty) Resistance R = 100, VR = 0.10 ln R For lognormal variables S Pf = ( ) 2 2
VR + VS
(1 year ) = 2.89
P (100 years)
f
Pf : (1 year ) 1.93E 3
P (1 year )
f
1.46 10 2 = ~7 1.93 10 3
Uncertainties in Load effects (S) and Resistance (R): Classification of uncertainties according to their nature
Normal Uncertainties or, Variability Fundamental (natural) Variability Example - Wave elevation/ - Loading/ - Load effects Lack of Knowledge e.g. - model uncertainty
time
- probability density function (fX(x) /distribution (FX(x)) Fundamental uncertainty in wave elevation and corresponding induced loads and load effects by stochastic methods Model uncertainty, X of a method: Estimate by obtaining a sample of: Predicted value (for a given set of parameters: PV True value (e.g. based on obs. or accurate analyses): TV Model uncertainty for observation i: Xi = TVi/PVi Establish statistics for X by a sample {Xi}n
wave-
NOTES: - Gross Errors are not considered in SRA as such - Unknown phenomena that can cause failures, cannot be treated with probabilistic methods simply because they are unknown !
T.Moan MARE WINT Sept.2013 33 T.Moan MARE WINT Sept.2013 34
c=18 m H=30m
70m
COV =
Design Wave Approach Kinematics : Stokes 5th order theory particle velocity particle acceleration Wave loading
Wave load model based on regular wave relating to API/ISO Mean uncertainty ~ 0.9 - 1.1 (1.0) COV ~ 0.25 - 0.35 Total uncertainty may be estimated by: F = .c.H where is the model uncertainty and the uncertainty in wave T.Moan MARE WINT Sept.2013 35 height is represented by H.
21
Estimate of uncertainty in the global wave load on jackets base shear force of the Magnus and Tern jackets:
Keulegan-Carpenter number
F measured(i)
Mean : CoV :
XR = 1.00+0.10
for
0 < 2.0
0.05 VX R = 0.08
for for
CeSOS NTNU
Tail sensitivity
LN
6 4 2 0 0 2 4
cov (S)
0.2 0.4
LN
6 4 2 0
cov (S)
0.2 0.4
LN : lognormal R and S
T.Moan MARE WINT Sept.2013 39
Relation between reliability measure, (Pf ) and Safety Factors (R , S ) in ULS design check
Semi-probabilistic design code:
14
15
R c / R SSc
Goal: The Implied P(R>D+L+E) Pft Pf depends upon the systematic and random uncertainties in R; D, L, and E
Reliability analysis:
pdf R and S modelled as random variables; e.g. by lognormal distributions R,S
Pf = P R S (
....... = (
ln (B R R S /B S )
ln ( R / S ) V R2 + V S2
)
1.2 1.4
T
LRFD
WSD
R = BR RC
S = BS SC
BR 1;BS < 1
V R2 + V S2
) = ( ) 10
- denotes mean value - denotes st. deviation V = / coefficient of variation T.Moan MARE WINT Sept.2013 41 (-) = standard cumulative normal distr.
Reliability-based code calibrations: Offshore oil and gas - NPD/DNV; API/LRFD; - Conoco studies of TLPs ; Wind turbines -IEC
CeSOS NTNU
Safety against fatigue or other degradation failure is achieved by design, inspection and repair Design criteria: FLS
D=
16
In-service Experiences
Fatigue behaviour
ni Dallowable Ni
Tubular joints
Time
Fatigue depends on local geometry - Cracks start to grow at hot spot points, with high stress concentration - Initial crack depth of 0.1 mm - driven by cyclic tensile stresses
Cracks can be detected and repaired. - Mean detectable crack depth: NDE: 1-2 mm Close visual inspection: 10-20 mm Fracture or ductile tearing under given extreme stress Total loss if the structure lacks residual strength after a member failure
T.Moan MARE WINT Sept.2013 44
Experiences with cracks in fixed offshore platforms (See Vrdal, Moan et al, 1997...)
