0 tayangan

Diunggah oleh Otso Cronvall

Efficient Procedure for Probabilistic Crack Growth Analyses.

- Doane Chapter 06
- DNV RP F108
- ME CAD-CAM_2010
- rp-f108
- Chapter #19 Solutions- Engineering Economy, 7 th editionLeland Blank and Anthony Tarquin
- Discrete random variables
- Fatigue Primer for Engineers
- CTOD
- Misurata University
- tugas 2 metolid
- 53214400-A-Fatigue-Primer-for-Structural-Engineers.pdf
- Model for Quantifying Risks (Table of Content)
- Modeling Discon XFEM
- Numerical and Experimental Analysis of the Directional Stability on Crack Propagation Under Biaxial Stresses
- A Distributed CSMA Algorithm for Throughput and Utility Maximization in Wireless Networks
- 11.1 Probabilistic Risk Analysis - Examples_with Notes
- 14.2016 Effect of a New Specimen Size on Fatigue Crack Growth Behavior in Thick-walled Pressure Vessels
- Wang - Industrial Challenges for Thin Wafer Manufacturing
- pm10_tisk
- sdarticle

Anda di halaman 1dari 41

RESEARCH REPORT

degradation mechanisms

Growth Analyses

Confidentiality: Public

1 (40)

Report’s title

Efficient Procedure for Probabilistic Crack Growth Analyses

Customer, contact person, address Order reference

State Nuclear Waste Management Fund (VYR), VTT Dnro: SAFIR 20/2017

Project name Project number/Short name

FOUND 2017 113412/ FOUND 2017

Author(s) Pages

Otso Cronvall (VTT), Paul Smeekes (TVO) 29/11

Keywords Report identification code

Interpolation, fracture mechanics, crack growth VTT-R-04910-17

Summary

This study concerns development of a new efficient probabilistic procedure for crack growth analyses.

The main target for it is probabilistic fracture mechanics based crack growth simulations, which involve

applying Monte Carlo simulation. To achieve sufficient accuracy, the number of computed crack growth

simulations needs to be at least several thousands for one analysed location, e.g. pipe cross-section. With

commonly applied weight and influence function methods such an analysis can take hours. For doing

analyses for a nuclear power plant piping system with hundreds of locations/welds this is way too long.

The developed procedure for applies interpolation between a large enough representative set of case

specifically computed crack growth simulations. These simulations cover the probabilistically

distributed input data parameters, which are here the depth and length of initial cracks. The interpolation

is based on the initial crack sizes, as picked randomly from the respective probabilistic distributions. It

is checked between which two actually simulated crack growth curves the randomly picked initial crack

size is located. The crack growth curve is then interpolated by keeping the relative distance from the two

nearest simulated curves the same as that for the initial crack size.

Two additional alternative efficient probabilistic procedures for crack growth analyses were developed

too. They are described in Appendix 1.

The purpose is to program in VTT the new procedure in the near future. It is expected that, once

programmed, the application of the procedure in probabilistic NPP piping component leak/break

analyses is much more efficient than using solely the weight/influence function method based solutions.

Confidentiality Public

Espoo 8.9.2017

Written by Reviewed by Accepted by

Senior Scientist Research Scientist Research Manager

VTT’s contact address

P.O. Box 1000, 02044 VTT

Distribution (customer and VTT)

SAFIR2018 Reference Group 5,

VTT Archive (2)

The use of the name of the VTT Technical Research Centre of Finland (VTT) in advertising or publication in part of

this report is only permissible with written authorisation from the VTT Technical Research Centre of Finland.

RESEARCH REPORT VTT-R-04910-17

2 (40)

Contents

Foreword .....................................................................................................................3

1 Introduction .............................................................................................................7

3 On theoretical background......................................................................................9

3.1 Weight/influence function methods and handbook solutions...........................9

3.2 Deterministic and probabilistic crack growth computation procedures ..........13

3.3 Distributions for initial crack sizes .................................................................16

3.4 Latin hypercube simulation ...........................................................................18

4 Developed probabilistic procedure for crack growth analyses ..............................20

References ................................................................................................................27

RESEARCH REPORT VTT-R-04910-17

3 (40)

Foreword

This report has been prepared under the research project FOUND 2017, and therein for WP2

Susceptibility of BWR RPV and internals to degradation mechanisms. The project is a part of

SAFIR2018, which is a national nuclear energy research program. FOUND WP2 work in 2017

was funded by the BeräkningsGrupp (BG), the State Nuclear Waste Management Fund (VYR)

and the Technical Research Centre of Finland (VTT), which is gratefully acknowledged. The

work was mainly carried out at VTT.

Espoo 8.9.2017

Authors

RESEARCH REPORT VTT-R-04910-17

4 (40)

Abbreviations

BWR Boiling water reactor

FCG Fatigue crack growth

FE Finite element

FFS Fitness-for-service

HAZ Heat affected zone

IASCC Irradiation accelerated stress corrosion cracking

IGSCC Intergranular stress corrosion cracking

LHS Latin hypercube simulation

MCS Monte Carlo simulation

NPP Nuclear power plant

NWC Normal water chemistry

PC Personal computer

PFM Probabilistic fracture mechanics

POD Probability of detection

PWSCC Primary water stress corrosion cracking

RPV Reactor pressure vessel

SCC Stress corrosion cracking

SIF Stress intensity factor

TGSCC Transgranular stress corrosion cracking

TVO Teollisuuden Voima Oyj

VTT Technical Research Centre of Finland Ltd (Teknologian tutkimuskeskus VTT)

VYR State Nuclear Waste Management Fund

WRS Welding residual stress

Latin letters

ak(tl) Time dependent crack depth in crack growth realisation [mm]

c Half of crack length [mm]

C Parameter in probability density function for crack aspect ratio [-]

ck(tl) Time dependent crack half-length in crack growth realisation [mm]

CFCG Parameter in the Paris-Erdogan equation [(mm/cycle)/((MPa m)nFCG)]

CSCC Parameter in the stress corrosion cracking equation [(mm/s)/((MPa m)nSCC)]

f0( ) Probability density function for crack aspect ratio [-]

fbg Tabulated geometry function values for the bending part [-]

fal(l0) Probability density function for initial crack length [1/mm]

fi Tabulated geometry function values [-]

fx(x) Probability density function for depth of manufacturing induced cracks in

NPP piping of stainless steel [1/mm]

G0 to G4 Influence coefficients for a stress function [-]

Gn Influence coefficients [-]

H Heaviside step function [-]

RESEARCH REPORT VTT-R-04910-17

5 (40)

KI Stress intensity factor, mode I [MPa m]

KI,max Maximum value of stress intensity factor, mode I [MPa m]

KI,min Minimum value of stress intensity factor, mode I [MPa m]

l Crack length [mm]

l0 Initial crack length [mm]

L1, L2 Straight distances from the randomly picked initial crack size to the nearest

simulations [mm]

L1a, L2a Projections to crack depth axis of the straight distances from the randomly

picked initial crack size to the nearest simulations [mm]

L1c, L2c Projections to crack half-length axis of the straight distances from the

randomly picked initial crack size to the nearest simulations [mm]

LHSl,m Identification of rectangles having the same probability in 2D space spanned

by the initial crack depth and length, with l, m = 1, 2, 3, …, nLHS

Mi Weight function coefficients [-]

n Exponent [-]

NCGR Number of crack growth realisations to be computed [-]

nFCG Exponent in the Paris-Erdogan and equation [-]

NGRID Number of grid points for the initial crack depth and length values [-]

Ni Weight function coefficients [-]

nLHS Number of intervals in LHS for the initial crack depth and length [-]

NLHS Number of rectangles having the same probability in 2D space spanned by the

initial crack depth and length [-]

nSCC Exponent in the stress corrosion cracking rate equation [-]

N Number of load cycles [-]

Nf Number of failed simulations [-]

Ns Size of the sample as total number of simulations [-]

Pf Probability of failure [-]

pn Pressure at crack tip [N/mm2]

p(x) Crack face traction [N/mm2]

Q Crack shape parameter [-]

Ri Inner radius of pipe cross-section [mm]

t Time [s, year]

tl Discretized time [s, year]

tw Wall thickness [mm]

x Coordinate through the crack starting from the cracked surface [mm]

~ Median of dimension variable of interest [mm]

X

Greek letters

m Parameter in probability density function for crack aspect ratio [-]

a Crack depth increment [mm]

KI Range of mode I stress intensity factor, i.e. extraction of maximum and

minimum values [MPa m]

t Time increment [day]

Angular crack front location coordinate [°]

C Perimeter of the crack [mm]

Parameter in probability density function for crack aspect ratio [-]

Fitted parameter in probability density function for initial crack length [-]

RESEARCH REPORT VTT-R-04910-17

6 (40)

of initial crack length [-]

y Scale parameter for probability density function [mm]

bg Global bending stress [N/mm2]

0 to 4 Polynomial coefficients of a stress function [-]

i Factors of the polynomial of fitted stress distribution as normalized through

wall [-]

(x) Total stress through wall [MPa]

y Shape parameter for probability density function [-]

RESEARCH REPORT VTT-R-04910-17

7 (40)

1 Introduction

This study is part of the SAFIR2018 research program project FOUND 2017. The overall

objective of this project is cross-disciplinary assessment of ageing mechanisms for safe

management and extension of operational lifetime of Finnish nuclear power plants (NPPs). This

involves developing deterministic, probabilistic and risk informed approaches in computational

and experimental analyses with education of new experts. FOUND is an acronym of Analysis

of Fatigue and Other cUmulative ageing to exteND lifetime.

This study concerns development of a new efficient probabilistic procedure for crack growth

analyses, which in particular would be intended for probabilistic fracture mechanics (PFM)

based analyses. These analyses involve applying Monte Carlo simulation (MCS). Typical

degradation mechanisms driving the crack growth for NPP piping components are stress

corrosion cracking (SCC) and fatigue induced crack growth (FCG). To achieve sufficient

accuracy, the number of computed crack growth realisations, i.e. simulations, need to be large

in PFM based analyses, at least of the scale of thousands for one analysed location, e.g. pipe

cross-section. In deterministic fracture mechanics based analyses, weight and influence

function methods are applied to develop component/structure specifically tailored solutions.

