Sei sulla pagina 1di 18

1985 The Royal Institution of Naval Architects.

Structural Reliability Analysis of Stiffened Panels


by_ D. K. Hart,* B.Sc.(Eng.), M.Sc., C.Eng. (Member), S. E. Rutherford,* B.Sc., D.I.C., Ph.D., C.Eng. (Member), and A. H. S.
Wtckham, * B.Sc.(Eng.)
Read in London at a meeting of the Royal Institution of Naval Architects on October 3, 1985, The President, Professor J. B.
Caldwell, O.B.E., Ph.D., D.Sc., F. Eng., in the Chair.
SUMMARY: The paper describes the structural reliability analysis of stiffened panels forming part of the hull envelope of two large oil
tanker cases designed in accordance with Lloyd's Register's current oil tanker Rules. Throughout the paper, structural reliability theory is
viewed as a potential tool to be effectively used in the development of classification rules and regulations.
The structural reliability analysis methods presented are based on advanced level2 structural reliability theory. It is demonstrated that
these methods can be used to evaluate the reliability of structures having time-varying strength and acted upon by combinations of time-
varying loads.
The Society's ultimate strength approach used to calculate the stiffened panels' strength is briefly presented.
The paper uses probabilistic reasoning to develop defmitions of those uncertain quantities which influence structural performance,
including corrosion diminution. Extreme value theory is used to derive probabilistic models of time-varying loading processes such as
wave-induced bending moment and hydrodynamic pressure.
The paper tentatively quantifies the influence of corrosion control, still-water bending moment control and steel mill rolling tolerance
control on the structural reliability of the stiffened panels.
1. INTRODUCTION
The basic aim of structural reliability analysis is the rational
treatment of uncertainties which affect structural performance
and the associated problems of rational decision making. It is
now widely recognised that some small risk of unacceptable
structural performance must be tolerated. In quantifying this
risk, structural reliability analysis can play an important part in
the development of classification requirements for the design,
construction and service of ships.
This paper can be considered as an initial step in the direction of
the quantification of acceptable risk. The paper investigates the
structural reliability of stiffened panels subject to corrosion. The
panels form part of the hull envelope of two ship cases which are
designed in accordance with the Society's current oil tanker
Rules. For each year of service, the notional probability of their
deck and bottom stiffened panels failing at the ultimate strength
limit states is calculated using advanced reliability analysis
techniques.
Particular attention is given to the influences of corrosion
control, still-water bending moment control and steel mill rolling
control. The study concerns itself wit}l axial stress and lateral
pressure arising from normal service loads. It does not cover the
reliability treatment of accidental loads, gross error, structural
redundancy, or failure at other limit states such as fatigue and
brittle fracture.
2. RELIABILITY THEORY
2.1 General
The concept of the reliability index f3 as a measure of structural
reliability and the probability of structural failure P r as a
quantification of risk is now well established. Within the field of
ship structural reliability f3 has generally been defined in terms of
the demand D and capability C, which are defined by their means
/lo. fAc and standard deviations o
0
, ac (I,
2

3
l. Thus,
(I)
*Lloyd's Register of Shipping
and
where <I> signifies the standard normal cumulative distribution
function.
(2)
Although this distribution-free approach offers attractive
simplicity, it is limited in its application and can only provide
approximate values of f3 and P r Key problems with the approach
are:
(i) In general, D and C are not normally distributed and may be
correlated.
(ii) The equations only apply to single independent failure
modes.
(iii) D and C are generally non-linear functions of a number of
basic variables. The distribution of a variable may well be
non-normal and also it may be truncated to reflect effective
controls and contraints.
(iv) The basic theory applies only to a single time-varying loading
process.
(v) The basic theory cannot treat time-varying strength
problems.
The problems with the above approach are overcome by the more
general theory adopted by the authors. The theory, which is
presented in the following sections, embraces advanced level 2
reliability analysis methods which incorporate systems reliability
and stochastic process theory. The presentation is restricted to
independent basic variables but can be extended with no loss of
generality to deal with correlated basic variables. Although the
theory presented does not provide a fully probabilistic solution
such as is furnished by the Monte-Carlo analysis presented in
Ref. 4, it does provide an accurate, efficient and powerful
analytical tool which is particularly suited to code development
applications. The general structural reliability theory used in the
present study is that compiled by Thoft-Christensen and Bakeri
5
l
and reviewed by Wickham<
6
l.
2.2 Basic Theory
Both the capacity of the structure and the demand imposed on it
are trcaied as functions of quantities whose values are to a greater
or s ~ r e'\tent uncertain; these uncertain quantities are termed
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
basic variables. The probability of failure Prof a structure is the
probability that the realisation of the basic variable vector results
in failure.
The basic variable space is an n-dimensional Cartesian space,
each axis of which corresponds to one of the basic variables X;,
for i = I ton. Each point in the basic variable space therefore
defines a realisation x of the basic variable vector. Each point in
the basic variable space also defines the value of a probability
density function fj((X) which is the joint probability density
function of the basic variables.
For each of the potential failure modes of the structure a safety
margin, Mi for the jth mode of failure, is defined in terms of the
basic variables by means of a failure function.
(3)
such that the probability of failure in mode j is equal to the
probability that Mi is less than zero. The probability is given by
the integral of the Joint probability density function of the basic
variables over that part Qi of the basic variable space within
which Mi is negative, i.e.
Pr = P(Mi<O) = f fX(X)dx
J Qj
(4)
The procedure for evaluating equation (4) is as follows. The
original basic variable space (or X-space) is transformed into a
standard normal variable space (or U-space) within which the
variables are all mutually independent. In general, this is achieved
by means of the Rosenblatt transformation given by
Hohenblicher and Rackwitz<7l. In the present study, because the
original X-space basic variables are all mutually independently
distributed, the required transformation is defined by
x; = Fx:(<l>(u;)) fori = I ton (5)
where <I> ( ) is the standard normal cumulative distribution
function, and Fx!( ) is the inverse of the cumulative distribution
function of the ith basic variable.
The probability of failure is approximated by the probability
content of the failure region obtained by linearising the
transformed failure boundary at the point of its closest approach
to the origin of the U-space using a truncated Taylor's series. The
distance from the origin of the U-space to the linearisation point
is defined by Hasofer and Lind<
8
l as the reliability index (3, see
Fig. I . The probability of failure is defined in terms of the
reliability index by equation (2).
Fig ' Lineari,;mion or f<ti lurc boundar: in
:itanclardi sed nmm<1l vnri<1b!e -;pace
The value of the reliability index is detemlined as follows. The
mapping functions given in equation (5) are combined with the
failure function such that the safety margin M of a structure is
defined as a function of the vector 0, i.e.
M = g(TI) (6)
Given a starting point u*, a better approximation to the point at
which the failure boundary in U-space is closest to the origin is
given by
/
j=n j = n
u;"+ out = (a _E af). (_L aiuj - g(TI*))
J= I J= I
fori = I ton
where
a, = o(g(U*))/ ou;
The reliability index (3 is given by
i=n
{J = ( E

when ou"' = 0 for j = I to n


i=l J
2.3 Treatment of Time-varying Loads
(7)
(8)
(9)
In this study, the Borges-Castanheta load process model<
9
l has
been used to define the time-varying loads in time-independent
form. This load process model assumes that each load process
consists of rectangular pulses of fiXed duration, see Fig. 2. The
intensity of the load within each pulse is a basic variable. The
Borges-Castanheta load combination rule, derived from
Turkstra's rule<
10
l, defines 2k - l combinations of non-time-
varying loads to replace a combination of k time-varying loads,
see Table I and Fig. 2. Extreme value theory is used to derive the
distribution functions for the maximum load corresponding to a
given number of repetitions of the original rectangular pulse
process, 1.e.
F,((x) = (Fx(x))' (10)
where F,(( ) is the distribution function of the maximum of r
repetitions of the variable whose distribution function for a single
repetition is Fx( ).
0 T
X
2
(t)
r
2
pulses over T
fx, (x
2
) t (time)
... ..
0 T
Combination 1 -(Maximum of r
1
repetitions of X
1
)+(1 repetition of X
2
)
Combination 2 -(Maximum of .2:.'. repetiti ons of X,) +( maxi mum of r
2
repet i ti ons of X, \ ' '
f ig. 2. Borges-Casranheia loacl combina<io n ruk for l\\' O
recta ngula r pulse processes, ' ';
1
:mel
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
TABLE I. Borges-Castanheta load combination rules for three
time-varying loads
Load No of Repetitions of:
Combination Load I Load2 Load3
rt rzlrt r/r
2
2
rl r/r
1
3
r2 r/r
2
4
rJ
Note: Total number of load pulses in reference period for loads
I, 2 and 3 are r
1
, r
2
and r
3
respectively, where r
3

r
1

2.4 System Reliability Analysis
If the structure has I potential structural failure modes and is
acted on by k time-varying loads, the reliability analysis
procedure described in Sections 2.2 and 2.3 must be carried out a
total of /.2k-l times.
In the present study, the virtual elementary failure modes
corresponding to the 1.2k-l analyses are serially related. The
overall failure region in the U-space is therefore defined by the
union of thei1 elementary failure regions Q;, see Fig. 3. The
overall probability of failure P(Q) is given by the Ditlevsen
bounds Ot.m, i.e.

