Escolar Documentos
Profissional Documentos
Cultura Documentos
AccEPTANcE CBtTEBtA
db ahi6 dim a
wdr
GInMjm
mR
Dfld-.)I T^'I
OT NDE AC.xt'rANcE
CRIIEId
ee didi
'c[
]r'arlorl
.c'.trdi{&$ d djcdi
or
wba
,'h@rip
.b iorans
h qwi, oi *h.h
'ebEe
(bD) d dqny d ddEr rcdi@, ,'i
&
NDE
$bjdbNDE
No
@id ii bveq,
d+tu s
dr*
d. $!dry
of
rb
baptuittt
oJ
tro\|'
rtu
obi{.
Jutur
wb4
Pdiol{dddipEgtqh4dgeloed
@3tret c4tue 4.d3 kt 8p8h3
obj.d dE
do
djEr de b te
{kbE
&tu
Apr'"ftFbd!@igdd'grcdty
s&rd6ii dja.
tu DE stmr d
obadiq idtrdiq 6
^fuldgntabrcvh,h6mEl'!dddstt9oi9
mEiddimrt..ffi@idi4bmdini4de.et
subld'iE
or
s@ry iedotr
rw
hdip'idoihv*dluNDE'ehiqEililihi6p@s
ouurjry q earue sids
ituotr
mE Ehbiriry
FlddofdddsinrlryNn6*6.htv.j].dby&mEfuqcbqdid
PMilyiol!ele!'lueo'iqdtyep{lfubewi$iilrifu@g?ldtdb}
fth9fuotr@mli.yo@fug!ler
Adotdfondddiol.fuofqd
iudF@d'bkdcrd .qbeqdd.
reh
d di6k
Ftrdo
in
ung
ud
by
mE b4d
by NDE
td
'e
lpdid
rE rpprid
E4
u.d!E
.P'6ailydi*dotr@Feb!d*4lrE
IlMb4&d!&dopMdME
ddqddi*edemEdjqii@d@ddmEb$id!r'ddd1o@3h
bvl.!&o.mEdbbily&ed
d'f'sldYdsEc@[illyddlbly
Page 5 of 19
Nordtest Project
141 5-98
GuroEunes roR DevetopMENT oF NDE AccEprnruce CRreRtn
1999-02-22
2.2 Symbols
x
xa
w(x)
i
f(x)
Defect size, also used for the combination of more size (e.g. height and length) and other
defect severity parameters (e.g. location), when relevant
lowable defect size
Defect severity, often an expression for the probability of a defect leading to failure
Defect number suffix
= dF(x/dx, defect distribution (often also commonly denoted defect probability denstty
function)
NDE response
g(E;x) Response distribution
G(E;x) Response cumulative distribution
& NDE sensitivity level
p Probability of detection, often p = p(x;l%), if detection when E>Eo then
p(x;Eo) = e(E; x)' dE = I - G(Eo;x)
J
Eo
A
pr
pa
ps
C
Further information on the definitions and used symbols can be found in Chapter 4.
3. REF'ERENCES
These guidelines rely on the principles laid down in the following two documents and a crossreference is made to them on relevant topics not covered here:
Nr TscHN
2.
Nr
Guidelines
for
REPORT 394.
ddmid
a si& l.rEth, biah. ad @s$bry
duds
!i!@!irgl!!
rE:
rldd)
tdli4Bdyrcwr&tdbidle.6!d3i{oyYgqi@nPkkftndd@
i[duEnfut6olnruhdhtgir'a
eME@i.oiviewdlyquld.ome
a s4!e
GE4r
PoD
vd ca &
lib!t$6dlftddpryltsdtlhsl@n
ud
nmron{dnaGtrdiE)dd
tuau.ffit
&. .M
'b
(iEa3e)
c6 (tuD
ryrEdb
dd a prdod
dd.d FEdry
Nordtest Project
41
Page 7 of 19
5-98
1999-02-22
it
deviations.