These have been obtained by fitting against the results from numerical 2D or 3D finite element

(FE) analyses. The weight/influence function method based solutions are accurate and

computationally much less laborious to use than to do FE analyses. Still, several thousands of

weight/influence function method based crack growth simulations takes a relatively large

amount of time, from one to several hours with a typical personal computer (PC). For doing

analyses for a NPP piping system, this is way too much, as the scope consists of hundreds of

locations to be analysed, which are mostly welds. Thus, a more efficient but still accurate crack

growth analysis procedure is needed for this purpose.

After this Chapter the structure of this report is as follows. The scope and objectives of the

study are described in Chapter 2. The weight/influence function method as well as deterministic

and probabilistic crack growth computation procedures are presented in Chapter 3. This is

followed with describing in detail the developed new efficient probabilistic procedure for crack

growth analyses in Chapter 4. The summary and conclusions are presented in Chapter 5. Two

additional alternative efficient probabilistic procedures for crack growth analyses were

developed too. They are described in Appendix 1.

RESEARCH REPORT VTT-R-04910-17

8 (40)

The objectives and scope of the study are described briefly in the following.

The purpose of this study is to develop and describe a new efficient probabilistic procedure for

crack growth analyses. In particular, it is intended for PFM based analyses of NPP piping

systems. Further, the structure and level of detail in describing the new procedure is such that

it would serve the programming of it as much as possible.

describing briefly the theoretical background of the new computation procedure, including

the weight/influence function method and FE analysis based solutions as well as

deterministic and probabilistic crack growth computation procedures,

describing in detail the new efficient probabilistic procedure for crack growth analyses.

The demonstration of the applicability and accuracy of the new computation procedure will be

carried out in the next phase of this study, in the near future.

RESEARCH REPORT VTT-R-04910-17

9 (40)

3 On theoretical background

The main issues concerning the theoretical background of this study are described in this

chapter. Those concern the basis for the developed new efficient probabilistic procedure for

crack growth analyses, which is described in chapter 4. The actual theoretical background is

larger than that presented in the following, because only the main issues are covered, but the

associated relevant background documents are referred to, where appropriate.

weight/influence function methods and handbook solutions,

deterministic and probabilistic crack growth computation procedures, as well as

distributions for initial crack sizes.

The characteristics of the weight/influence function methods and handbook solutions used to

compute the stress intensity factor (SIF) values for cracks in load bearing components are

described in the following. For realistic cracks in actual load bearing components it is necessary

to compute the SIF values in several points over the crack front, at least in the crack tip and

edge/surface points. There are thorough handbooks available on SIF and other fracture

mechanics issues, see e.g. Anderson [1].

When performing a fracture mechanics based analysis for a cracked structure, the computed

SIF values apply only to one particular set of boundary conditions. Different loading conditions

cause different SIFs for that geometry. It turns out, however, that the solution to one set of

boundary conditions contains sufficient information to derive SIFs for any other boundary

conditions on that same geometry [1].

The weight functions, developed by Bueckner [2] and Rice [3], are first-order tensors that

depend only on the geometry of the cracked structure. Given the weight function for a particular

configuration, it is possible to compute SIFs for any boundary conditions. Moreover, the

principle of superposition is applicable too, so that any loading configuration can be represented

by appropriate tractions applied directly to the crack faces. Thus, the crack opening Mode I SIF,

KI, for a two-dimensional cracked structure can be derived from the following expression [1]:

KI p x h x dx (3.1-1)

C

where p(x) is the crack face traction (equal to the normal stress acting on the crack plane when

the structure is un-cracked), h(x) is the weight function, C is the perimeter of the crack, and x

is the coordinate through the crack starting from the cracked surface. The weight function h(x)

can be interpreted as the stress intensity resulting from a unit force applied to the crack face at

x, and equation (3.1-1) represents the superposition of the KI values from discrete opening

forces along the crack face.

The weight function concept extends well to three-dimensional structures. In their early work

on the weight functions, Rice [4] extended the theory to three dimensions, while Bueckner [2]

considered combined Mode I/II loading, and they both allowed for anisotropy in the elastic

properties. Subsequent researchers [5,6,7] have shown that the theory applies to all linear-

elastic structures that contain an arbitrary number of cracks.

RESEARCH REPORT VTT-R-04910-17

10 (40)

The influence coefficients are related to the weight functions. A surface breaking crack of depth

a with power-law crack face pressure is shown in Figure 3.1-1. This crack face pressure is

expressed as [1]:

n

x

p x pn (3.1-2)

a

where pn is the pressure at x = a, and n is a non-negative integer. For the special case of uniform

crack face pressure, n = 0. For this loading KI can be written in the following form [1]:

a

KI Gn pn (3.1-3)

Q

where Gn is the influence coefficient, and Q is the crack shape parameter, which is based on the

solution of an elliptical integral of the second kind. The value of the influence coefficient is a

function of geometry, crack dimensions, and the power-law exponent n.

Figure 3.1-1. Power-law stress distribution applied to the crack faces in a two-dimensional

structural model [1].

For a more general crack analysis case, the prevailing stress is expressed as a polynomial

function, typically of 4th order, as follows:

2 3 4

x x x x

x 0 1 2 3 4 (3.1-4)

tw tw tw tw

where tw is the wall thickness and 0 to 4 are the polynomial coefficients. The application of

the principle of superposition leads to the following expression for KI [1]:

RESEARCH REPORT VTT-R-04910-17

11 (40)

2 3 4

x x x x a

KI 0 G0 1G1 2 G2 3G3 4 G4 (3.1-5)

tw tw tw tw Q

where G0, G1, G2, G3 and G4 are the influence coefficients. While the influence coefficient

approach is useful, it has limitations, as it requires that the stress distribution is fitted to a

polynomial. There are many cases where a polynomial expression (fourth order or lower) does

not provide a good representation of the actual stress field.

The weight function method allows to compute SIF solutions for arbitrary loading. Let us

consider a surface breaking crack of depth a, subject to a normal stress that is an arbitrary

function of coordinate x. Now, the equation for KI can be given as [1]:

a

KI h x, a x dx (3.1-6)

0

where h(x,a) is the weight function. The considered flaw postulate is a semi-elliptic surface

breaking crack, see Figure 3.1-2 below. For the deepest point, i.e. tip, of a semi-elliptic crack

( = 90°), the weight function can be expressed as [8,9]:

12 32

2 x x x

h90 1 M1 1 M2 1 M3 1 (3.1-7)

2 a x a a a

where the coefficients M1 to M3 depend on the geometry and crack dimensions. The

corresponding expression for the free surface ( = 0°) is given as [8,9]:

12 32

2 x x x

h0 1 N1 1 N2 1 N3 1 (3.1-8)

x a a a

Figure 3.1-2. A semi-elliptic surface breaking crack with associated geometry parameters [1].

Both equation (3.1-7) and (3.1-8) contain three unspecified coefficients. These weight function

coefficients Mi and Ni can be inferred from the reference KI solutions for two load cases on the

configuration of interest. The choice of reference load cases is arbitrary, but it is convenient to

use uniform and linear crack face pressure, i.e. n = 0 and n = 1. The corresponding influence

coefficients for these load cases are G0 and G1. Applying in addition boundary conditions and

other necessary information, the following expressions are derived for the weight function

coefficients as a function of the influence coefficients [1]:

RESEARCH REPORT VTT-R-04910-17

12 (40)

2 24

M1 3G1 G0 (3.1-9a)

2Q 5

M2 3 (3.1-9b)

6 8

M3 G0 2G1 (3.1-9c)

2Q 5

3

N1 2G0 5G1 8 (3.1-10a)

Q

15

N2 3G1 G0 15 (3.1-10b)

Q

3

N3 3G0 10G1 8 (3.1-10c)

Q

where for equation (3.1-9) the influence coefficients G0 and G1 are evaluated at = 90° and for

equation (3.1-10) they are evaluated at = 0°, respectively. From these two equations it is also

possible to solve the influence coefficients as a function of the weight function coefficients.

Nonmandatory Appendix A KI solutions for surface breaking cracks in straight pipes and plates

using both the influence coefficients and the weight function coefficients.

A commonly applied approach in FFS guidelines to produce KI solutions for cracks in most

typical industrial components is by fitting against finite element (FE) analysis results. These

components are straight plates and pipes. The covered flaw postulates include semi-elliptic and

infinite surface breaking cracks and elliptic embedded cracks. To describe the growth process

of a crack postulate it is modelled with different sizes and aspect ratios (the latter corresponding

to a/c values, see Figure 3.1-2 for illustration of a and c).

For surface breaking semi-elliptic crack postulates in straight plates and axially orientated

surface breaking semi-elliptic crack postulates in straight pipes the KI solution equation

typically has the form:

N

KI a i fi (3.1-11)

i 0

where N is typically 5, i are factors of the polynomial of fitted stress distribution as normalized

through wall, and fi are the tabulated geometry function values, as obtained by fitting against

the underlying FE analysis results. For circumferentially orientated surface breaking semi-

elliptic crack postulates in straight pipes the KI solution equation typically has the form:

N

KI a i fi bg f bg (3.1-12)

i 0

RESEARCH REPORT VTT-R-04910-17

13 (40)

where N is typically 3, i and fi are the same as for equation (3.1-11), whereas bg is the global

bending stress and fbg are the tabulated geometry function values for the bending part, as

obtained by fitting against the underlying FE analysis results.

The commonly applied FFS guidelines presenting KI solutions using the format of equations

(3.1-11) and (3.1-12) include:

the R6 Method, Revision 4 [11],

the SSM-R-2008:01 handbook [12],

the SINTAP Procedure [13,14], and

the FITNET Procedure [15].

Some commonly applied FFS standards/guidelines use to some extent different but still simple

form of KI solution equations for cracks in typical industrial components, as fitted against FE

analysis results. Therein, the geometry function data is given both as fitted functions and as

tabulated values, depending on the crack case in question. These FFS standards/guidelines

include:

the British Standard BS 7910: 2015 [16], and

the API 579 Procedure [17].