+
i=n
E max ((P(Q;) -
i=2
(II)
i = n i = n
P(Q) E P(Q;) - E max (P(Q; n Qi)) (12)
i = 1 i = 2 j<i
where P(Q;) is the probability of failure in the ith virtual
elementary failure mode. P(Q; n Q) is found by numerical
integration of the bivariate normal probability density function,
i.e.
I oo oo
P(Q; n Q) = 1r f f I/(2(1-ety'
5
).
(J; (Jj
(13)
and Q;i is the correlation coefficient between the ith and jth virtual
elementary failure regions. The reliability index is found from
equation (2).
Fig. 3. Overall failure region Q for three serially related failure
modes
Within the present study, the structural reliability has been
evaluated over intervals of one year for each year of service up to
the 20th year. These annual reliabilities and probabilities of
failure are calculated using the imposed stiffened panel loads and
strengths appropriate to the end of the year being considered.
Thus, the annual values of (J and P r reflect the effects of
corrosion diminution of both the hull girder and the stiffened
panel. An approach formulated by Wickham(l
3
>, which is an
extension of the above theory, can be used to evaluate the
cumulative probability of failure.
3. PROBLEM DEFINITION
3.1 General
The deck and bottom stiffened panels of the two large oil tanker
cases used in the study are located in the cargo/ballast centre
tanks which are situated amidships. As shown in Fig. 4, axial
stress is assumed to act on a deck panel, and both axial stress and
lateral pressure to act on a bottom panel. No other load
components, e.g. in-plane shear and transverse stress, are
considered to act on the panels as their influence on the analysis is
judged to be small.
P, axial load

a) Deck member
'-1 1--

p, lateral pressure
b) Bottom member
Fig. 4. Stiffened panel beam-column model
3.2 l..oad Sources
In general, the sources of axial load on the stiffened panels are
still-water bending moment (SWBM), low frequency vertical
wave-induced bending moment (VWBM), the ship's own wave
train, thermal loads, and high frequency springing and slamming
loads. However, for a large oil tanker, attention can be restricted
to SWBM and VWBM as the likelihood of the other loads having
a significant effect at the same time as an extreme value of
VWBM is considered small. The axial load on the stiffened
panels is calculated using simple bending theory, i.e.
a = a, + Ow = (SWBM + VWBM)/Z (14)
where Z is the deck or bottom hull section modulus, as
appropriate. The axial load P on a single stiffener and attached
plating member is given by P == oA, where A is the section area of
the member.
As ship motions are not included in the present study, it is
necessary to approximate the net lateral pressure on the bottom
panel. A conservative value of reliability is calculated by
assuming the tank containing the panel is always empty. Then,
the net lateral pressure is given for both load and ballast
conditions by
P = P, + Pw
(15)
where p, is the still-water pressure and p,, the maximum (pt<<k)
wave-induced pressure.
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
In the study SWBM, VWBM and lateral pressure are considered
as time-varying loading processes. Two sets of these processes,
corresponding to load and ballast voyages, are analysed. During
each year of service, the ship is assumed to be operated for 6
months in both load and ballast conditions.
3.3 Strength of Stiffened Panel
To assess the strength of stiffened panels a beam-column
approach is used, where the panel is idealised as a single stiffener
together with an effective width of plating. The overall axial
strength is obtained from a strut formulation in which the
individual plate and stiffener strengths provide limiting extreme
fibre stresses. Full details of the formulation are given in Ref. 14.
Expressions for obtaining the effective width factor K and the
plating strength oprtl (Refs. 14 and 15) are dependent on the plate
deformation 60 and the level of residual stress a,:/B is evruated
by combining the fabrication tolerance (b/200) oy! o
0
with
deformation which occurs under lateral pressure. The residual
stress is obtained from the following expressions which allow for
the reduced influence of o, in stocky panels:
For a< 10 o, = 0
For 10 ~ ~ 1-6 o, = oy(a-1)/6
For a>l6 a,=01ay
where: a = (b/t) VTa7EJ
oy = yield stress
b = width of plating
t = thickness of plating
o
0
= the yield strength of mild steel.
A number of alternative expressions for fabrication tolerance
have been suggested by other authors, and the use of these in the
plate strength equations has been investigated. Over the range of
slenderness generally associated with ship structures, most
expressions were found to produce similar results. The value of
residual stress was selected on the basis that higher initial levels
will be reduced by 'shakedown' in service. Percentage errors
quoted in Section 4.6 were obtained using these fabrication
values, and good correlation was achieved.
The strength of the stiffening members a, is obtained by solving
the following expression from Ref. 16, where oE represents the
critical strength of the stiffener allowing for contraint at the
web/plate junction:
The value Y), which allows for imperfections, is obtained from
empirical relationships given in Ref. 16.
(16)
The overall initial deformation Llo is assumed to take the form of
a single half-wave along the length L of the panel. Magnitudes
are consistent with Merrison tolerances, these being L/900 when
the bow causes compression in the plate and L/1200 when it
causes compression in the stiffeners; these values have been used
to produce the percentage errors quoted in Section 4.6. As it is
possible for either mode to occur during fabrication, two strength
calculations are necessary, one involving plate compression
failure for positive Llo \lnd the other involving stiffener
compression failure fot negative Ll
0
The possibility of tensile
stiffener failure occurring under compressive loads is small and
has therefore been neglected.
The maximum compressive force P c acting through the centroid
of the cross-section, can be obtained for the plate failure mode
from:
0 =--+ --- +
K.Pc K.yr [Pc.Pe.Ll MQ]
pm A< Ie (P e - p J
where: Ae = effective cross-sectional area.
Ie = effective second moment of area.
I\ = Euler load.
(17)
Yp = separation between effective centroid and middle-
plane of plate.
296
The term A is a measure of the load eccentricity experienced at
mid-span and is obtained by summing the imperfection level
Ll o = L/900, the shift of the neutral axis resulting from loss of
plate effectiveness, and the overall deformation resulting from
lateral pressure application. A correction factor of the form
(10 -02 vo;ra;J, where oe is the Euler stress, is then applied to
improve the solution for low slenderness beam-columns; this
correction has been shown to produced better correlation with
test data. The term MQ represents the moment at mid-span
resulting from lateral pressure and is calculated on the
assumption that the ends of the panel are rotationally fixed.
For compressive stiffener failure, Pc is obtained from:
0 = Pc_li. [Pc-Pe.Ll + MQ]
' A I <Pe-PJ
where: A = total cross-sectional area
I = total second moment of area
y, = separation between the centroid and the extreme
stiffener fibres.
(18)
The term Ll is obtained by summing the fabrication imperfection
- L/1200 with the lateral pressure deformation, and factoring
the result as decribed above. The use of total section properties
rather than effective values has been shown to improve results for
this mode of failure (Ref. 14).
For members in compression, P cis taken as the lesser of the two
values obtained above. When uniaxial tension is applied in
conjunction with moderate levels of lateral pressure, it is assumed
that full yield can be sustained and P cis calculated from
Pc=oy.A.
In deriving section properties, the effect of uniform corrosion is
allowed for by taking each thickness as
t = t
0
- rn (19)
where t
0
is the initial nominal thickness, r the annual corrosion
rate and n the number of years in service.
3.4 Failure Criterion
A stiffened panel is considered as failed if, for a particular failure
mode and load condition or combination, P c ~ P.
4. UNCERTAIN QUANTITIES
4.1 General
Within the context of structural reliability analysis, uncertainties
are classified as being physical, model and statistical. Physical
uncertainties include such quantities as dimensions, material
properties and loads. Model uncertainties are associated with the
accuracy of analytical or numerical methods. Statistical
uncertainties reflect a lack of knowledge concerning the statistical
distributions of physical and model uncertainties. Physical and
model uncertainties required for the reliability analysis of the
stiffened panels are developed in the following sections.
However, it should be noted that appropriate actuarial
information is of limited quality and quantity, and the definitions
of uncertainties should be considered as reasonable
approximations.
Initial imperfections of the plating and stiffeners, residual
stresses, Young's modulus and Poisson's ratio are quantities
whose variability is included in the strength model uncertainty.
Consequently, in the reliability analysis they are assigned nominal
values. The model uncertainty on structural response reflects
departure from simple bending theory in the calculation of
longitudinal stress acting on the stiffened panel. Subjectively, the
uncertainty is given a mean of 1 0, a standard deviation of 0025
and is normally distributed.
4.2 Still-Water Bending Moment
A statistical definition of maximum hull girder still-water
bending moments imposed on large oil tankers0
7
) is used to
establish the departure values in Table II for load and ballast
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
TABLE II. Statistical definition of maximum still-water
bending moments at departure
Voyage Moment Mean Standard cov
Type Type Deviation
Load Sagging 067M, 015M, 220Jo
Ballast Hogging 067M, 015M, 22%
voyages. The values reflect the use of loading instruments to
control still-water loads. A significant through voyage variation
in SWBM exists0
8
J of approximately 03M, (sagging), but this
effect is neglected in the present study. Other sources(l9,
20
> have
suggested higher values of covs than those in Table II, but these
sources appear to assume maximum sagging and hogging
moment co-existence and, therefore, are considered
inappropriate. Maximum SWBMs are assumed to follow normal
distributions which are truncated at 0 (values cannot be negative)
and at M, to reflect the effective use of on-board-ship means of
limiting still-water loads. In the analysis the maximum SWBM
pulse duration equals the voyage duration, which for convenience
is taken as one month. Also required is the distribution of
minimum values. This distribution is not known but it is
approximated with little error by taking the minimum SWBMs as
zero.
4.3 Vertical Wave Bending Moment
In the present study, the long-term distribution of vertical wave
bending moment at amidships is evaluated for deep draught and
normal ballast loading conditions, rather than for the full range
of loading conditions, using the Society's standard linear strip-
theory and spectral analysis package(2t,
22