4. 1.4 Acceptance
criteria
1.
Determination of an NDE sensitivity level to facilitate detection of all defects which might be
unacceptable
2.
After having applied the acceptance criteria an NDE system will have a certain probability of
rejecting a defect. If p,p(x) denotes the probability of a defect being rejected when it has been
detected (with a probability p), the total Probability of Rejection (PoR) for a defect amounts to
P'P.ld.
If rejection is based on sizing, sizing is done with a systematic error of e and a random
error of 6
with a cumulative distribution H(6) and rejection is based on measuring a size larger than xo, the
probability of rejecting a detected defect of size x amounts to
P,p(x)=l-H(xo-x-e)
By properly selecting
xo
o.s
.ct
oo
o.o
*
0.2
If NDE
system and
assessment
ieobo'N*]rqbeElddbfu6
d'bsdFrdi@d&r*a'tudturorddd@ tudobsME@iobjdr
FIddb4ysFdcqldlilylkeljtdi
sg
Ftu
!tu qrltt
int:lqd
by
4/rals tb
djuniig
B'
(PoF)
b Ernc
tu3brd
ib!
prd4,
nnfuh&cddmEElj?[ttydoM
s,iqlbsbeole'Ydue'fucebd6by
sG3.rui&'ddeidu)@deqldcddfu
sidqco!ffdd.rd.borlEdd
bodllsfu&ddtlidili'@
$ffitudy hid. h qutql @o.d w b d.Fn6ror e 4drcdh de dbb'v
'FJFbLsermtrlo's.loFb'hso
Nordtesl Proisct
141
$98
P96 9 of 19
1999-t2-22
T.6dnq obloct
D.f.ctg
. Mt6dat(s)
.ryp(s)
Tectlno condldons
NDEtechnlque
NDE pllonnel
. Educallon
pnnciPls
. Equli.nnt lncl.
.EqulFnent
liquids, including
cl6nlng
Clsaning elc.
.Collngs nd
in
For calculations averaging a quantity z(x) over all xi it can then in cases be convenient and
releyant to replace the su[unation over the xi by an integral over f(x) accoding to the following:
N
N.Jz(x).f(x)
!z(x,)
=
i=l
dx
when it is advantageous for analysis to assume a form of a defe{t size distribution, the lognormal
or the Weibull distdbution might be suitable. For assumption on defect contents, previous
experiencg as also occasionally reported in open literatuF, from applications similar to the one
in question, can be used,
An evaluation of defect severity should ideally yield an estimate for a defect of size (and other
significant parameters) x in terms of a Fobability w(x) for leading to failure One should then
rclate to w(x) for fomulating NDE acceptance cdteria.
In most practical cases today (1999) this routc is, however, not followed, but an allowable defect
size x" calculated (or experimentally determined) based on certain assumptions and utilising
safety factors on important defect significance input panmeters. Such a calculated allowable
yalue will then correspond to a certain (mostly ulknown and low) probability for failure of
w. = w(xJ
xu and not
criteria.
Nordlest Proisct
41
Page '10 ot 19
5-98
1999-02.22
When NDE is perfomed on an object and discrete acceptance criteda per corded indication are
used, then the defect content after NDE and corective actions is described by a defect density
n""""p,"d(x;A) =
n(x) (1
- P(x;A))
when the initial defect density is n(x) and the accptance criteria A. The total number of defects
rcmaining in the object is reduced ftom N to
N.**tA)
Jn*no(x;A).ax
00
Jn(x).(l-p.(x;A)).dx
N Jf(x)(1-p.(x;A)).dx
If
w(x) is the probability that a defect of size x may lead to an unacceptable condition, the
prcbability that any ofN defects would lead to an unacceptable condition without NDE is
N
w"
=l-II(1-w(x,))
i=l
which yields
ln(t-wb)
w"
=r-
ih(li=l
"-n[ir"fr
- *t-, ))
- "-n[N'
jit-l
r"tr
- *<"il
a,.