To summarize, the SIF values for cracks in industrial components are nowadays most often

computed with the weight/influence function methods and FE analysis based solutions. The

collections of the applicable case specific SIF solutions are presented in several FFS standards

and guidelines, such as those mentioned above. Of the available SIF computation procedures,

these are the most efficient, while still being accurate enough. These procedures are relatively

easy to use, and for stationary crack cases the SIF results are obtained very quickly. Unlike with

FE analyses, as it usually takes hours or days just to construct the analysis model. Obviously,

for cases with complex geometry, three-dimensional FE analyses are often the only possible

way to obtain the needed SIF solution.

For PFM based analyses applying MCS, these SIF computation procedures quickly become too

laborious and time consuming to use. This is because at least several thousands computed crack

growth realisations for one analysis case are needed. Each such realisation is computed

incrementally, thus, to maintain the sufficient accuracy the SIF values over the crack front are

typically computed hundreds of times to obtain the needed crack growth curve.

procedures

Deterministic crack growth computations

Deterministic crack growth computations are most often done using weight/influence function

methods and handbook procedures. For components/structures with more complex geometry

and/or several different materials, FE analyses are usually used. For NPP components, the most

often considered degradation mechanisms driving the crack growth are SCC and FCG. They

come in different forms. For SCC, the crack propagation can be intergranular (IGSCC) or

transgranular (TGSCC), and on the other hand accelerated by specific environments, such as

primary water (PWSCC) and irradiation accelerated (IASCC). For FCG, the crack propagation

can be caused by cycling of mechanical or thermal loads, as well as by their combination, and

in terms of cycling rate, there is low-cycle and high-cycle loading.

RESEARCH REPORT VTT-R-04910-17

14 (40)

To model SCC, the rate equation [18] is used most often. Therein, the parameter driving the

crack growth is KI. The popularity of using this equation for NPP components is mainly due to

the available data for the material, temperature and environment specific constants. The SCC

rate equation for the crack tip is given as [18]:

da

C SCC K InSCC (3.2-1)

dt

where:

a is crack depth, in mm,

t is time, in years, whereas

CSCC and nSCC are the material, temperature and environment specific constants, with CSCC

in (mm/year)/[(MPa m)nSCC] and nSCC being dimensionless.

Usually this equation is applied incrementally over a small enough time increment. Thus, da

becomes ai-ai-1 = a, and dt becomes ti-ti-1 = t. The rate equation can be applied also to other

points in the crack front than the crack tip. Then the direction of the crack growth is taken as

the normal to the tangent of the crack front curve at the point in question. The ending criterion

of the computation is that the crack depth reaches some pre-defined value or the opposite

surface of the component wall.

To model FCG, the Paris-Erdogan equation [19] is used most often. Therein, the parameter

driving the crack growth is the range of KI, i.e. KI = KI,max-KI,min., where KI,max is the maximum

KI value and KI,min is the minimum KI value, respectively. The popularity of using this equation

for NPP components is mainly due to the available data for the material, temperature and

environment specific constants. The FCG rate equation for the crack tip is given as [19]:

da

C FCG K InFCG (3.2-2)

dN

where:

N is the number of load cycles, dimensionless, whereas

CFCG and nFCG are the material, temperature and environment specific constants, with CFCG

in (mm/load cycle)/[(MPa m)nFCG] and nFCG being dimensionless.

Obviously, the load cycles occur as a function time, which information gives the rate and

duration of the load cycling sequence. Usually this equation is applied incrementally over one

load cycle. Thus, da becomes again a, and dN becomes 1. If necessary to speed up the

computation, dN can be also several load cycles. This approach is called load cycle lumping,

and it is to be taken into account accordingly in the value of CFCG. The Paris-Erdogan equation

can be applied also to other points in the crack front than the crack tip. Then, the direction of

the crack growth is taken as the normal to the tangent of the crack front curve at the point in

question. The ending criterion of the computation is, again, that the crack depth reaches some

pre-defined value or the opposite surface of the component wall.

As mentioned earlier, in the PFM based analyses applying MCS at least several thousands of

crack growth realisations are computed for each analysis case. In such probabilistic crack

RESEARCH REPORT VTT-R-04910-17

15 (40)

growth computations, one or more of the input data parameters are considered as having a

probabilistic distribution. The values for these input data parameters are then picked at random

from the respective distributions for each crack growth realisation.

exponential,

normal,

log-normal,

Poisson,

Weibull,

Gamma,

Gumbel, and

linear.

Typical distributed initial data parameters in the PFM analyses are the depth and length of the

initial crack size. This issue is described in more detail in Chapter 3.3. Other PFM input data

parameters with distributed characteristics include crack growth equation parameters, fracture

toughness, and load cycling rate. To achieve sufficient accuracy, the number of the needed

crack growth realisations, i.e. simulations, increases exponentially with the number of input

data parameters taken as distributed.

Nf

Pf (3.2-3)

Ns

where:

Nf is the number of the failed simulations, dimensionless, and

Ns is the size of the sample, dimensionless.

Here Ns is the number of the crack growth simulations. The inaccuracy of a MCS result can be

evaluated with the following simple equation [22]:

1 Pf

inaccuracy [%] 200 (3.2-4)

N s Pf

The reliability of this formula is 95 %, which means that the probability for the inaccuracy of

the MCS result being less than the value given by equation (3.2-4) is 95 %. The failure state

can be chosen as appropriate for the case in question, e.g. for NPP pipe components it can be

leak (crack has just grown through wall) or break (plastic collapse or total/guillotine break).

The main drawback with this plain MCS is the computational effort involved [23]. This problem

is emphasized when it is attempted to assess such probabilities that, based on the underlying

data, are known to be relatively small, such as pipe breaks. A variety of techniques have been

developed to reduce the number of samples required. One recommendable such technique for

PFM analysis purposes is Latin Hypercube simulation (LHS). This technique is based on

stratified sampling of probability distributions. The approach is to divide each distribution into

Ns equal probability intervals, where Ns is here the number of simulations. For each simulation

RESEARCH REPORT VTT-R-04910-17

16 (40)

run, one parameter value is selected from each distribution, each region being selected only

once.

VTT is using for Pf computations the discrete time Markov procedure, see VTT Research Notes

2382 [24]. This procedure, which has also been developed further by VTT, divides the crack

growth realisations into a discrete number of so called degradation states. The Pf computations

are carried out using degradation potential and inspection matrices, the latter corresponding to

the effect of yearly inspections. The Markov procedure allows efficient computation process

applying LHS, as well as practical incorporation of the effect of inspections. For NPP piping

components, the probability of detection (POD) of crack like flaws is typically expressed with

POD functions, that are often material type and degradation mechanism specific. It is assumed,

that when a crack is detected in inspections, the pipe component in question is either repaired

or replaced. Which returns it to the intact state. For POD functions applicable to NPP piping

components, see e.g. the report on the PFM code WinPRAISE [25], report NUREG/CR-3869

[26] and SKI Report 2005:03 [27]. However, it would be beyond the scope of this study to go

further in this issue.

The Markov procedure applied by VTT for the degradation potential analyses of NPP piping

components is summarised as follows [24]:

1. Construction of degradation matrix transition probabilities from crack growth simulation

results and database analysis of crack initiation frequencies.

2. Model for inspection quality, as based on applicable POD functions, which are in turn used

to construct inspection matrix transition probabilities.

3. Markov model to calculate pipe leak/break probabilities for chosen inspection programs,

including the case of no inspections.

4. Writing the time or load cycle dependent leak/break probability results to a separate file.

To be able to carry out deterministic or probabilistic crack growth analyses, part of the

necessary input data are estimates of the initial crack sizes. Probabilistic distributions for initial

crack sizes, the issue here, are used in the PFM and MCS based crack growth simulations. These

distributions are typically expressed with commonly used distribution functions, see Chapter

3.2 for those. The values for the initial crack depth and length are picked at random from the

respective distributions for each crack growth simulation.

As the data in the NPP component degradation databases concern only grown cracks, the sizes

of the initial cracks have to be assessed recursively. This is not a straightforward task, and

consequently, there are not many applicable estimates available for initial crack sizes. Some

selected probabilistic distributions for sizes of initial cracks in NPP pipe components are

presented in the following.

Khaleel and Simonen [28] have developed distributions for the sizes of the manufacturing

induced cracks in NPP piping components of stainless steel. They were determined as

circumferentially orientated semi-elliptical cracks opening to inner surface. Log-normal

distribution is used for the crack depth, denoted as fx(x) [-], with associated tabulated parameter

data given in ref. [28]. This log-normal distribution equation is:

RESEARCH REPORT VTT-R-04910-17

17 (40)

2

1 ln x y

fx x exp 2 (3.3-1a)

2 yx

2 y

and:

~

y ln X (3.1-1b)

where:

x is the variable of interest, in mm,

y is shape parameter, dimensionless,

y is scale parameter, in mm, and

~

X is the median of x, in mm.

Exponential distribution is used for the crack length, by giving it as crack aspect ratio, [-].

Then, = c/a, where:

c is half of the crack length, in mm, and

a is the crack depth, in mm.

0, 1

2

f0 C 1 ln , 1 (3.3-2)

12

exp

2 2 2 m

= 0.5382,

C = 1.419, and

m = 1.136.

This aspect ratio distribution is assumed to be independent of the flaw depth. A common

distribution is used for all pipe sizes addressed in the computations.

The depth and length of SCC induced initial cracks according to the PFM code WinPRAISE

[25] are presented in the following. These initial cracks are determined as circumferentially

orientated semi-elliptical cracks opening to inner surface. The length of the initial cracks is

determined as log-normally distributed, with a median value of 1/8 inches (3.175 mm) and a

shape parameter value of 0.85, whereas the depth of the initial cracks is determined as 0.001

inches (0.0254 mm).