23
>.
The basic formulation is
T=max [ H=max 2 1
Px(x) = E p(I) E p(HIT) e- ;2mo ilH ilT
T=min H=O
(20)
where m
0
is the peak-to-peak variance of response X in seas of
wave height Hand wave period T, and p(T) and p(HIT) are the
appropriate probability and conditional probability densities<
24
l.
The following conditions are assumed:
Sea state North Atlantic (Ref. 25 all seasons data for
areas 1, 2, 6 and 7)
Wave spectrum
Wave type
Wave direction
Ship speed
ISSC (1967) type
Long-crested seas
Head seas
Service speed.
The calculation can be readily modified to reflect a particular
service proftle, short-crested seas and all ship headings.
The long-term distribution provides the probability of exceeding
a peak value of VWBM per wave encounter and approximates
closely to an exponential distribution, which is conveniently
defined by using the value of long-term VWBM at w-s
probability of exceedance. The maximum of a large number of
realisations of an exponential distribution is distributed in
accordance with a Type I extreme-value distribution, see Fig. 5.
Therefore, the extreme distribution function of the maximum
value of wave bending moment in n wave encounters is given by
the general equation
Fx(x) = exp(-exp(-a(x-u)))
where u and a are measures of location and dispersion
respectively.
(21)
The number of wave encounters in a 20 year service life is taken
as !OS. Hence, a value of n = 571 is taken to determine the
distribution of the maximum value of VWBM over a pulse
duration of one hour. The model uncertainty on VWBM is taken
as being normally distributed with a mean value of 08 for
,.,.,
1
\ ~
0 -
z
..._ ...,.
~ 0 5 \ ~
:0 \
i ., 8!
~ .c:l
c: _,
'i 0 '-...,1
~ 0
X
w
1,0
VWBM
Mw (Rule)
Fig. 5. Probability distributions of VWBM for various service
periods
sagging moment and 07 for hogging moment. A nominal
standard deviation of 005 is taken for both cases.
4.4 Lateral Pressure
Lateral pressure is also evaluated at deep draught and normal
ballast loading conditions. Thus, the static pressure p, of
equation (15) is a deterministic quantity. The dynamic pressure<
26
l
is, however, treated as a stochastic process and its distribution is
evaluated in a similar manner to VWBM. An error in dynamic
pressure has a small influence on resultant reliability and
consequently its model uncertainty is neglected.
4.5 Hull Section Modulus
Initially, the hull section modulus at the deck or keel is a function
of the sectional dimensions and the as-built scantlings of
longitudinal components. Whilst the ship is in service, the section
moduli decrease due to corrosion of the longitudinal components
and, as a result, stresses applied to a stiffened panel increase. It is
reasonable to assume that the longitudinal components are large
in number and mutually independent; the cov of section modulus
is then negligibly small. Hence, the section modulus becomes a
time-varying deterministic quantity which is calculated from time
in service and the mean of the values of scantlings, dimensions
and corrosion rates. Typical variations in the midship section
moduli at deck and keel are shown in Table III, and it can be seen
that the annual rates of reduction in the section moduli are
practically constant over a 20 year period.
TABLE III. Typical percentage variation of deck and bottom
section moduli over 20 years
'cc' Non-'cc'
Scantlings Scantlings
Deck Keel Deck
Nominal 100 100 100
Mean(i) 9897 9899 9902
After I year 9886 9882 9839
After 5 years 9848 9813 9592
After 10 years 9789 9728 9281
After 15 years 9735 9642 8972
After 20 years %81 9557 8662
Note:
(i) based on Table VII
4.6 Strength Model Uncertainty
A histogram of the percentage error in predicted ultimare
compressive strength0
4
l is reproduced in Fig. 6. The error b
Keel
100
9902
9856
9674
9445
9216
8987
297
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
evaluated as the difference between predicted and test strength,
and expressed as a percentage of test strength. The error has a
mean of -01 OJo and a standard deviation of 963 OJo. A normal
distribution having the same mean and standard deviation, also
shown in Fig. 6, fits the data well. The above normal distribution
has therefore been used for the strength model uncertainty.
30
28
26 ..
Underestimate
of strength
Overestimate
...
of strength
24
22
"'20
c:
0
;; 18
>
)j; 16

014
j 12
"' z
10
8
s
4
2
-30 30
Percentage error
Fig. 6. Variability of strength model uncertainty
Any bias or scatter in the test results that did not reflect the actual
physical behaviour of the test specimens will have influenced the
values calculated for both the mean and variance of the model
uncertainty. It is not possible to estimate the extent to which this
may have happened. However, provided that the test results were
unbiased overall, this will simply have slightly exaggerated the
calculated variance.
4.7 Yield Stress
Yield stress data<
27
> for a modem steel mill are summarised in
Table IV. As the commercial practice is to use high test strain
rates, the mean values in the table are the sample means reduced
by 20 N/mm
2
to reflect ship service conditions. A mean value of
270 N/mm
2
and a standard deviation of 20 N/mm
2
, i.e. a cov of
741 07o is adopted for mild steel yield stress in the development of
BS540(}(
28
). The reference also states that the mean is typically 2
standard deviations above the specified minimum yield stress.
Refs. 2 and 29 suggest covs of 60Jo to 80Jo. The values in Table V
are selected for the present study. A normal distribution rather
than a lognormal distribution is adopted because a normal
distribution is more appropriate for steel from a single mill. The
nature of acceptance testing is such that it is considered
inappropriate to truncate the distribution of yield stress.
4.8 Dimensions of Stiffened Panel Elements
The variability of the dimensions of elements of a stiffened panel
from Ref. 31 is given in Table VI. As the variability is small, the
values are treated in the analysis as deterministic and nominally
valued.
4.9 Initial Thickness of Stiffened Panel Elements
Table VII is based on as rolled thickness of plates from a modern
steel mi!l<27l, and Table VIII on data from an older study(3'l. The
data of the modern steel mill are obtained at the rate of one
measurement per rolled plate for the purpose of quality control,
and it is realistic to truncate the distribution of plate thickness
from this source at the specified minimum given in the Societv's
Rules, sec Table IX. In both case;, the thickne<os is tab?n to be
normally distributed.
TABLE IV. Variability of yield stress for a modem steel mill
Steel Specified Mean Standard cov
Grade Minimumm Deviation
N/mm
2
N/rnm
2
N/rnm
2
OJo
A 235 277 225 812
A*<"J 235 276 191 692
B 235 272 17-2 632
D 235 263 127 483
E 235 270 151 559
AH32 315 337 203 602
A+A*+B 235 276 19-6 710
Notes:
(i) as per the Society's Rules
(ii) *means low Si-Al deoxidization
TABLEV. Statistical definition of yield stress for a modern
Steel Type
Mild steel
HTS
steel mill
Mean COV
70Jo
60Jo
TABLE VI. Statistical variability of element dimensions
Uncertainty Mean Standard
coy(il
Oversize Deviation
mm mm OJo
Depth of stiffener 069 166 026
web
Breadth of 014 125 057
stiffener face flat
Length of panel 04 125 002
Spacing of 04 125 013
stiffeners
Note:
(i) Stiffener: depth 650mm, flat 220rnm
Panel: length 5 ,270mm, stiffener spacing 950mm
TABLE VII. Variability of plate thickness from a modem steel
mill
Nominal Mean Standard COY
Thickness, Undersize Deviation
t,mm mm rnm OJo