After NDE and conective actions the probability that any defect would lead to an unacceptable
condition is reduced to
*<ot
=,-",.0[ittl-
w(x,).(1-p,(x,;Alil)=
ll
Page
ot 19
1999-02-22
On the steps to be undertaken during development of NDE acceptance criteria rcference is made
to Sections 4. I .4 and 6. 1 , l.
4.7.2 Probabilistic acceptance criteria
If
Wo is an acceptable limit for the probability of failure of an object, the relation for W(A)
deriyed at in Section 4.6.2 can, in principle, be used to tune the acceptance criteria A, so that
w(A)<wo
The problem with dircctly doing this is that neither the numbq of defects, N, nor their
distribution f(x), or ever the type of distribution, Ne a prio'|kno.[n. An assessment would have
to be based on an assumed number of defects and an assumed distribution based on previous
experience.
If, however, the NDE performed would give information on the defect conteots of an object (see
Section 4.8) this relation could be used to determine whether an object is acceptable or not. In
fact one would then not have to rely on Fedelermined acceptance cdteria, but decide on these
based on the outcome of the llDE and the knowledge of defect seyedty. When following this
approach, it would be advantageous to do the NDE with a high sensitivity to gain as much
information on the object as possible. The setting of the sensitivity would then have to be
properly balanced against the probability for false calls (see atso Section 4.9).
For certain applications, however, such a /atal evaluation
for an object is not practical,
^pproaah
This applies for instance dudng offshore pipelaying, where
a decision on the integrity of a weld
has to be decided upon before the weld is lowered onto the seabed, precluding an assessment of
the pipeline as a whole for decision on acceptance or rejection. F this case a solution per weld
has to be sought.
If
1-II(l-wr)<wo
FI
If defect sevedty w(x) is not known from an Eryineering Critical Assessment (ECA), but only a
determined allowable defect size xn (see Section 4.5), it is very difficult to assess the
probabilities involved including the requirernents to PoD and PoR. It must, however, be assumed
that the failure probability related to x. is low and that the safety factors involved provide for an
extra safety rnargin, allowing PoR values less than (this should be known to personnel
performing the ECA), The assessment of PoD and PoR will then also be subject to an
engineering judgment by an NDE expert. In general one should aim for high, but not
Pag 12 ol19
1S-O2-22
Frorn the fomulae for W(A) deducted in Section 4.6.2 arother, quite expected, observation is
clearly evident: The largq the number of defects, or the larger the object, the larger is the
probability of failure, or, - many small defects can be as dangerous as one large one. This rofurze
elct is usually completely disregarde4 when acceptance criteria are fomulated for individual
defects, but some common criteda contain a cumulative clause (e.g, the total length of defects in
a specified length of weld shall not exceed a cedain value), taking this, at least partly, into
account.
Simplified aoproach
of observations of a quantity xi for i = l, 2, ..., n and the observed xis are all
assuned to be outcomes of a distribution f(x), then the parameters of f(x) can, accoding to the
maximun likelihood principle,be detemircd, by maximizing the likelihood expression
If
ni is the number
flf(x,)''
i=l
If now NDE
!n,.lnf(x,)
i=l
of sizes xr, x2, ..., xi, ..., x, in an object and the conesponding,
known probabilities of dete{tion are p(x) the estimated 'expected number' of defects of size xi is
l/O(x), The parameters of an expe{ted defect size distribution f(x) can then be determined by
maximizing the expression
has rcyealed n defects
lnf(x,)
p\-)
N=$p(x,I
?i
Pg 13 ol 19
1999-02-22
No"-*
= Jn(x). p(x). dx = N
.J(x). p(x).dx
r*."-c)
If
"(.).
pGy'i"<fl
f(x) is assurned known, but with unknown pammeter, p(x) is known, and the NDE
has included sizing (and other characterization) giving the sizes xi b = 1,2,...,Naa-r"a), the
parameters of f(x) can be determined by maximizing the likelihood expression
the form of
N.*
flrup.w)f
1-
[nt.p(t
If
the sizes xj have not been detemined by the NDE, a similar approach can be followed for the
observed NDE indications E (se the relation given in Section 2.2 baween PoD and the
distribution of NDE indications).