The depth and length of the SCC induced initial cracks according to PFM code NURBIT [29,30]

are presented in the following. They are circumferentially orientated cracks breaking to inner

RESEARCH REPORT VTT-R-04910-17

18 (40)

surface. The depth of the initiated cracks is taken to be 1.0 mm. The probability density function

for the initial crack length, fal(l0), was estimated from a total of 98 IGSCC cases in Swedish

stainless steel girth welds in straight pipes, as collected from their nine boiling water reactor

(BWR) units. In most cases, IGSCC was confirmed by metallographic evaluations.

A correction was made for the observed crack lengths in order to obtain the initial crack lengths

by subtracting an amount of crack growth at each crack tip equal to the observed crack depth.

This procedure was motivated by the observation that once the crack has been initiated the

absolute crack growth in the depth and length direction is of the same order. The following

truncated exponential distribution for fal(l0) [1/mm] was fitted to this sample:

l0 1

f al exp 1 e H 2 Ri l0 (3.3-3)

2 Ri 2 Ri

where:

is a fitted model parameter, dimensionless,

l0 is initial crack length,

Ri is inner radius of pipe cross-section, in mm, and

H is the Heaviside step function, dimensionless.

The last two factors in equation (3.3-3) are due to the truncation. The parameter was chosen

with 0 equal to 9.380 so that the mean values of the observed and fitted distributions coincided.

This corresponds to a mean value for 1/ 0 of 10.66 % of the inner pipe circumference.

Models that contain a large number of parameters can be difficult to analyse by using MCS

methods. The difficulty arises in trying to obtain a representative sample of parameter values

from their distributions. Having a large number of parameters requires a large number of Monte

Carlo simulations to produce defensible results. One approach to obtaining a representative set

of model output data is to use Latin hypercube simulation (LHS) to obtain a representative set

of input samples to evaluate using the model.

LHS is based on stratified sampling of probability distributions. The approach is to divide each

distribution into n intervals of equal probability. In the simulations, one point from each interval

is sampled, so that the interval specific location is chosen randomly. Thus, if each distribution

is divided into two parts and there are p parameters, there will be 2 to the power of p sampling

intervals. When a random approach is used for selection, then the order of intervals from which

the points are sampled is also selected randomly [32]. An alternative is to use the median of

each interval in the analysis [33].

LHS generally gives better results in calculating the tails of a probabilistic distribution and

requires fewer simulations relative to plain MCS [33].

Ding et al. [34] have applied LHS for the calculation of the fracture probabilities of NPP

components. Their application is useful from the viewpoint of this study. As a starting point,

the probabilistic variables in the work equation can have values between 0 and 1. Then, the

region between 0 and 1 is divided uniformly into N non-overlapping intervals for each

probabilistic random variable, where N is the number of random numbers that needs to be

RESEARCH REPORT VTT-R-04910-17

19 (40)

generated for each random variable, i.e., the number of the simulations. The N non-overlapping

intervals are selected to be of the same probability of occurrence. Then, N different values in

the N non-overlapping intervals are selected randomly for each probabilistic random variable,

i.e., one value per interval is generated. This constrained sampling scheme leads to accurate

results with considerably smaller number of simulations than with the MCS.

Thus, when applying LHS the distribution functions for initial crack sizes described in Section

3.3 need to be discretized to regions having the same probability. Let us consider a two

dimensional (2D) space spanned by the depth and length of the initial crack sizes. For a set of

initial crack size distributions this space needs to divided into a number of non-overlapping

rectangles that amass equal probability.

RESEARCH REPORT VTT-R-04910-17

20 (40)

analyses

The developed efficient probabilistic procedure for crack growth analyses applies interpolation

between a set of case specifically computed crack growth simulations. This development is

mainly provided by the author from Teollisuuden Voima Oyj (TVO), and to a lesser extent by

the author from VTT.

The crack growth simulations should cover all input data parameters considered as

probabilistically distributed. The ones considered here are the depth and length of the initial

cracks. They are considered simultaneously here with a relatively small number of crack growth

simulations, covering the range of validity of the given underlying probabilistic distributions.

Then, it is checked for each crack growth simulation that at which point the corresponding SIF

value equals the material and degradation mechanism specific threshold value. These values

describe such a curve under which no crack growth takes place.

After that, the majority of the crack growth simulations, or rather realisations, are obtained by

interpolation. This is based on the initial crack sizes, as picked randomly from the respective

probabilistic distributions. It is checked between which two actually simulated crack growth

curves the randomly picked initial crack size is located. The crack growth curve is then

interpolated by keeping the relative distance from the two nearest simulated curves the same as

that for the initial crack size.

The obtained results are used for the degradation state probability computations with the

Markov application, up to the failure state, corresponding to pipe leak or break. More input data

parameters can be considered distributed too, and they would be treated correspondingly in the

analyses. Compared to any existing Monte Carlo procedure, involving at least several thousands

of crack growth simulations applying fracture mechanics, the needed computational work using

simple interpolation is much smaller. The number of the needed actual crack growth simulations

is less than 1 % of that needed with any Monte Carlo procedure.

The steps of the interpolation based probabilistic procedure for the crack growth analyses are

described in detail in the following. A flow chart of the procedure is presented at the end of this

section, see Figure 4.1-2. A computational example concerning the interpolation part of the

procedure is presented in Appendix 2.

Check the scope of the probabilistic distributions for depth and length of initial crack.

To allow applying LHS, select the number of intervals, nLHS, to consider both the initial

crack depth and length, it should be at least 40.

Divide the 2D space spanned by the initial crack depth and length into nLHS×nLHS = NLHS

rectangles having the same probability.

Select values from both distributions:

o Choose the number of grid points for the initial depth and length values, denoted as

NGRID. For the initial depth values the length coordinate is 1 %, and for the initial

length values the depth coordinate is 1 %, respectively.

RESEARCH REPORT VTT-R-04910-17

21 (40)

o Recommendation 1: Take 1, 10, 20, 30, 40, 50, 60, 70, 80, 90 and 99 % points. This

results with NGRID = 11+11-1 = 21 cases to be simulated, see Figure 4.1-1 for

illustration.

o Recommendation 2: Take 1, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75,

80, 85, 90, 95 and 99 % points. This results with NGRID = 21+21-1 = 41 cases to be

simulated.

For the given pipe geometry, material properties, loads and degradation mechanism

equation, perform fracture mechanics based deterministic crack growth simulations for the

NGRID cases, and save the results.

Step 2: Computation of the crack size curve corresponding to the SIF threshold

Check for each of the NGRID crack growth simulations at which point the corresponding SIF

value equals the material and degradation mechanism specific threshold value, which data

is saved.

These data points are connected with straight lines to describe such a curve under which no

crack growth takes place, see Figure 4.1-1 for illustration of Recommendation 1.

a [mm]

10,1

10,0

9,1

9,0

ai+1(tl), ci+1(tl)

…

3,0 3,1

ak(tl), ck(tl)

2,0 2,1

L2 ai (tl), ci(tl)

L1

1,0 1,1

1,2

1,3 1,9

1,10

0,0 0,1 0,2 0,3 … 0,9

0,10

c [mm]

Figure 4.1-1. Illustration of a crack growth analysis set-up, with two simulation curves shown

with continuous lines and one crack growth realisation with a dashed curve.

RESEARCH REPORT VTT-R-04910-17

22 (40)

For increased computational efficiency, LHS is used, as based on ref. [34]. However, a more

straightforward modification of the procedure is applied here.

For each crack growth realisation:

o Pick at random values from the probabilistic distributions for depth and length of

initial crack, one value from each.

o Check is the initial crack size below the crack size curve corresponding to SIF

threshold or matching it, and if it is, the crack does not grow, and if it is above the

curve, continue as follows.

o Check between which two already computed crack growth simulations the initial

crack falls, see Figure 4.1-1, where a is crack depth and c is half of crack length l.

o Check inside which of the NLHS rectangles the initial crack falls. These rectangles

are denoted as LHSl,m, where l, m = 1, 2, 3, …, nLHS.

o When the initial crack has fallen to such a rectangle for which a crack growth

realisation has already been computed, it is taken to equal the existing one, i.e. in

this case no computations are made.

o When the initial crack has fallen to such a rectangle for which a crack growth

realisation has not been computed, interpolate the crack growth realisations based

on the two crack growth simulations by using equations (4.1-1a), (4.1-1b), (4.1-2a)

and (4.1-2b).

o When all of the NLHS rectangles have been tried at least once, the computation is

stopped. The number of computed realisations is denoted as NCGR. The minimum

value for NCGR is 5000.

o Save the results.

For the crack growth realisation example presented in Figure 4.1-1, the following

discretized time dependent equations for the crack depth, ak(tl), and half-length, ck(tl), are

derived:

L1a

a k tl ai tl ai 1 tl ai tl (4.1-1a)

L1a L 2a

L1c

ck tl ci t l ci 1 tl ci t l (4.1-1b)

L1c L 2c

where:

o sub-index i is for the simulated crack growth curves, with i = 1 for the curve nearest

to the c axis, see Figure 4.1-1,

o for the crack growth realisation number k the discretized time tl runs as tl = t0, t1, t2,

…, tend., whereas L1a, L2a, L1c and L2c are obtained with equations (4.1-2a) to

(4.1-2f) below.

Special case 1: When L1a = L2a = 0 and L1c & L2c 0, the equation for ak(tl) becomes:

L1c

a k tl ai tl ai 1 tl ai tl (4.1-1c)

L1c L2c

Special case 2: When L1a & L2a 0 and L1c = L2c = 0, the equation for ck(tl) becomes:

RESEARCH REPORT VTT-R-04910-17

23 (40)

L1a

ck tl ci tl ci 1 tl ci t l (4.1-1d)

L1a L 2a

Note, that tl can be replaced with the number of load cycles, Nl, when necessary, where Nl

= N0, N1, N2, …, Nend.

The lengths L1 and L2 correspond to the minimum distances from the initial crack size ak(t0)

& ck(t0) to the two simulated crack growth curves between which it is located, see Figure

4.1-1.