15 0174 0107
H<i)
0194 0118
06(ri)
0342 0178
05(iil)
Notes:(i) t=10mm;(ii) t=20rnm;(iii) t=375rnm
TABLE VIII. Variability of plate thickness from an older steel
mill
Nominal Mean Standard cov
Thickness, Undersize Deviation
t, rnm rnm rnm
15 - 0033 023 23(i)
0125 0245 12(ii)
0133 0357 HJ(liil
Notes: sec Table VJI
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
TABLE IX Maximum pennissible thickness minus tolerance of
plates
Nominal Thickness, Minus Tolerance
t,mm mm
15 04
Oo02t+01
t>45 10
4.10 Corrosion Diminution of Stiffened Panel F.Jements
The frequency distributions of annual rates of general corrosion
diminution for deck and shell plating of large oil tankers without
'cc' notation* are given in Fig. 7. The distributions are obtained
from a limited review of a number of thickness determination
surveys carried out in accordance with the Society's Rules. Each
figure also shows an exponential frequency distribution plotted
with a mean equal to the sample mean. Thus, it is considered
appropriate for non-'cc' and 'cc' ships that the uniform
corrosion diminution rates of deck and bottom plates are
modelled by exponential distributions with means as in Table X.
The corrosion of a stiffener is also assumed to be exponentially
distributed with a mean rate of 15 times that for plates to allow
for two-sided corrosion.
0,40
0,40 0,50 0,60
Corros1on d1m1nution rate (mm/year)
Deck plating
0,30
!
0.20
'"
..,
0:
0,10

0 0,10 0,20 0,30 0,40 0,50 0,60
Fig. 7.
Corrosion diminutiOn rate (mm/yearl Side and
bottom plating
Annual corrosion diminution of hull plating of large oil
tankers without 'cc' notation
* Oil tankers with 'cc' notation have reduced scantlings in
association with approved methods of corrosion controL
t Long-term JO
8
probability of exceedance.
'*Figures in brackets indicate reduced scantlings associated wilh
Rule 'cc' notation.
TABLE X. Annual corrosion diminution
Component 'cc' Notation Ship Non-'cc' Notation Ship
mm/year mm/year
Deck Plating 001 015
Shell Plating 003 007
If the annual rate at which the steelwork corrodes is unaffected
by renewal, and if the permissible diminution prior to renewal is v
then the cumulative distribution function for the amount of
diminution D after n years is given by
J= 00
F
0
(d) "" I.: (FR((jv + d)/n) -- FR(jv/n))
j=O
(22)
where FR( ) is the cumulative distribution function for the annual
corrosion rate. This expression assumes that the steelwork is
replaced as soon as it reaches its minimum permissible thickness.
The pattern of corrosion is influenced by such factors as tank
contents and usage, region of tank, corrosion prevention
measures and ship age. For further insight into corrosion, the
reader can refer to other literature on the subject<
20
32, 33>.
However, the proposed probabilistic modelling given here is
considered applicable to deck and bottom panels within
combined cargo and ballast tanks.
5. RELIABILITY ASSESSMENTS
5.1 General
Alternative designs of a traditional 317,000 tonnes deadweight oil
tanker are used in the study. The two cases are identical except
that one has reduced scantlings to reflect Rule allowances for 'cc'
notation corrosion controls. They have the following particulars:
M,
VWBMt
Dynamic pressure t
:6,475,700 kNm
: 11 ,920,800 kNm deep draught
13,270,800 kNm normal ballast
: 00348 N/mm
2
deep draught
00485 N/mm
2
normal ballast
Transverse frame spacing: 5,270 mm
Longitudinal spacing : 950 mm
Deck plating thickness : 225 (215)**mm
Deck longitudinal : 400 mm x 282(27) mm
Bottom shell thickness : 23 (22) mm
Bottom longitudinals
Minimum yield stress
Steel supplier
:650 mm x 158(15) mm web
220 mm x 315(30) mm face
:353 N/mm
2
: Modern mill
The results of the study are presented in the following sections.
The ultimate tensile strength of the stiffened panels on the
reliability assessment is negligible since compression failures
dominate the analysis of deck and bottom panels during both
load and ballast voyages.
5.2 Influence of Corrosion Control
Fig. 8 shows the predicted variation in the annual reliability index
and annual failure probability of the deck and bottom stiffened
panels for both the 'cc' and non-'cc' cases over a 20 year service
period. The change in annual reliability is solely due to the
modelling of the corrosion diminution and renewal of"
As expected, at the start of the ships' lives, the stiffened panels ot
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
Number of years of service
Fig. 8. Influence of corrosion control on annual reliability of
stiffened panels
the non-'cc' case are safer than the stiffened panels of the 'cc'
case. The cross-over occurs after 19 years for the bottom panels
and after 8 years for the deck panels. After 20 years, the annual
reliabilities of the bottom panels of the 'cc' and non-'cc' ships are
similar. In comparison, the reliability of the non-'cc' ship's deck
panel is considerably below that of the 'cc' ship's deck panel.
Fig. 8 also shows the effect on reliability when no renewals are
effected. For the 'cc' notation ship, the influence of renewals on
the stiffened panel reliability is very small and the effect cannot
be shown in the figure. Only for the deck panel of the non-'cc'
ship does renewal significantly improve the value of reliability in
the later years of the ship's service.
On average, over 20 years the annual reliability of the bottom
and deck panels of the 'cc' notation ship is given by {3 = 3 65
(Pr = 13 X IO- ") and {3 == 22 (Pr == 14 X 10-2) respectively. Thus,
the theory predicts approximately 1 in 7, 700 of the bottom panels
and I in 70 of the deck panels will fail each year. However, these
rates reflect the conservative nature of the reliability analysis, and
are not borne out by damage records. The difference in the
reliability of the deck and bottom panels indicates the need to
study the relative reliabilities of other ship cases and also the need
to consider other aspects of the design problem, such as
transverse stress and grillage bending.
5.3 Influence of SWBM Control
The effect of loading controls is inherent within the distributions
of SWBM used in the study. One effect of loading contro_!ls to
ensure that SWBMs do not exceed their permissible value M,.
Thus, the tails of the basic SWBM distributions are removed at
the value of SWBM corresponding to 2 2 standard deviations
above the mean. Without the truncation l40Jo of SWBM
realisations would theoretically exceed M,. For the 'cc' notation
case, the presence of the tails has the effect of slightly reducing
the calculated value of the reliability of the deck and bottom
stiffened panels, as shown in Fig. 9. The effect would have been
more significant if the cov of maximum SWBM had been
somewhat larger than the 220Jo value used in the stud.Y:_ Fig. 9 also
shows that to use a value of SWBM always equal to M, (sagging
for load and hogging for ballast voyages) leads to the calculated
value of reliability being considerably underestimated.
5.4 Influence of Steel Mill Rolling Control
The marked similarity in the calculated values of reliability
achieved using steel from the modern and older steel mills is
shown in for the deck and bottom st iffened panel s of the
cc notation shir case. The differences in calculated reliability
JOO
p
5
4
:.0 3
1l

.....
-i 2
0
pf
10-
6
10_, Bottom panel
10-
SWBM ;;;;;, truncated
SWBM truncated
10_, _______ Ms
-------
Deck panel
-l s
10
SWBM non- truncated = = m -= ""
0 5 10 15 20
Number of years service
Fig. 9. Influence of SWBM definition on annual reliability of
stiffened panels in 'cc' notation ship