If only
the form of p(x) is known (or assumed), but not its parameters, the maximum likelihood
principle can be used to detemine also these paranrete$.
If
As for maximum likelihood estirnation, the distribution type for f(x) is assumed known, whereas
the palametels of this distribution are unknown. The task is to estimate these distribution
paramers. For the purpose of estimating a parameter e in the density function f(x) of the defect
size x, it is assumed that e is a stochastic variable O with known distribution function Fo(o) and
coresponding density function fo(o) = dFe/do. Fo(e) is denoted thep/i,rr distribution of @ and is
based on prior knowledge and experience:
Fo(O) = Fo. p'i.i(Q) = P[O < 0]
The NDE leads to a set of observed defects. The information inherent in these observations is
used in conjunction with the prior information about e to produce so-called Bayesian estimates
of g. Two cases are considered for the oulcome of the NDE:
(l)
(2)
The po.rteliol distribution of @ is a conditional distribution, - given the prior information and the
sample data from the NDE:
Nordlest Proisct
141
Page 14 ot 19
1999-02-22
5-98
For Case (l )i
= xN
)]=
PP<O
r"",,,n.,(e)=rP<e lx
<
xol=
p[o<e
nx
<
P[x. *ol
^o]
Methods are available to calculate the probabilities in these expressions. Note that x0 is the
magnitude of a non-detectable defect, whose distribution is given by the probability of detection,
p(x), which is assumed known.
The estimator 6 for 0 is chosen as the mean value of the posterior dist.ibution of O, hence for
Case (1):
6
=E[@
l(x,
= x,
^x,
= x2
n...nX*
xp)l
6=s[olx<xo]
4,9 Cost-effectiveness evaluation
Cses
Crar*
Cx*6,
The expected total costs related to NDE can then be represented by a cost function (here given in
a simplified fashion) as function of NIDE efforts A (representative for applied acceptance criteria
including NDE sensitivity setting):
Pago 15 ol 19
1999-02-22
If
9
'E
{/l tnasns
.a
dp=
inspscli@ eno
dp"
Cr.{,i.
CF,jr*"
- CR
>0
u,*tv
-9
- o
Costopinlm3lope
co3t opliduin Polnr
Probblllty of
5.
fals cll
techniques
5.1 Introduction
Quality control NDE is aimed at checking whether an acceptable workmanship level is met or
not. If this workmanship level is not meq there is a chance, in the long run, to have unacceptable
defects introduced into the object(s) in question. violating the quality requirement should
thereforc lead to corrective actions. This is often done as a punishment by forcing manufacturers
and their personnel (welders, etc.) to do corrective actions by rcpair or replacement. Morc
seldom the violation of the workrnanship standards leads to a fitness-for-purpose assessment of
the findings made, which could avoid damaging repairs to be done.
Quality levels have often ben developed over time, and are as such empirically based, often as
expreEsed by the used acceptance criteria f traditional NDE techniques. Altematively an
engineering judgement is used to define quality levels gxpressed by defect severity parameters
like size, type and location as is done in ISO 5817 / EN 25817 Atc-welded joihts in steel Guidance on qualiE levels for imperfections. In both cases the quality level definition is thus
somewhat arbitrary. This fact can also be taken into account when formulating NDE quality
confol acceptance criteria.
5.2 Principles
A quality or workmanship level is something that should be met on an 'average' in the long run.
The requirements to the applied NDE and acceptance criteda are thus also not more stringent
than to reject defects coresponding to the quality level on an average, i.e. in 5070 of the cases.