Data basis
- 30 North Sea platforms, with a service time of 5 to 25 years - 3411 inspections on jackets - 690 observations of cracks The predicted frequency of crack occurrence was found to be 3 times larger than the observed frequency On the other hand: - Cracks which are not predicted, do occur. Hence, 13 % of observed fatigue cracks occurred in joints with characteristic fatigue life exceeding 800 years; due to - abnormal fabrication defects (initial crack size 0.1 mm !) - inadequate inspection
T.Moan MARE WINT Sept.2013 45
In-service experiences:
In-service experiences:
Corrosion
Splash zone: Corrosive environment / difficult access affects ultimate and
fatigue strength - plate thinning effect - crack growth rate no or damaged coating or cathodic protection corrosion rate for general corrosion: 0.1 1.0 mm/year;
Fatigue loading - Weibull distribution of stress ranges (with shape and scale parameters B and A) Fatigue resistance - SN approach - Fracture mechanics model:
Time
Stress,
Prob. density
fx(x)
xc
T.Moan MARE WINT Sept.2013 47
Fatigue failure expressed by SN formulation Load effect (stress range), S Fs(s) = 1 exp(-(s/A)B) Total number of cycles in period, N0 Resistance, SN formulation (simplified) N = KS-m K, m: material, local geometry dependent Cumulative damage: n N N m N D = i = 0 E ( S m ) = 0 Am 1 + = 0 Seq m Ni K K B K where sref = A(lnNref)1/m often Nref = N0(108 in 20 years e.g. or wave loads)
T.Moan MARE WINT Sept.2013 49
D=
m ni N o so = m +1 B Ni K ( ln N o )m B
Pf = P [ D ]
Pf = ( ) K i ( ln N o ) m N o so m +1 B = 2 2 2 V + m Vso + VK
m B
)
T.Moan MARE WINT Sept.2013 50
Fs(s) = 1 exp(-(s/A)B)
Cumulative f ailure probability 1.0E-01 Cumulative, stdv(lnA )=0.15 Cumulative, stdv(lnA )=0.3 A nnual f ailure probability A nnual, stdv(lnA )=0.15 1.0E-02 Failure probability A nnual, stdv(lnA )=0.3
Probability of failure in interval t +t, given it has survived up to t Suitable for time-dependent problems i.e. implicitly accounts for degradation of structure!
hR (t ) =
F(t) f(t)
1E-01
Failure probability
1E-02
1E-03
f (t ) 1 F (t )
1E-04
1.0E-03
1E-05
10 Time
15
20
1.0E-04
Implied reliability
Ca allow se able
1 2 1 0.33
Service life Pf
101 102
D=
1.0E-05
ni Dallowable Ni
Then T.Moan MARE WINT Sept.2013 51 hR(t) = Annual hazard rate or annual failure probability for fatigue
pf = P [g( x ) 0] =
g( x ) 0
f (x)dx = ( )
Pf = ( )
M (t) = a f a(t )
Final crack size (plate thickness)
C, m lnA, B NO,
aupd(ti) ti
ted dic Pre c an me
a(ti)
a0
k rac
size
Plate thickness
17
Current Field Measurement) Magnetic Particle Insp. Eddie Current In-service MPI & EC (Moan et al, 1997)
0.70 (3.5)
0.21 (1.05)
Pf,up (t) = P[ac a(t) 0 | aD a(t) 0] = P[F |IE] = P[F IE]/ P[IE] ac = critical crack size aD = detectable crack size where FAD (a) = POD(a)
Known outcomes in-service vs uncertain outcomes at the design stage Updating late in the service life has larger influence
CeSOS NTNU
Target level, pfsysT for global failure and pfT (T) for fatigue reliability of a specific joint
P [Fi]
18
x P [FSYS(U)|Fi] = pfsysT
Conditional annual ultimate failure probability Target Level for global failure of the structure; Depending on the potential of - Fatalities - Pollution - Property loss
; depending on the
consequences of failure
P [FSYS(U)|Fi] = (-FSYS|Fi)
12
16
20 Time (years)
Inspection sceduling for a welded joint based upon no detection of crack during inspection
T
T.Moan MARE WINT Sept.2013 57
Pf = ( ) 101.2 1.4
0.85 0.7 log Pf
Extension of method:
- consideration of other inspection events; - effect of corrosion etc T.Moan MARE WINT Sept.2013 58 - many welded joints , i.e. system of joints
CeSOS NTNU
Long term fatigue analysis of multi-planar tubular joints In an offshore jacket wind turbine
Two-parameter Weibull model: Fs(s) = 1 exp(-(s/A)B) Characteristic fatigue damage for different model
Characteristic fatigue damage Tubular Joints Raw data Leg1Joint2 Leg1Joint3 Sur12Joint1
Contribution to cumulative fatigue damage of wind loads, wave loads and interaction of wind and wave loads
Fatigue reliability analysis of multi-planar welded tubular joints considering corrosion and inspection
Fatigue reliability results:
Weibull model
0.0881 0.2713
0.0859 0.2714
0.2142
0.2069
0.2213
Reliability index for welded joints in jacket as a function of time. No inspection and repair. Rcorr = 0.1 .
Leg2Joint2
Reliability index for welded joints in jacket as a function of time. Rcorr = 0.1 , t pt = 5 years, a0 = 0.11 mm, aD = 2.0 mm, and aR = 0.11 mm.
0.0876
0.0836
0.0889
D=
ni
i
d = 1 / FDF
P FF ( i ) PfsysT FSYS FF ( i ) P
Inspection Inspection Failure No in splash inside/ Consequence inspection zone topside Static determ. 10 (10) Fulfills ALS with one 4 (3) member failed 5 (3) 2 (2) 3 (2) 1 (1)
(2) Check the effects of global model on the contact force calculation.