.=! 4
:!
0
i!::
:.0
.l3
3
0
ct
1 0-
6
Bottom pane I
10
_, Older mill
10
-= -- NOI!!!!!..a! thickness
Modem milt (with and without O.C.)
Deck panel
.....
"'
_,- t'Jorninal thickness
10 . -
"'
2
"0
.
Modem mill (with and without O.C.) and Older mill
i':
:.0
"'
-.;
a:
0
0 5 10 15
Number of years service
Fig. 10. Influence of steel mill rolling control on rumual
reliability of stiffened panels in 'cc' notation ship
20
are entirely due to the different thickness distributions, which
reflect production and quality controls assigned to each mill, and
are so small that they can only be shown in Fig. 10 for the bottom
panel. As the same yield stress distribution is used for both mills,
the reliability attributed to the older mill is somewhat over-
estimated as its yield stress values are likely to have been slightly
less than those of the modem mill.
Fig. 10 also shows that the quality control checking of plate
thickness, as modelled by truncating the thickness distributions,
has no measurable effect on the reliability of the stiffened panels
which are fabricated with steel from the modern steel mit!. Also
evident from Fig. JO is that the stiffened panel reliabi!iry is only
slightl y over-estimated by using thi cknesses instead of
the statistical distr ibutions.
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
6. CONCLUDING REMARKS
This paper has demonstrated the use of advanced level 2
structural reliability analysis techniques for determining the
inherent level of safety in the Society's current classification
requirements for design, construction and service. It has also
highlighted the potential of these techniques as the means of
predicting, in advance of implementation, the implications of
future changes in requirements.
The paper has presented the results of the analysis of selected
deck and bottom stiffened panels of two large oil tanker designs,
and shown for the particular limit states considered that:
(i) Under the simplified loading of the study, the bottom panels
are significantly safer than the deck panels; but the effect of
additional loads on the panels needs to be investigated.
(ii) The annual reliability of the stiffened panels can reduce
considerably as the steelwork of the ship steadily corrodes,
even with steel replacement.
(iii) Truncation of the distribution of maximum still-water
bending moment has a small effect on the calculated
reliability, and taking SWBM always at the permissible level
leads to a considerable underestimation of reliability.
(iv) The reliability of the stiffened panels fabricated using steel
from the modem and older steel mills is similar; and the
quality control of thickness, as modelled by truncating the
thickness distributions, has little effect on the predicted
reliability.
It should be noted that the findings are of a preliminary nature,
reflecting the assumptions and approximations of the reliability
analysis. Also, a broad range of oil tankers requires investigation.
The notional values of reliability presented in this paper are only
suitable for comparative assessments of structural performance.
Thus, the aim of the authors is to use the results of reliability
analysis for rule calibration so as to provide the Society with
additional justification for its choice of design criteria and safety
factors. This will require extensive development of structural
reliability theory to cover the complexity of ship loads and their
structural response. Clearly, an improved factual knowledge of
structural uncertainties and a better understanding of their
influence on structural performance will greatly assist in the
development.
NOTATION
COV
M
Mw
M,
Pr
Q,Qi
SWBM
ui
VWBM
{3
coefficient of variation
safety margin (Capability-Demand)
Rule design VWBM
Rule permissible SWBM
probability of failure
failure region, elemental failure region
still-water bending moment
standard normal basic variable
vertical wave bending moment
reliability index
yield stress, specified minimum yield stress
REFERENCE'S
1. Faulkner, D. and Sadden, J. A.: 'Toward a Unified
Approach to Structural Safety', Trans. RINA, Vol. 121,
1979, ppl-- 28.
2. StianE.en, S. G. et al: 'Reliability Methods in Ship
Structures', Trans. RINA, Vol. 122,1980, pp38l-397.
3. Hughes, 0.: 'Ship Structural Design: A Rationally-Based
Computer-Aided Optimization Approach', John Wiley &
Sons, 1983.
4. Goodman, R. A. and Mowatt, G. A.: 'Application of
Strength Research to Structural Design', International
Conference on Steel Plated Structures, Crosby Lockwood
Staples, 1977.
5. Thoft-Christensen, P. and Baker, M. J .: 'Structural
Reliability Theory and its Applications', Springer-Verlag,
1982.
6. Wickham, A. H. S.: 'A Review of Structural Reliability
Theory', Hull Structures Report No. 84/33, Lloyd's
Register of Shipping, 1984.
7. Hohenblicher, M. and Rackwitz, R.: 'Non-Normal
Dependent Vectors in Structural Safety', Jnl. Eng. Mech.,
ASCE, Vol. 107, 1981.
8. Hasofer, A.M. and Lind, N.C.: 'An Exact and Invariant
First Order Reliability Format', Jnl. Eng. Mech., ASCE,
1974, pp1ll-121.
9. Ferry Borges, J. and Castenheta, M.: 'Structural Safety',
2nd Edition, Laboratorio Nacional de Engenharia Civil,
Lisbon, 1972.
10. Turkstra, C. J.: 'Application of Bayesian Decision Theory.
Study No. 3: Structural Reliability and Codified Design',
Solid Mechanics Division, University of Waterloo,
Waterloo, Canada, 1970.
11. Ditlevsen, 0.: 'Narrow Reliability Bounds for Structural
Systems', Jnl. Struct. Mech., Vol. 7, 1979, pp435-451.
12. Ditlevsen, 0.: 'Reliability Bounds for Series Systems with
Highly Correlated Failure Modes', DCAMM Report No.
207, Technical University of Denmark, Lyngby, Denmark,
1981.
13. Wickham, A. H. S.: 'Reliability Analysis Techniques for
Structures with Time-Dependent Strength Parameters',
ICOSSAR '85, 4th International Conference on Structural
Safety and Reliability, 1985 (to be published).
14. Rutherford, S. E.: 'Stiffened Compression Panels-The
Analytical Approach', Hull Structures Report No. 82/26,
Lloyd's Register of Shipping, 1984.
15. Chatterjee, S. and Dowling, P. J.: 'The Design of Box
Girder Compression Flanges', Steel Plated Structures; An
International Symposium, Crosby Lockwood Staples,
London, 1977.
16. Rogers, N. A.: 'Outstand Failure in Stiffened Steel
Compression Panels', Cambridge University Internal
Report, CUED/C-Struct/TR.54, 1975.
17. Guedes Soares, C. and Moan, T.: 'Statistical Analysis of
Still-Water Bending Moments and Shear Forces in Tankers,
Ore and Bulk Carriers', Norwegian Maritime Research,
No. 3/1982, pp33-47.
18. Tozer, D. R. and Hart, D. K.: 'A Review of Still-Water
Loads and Their Uncertainties', Hull Structures Report No.
84/31, Lloyd's Register of Shipping, 1984.
19. Mano, H. et al: 'Statistical Character of the Demand on
Longitudinal Strength (2nd Report). Long-Term
Distribution of Still-Water Bending Moment', Jnl. of
SNAJ, VoL 142, Dec. 1977, pp255-263.
20. Akita, Y.: 'Lessons Learned from Failure and Damage of
Ships', Proc 8th ISSC Joint Session I, 1982.
21. Blixell, A.: 'LR2570 Users' Manual-Calculation of Ship
Responses in Regular Waves by Strip Theory'. R & T A
Report No. 5116, Lloyd's Register of Shipping, 1972.
22. Robinson, D. W.: 'LR257L Users' Manual-Calculation of
Non-Dimensional Variance of Wave-Induced Ship
Responses', Development Unit Report No. 37. Lloyd's
Register of Shipping, 1973.
"_l{jl
.)\_y i
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
23. Robinson, D. W.: 'LR257X Users' Manual-Long-Term
Predictions of Wave-Induced Ship Responses',
Development Unit Report No. 28, Lloyd's Register of
Shipping, 1974.
24. Robinson, D. W.: 'Sea State Distributions for Use in the
Calculation of Long-Term Wave-Induced Ship Responses',
Development Unit Report No.4, Lloyd's Register of
Shipping, 1972.
25. Hogben, N. and Lumb, F. E.: 'Ocean Wave Statistics',
HMS0,1967.
26. Dawkins, R. A.: 'A Method for Calculating the
Hydrodynamic Pressure on a Ship's Hull Moving in a Sea-
Way', Hull Structures Report No. 84/32, Lloyd's Register
ofShipping, 1984.
27. Wickham, A. H. S. and Hart, D. K.: 'A Review of
Structural Uncertainties', Hull Structures Report No.
84/22, Lloyd's Register of Shipping, 1984.
28. Flint, A. R. et al: 'The Derivation of Safety Factors for
Design of Highway Bridges', Design of Steel Bridges,
Granada, 1981, pp11-36.
29. Daidola, J. C. and Basar, N. S.: 'Probabilistic Analysis of
Ship Hull Longitudinal Strength', SSC-301, 1981.
30. 'Japanese Shipbuilding Quality Standard, Background
Document', Committee of Shipbuilding Research
Association of Japan, 1979.
31. Baker, M. J .: 'Variability in the Strength of Structural
Steels-A Study in Structural Safety. Part 1: Material
Variability', CIRIA Technical Note 44, Construction
Industry Research and Information Association, 1972.
32. Herring Jr, L. C. and Titcomb, A. N.: 'Investigation of
Internal Corrosion and Corrosion Control Alternatives in
Commercial Tankships', Report No. SSC-312, Ship
Structure Committee, July, 1981.
33. Viner, A. C. and Tozer, D. R.: 'Influence of Corrosion on
Ship Structural Performance', Hull Structures Report No.
84/92, Lloyd's Register of Shipping, (Contribution to
Report of Committee III.3, 9th ISSC, 1985).
DISCUSSION
Mr A. C. Viner, B.Sc. (Member): I should like to
congratulate the authors on this fine paper. I know how much
hard work has gone into its preparation.
Classification Society Rules have developed over the course of
many years by a variety of methods, and service experience
has been incorporated through feedback. However, the Rules
do contain some inconsistencies and it is not always clear what
margins of safety are implied. The emergence of structural
reliability methods provides the opportunity to establish a
consistent technical framework to the Rules and direct
calculation procedures. The authors have made an excellent
start in this direction.
I am glad to see that the authors stress the notional nature of
the probabilities of failure and reliabilities which they have
derived. Since many assumptions and approximations have
been made, the probabilities of failure must not be regarded as
absolute and can only be used in a relative sense.
For merchant ships, it is essential to take into account the
effects of corrosion in structural reliability analysis. In his
treatment of time-dependent strength in Ref. 13, Mr Wickham
has made a notable addition to reliability theory. Would the
authors explain why they have chosen to consider annual
reliabilities and probabilities of failure in the present paper
rather than the cumulative probability of failure as proposed in
Ref. 13?
With regard to uncertainties, there is still a need to collect
much 01ore data. The authors have made good use of what
data are available at present. but I wonder whether their
proposal for yield stress is not a little optimistic. Ductile