To obtain L1, find such (ai(tl), ci(tl)) that:

2 2 12

L1 min ai tl a k t0 ci t l ck t0 (4.1-2a)

To obtain L2, find such (ai+1(tl), ci+1(tl)) that:

2 2 12

L2 min ai 1 tl a k t0 ci 1 tl ck t0 (4.1-2b)

Now, L1a, L2a, L1c and L2c are obtained as follows:

L1a ai t l min

ak t0 (4.1-2c)

L1c ci tl min

ck t 0 (4.1-2d)

L2 a ai 1 t l min

a k t0 (4.1-2e)

L2 c ci 1 tl min

ck t 0 (4.1-2f)

Note, that the two simulations typically grow to leak/break state after different times. This

is to be taken into account in the computations by letting the stopped simulation to have the

end state crack dimensions until the crack growth realisation computation stops to the

leak/break state.

Do the Markov analysis run using as input data the crack growth realisations computed in

Step 3.

Save the obtained time or load cycle dependent degradation propagation and leak/break

probability results.

RESEARCH REPORT VTT-R-04910-17

24 (40)

Start

Check the scope of the probabilistic distributions for depth and length of initial crack.

To allow applying LHS, select the number of intervals, nLHS, to consider both the initial

crack depth and length, it should be at least 40.

Divide the 2D space spanned by the initial crack depth and length into nLHS×nLHS = NLHS

rectangles having the same probability.

Choose the number of grid points for the initial depth and length values, NGRID. For the

initial depth values the length coordinate is 1 %, and for the initial length values the depth

coordinate is 1 %, respectively.

o Recommendation 1: Take 1, 10, 20, 30, 40, 50, 60, 70, 80, 90 and 99 % points. This

results with NGRID = 11+11-1 = 21 cases to be simulated.

o Recommendation 2: Take 1, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75,

80, 85, 90, 95 and 99 % points. This results with NGRID = 21+21-1 = 41 cases to be

simulated.

For the given pipe geometry, material properties, loads and degradation mechanism

equation, perform fracture mechanics based deterministic crack growth simulations for the

NGRID cases, and save the results.

Save the results.

Check for each of the NGRID crack growth simulations that at which point the corresponding

SIF value equals the material and degradation mechanism specific threshold value.

Save the resulting data.

Connect these threshold points with straight lines to describe such a curve under which no

crack growth takes place.

Figure 4.1-2. A flow chart of the steps of the interpolation based probabilistic procedure for

crack growth analyses.

RESEARCH REPORT VTT-R-04910-17

25 (40)

Pick at random values from the probabilistic distributions for depth and length of initial crack

threshold crack size

no curve ? yes

computed crack growth simulations the

initial crack falls

Save the result

Check inside which of the NLHS rectangles the initial crack falls. These rectangles are denoted as

LHSl,m, where l, m = 1, 2, 3, …, nLHS.

When the initial crack has fallen to such a rectangle for which a crack growth realisation has

already been computed, it is taken to equal the existing one, i.e. no computations are made.

When the initial crack has fallen to such a rectangle for which a crack growth realisation has

not been computed, interpolate the crack growth realisations based on the two mentioned

crack growth simulations, by using equations (4.1-1a), (4.1-1b), (4.1-2a) to (4.1-2f).

Save the results.

When all of the NLHS rectangles have been tried at least once, the computation is stopped.

The minimum value for the number of computed realisations, NCGR, is 5000.

yes Nk NTOT ?

no

Do the Markov analysis run using as input data the results from Step 3.

Save the degradation propagation and leak/break probability results.

End

RESEARCH REPORT VTT-R-04910-17

26 (40)

This study concerns development of a new efficient probabilistic procedure for crack growth

analyses. The main target for it is PFM based crack growth simulations, which involve applying

MCS. To achieve sufficient accuracy, the number of computed crack growth simulations needs

to be at least several thousands for one analysed location, e.g. pipe cross-section. With

commonly applied weight and influence function methods such an analysis can take hours. For

doing analyses for a NPP piping system with hundreds of locations/welds to be analysed, this

is way too long. This motivated the development of a more efficient but still accurate crack

growth analysis procedure.

To provide the relevant starting points, the theoretical background is described first. This

includes:

weight/influence function methods and handbook solutions,

deterministic and probabilistic crack growth computation procedures, together with

describing the modelling of two degradation mechanisms relevant for NPP piping, namely

SCC and FCG, as well as

distribution models for initial crack sizes.

The developed efficient probabilistic procedure for crack growth analyses applies interpolation

between a large enough representative set of case specifically computed crack growth

simulations. These simulations are to cover all input data parameters considered as

probabilistically distributed. Here they are the depth and length of initial cracks. To consider

both of them simultaneously, it is sufficient to compute altogether at maximum approximately

40 crack growth simulations. The interpolation is based on the initial crack sizes, as picked

randomly from the respective probabilistic distributions. It is checked between which two

actually simulated crack growth curves the randomly picked initial crack size is located. The

crack growth curve is then interpolated by keeping the relative distance from the two nearest

simulated curves the same as that for the initial crack size.

The obtained results are used for degradation state probability computations applying the

discrete time Markov procedure, up to the failure state, corresponding to pipe leak or break.

In VTT, the crack growth simulations are done with probabilistic VTTBESIT code, whereas

the Markov computations are carried out with a special application, see ref. [24] for more details

on these analysis tools.

Two additional alternative efficient probabilistic procedures for crack growth analyses were

developed too. They are presented in Appendix 1.

The purpose is to program in VTT the developed more efficient probabilistic procedures for

crack growth analyses in the near future. This will include the demonstration of its applicability

and accuracy. It is expected that, once programmed, the application of the procedure in

probabilistic NPP piping component leak/break analyses is much more efficient than using

solely the weight/influence function method based solutions.

RESEARCH REPORT VTT-R-04910-17

27 (40)

References

1. Anderson, T. FRACTURE MECHANICS - Fundamentals and Applications, Third

Edition. CRC Press, Taylor & Francis Group, 2005. 630 p.

2. Bueckner, H.F. A Novel Principle for the Computation of Stress Intensity Factors.

Zeitschrift für Angewandte Mathematik und Mechanik, Vol. 50, 1970, pp. 529-545.

3. Rice, J.R. Some Remarks on Elastic Crack-Tip Stress Fields. International Journal of

Solids and Structures, Vol. 8, 1972, pp. 751-758.

4. Rice, J.R. Weight Function Theory for Three-Dimensional Elastic Crack Analysis. ASTM

STP 1020, American Society for Testing and Materials, Philadelphia, PA, 1989. pp. 29-

57.

5. Vainshtok, V.A. A Modified Virtual Crack Extension Method of the Weight Functions

Calculation for Mixed Mode Fracture Problems. International Journal of Fracture, Vol.

19, 1982, pp. R9-R15.

6. Sha, G.T., Yang, C.-T. Weight Function Calculations for Mixed-Mode Fracture Problems

with the Virtual Crack Extension Technique. Engineering Fracture Mechanics, Vol. 21,

1985, pp. 1119-1149.

7. Sham, T.-L. A Unified Finite Element Method for Determining Weight Functions in Two

and Three Dimensions. International Journal of Solids and Structures, Vol. 23, 1987, pp.

1357-1372.

8. Shen, G., Glinka, G. Determination of Weight Functions from Reference Stress Intensity

Solutions. Theoretical and Applied Fracture Mechanics, Vol. 15, 1991, pp. 237-245.

9. Zheng, X.J., Kiciak, A., Glinka, G. Weight Functions and Stress Intensity Factors for

Internal Surface Semi-Elliptical Crack in Thick-Walled Cylinder. Engineering Fracture

Mechanics, Vol. 58, 1997, pp. 207-221.

10. ASME Boiler and Pressure Vessel Code, Section XI, Rules for In-service Inspection of

Nuclear Power Plant Components, Division 1. American Society of Mechanical Engineers

(ASME), New York, US, 2015.

2004 update of 2001 edition. British Energy (BE).

12. Dillström, P. et al. 2008. A Combined Deterministic and Probabilistic Procedure for

Safety Assessment of Components with Cracks – Handbook. SSM Research Report

2008:01, Swedish Radiation Safety Authority (Strålsäkerhetsmyndigheten, SSM).

Stockholm, Sweden, 2008. 27+196 p.

13. SINTAP - Structural Integrity Assessment Procedures for European Industry - Final

Procedure: November 1999. Project funded by the European Union (EU) under the Brite-

Euram Programme: Project No. BE95-1426, Contract No. BRPR-CT95-0024.

14. Barthelemy, J., Y., Janosch, J., J. Structural Integrity Assessment Procedures for European

Industry; SINTAP; Task 4; Compendium of Residual Stress Profiles; Final Report:

RESEARCH REPORT VTT-R-04910-17

28 (40)

18.5.1999. Project funded by the European Union (EU) under the Brite-Euram

Programme: Project No. BE95-1426, Contract No. BRPR-CT95-0024. 40+18 pages.

European Fitness-for-Service Thematic Network – FITNET, Germany, January 2008.

acceptability of flaws in fusion welded structures. The British Standards Institution (BSI),

2015. 492 p.

17. Recommended Practice for Fitness-for-Service 579 (API 579). American Petroleum

Institute (API), Washington, DC, U.S., 2000.

18. Congleton, J., Craig, I., H. Corrosion Fatigue. In: Corrosion Processes, Editor Parkins, R.,

N. Applied Science Publishers, 1982.

19. Paris, P.C., Erdogan, F. A Critical Analysis of Crack Propagation Laws. Journal of Basic

Engineering, Vol. 85, 1960. pp. 528-534.

Sensitive Structures. In: Provan, J.W. (ed.). Probabilistic Fracture Mechanics and

Reliability. 1987, Martinus Hijhoff Publishers. pp. 1-45.

21. Ang, A., H-S., Tang, W., H. Probability Concepts in Engineering Planning and Design,

Volume II, Decision, Risk, and Reliability. John Wiley & Sons. U.S.A., 1984. 530+33 p.

22. Millwater H., Monte Carlo Simulation. In the course: 11th Annual Short Course on

Probabilistic Analysis and Design, Computational Methods and Applications. September

1115, 2000. Southwest Research Institute (SwRI), San Antonio, Texas, U.S.