Gear contact fatigue analysis of a wind turbine drive train under dynamic conditions
Model for crack propagation:
- Subsurface initiated pitting. Linear elastic fracture mechanics (LEFM) model is used:
da = C ( K II _ eff dN
(a) Contact model of two gear flanks and (b) Equivalent model of two cylinders (Glodez, et al.,1997)
determined by experiments;
KII _ eff : effective Mode II (shear) stress intensity factor range ; dependent on Hardness ; microstructure; friction; crack closure
Uncertainties: - loading: contact pressure - Model uncertainties in aerodynamic loads, global and local struct. analysis
FORM/SORM
Combined FORM and MC Estimate approximate solution by FORM Improve solution by MC, focused on the most important area
T.Moan MARE WINT Sept.2013 65
Offshore structures are designed according to approaches which are: Goal-based; rather than prescriptive Probabilistic; rather than deterministic First principles; not purely experiental Integrated total; not separately (i.e. system consideration) Balance of safety elements; not hardware only NPD Guidelines for Quantitative Risk Analysis(1981) NPDs Accidental Collapse T.Moan (1984) MARE WINT Sept.2013 Limit State (ALS) 69 UK Safety Case (1992)
Fault tree
Fault tree
Unacceptable
T.Moan MARE WINT Sept.2013 70
Acceptable Acceptable
TLP, EC 5
Wilkinsonetal.,2011
1
Fault occurs
0.5
-0.5
Pitchsystem
-1
-1.5 -200
Blade seize: imbalance loads Shutdown loads: impulse from aerodynamic braking can lead to pitch vibrations What about sensor faults? Does changing the shutdown pitch rate help? Possible instability for TLPWTs (idling with one blade pitched Jonkman and Matha, 2010)
T.Moan MARE WINT Sept.2013 71
8 Environmental loads on platform in abnormal floating position T.Moan MARE WINT Sept.2013 72
Accidental Collapse Limit State relating to structural strength (NPD,1984, later NORSOK)
P, F A
Estimate the damage due to accidental event (damage, D or action, A) at an annual probability of 10-4 - apply risk analysis to establish design accidental loads
P, F
Survival check of the damaged structure as a whole, considering P, F and environmental actions ( E ) at a probability of 10-2
Target annual probability of total loss: T.Moan MARE WINT Sept.2013 73 10-5 for each type of hazard
74 Account of all measures to reduce the probability and consequences of the hazards
T.Moan MARE WINT Sept.2013
Ship collisions
Types and scenarios according to type of ships and their function: offshore site related ships (supply vessels, offshore tankers, ) floating structures (storage vessels, drilling units, crane barges..)
Oseberg B Submarine U27
Collisions do occur.
- collision events with trading vessels (with a probability of exceedance of 10- 4 ) site specific events identified by risk analysis T.Moan MARE WINT Sept.2013 76
PC
N ij
annual impact frequency for a given platform in a given location annual number of vessels with a size (j) in route (i) navigation group (k) in route (i) is on collision course Centre line platform
External mechanics
Rs Ri
PCC ,ij
Es,s dws Ship
Probability density of ship position PFR , jk probability that a vessel with a size (j) in MARE WINT Sept.2013 77 in navigation group (k)T.Moan does not succeed avoiding the platform
Internal mechanics
28
29
(463)
70
m 56
Broad-side and end view. Deck model indicated by dashed line Broad side loading
Nonlinear FEM
-General purpose (ABAQUS.) -Special purpose (USFOS)
T.Moan MARE WINT Sept.2013 79
Ultimate strength Ultimate strength Fult / FH100 Residual strength Fult(d) / Fult
IEC 6400-3
Ship collision:
Maximum size of service vessel and limiting operational conditions to be specified by designer: Vessel speed not less than 0.5 m/s Kinetic energy based on 40% added mass sideway 10% added mass bow/stern Impact energy < 1 MJ (small) Max. force may be assumed 5 MN - includes dynamic amplification
T.Moan MARE WINT Sept.2013 82
(i) P A jk
is determined by risk analysis while the other probabilities are determined by structural reliability analysis.
P [ FSYS | D ] Is determined by due consideration of relevant action and their correlation with the haazard causing the damage
T.Moan MARE WINT Sept.2013 81
2.3 Standard directives for approval practice Number 4: The structures shall be designed and configured in such a way that in the event of collision with a ship, the hull of the ship shall be damaged as little as possible.
Design Collision Event A single-hull Suezmax tanker with 160,000 tdw The ship is drifting sideways at 2 m/s Ship displacement 190,000 tons Kinetic energy (40% added mass) = 530 MNm
T.Moan MARE WINT Sept.2013 83
Mass of 5 MW turbine 450 tons Drop height 60 m Energy 275 MJ Nacelle may fall through the tank!
(Source: J.Amdahl, Tekna seminar, NTNU, January 2012)
30
Concluding remarks
Experiences regarding - failures and accidents and - life cycle safety management for oil and gas installations can serve as a basis for structures in other offshore industries, notably wind turbines, - when the differences between the oil and gas and the other industries are recognised In particular - normal uncertainty and variability in structural performance as well as possible gross errors in fabrication and operation should be properly considered in the decision process
Thank you!
CeSOS NTNU