failure predictions for stiffened plating arc, of course, very
sensitive to yield stress and the statistical values chosen will
have a major influence on results.
It is important to consider what definition of yield stress is
most appropriate for structural design. There were three cases
of hull girder ductile collapse a few years ago and
investigations to correlate hull ultimate bending moments with
numerical methods indicated that it was appropriate to take the
static yield stress value (i.e., at zero strain rate) if known, or
alternatively the specified minimum yield stress. For one
investigation, tensile test specimens were cut from the deck
plating some way from the damaged area. As shown by Table
XI, the mean static yield stress was found to be nearly the
same as the specified minimum whilst the mean mill test value
was 45 N/mm
2
higher. The authors' proposed statistical mean
falls midway between.
TABLE XI
Specified minimum (aymin)
Mean mill test
Mean static yield
I 06 llymin
343
388
342
364
Mill tests are conducted at relatively high strain rates and may
be considered to give upper yield stress values which,
however, tend to be sensitive to the testing machine used,
axiality of loading, surface finish and geometry of test piece.
There seems to be a need for a systematic investigation to
relate mill test results to yield stress values at zero or low
strain rates. Would the authors agree with this?
The paper deals with the ductile failure modes of stiffened
panels forming part of the hull envelope. However, it does not
mention overall failure modes of the hull girder. Do the
authors consider it necessary to include such modes in a
structural reliability analysis?
Mr P. A. Frieze, M.Sc., Ph.D. (Member): I would like to
congratulate the authors on this paper. I find the discipline of
reliability analysis extremely useful, as it brings together all
aspects of the design problem: loading, response and strength.
So a contribution in this area is always welcome, particularly
when it contains an element of advancement of reliability
analysis techniques and some results.
My remarks will be directed primarily at the structural aspects
of the paper and will be in the form of comments and
questions. Some of these will be detailed because I have
recently been involved in re-examining the partial safety
factors in the new steel bridge code, BS 5400: Part 3, so it
can be used for offshore designs<34). This code of course
contains clauses concerning stiffened panels, so it is directly
relevant to the present paper.
In examining stiffened plating, the authors have quite rightly
adopted an effective width approach. When calculating the
section modulus of the ship cross-section, however, they
appear to have used the gross cross-section. In doing so, they
will underestimate the stress in the deck and bottom plating as
determined from equation (14). I appreciate that effective
width is a load-dependent problem, and that they are already
considering the time varying nature of the section modulus, but
as they are examining the extreme events, the effective width
at failure of the stiffened plating will be more appropriate than
the gross width. The latter is probably more appropriate in
relation to tensile stresses, but there is also a Joss of
effectiveness which occurs in that plating when residual
stresses are present.
I was a little surprised to find that considerable effort has been
spent on deriving a strength formulation for stiffened plating
which accounts for initial distortions in both the plating and the
stiffeners, and residual stresses in the plating, and yet, in the
reliability analysis, these variables are only assigned nominal
values. While appreciating that the variabilities generated by
these parameters are accounted for in the strength model
uncertainty. useful lessons can be learned concernim1. their
importance by examining them in more detail. -
For example, for stiffened plates designed in accordance with
BS 5400: Part 3, the variation with slenderness of the different
sensitivity factors, as calculated using the CIRlA 63 reliability
analysis algorithm, has been found<34l. In this particular
formulation, rather like that of the authors', initial stiffener
distortion is a function of length. It is in the range of
slenderness 0 7 to I 0 where the effect of initial distortions
can be expected to have their greatest impact. However, in this
slenderness range it was found that there was very little
sensitivity to that variable as expressed through the length
parameter. It was only at large slenderness where the
sensitivity to length was significant and this appeared to be due
to the influence of elastic buckling and not initial distortions.
From this it can be concluded that initial stiffener distortions
are relatively unimportant, and probably from design and
fabrication viewpoints, can be treated with less respect than
hitherto has been the case.
In examining the effect of imperfections on stiffened panel
strength, the authors appear to have used tolerance values for
conducting comparisons with test data. I would have preferred
to see measured model values used for corroborating the
strength model, and then tolerance values used for establishing
a design curve, for example. Possibly better, since thickness
and yield stress survey results are available, why not survey at
least the geometrical imperfections (I think there is a
suggestion that they might be coming up anyway), and use
these values for establishing typical strengths? I wonder if the
authors could explain why they have adopted this particular
approach of using tolerance values in the strength model
formulation rather than measured values that would have been
available with the test data.
Reference has been made to the shakedown of residual stresses
in service. A numerical study has been conducted into this
effect(3SJ in which a tension pre-load was used to simulate the
shakedown of residual stresses. An improvement in strength
resulted although it was not very significant and obviously will
not affect the reliability results significantly. So the authors'
assumption that it can probably be ignored seems correct.
In considering stiffened plate failure, only compression failure
of the outstand has been considered, the possibility of tension
failure being considered to be too small. I would have thought
though with slenderness in excess of 1 0 to 1 2, tension
failures will dominate the plate mode of failure. It should
affect the results but this slenderness may be outside the
practical range for merchant ships, although it will be relevant
for naval vessels.
In allowing for the interaction between in-plane and out-of-
plane behaviour, equation ( 17), the authors have suggested a
simple solution involving, I believe, a pin-ended model for
axial load effects and the clamp ended conditions for lateral
load effects. This obviously eliminates the problem of
identifying which appropriate buckling load to use, either the
Euler Load or four times Euler, although it is not consistent
with the more usual way of handling the lateral load effect
which would be to multiply MQ by the first order
amplification factor Pe/(Pe -Pc). Have the authors examined this
particular approach and found it to be a satisfactory way of
dealing with the interaction problem, or maybe they have not
examined enough test data involving lateral pressure. It would
be interesting to know how good this model is over the full
range of lateral pressure.
When considering the load eccentricity at mid-span, to achieve
good correlation with test data, the authors have introduced a
correction factor which appears to be quite significant for low
slenderness stiffened panels. Have the authors been able to
identify the physical reasoning behind that? I would be
interested to know what it might be.
In their presentation, the authors covered some aspects of
deriving bias and uncertainty in relation to modelling error, the
wave loading and the overall structural analysis. However,
they all seem to have been assessed rather subjectively or have
the authors more information concerning these aspects which
they could make available? Such information would be useful
to people like myself who are trying to apply these methods.
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
In tackling stiffened panel model uncertainty, the authors
appear to have used the ratio of the difference between
predicted and test data, non-dimensionalised with respect to the
test value. The approach in the CIRIA 63 algorithm seems to
suggest one should non-dimensionalise the test value with
respect to the predicted value. This may not affect the
coefficient of variation obtained for the model uncertainty very
much but may affect the bias, generating an inverse
relationship. Have the authors examined this to see if it has an
. influence on their results?
With regard to the wave loading model, the wave loading
conditions generating hogging and sagging will clearly be
different. The process by which the dynamic pressure is
assessed, Section 4.4, appears to ignore this, i.e., there is no
correlation between dynamic pressure and vertical bending.
Maybe I have misinterpreted the paper so could the authors
explain whether they have examined this correlation and found
it to be important or otherwise.
On the point of yield stresses which has already been raised by
Mr Viner, it is a perennial problem. I would like to see the
authors identify whether the yield stresses conducted in
connection with the model tests were carried out at a static rate
or at a higher rate. The latter would be more consistent with
practice in which a tensile test is conducted at a specified rate
and then failures might generally be expected to occur under
quasi-static conditions. However, I am not sure whether ships
suffer quasi-static failures in the real sense of the word.
Mr Viner mentioned that static yield stress values were
obtained after five minutes of no elongation of the test
specimen. Under the conditions being considered here, i.e.,
ultimate load, it is unlikely that a ship would experience five
minutes of stationary conditions. This time span is more likely
to be measured in seconds. Therefore the strain rate that is
applicable at failure in the ultimate bending case is probably
not the static one but somewhere between the normal and static
rates.
There is an additional problem which is probably more
relevant to the offshore industry than to merchant shipping,
and that is the American use of quite a different strain rate
when doing tensile tests than the UK, European and even
Japanese industries. The USA yield rate is about Y,
6
in/in/min
which is almost one yield strain per second. It can give up to