23. Health and Safety Executive Hazardous Installations Directorate. Probabilistic Methods:

Uses and Abuses in Structural Integrity. Prepared By BOMEL Limited. England, 2001.

103+117 p.

24. Cronvall, O., Männistö, I., Simola, K., Development and testing of VTT approach to risk-

informed in-service inspection methodology, Research Notes 2382, Technical Research

Centre of Finland (VTT), Espoo, Finland, 2007. 43 p.

25. Khaleel, M., A., Simonen, F., A. Evaluations of Structural Failure Probabilities and

Candidate Inservice Inspection Programs. Report NUREG/CR-6986, U.S. Nuclear

Regulatory Committee (US NRC), March 2009. 194 p.

26. Simonen, F., A., Woo, H., H. Analyses of the Impact of Inservice Inspection Using a

Piping Reliability Model. NUREG/CR-3869, Topical Report. U.S. Nuclear Regulatory

Commission (USNRC), July 1984. 59 p.

27. Jelinek, T., Tidström, L., Brickstad, B. Probability of Detection for the Ultrasonic

Technique according to the UT-01 Procedure. SKI Report 2005:03, Swedish Nuclear

Power Inspectorate (SKI). Sweden, 2005. 38 p.

28. Khaleel, M., A., Simonen, F., A. Effects of alternative inspection strategies on piping

reliability. Nuclear Engineering and Design 197 (2000) 115–140.

RESEARCH REPORT VTT-R-04910-17

29 (40)

29. Brickstad, B. The Use of Risk Based Methods for Establishing ISI-Priorities for Piping

Components at Oskarshamn 1 Nuclear Power Station. SAQ/FoU-Report 99/5, SAQ

Control AB, Sweden, 1999. 83 p.

30. Brickstad, B. Appendix D1, A Short Description of the NURBIT Piping Reliability

Program for Stress Corrosion Cracking Analyses. NURBIM Project, WP-4, Review and

benchmarking of SRMs and associated software, May 2004. 17 p.

Utgåva: 4, ABM Bilaga 7A, Materialdatabok MD-01 Revision 3,

Spänningskorrosiontillväxt i BWR Miljö, T-SEK 41/99. Sweden, 2001-04-01. 246 p.

32. Inman, R., L. 1999 Latin Hypercube sampling. Encyclopedia of statistical sciences,

Update volume 3, Wiley, New York, U.S.A. pp. 408-411.

El-Shaarawi, A., H., Piegorsch, W., W. John Wiley & Sons, 2002, U.K. pp. 2283-2297.

34. Ding, K., Zhou, Z., Liu, C. Technical Note - Latin hypercube sampling used in the

calculation of the fracture probability. Reliability Engineering and System Safety 59

(1998) 239-242.

35. Pal, S., Ibrahim, R. N. and Singh Raman, R. K. Studying the effect of sensitization on the

threshold stress intensity and crack growth for chloride stress corrosion cracking of

austenitic stainless steel using circumferential notch tensile technique. Engineering

Fracture Mechanics, vol. 82, pp. 158-171, 2012.

36. Pal, S. and Singh Raman, R. K. Determination of threshold stress intensity for chloride

stress corrosion cracking of solution-annealed and sensitized austenitic stainless steel by

circumferential notch tensile technique. Corrosion Science, vol. 52, pp. 1985-1991, 2010.

37. ASME Boiler and Pressure Vessel Code, Section II. 2015 Edition.

38. Nuclear power plant units Olkiluoto 1 and Olkiluoto 2. Teollisuuden Voima Oyj (TVO),

PDF Publication in English, from http://www.tvo.fi, 24 November 2011. 51 p.

39. Roark, R., Young, W. Formulas for Stress and Strain. Fifth Edition. McGrawHill Book

Company, U.S.A., 1975. 624.

40. Dillström, P. et al. 2008. A Combined Deterministic and Probabilistic Procedure for

Safety Assessment of Components with Cracks – Handbook. SSM Research Report

2008:01, Swedish Radiation Safety Authority (Strålsäkerhetsmyndigheten, SSM).

Stockholm, Sweden, 2008. 27+196 p.

RESEARCH REPORT VTT-R-04910-17

30 (40)

growth analyses

The developed two alternative efficient probabilistic procedures for crack growth analyses are

described in the following. The first one of them applies interpolation between a large enough

representative set of case specifically computed crack growth simulations. The second one is

based on picking the nearest crack growth simulation. These developments are provided by the

author from VTT.

These simulations should cover all input data parameters considered as probabilistically

distributed. Typical examples of such parameters are the depth and length of initial cracks. To

consider both of them simultaneously, it is sufficient to compute altogether less than 150 crack

growth simulations, covering the range of validity of the given underlying probabilistic

distributions. Then, the majority of the crack growth simulations, or rather realisations, are

obtained by interpolating between the results from the actual simulations.

The interpolation is based on the initial crack sizes, as picked randomly from the respective

probabilistic distributions. In alternative procedure 1 it is checked between which four actually

simulated crack growth curves the randomly picked initial crack size is located. The crack

growth curve is then interpolated by keeping the relative distance from the four nearest

simulated curves the same as that for the initial crack size.

The obtained results are used for degradation state probability computations with the Markov

application, up to the failure state, corresponding to pipe leak or break. More input data

parameters can be considered distributed too, and they would be treated correspondingly in the

analyses. Compared to any existing Monte Carlo procedure, involving at least several thousands

of crack growth simulations applying fracture mechanics, the needed computational work using

simple interpolation is much smaller. The number of needed actual crack growth simulations is

less than 10 % that needed with any Monte Carlo procedure. The accuracy of the new procedure

is very good too.

The two procedures are described in more detail in the following. They concern computations

for one crack analysis case at a time. The input data parameters considered as probabilistically

distributed are the depth and length of the initial crack.

The alternative procedure 2 is more simple than the interpolation procedure based alternative

procedure 1. As it is also to some extent less accurate, its usage requires a larger amount of

crack growth realisations. This procedure is described after the alternative procedure 1.

The steps of the interpolation based alternative probabilistic procedure 1 for crack growth

analyses are described in detail in the following.

RESEARCH REPORT VTT-R-04910-17

31 (40)

Check the scope of the probabilistic distributions for depth and length of initial crack.

Select values from both distributions at 1, 10, 20, 30, 40, 50, 60, 70, 80, 90 and 99 % points,

and make all possible combinations, which results with 11×11 = 121 cases.

Check for all initial cracks is the corresponding SIF value less than or matching the material

and degradation mechanism specific threshold value. No simulations are carried out for the

initial crack cases with SIF value below the threshold value or matching it. However, these

cases are included in the further analyses by having them keep the initial crack size through

the considered time span.

For the given pipe geometry, material properties, loads and degradation mechanism

equation, perform fracture mechanics based deterministic crack growth simulations for the

121 cases, and save the results. For some illustrative examples of the crack growth

simulations, see the five continuous curves in Figure A1.1-1.

a [mm]

…

9,0 9,1 9,2 9,3 9,9 9,10

… … … … …

3,0 3,1 3,2 3,3 3,9 3,10

…

2,0 2,1 2,2 2,3 2,9 2,10

…

1,0 1,1 1,2 1,3 1,9 1,10

… X

c [mm]

Figure A1.1-1a. Illustration of a representative set of crack growth simulations, with some

simulation curves shown with continuous lines, and the set-up for one crack growth realisation

with a dashed curve. For more details of the set-up, see Figure A1.1-1b.

RESEARCH REPORT VTT-R-04910-17

32 (40)

a [mm]

… …

…

L4 L1a

L4a L1

X

L3a L3 L2 L2a

…

0,9 L3c L2c 0,10

c [mm]

Figure A1.1-1b. Detail of the set-up for one crack growth realisation.

To allow applying LHS, select the number of intervals, nLHS, to consider both the initial

crack depth and length, it should be at least 40.

Divide the 2D space spanned by the initial crack depth and length into nLHS×nLHS = NLHS

rectangles having the same probability.

For each crack growth realisation:

o Pick at random values from the probabilistic distributions for depth and length of

initial crack.

o Check is the SIF value of the initial crack size less or matching the material and

degradation mechanism specific threshold value, and if it is, the crack does not grow,

and if it is higher than the threshold value, continue as follows.

o Check between which four already computed crack growth simulations the initial

crack falls, see Figure A1.1-1, where a is crack depth and c is half of crack length l.

o Check inside which of the NLHS rectangles the initial crack falls. These rectangles

are denoted as LHSl,m, where l, m = 1, 2, 3, …, nLHS.

o When the initial crack has fallen to such a rectangle for which a crack growth

realisation has already been computed, it is taken to equal the existing one, i.e. in

this case no computation is made.

o When the initial crack has fallen to such a rectangle for which a crack growth

realisation has not been computed, interpolate the crack growth realisation based on

RESEARCH REPORT VTT-R-04910-17

33 (40)

the four crack growth simulations by using equations (A1.1-2a) to (A1.1-2f) and

(A1.1-1g).

o When all of the NLHS rectangles have been tried at least once, the computation is

stopped. The number of computed realisations is denoted as NCGR. The minimum

value for NCGR is 5000.

o Save the results.

For the crack growth realisation example presented in Figure A1.1-1, the following time

dependent equations for the crack depth, ak(tl), and half-length, ck(tl), can be derived:

a k tl a1,10 tl a0,10 tl a0 , 9 tl a1, 9 tl (A1.1-1a)

La La La La

L1c L 2c L3c L 4c

ck tl c1,10 tl c0 ,10 tl c0 , 9 t l c1,9 tl (A1.1-1b)

Lc Lc Lc Lc

where for crack growth realisation number k the discretized time tl runs as tl = t0, t1, t2, …,

tend, whereas the sub-indexes for a and c in the right side of the equation correspond to the

four points shown in Figure A1.1-1b.

Note, that tl can be replaced with the number of load cycles, Nl, when necessary, where Nl

= N0, N1, N2, …, Nend.