Ed'y q rd9
pads (i4
@ei.i@6budb.bsddlFcifidd&
rqd@
9F)
rrc{h4e
or
6. Fliditiry
or
s.
6a&eMca{r.orqb; d
dl)niqErol![jdDlFjffcdEie
7..]l4t4ilhoBlfub'd.1v|@
in
turtr
2,
evildi
lq 4te,c
NM
d.t4l'ebqELqldbqhgluda5,
Fmdrd4mEacidi.lo
PjdPt*qidv!Jdd!d,dcbbdbid
dkq v6ch dghr d b u lourBd! @drh d e objd, q ! 4h Fobb ni 6r
@Ad.dbocn* r&is3iy6r&iib
.lomof'dbiwoff.il6nridioiof&le'
l.td'y'Me6,.tiFl!d4din
b4!B 4h e bed otr e Eb iy upddn3
utd rbgh DE
ididiom
NDE
dr, d lEpr
Gr
dd
Fn
e@t d
rren
i@fut{ndudngdishPipdqb&ykdrnfunby}ddngdh!d8
d & piF ib rhr s. A dadd oi Eer@. d rtd'ioi of .*d srd b ro h d lihn
sNilpdoE@disdddl!!46g.
djd
7.2)
!cDr'.dd, r
3.U$qFn@ftdtl.leim
ftdq3d(hadblsrc3ikibch4tr1
A@'qlddldnucffu$.id.@@..pp!d'fuFre6abi1kd.dfut@bdoldc
DE vhi.h .h b fdo,!d. vb a drd ffiry alsnqr o y tdd jrsdL d.id
lLB ( S.io 4t ad rb DE idld d4d. dr'8 bbqE n o![d dov.
ftsrpp{dlbiN!$e6l]ovi4sbpsi
lhrdebhgidnscd'ialA!G!4
ud3 b BS m64e' q rd e
2UgdastrEtyhi3hMEjddviqbB.Irdd*d'dd!
3. u! d trEld d.rd sir b j!d8" @ 3!r ft Ejdid
lis.
d'i
(E
y cro
2)
dro
Page '!8 o, 19
1999-92-22
well above those aphieved by an empirically accepted quality control NDE technique (illustrated
below). For sizing tolerances the 95 or 99 enor distribution quantile limits may be used.
Such fltness-for-purpose qiteria are often" as ar! additional safety measure, supplemented by
specific quality or workmanship cdteda to maintain weld quality at a prcdeftned, acceptable
level. Violating these quality criteria may then lead to stricter acceptance criteria to be used for
consecutive welds until the weld quality has bee[ re-established.
Figures are given below illustrating the above approach, PoD adequacy and assessment of sizing
accuracies and limits.
14
'12
x50
Er0
E
8
E6
o
80 100
120
Probabilistic routines for developing optimised NDE fitness-for-purpose criteria may become
very complicated (see Section 6.1) and are not yet (1999) fully developed in a simplified form,
but approaches can be used or established based on the guidance given in Chapter 4, especially
Sections 4.4 to 4.9 (see also Section 6,1.1). Corrunercially available probabilistic softw:re
packages are available for doing the required calculations and optimisation. A description of
such software does, however, fall beyond the scope of these guidelines, and reference is made to
appropriate literature on the topic, including information from relevant software companies.
Norcltest Project
41
$98
Page 19 of 19
't99go2-22
PrcbaHlitv of dotectton
l-0-
AIJT:
FoDdm
G s%6iddE
-Rad'.sdiy
FoD
tuDo mns
@re
inn
0,1
lor
.
.
ToFD bsod
990 quanlil
-950/0
fot
quliry
in
obsoMfons
tleishl
qlantloln sitno
-lvleasured
ccepled
obs6Mto.s
Heisht=
'
omr dirributon
sitno ercr
'*i
1",
F.;
@...o..