about a 35% increase in yield stress compared with the static
value. In Japan and the UK it is one tenth of this rate which
results in increases of IO% to 15%. Thus it is possible for US-
qualified steel to fail the UK test.
My final comment relates to plate rolling tolerances which
have already been brought out in the comparisons in Tables
VII and VIII. In view of the results found by the authors,
there appears to be little to be concerned about arising from
the ability of modern mills to roll to tight tolerances and
consistently undersize. However, in the light of this, has any
thought been given to reducing the negative tolerance so that
less undersized plate will be produced?
REFERENCES
34. Frieze, P.A. and Plane, C. A.: 'Partial Safety Factor
Evaluation for Some Aspects of Buckling of Offshore
Structures. A Pilot Study based on BS5400: Part 3'.
Prepared by University of Glasgow for UK Department of
Energy (to be published by HMSO).
35. Frieze, P.A. and Drymakis, E: 'An Examination of
Shakeout in Stiffened Plates'. In Behaviour of Thin-
Walled Structures, ed. Rhodes and Spence, Elsevier
Applied Science, London, 1984, pp 69-92.
Mr D. W. Chalmers, O.B.E., R.C.N.C. (Fellow): I must
apologise to the authors because I did not get time to read the
paper until I was in the train this morning. Dr Frieze has
kindly brought out almost all the points I might have brought
out anyway which helps to speed things up. r have two
questions, however, and one comment.
The first question relates to corrosion. There seems to be an
implicit assumption that the corrosion rate is uniform through
30i
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
the life of the ship with probablistic variations. The experience
we have on modem warships is that with good protection
schemes, corrosion is very slow indeed, and is restricted to
local pitting which intuitively has very little effect on overall
strength. In old ships, corrosion appears to accelerate, possibly
because the ship as a whole becomes more difficult to
maintain. Perhaps the authors could comment on their
experience with merchant ships.
The other question relates to Section 4.1 where at the bottom
the authors say: 'Subjectively, the uncertainty is given a mean of
10, a standard deviation of0025 and is normally distributed'.
I am very unclear as to why this should be so. When you think
of what is happening to the longitudinal stress and how it is
dependent on the second moment of area and the neutral axis,
both of which are acting in a somewhat uncertain manner as
failure is approached, perhaps the authors could provide some
justification for the statement.
My final comment, and I hasten to say that I am not criticising
the authors in any way as this is a very well argued and
presented paper, but I always feel uneasy when I see very
mathematical 'level 2' analyses presented, that they will never
be of real use to the designer. There are so many uncertain
uncertainties: so many intuitive assumptions, and although the
authors have nibbled away at a few of them, there is an awful
long way to go. When one thinks that what actually causes
ships to be lost is the unexpected phenomenon, which by its
very nature is unpredictable, I just wonder about the real
practical use to the designer.
Mr R. A. Goodman, B.Sc., Ph.D. (Fellow): In this paper
the authors have presented a new approach for tackling a wide
range of practical ship structure analysis problems. The general
method is now used and accepted for dealing with many types
of structures, but it has only recently begun to be used for the
analysis of ship structures. What the authors have done is
taken the latest state-of-the-art of a number of techniques,
extended them and produced an advanced reliability analysis
method which allows at least two characteristics which are
important in ship structures work to be comprehensively
accounted for, namely, time dependent loads and time
dependent strength. Allowing for time dependent strength is
particularly important because it means we can at last begin to
investigate the effect of corrosion in a rational way.
In a paper which contains much which is new to the
shipbuilder, I think the ability to handle the effect of corrosion
is the most important. Although the advanced reliability
method presented allows us to examine ship operation and
shipbuilders' controls and constraints, (e.g., still-water bending
moments and plate rolling tolerances), the ever-present costly
problems of how much corroded material to renew and when
to renew it can now be studied systematically. I am sure the
long-run economic benefits of this work will be considerable.
Calculations of the type presented are complex and require
expertise. It is very unlikely that such calculations will be
required to be carried out by naval architects when designing
ships' structures. The advanced structural reliability analysis
techniques presented come into their own for the development
of structural design codes such as Classification Society
requirements for design, construction and service. I fully
endorse the authors' view that the techniques have great potential
for predicting the implications of future changes in requirements.
At the moment, there is a great deal of discussion about
reliability methods. The methods are complex; they require
great expertise; they need specialists to handle the data; there
are many uncertainties; there are many models available. I am
not clear about where we go from this point. I would welcome
the authors' views on the next step to be taken.
Finally. I would like to thank and congratulate most warmly
the three authors for an excellent paper; a paper which I am
sure will become a mile-stone in the development of ship
structural analysis methods.
Professor D. Faulkner, B.Sc., Ph.D., R.C.N.C, F.Eng.
<Fellow) (read by the Secretary): This is a valuable addition w
our Transaerions in bringing into focus an important discipline
which is at long last beginning to impinge on design and which
should also be central to inspection policy procedures. As I
have not had the opportunity to study the paper in depth, I will
confine my remarks to a single question against the
background of my own belief that the variability in safety
levels of all ships is very wide. Have Lloyd's Register, for
example, examined the safety indices which are implicit within
their rules for the bottom structure of transversely framed
merchant ships around 100 m in length? In the 1981 paper I
wrote for the SNAME/SSC 'Extreme Load and Response'
Symposium, I suggested that such an analysis would show a
very much lower safety level than is implicit for the upper
decks of the same ships. It has recently been brought to my
attention that other people, including a ship classification
society, also appear now to have reached the same conclusion.
If this is also true for Lloyd's Register, then perhaps a further
attribute of reliability theory should be for the examination of
present rules to ensure that safety levels are more uniform and
structures more efficiently designed. In such a task, I suggest
that the more uniform safety level is targeted at the lower end
of those levels implicit in current rule designs unless there has
been evidence that these have been unacceptably unsafe. In this
way, steel weight may be saved and, perhaps in many cases,
construction costs also.
Mr H. Lackenby, M.Sc., D.Sc. (Fellow): I would just like to
make a very brief comment.
I note that one of the factors not taken into account in the
work has been fatigue. This, of course, is a highly time
varying property and I would like to ask whether it is intended
to include it in due course. As far as the present paper is
concerned, a very good answer would be that the stresses
inherent in the work are below the fatigue limit, but I doubt
whether this is the case. Perhaps the authors may care to
comment.
Mr G. H. Sole, B.Sc. (Member): First I wish to echo the
congratulations of the previous contributors on what I think is
a very fine paper, and an excellent presentation.
One of the most interesting findings reported in the paper is
the large difference in reliability between the deck and bottom
panels. Perhaps one of the origins of this appears to lie in the
Ship Rule Requirements of both Lloyd's Register and other
Classification Societies, in which the bottom longitudinal
structure is frequently designed on the basis of a lateral
pressure rather than a compressive buckling capability. Thus,
these designs are effectively based on a maximum permissible
stress concept, which ensures that the local stresses at the
welded connections of plate to the longitudinal, and the
longitudinal to frame, do not exceed certain levels. Used in
this sense, the permissible stress concept can be interpreted as
a rather simple fatigue criterion, which brings me to the point
just raised by the previous contributor, that consideration of
other failure modes such as fatigue may well reduce the
difference in reliability between the deck and bottom panels.
Expressing this another way, whereas a moderate lateral
pressure generally does not have a very significant influence
on the ultimate compressive stress of a longitudinal, as we
have seen from some of the slides previously shown, it will,
nevertheless, have a very large influence on the stresses at the
connection to the supporting members.
Having said that, the authors have correctly referred to the
possibility that other load components such as transverse stress
and grillage bending may well narrow this difference. A
further influence of these additional load components may also
be to produce a pronounced variation of reliability over the
bottom shell, such that the probability of a compressive failure
in the mid-tank region could be greater than that for panels
around the tank boundaries.
One further comment, referring now to the loading, and
specifically the vertical wave bending moment, it is inevitable
that many assumptions have to be made, and perhaps the
authors may like to comment on the sensitivity of the final
results to some of these assumptions, such as, for example, the
seagoing conditions and the statistical values selected for model
uncertainty.
The President, Professor J, B. Caldwell: This has been a
very interesting evening and I am sure you would want me to
conclude it by conveying our congratulations and thanks to our
three authors. We are, I believe, in the process of building a
new methodology of structural design and assessment based on
concepts of reliability; but one can view that process in two
different ways.
One school of thought would say that this is the way we must
go. It makes sense to try to represent the world in a stochastic
way, and then to base decisions about assessment and design