For the example of Figure A1.1-1, the equations for the length parameters L1a to L4a, L1c

to L4c, as well as La and Lc are:

L2 a a0,10 t0 ak t0 (A1.1-2b)

L3a a 0 ,9 t 0 a k t0 (A1.1-2c)

L 4a a1,9 t0 a k t0 (A1.1-2d)

L2 c c0 ,10 t0 ck t0 (A1.1-2f)

L3c c0 , 9 t 0 ck t 0 (A1.1-2g)

L 4c c1,9 t0 ck t 0 (A1.1-2h)

The general equations for ak(tl) and ck(tl) are derived next. By denoting the row number as

i and the column number as j, with both running as 0, 1, 2, 3, …, 10, equations (A1.1-1a)

and (A1.1-1b) are written in general form for the crack growth realisations as:

RESEARCH REPORT VTT-R-04910-17

34 (40)

L1a L2 a L3a L 4a

a k tl ai 1, j 1 tl ai , j 1 tl ai , j tl ai 1, j tl (A1.1-3a)

La La La La

L1c L2 c L3c L 4c

ck tl ci 1, j 1 tl ci , j 1 t l ci , j tl ci 1, j tl (A1.1-3b)

Lc Lc Lc Lc

where the general equations for the length parameters L1a to L4a and L1c to L4c now

become:

L1a ai 1, j 1 tl a k tl (A1.1-4a)

L2 a ai , j 1 tl a k tl (A1.1-4b)

L3a ai , j tl ak tl (A1.1-4c)

L4 a ai 1, j tl a k tl (A1.1-4d)

L1c ci 1, j 1 tl c k tl (A1.1-4e)

L2 c ci , j 1 tl ck t l (A1.1-4f)

L3c ci , j t l c k tl (A1.1-4g)

L4 c ci 1, j tl ck t l (A1.1-4h)

whereas La and Lc are still computed with equation (A1.1-2i) and (A1.1-2j).

Note, that the four simulations typically grow to leak/break state after different times. This

is to be taken into account in the computations by letting the stopped simulations to have

the end state crack dimensions until the crack growth realisation computation stops to the

leak/break state.

For each crack growth realisation the set-up configuration of L1 to L4 and their projections

to a and c axes are the same, i.e. L1 corresponds to the straight distance from the randomly

picked initial crack size to the simulation nearest at the top right side of it, L2 corresponds

to the straight distance to the simulation nearest at the bottom right side of it, L3 corresponds

to the straight distance to the simulation nearest at the bottom left side of it, and L4

corresponds to the straight distance to the simulation nearest at the top left side of it.

Do the Markov analysis run using as input data the crack growth realisations computed in

Step 2.

Save the obtained time or load cycle dependent degradation propagation and leak/break

probability results.

RESEARCH REPORT VTT-R-04910-17

35 (40)

The steps of the alternative probabilistic procedure 2 for crack growth analyses are described

in detail in the following. The procedure is based on picking the nearest simulation.

o Pick at random values from the probabilistic distributions for depth and length of

initial crack.

o Check in terms of initial crack depth and length that between which four already

computed crack growth simulations the values fall, see Figure A1.1-1.

o Compute the distances L1 to L4 from the initial crack location to the four mentioned

simulation starting points by using equations (A1.1-2c) to (A1.1-2f).

o Choose the crack growth realisation to be the nearest of these simulations, i.e. the

one with the minimum value of L1 to L4.

o When all of the NLHS rectangles have been tried at least once, the computation is

stopped. The minimum value for NCGR is 10000.

o Save the results.

Do the Markov analysis run using as input data the crack growth realisations computed in

Step 2.

Save the obtained time or load cycle dependent degradation propagation and leak/break

probability results.

RESEARCH REPORT VTT-R-04910-17

36 (40)

In the following is presented a step by step computational example concerning the interpolation

part of the probabilistic procedure for the crack growth analyses. For a detailed description of

the procedure, see Section 4.

The initial crack size distributions are taken according to PFM code NURBIT [29,30], see

Section 3.3 for a more detailed description of these distributions. The depth of the initiated

cracks is 1.0 mm, whereas the initial crack length is given as an exponential probability

density function, see equation (3.3-3).

The initial crack lengths are considered at 1, 10, 20, 30, 40, 50, 60, 70, 80, 90 and 99 %

points, while the initial crack depth is kept at 1.0 mm. Thus is created a grid of points for

the initial depth and length values, see Table A2-1 for this data.

To provide an example of the application of LHS, the selected number of intervals, nLHS,

for the initial crack lengths is 20, so that the probability amassed by each interval is equal.

For the computation of these intervals see equation (A2-1), whereas the resulting interval

data is presented in Table A2-2.

The selected pipe geometry is: outer diameter = 200 mm, wall thickness = 10 mm.

The selected pipe component base material is: austenitic stainless steel SA-376 TP304.

The necessary material properties, loads and stresses are presented in Tables A2-3 to A2-5.

The considered environment is BWR primary circuit under normal water chemistry (NWC),

as corresponding to the conditions in units OL1 and OL2 of TVO.

The considered degradation mechanism is SCC. The propagation of SCC is computed with

equation (3.2-1), using the deterministic version of the VTTBESIT code and best estimate

values from ref. [31] for the material, temperature and environment specific constants CSCC

and nSCC.

The considered location of the initial crack is the heat affected zone (HAZ) of the pipe butt

weld.

The KI threshold for the onset of SCC in sensitized austenitic stainless steel is according to

refs. [35,36] 8.0 MPa m.

The considered loads correspond to 100 % normal BWR operation with locally confined

welding residual stresses (WRSs).

All results are saved.

Table A2-1. The depth, a0, and half-length, c0, of initial cracks at 1, 10, 20, 30, 40, 50, 60, 70,

80, 90 and 99 % points of the initial crack size distributions according to NURBIT [29,30].

location = 1 % 10 % 20 % 30 % 40 % 50 % 60 % 70 % 80 % 90 % 99 %

a0 [mm] = 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

c0 [mm] = 2.8 28.3 56.5 84.8 113.1 141.4 169.6 197.9 226.2 254.5 279.9

For application of LHS, the equation derived for the computation of the start and end points of

the 20 intervals, Xm [mm], for the initial crack lengths so that the probability amassed by each

interval is equal, is as follows:

RESEARCH REPORT VTT-R-04910-17

37 (40)

B

Xm log e exp BX m 1 B (A2-1)

20 A

with:

l0 l0

A 1 e and B

2 Ri 2 Ri

where:

each interval is spanned by points Xm and Xm-1, with m = 0, 1, 2, …, 20, and

the physical meanings of , l0, Ri and H are explained in the connection of equation (3.3-3).

Table A2-2. The computed locations of the 20 intervals for the initial crack lengths, so that the

probability amassed by each interval is equal.

No. = 0 1 2 3 4 5 6 7 8 9 10

location [mm] = 0.0 1.5 3.2 4.9 6.7 8.7 10.8 13.0 15.4 18.0 20.9

No. = 11 12 13 14 15 16 17 18 19 20

location [mm] = 24.1 27.6 31.6 36.3 41.8 48.5 57.2 69.4 90.3 282.7

Table A2-3. The necessary material properties of austenitic stainless steel SA-376 TP304, as

taken from Section II [37] of the ASME code.

temperature [°C] 20 300

yield strength [MPa] 207 129

tensile strength [MPa] 517 437

Young’s modulus [GPa] 195 176

Table A2-4. The process loads that the examined pipe component is exposed to under 100 %

normal operation, from TVO ref. [38].

Load Value

temperature [°C] 286

pressure [MPa] 7.0

Table A2-5. The stresses experienced by the examined pipe component. The axial membrane

stress is computed with an analytical equation, whereas the WRSs are taken from FFS guideline

recommendations, see the associated refs. in the table. The through-wall distribution of the

WRSs is linear.

Stress component Value Ref.

membrane from pressure [MPa] 30 [38]

axial WRS at inner surface [MPa] 173 [39]

axial WRS at outer surface [MPa] -173 [39]

Total stress at inner surface [MPa] 203

Total stress at outer surface [MPa] -143

RESEARCH REPORT VTT-R-04910-17

38 (40)

Step 2: Computation of the crack size curve corresponding to the SIF threshold

It was checked for each of the 11 crack growth simulations that at which point the

corresponding KI value equals the material and degradation mechanism specific threshold

value, which data is saved.

For all simulations the maximum KI value of the initial crack was more than 9.9 MPa m,

which exceeds the KI threshold value of 8.0 MPa m.

Thus, in this case it was not needed to construct a curve connecting the simulation result

points corresponding to KI threshold.

This step was largely omitted here, as for illustration only two crack growth realisations

were computed.

For the two crack growth realisations:

o Values from the probabilistic distributions for depth and length of initial crack were

picked. These are:

ak,0 = 1.0 mm, ck,0 = 18.09 mm,

ak+1,0 = 1.0 mm, ck+1,0 = 39.58 mm.

o It was checked between which two already computed crack growth simulations the

initial crack falls.

o The crack growth realisations were computed by interpolating between the two

neighbouring crack growth simulations by using equations (4.1-1a), (4.1-1b),

(4.1-2a) and (4.1-2b).

o The results were saved.

The crack growth simulation results starting from initial half-length points of 1, 10, 20, 30,

40, 50, 60, 70, 80, 90 and 99 %, while the initial crack depth is kept at 1.0 mm, see Figure

A2-1.

The two crack growth realisation results for the two computed cases, see Figures A2-2 to

A2-4.

All these results are shown in terms of crack depth as a function of crack half-length,

whereas for comparison some detailed results are shown also in terms of crack depth or

crack half-length as a function of time.

Note, that as the simulations grew to pipe leak state after different times, it was taken into

account in the crack growth realisation computations by letting the stopped simulations to

have the end state crack dimensions until the realisation computation stopped to the leak

state.

RESEARCH REPORT VTT-R-04910-17

39 (40)

12.0

10.0

crack depth [mm]

8.0

6.0

a1, c1 a2, c2 a3, c3 a4, c4

4.0 a5, c5 a6, c6 a7, c7 a8, c8

a9, c9 a10, c10 a11, c11

2.0

0.0

0 50

100 150 200 250 300

crack half-length [mm]

Figure A2-1. The crack growth simulation results starting from initial half-length points of 1,

10, 20, 30, 40, 50, 60, 70, 80, 90 and 99 %, while the initial crack depth is 1.0 mm.

12.0

a1, c1

10.0 a2, c2

a3, c3

crack depth [mm]

8.0 ak, ck

ak+1, ck+1

6.0

4.0

2.0

0.0

0 20 40 60 80

crack half-length [mm]

Figure A2-2. Detail showing two crack growth realisations and three simulations. Therein,

realisation ak, ck has been interpolated according to simulations a1, c1 and a2, c2, whereas

realisation ak+1, ck+1 has been interpolated according to simulations a2, c2 and a3, c3,

respectively.

RESEARCH REPORT VTT-R-04910-17

40 (40)

12.0

a1, c1

10.0 a2, c2

a3, c3

crack depth [mm]

8.0 ak, ck

ak+1, ck+1

6.0

4.0

2.0

0.0

0100 50 150 200

time [year]

Figure A2-3. Detail showing the crack growth realisations and simulations of Figure A2-2 in

terms of crack depth as a function of time.

80

a1, c1

a2, c2

crack half-length [mm]

60 a3, c3

ak, ck

ak+1, ck+1

40

20

0

0 100 50 150 200

time [year]

Figure A2-4. Detail showing the crack growth realisations and simulations of Figure A2-2 in

terms of crack half-length as a function of time.

As can be seen from Figures A2-2 to A2-4, the computation of the crack growth realisations by

interpolation well describes the crack growth paths falling between the crack growth

simulations. As the interpolation equations are considerably simple, this method is

computationally very efficient. It can be argued that with the used initial crack size distributions

the smallest size was too large for the simulations, being: a0 = 1.0 mm, c0 = 2.83 mm. More

suitable values for the smallest initial crack half-length could be e.g. 0.5 or 0.1 % values,

corresponding to 1.41 or 0.28 mm, respectively. The maximum initial crack half-length could

be kept at 99 % value, or changed correspondingly to 99.5 or 99.9 % value, respectively.

- Doane Chapter 06Diunggah olehThomasMCarter
- DNV RP F108Diunggah olehVicknesh Thanabal
- ME CAD-CAM_2010Diunggah olehRaja Raj Raj
- rp-f108Diunggah olehsuzilamatsalleh
- Chapter #19 Solutions- Engineering Economy, 7 th editionLeland Blank and Anthony TarquinDiunggah olehMusa'b
- Discrete random variablesDiunggah olehChandra Sekhar D
- Fatigue Primer for EngineersDiunggah olehstaplesjf
- CTODDiunggah olehdani_ong
- Misurata UniversityDiunggah olehDustin Ellis
- tugas 2 metolidDiunggah olehalya
- 53214400-A-Fatigue-Primer-for-Structural-Engineers.pdfDiunggah olehSandro De Carvalho
- Model for Quantifying Risks (Table of Content)Diunggah olehWalter Wong
- Modeling Discon XFEMDiunggah olehHarish Lambadi
- Numerical and Experimental Analysis of the Directional Stability on Crack Propagation Under Biaxial StressesDiunggah olehAPRIL_SNOW
- A Distributed CSMA Algorithm for Throughput and Utility Maximization in Wireless NetworksDiunggah olehMuthu Maha Raja
- 11.1 Probabilistic Risk Analysis - Examples_with NotesDiunggah olehFlora Lin
- 14.2016 Effect of a New Specimen Size on Fatigue Crack Growth Behavior in Thick-walled Pressure VesselsDiunggah olehArun Kumar
- Wang - Industrial Challenges for Thin Wafer ManufacturingDiunggah olehrini44f
- pm10_tiskDiunggah olehmshameli
- sdarticleDiunggah olehkazitani_nabil
- ctsunindexDiunggah olehvenkat_venkat
- pipeline_group_somerday_ms.pdfDiunggah olehAnonymous eTLUmY
- Markov ChainsDiunggah olehAna-Maria Babanica
- materials-11-00499Diunggah olehMisael Souto
- Eco StatDiunggah olehRani Gil
- Chapter 4 - Probability - The Study of RandomnessDiunggah olehyakamoto1
- 1998 Material Ease 5Diunggah olehAnonymous T02GVGzB
- MohanDiunggah olehmohan_rapaka
- Introduction to PrabilityDiunggah olehKaleemullah Bhatti
- 60487_STAT 151_WeaverDiunggah olehContinuing Education at the University of Vermont

- Nulife-08-23 - Scc Pilot Study Ra-3Diunggah olehOtso Cronvall
- Report_VTT-R-08651-12Diunggah olehOtso Cronvall
- Report_VTT-R-02199-10_FRAS-1-1-3_26Apr2010Diunggah olehOtso Cronvall
- SMiRT20 Manuscript VTTORC Div II Paper 1797Diunggah olehOtso Cronvall
- Cracking in Swedish BWR internals - Nuclear Engineering and Design 1990Diunggah olehOtso Cronvall
- Design of Nuclear Reactors Neighbouring the Nordic Countries, 1994.pdfDiunggah olehOtso Cronvall
- NUREG-1903-Seismic Considerations for the Transition Break SizeDiunggah olehOtso Cronvall
- On Hawking’s Final Theory on the Big Bang -- PBS NewsHour May2018Diunggah olehOtso Cronvall
- NULIFE-08-33 - Load History Effects on Crack Driving Force – Contribution to EG2-EG3 RPV Case Study - Dec2008Diunggah olehOtso Cronvall
- NULIFE-07-23 - Identification of Priority R-et-D and Harmonisation AreasDiunggah olehOtso Cronvall
- NULIFE-08-32 - Improved Methods for Dealing With Combined Primary and 2ndary Stresses in Defect AssessmentsDiunggah olehOtso Cronvall
- NULIFE-07-21 - Guideline for the Development of High Quality SCC Growth DataDiunggah olehOtso Cronvall
- NUREG-0313_Rev-2_Material Selection and Processing for BWR PipingDiunggah olehOtso Cronvall
- Beyond RPV Design LifeDiunggah olehOtso Cronvall
- A New Engineering Method for Prediction of Fracture Toughness Temperature Dependence for RPV SteelsDiunggah olehOtso Cronvall
- Accurate Prediction of Residual Stress in Stainless Steel Welds--2012Diunggah olehOtso Cronvall
- NUREG-CR-2815 - Probabilistic Safety Analysis Procedures GuideDiunggah olehOtso Cronvall
- NUREG-CR-5378 Aging Data Analysis and Risk Assessment EtcDiunggah olehOtso Cronvall
- EPRI-R-3002010332-MRP-418 Use of Master Curve for Pressure-Retaining MaterialsDiunggah olehOtso Cronvall
- NUREG-CR-4407--Pipe Break Frequency Estimation for Nuclear Power Plants- 1987Diunggah olehOtso Cronvall
- NUREG-CR-5587 Approaches for Age-Dependent PSAs EtcDiunggah olehOtso Cronvall
- Nureg Cr 5632 Agingpra SmithDiunggah olehWilliam Ivans
- 3445605994528Diunggah olehdudumuitoloco
- NUREG-1513 Integrated Safety Analysis Guidance DocumentDiunggah olehOtso Cronvall
- NUREG_SR_1482 REVISION 1 GUIDELINES FOR IST.pdfDiunggah olehSRVfiv
- Ml 041040063Diunggah olehNathan Block
- NUREG Sensitivity and Uncertainty Analyses in Aging Risk-based PrioritizationsDiunggah olehOtso Cronvall
- Nureg-0619 Bwr Feedwater Nozzle and Control Rod Drive Return Line Nozzle CrackingDiunggah olehOtso Cronvall

- System Development Life Cycle AssignmentDiunggah olehAugustine Barlow
- Manual NiktoDiunggah olehMade Urip Sumaharja
- ERP and Related TechnologiesDiunggah olehsaudgazali
- XXXCupid.com ReviewDiunggah olehxxxcupid-review
- Power BI Desktop - User Guide.docxDiunggah olehglen s
- DFM-Checklist-ExcelVersion.xlsDiunggah olehMark Wilkinson
- RBS 2111 VietnamDiunggah olehKhangTran
- Pack 1 - Tabs Updated EDITABLEDiunggah olehCalin Ionela-Claudia
- MOODE 2.5 eBookDiunggah olehMark Repp
- M3_Non Linear Data StructuresDiunggah olehresmi_ng
- CETE-ScilabTEC2013Diunggah olehmariomato
- Giaotrinhphp2009 Voduytuan Final 111201213145 Phpapp02Diunggah olehYure Pereira
- Network Lab Man AualDiunggah olehrajapoornima
- Evaluation of Individual ProjectsDiunggah olehVishnu N Krishna
- 10501.ADC Lab Course HandoutDiunggah olehAbhay Chaudhary
- 2017 NESA CatalogDiunggah olehJuan Guerra
- Phi SonDiunggah olehChiahsin Chen
- 2-3-1366022311-7.AVR Embedded.full.pdfDiunggah olehHarshal Patil
- A Survey and Comparative Study of Vehicular Traffic Control System (VTCS)Diunggah olehsairayousaf
- Solving Timetable Scheduling Problem by Using Genetic AlgorithmsDiunggah olehJAGuerrero
- Instructional Materials pptDiunggah olehDepEdResources
- Segment Based Visual Cryptography For Key DistributionDiunggah olehijcses
- Minimize Inrush Current in Power TransformersDiunggah olehaocalay
- Computer ArchitectureDiunggah olehprmahajan18
- 94830-186831-1-PB.pdfDiunggah olehfpopescu1425
- Practice 12Diunggah olehHansa Boricha
- Adafruit CatalogDiunggah olehgregguerrero
- SAP S4 HANA MM CertificationDiunggah olehYong Benedict
- Business Statistics Vol 1 for OnlineDiunggah olehAmity-elearning
- Rule Book 1Diunggah olehdreccox