on that essentially probabilistic nature of the world. This can
be a logical and in some ways quite elegant way of looking at
reality.
But there is another view which says that this whole new
edifice of methodology might collapse like a pack of cards
because, when you start looking at structures in this way, you
have to consider a great many possible events and to each of
them assign uncertainty levels and probabilities; and, when you
multiply up all the work that has to be done and recognise the
uncertainties that are embedded in that work, not least the
uncertainty as to how a structure might in fact fail (because
often the failures occur from rather unquantifiable,
unpredictable mistakes), then one may begin to wonder really
whether this sort of approach to design is indeed the right one.
As an example of these uncertainties, it has always worried me
a little that we proceed from the reliability analysis of a single
element of a small stiffened panel to make statements about the
reliability of an assembly of such elements across the whole
bottom or deck of a ship. Presumably, between the various
elements that constitute the deck or constitute the bottom, there
will be a degree of variability; so that the reliability of the
deck as a whole may be significantly different from the
estimated reliability of a single element of it. Could the
authors comment on that?
The authors have made a genuine contribution to the
development of reliability methods, not least in the handling of
those time dependent phenomena which can be so important. I
happen to have with me some Japanese analyses of structural
failures in ships, which refer to the fact that 80% of ship
damages reported in 1981 were found to be caused directly by
corrosion, and indirectly by reduced thickness. The ability to
handle corrosion and time dependent capability is clearly a
very essential ingredient in reliability assessment. That has
been, in my view, one of the main advances which our authors
presented to us tonight.
As Dr Goodman and others have said, this is a paper that will
be seen as a very valuable contribution to our Transactions.
May we therefore congratulate the authors, not only for their
paper, but for the way in which they have responded to quite a
tough discussion. We look forward to reading their considered
views on some of the detailed questions they have been asked;
but in the meantime let us show our appreciation.
The vote of thanks to the authors, proposed by the President,
was carried with acclamation.
WRITTEN DISCUSSION
Dr Ing A. Pittaluga (Member): I am particularly pleased to
have the opportunity to discuss this excellent paper, since I
was actively involved in the development of a similar study by
Committee V2 of ISSC 1985, Ref. 36, and I wish to
congratulate the authors on their very comprehensive work. I
particularly enjoyed the inclusion in the analysis of corrosiOn
and corrosion control, which is often an overlooked item.
I have two comments to offer:
(a) The authors have adopted the Borges-Castanheta model for
the combination of time-varying loads which is derived from
Turkstra's rule. This rule has been shown to be applicable to
statistically independent load processes among which may be
considered SWBW and VWBM, at least as a first
approximation.
It is not applicable to different types of wave loading, such as
VWBM and wave pressure, since they may be strongly
correlated. In the authors case this fact wa& probably of little
consequence, since wave pressure has little impact on
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
reliability, as they confirm, but might be of importance in
more general cases. I think that the method proposed in Ref.
37, which has been adopted by the ISSC Committee, is of
more general application.
(b) The second comment is almost philosophical. The authors
state that taking SWBW always at the permissible level leads
to a calculated reliability considerably lower than if its
statistical distribution is considered, and I fully agree with this.
On the other hand, the distribution of SWBM is not a
consequence of natural randomness of the environment or of
the building process, but is often a consequence of free choices
of the Owner and of the ship Master, who could decide to run
the ship always at the permissible level. So, arising from this,
are ships to be designed for the worst usage allowed, or have
we to take into account that many users will probably handle
them in a gentler way?
I think that regulatory bodies will have much to ponder on this
question.
REFERENCES
36. International Ship Structures Congress, Report of
Committee V2, Genova, September 1985.
37. Ferro, G. and Cervetto, D.: 'Reliability of Marine
Structures under Dynamic Loadings', Tech. Bull. N. 85,
Registro Italiano Navale, Genova, Jan. 1985.
Professor C. V. Betts, M.A., M.Phil, R.C.N.C. (Fellow): It
is good to see that Lloyd's Register is adopting and developing
structural reliability theory. With their renowned data gathering
resources, we may at last see major progress made in
collecting the necessary statistical information needed on
structural uncertainties.
The advanced level 2 method put forward by the authors
would seem to have certain advantages and the paper has
demonstrated some of these. However, the high hopes
engendered by the early part of the paper were, for me, rather
dissipated by the end and I was left wondering if the state of
the art had really moved forward that much over the past 5-10
years. Quantified reliability predictions are still subject to a
number of major assumptions and simplifications and can still
only be used in a purely comparative sense. Indeed in Section
5.2, the predicted failure rates of stiffened panels are
unrealistically high by orders of magnitude (as the authors
readily admit). Do the authors have an idea why this is so?
Many of the simplifications made in their analysis (such as
neglect of some dynamic effects; neglect of in-plane shear and
transverse stress; use of total section properties) would be
expected to lead to underestimates of failure rate rather than
the opposite. Have any sensitivity studies been done that might
cast light on this?
There seems little doubt that we shall have to accept that ship
structural reliability analysis will be a comparative rather than
an absolute tool for some time to come. However, one wishes
to explain results achieved, if only in a qualitative sense, in
order to have reasonable faith in the methods used. Subject to
that, I would wish the authors every success in their further
work in the field and hope that they will give us further
progress reports from time to time.
Professor E. Lehmann and Mr C. Ostergaard: The authors
are to be commended for having applied advanced level II
reliability theory to the important problem of time varying
strength of ship structural components. Their procedure seems
to be practical and should in principle be recommended for
future use in all cases where risk analysis offers an improved
decision basis for structural design.
The authors' optimism about the suitability of reliability
analysis as an initial step in the direction of quantification of
acceptable risk (see Introduction) or as a tool for comparative
assessment of structural performance (see Concluding
Remarks) cannot be shared by these discussers without
reservations. One of our principal reservations stems from the
fact that calculated (notional) risk values depend on all the
details of the applied methods (including stochastic and
STRUCTURAL RELIABILI7Y ANALYSIS OF STIFFENED PANELS
Mrf/ Mp t
Elastic
-1
' 0
y ~
Bottom longitudinal
p = a A
y y e
P/ = reduced yield load due to bending moment
Mp = full plastic bending moment at the supports
MP' = full plastic bending moment including longitudinal load
effects at the supports.
Fig. 11 . Interaction curves between bending moment and
axial load at clamped supports
mechanical models as well as data) and any certifying authority
would have to prescribe acceptable methods together with
acceptable risk values in order to establish an objective safety
measure. This has not been achieved with the traditional (level
I) safety factor approach in the past and we do not see
indications that it will be achieved with any advanced (level II)
reliability index approach in the future.
Nevertheless, we believe that reliability analysis has the
potential for significant improvements in our present design
procedures, as the authors have exemplified. Our criticism is
confined to a few details which we think, however, have
practical relevance.
With regard to the mechanical model, equations (17) and (18),
we should like to mention that the Euler load is much greater
than the compressive yield load in the example. Thus,
structural failure is not dominated by a buckling failure mode,
which would have the described consequence of a plastic hinge
formation at the midspan between the supports of the panel,
but by a yield moment failure mode at the supports themselves
because the fixed end moment is twice the midspan moment.
This is best explained by the interaction curves of Fig. 11.
A neglection of the influence of shear loads on the ultimate
bending moment is probably correct, provided the structural
design ascertains that the shear load is effectively distributed
into the supporting transverse frame structure. We are
surprised that corrosion significantly decreases reliability in the
case of the considered stocky structures, Fig. 8, since
structural imperfections should have only a minor influence in
such cases. Some further explanation will perhaps be desirable.
With regard to the vertical wave bending moment we think
that model uncertainties have been underestimated rather
significantly. Referring e.g. to Kaplan<
38
J the coefficient of
vanation should be of the order of 0 l (0 065 or 0 0714 in
the paper) and a spectral shape variability should be included
with an additional coefficient of variation of the order of 0 15
(completely neglected in the paper) . Because of the strong
influence of wave bending on the estimated reliability indices.
we think that the presented results underestimate risks of the
considered failure modes. In view of our comment on the
mechani cal model we conclude that the assumptions of head
seas only with no reduction of the service speed in bad
306
weather yield overestimates of the probabilities of failure and
the calculations should in fact be modified to reflect a more
realistic service profile, or, reversely , a VWBM value at a
probability of exceedence level less than w-s should be
considered. Eventually, any correlation of the vertical wave
bending moment with the still water bending moment must be
ignored when applying the Borges-Castanheta load combination
rule, see Section 2.3 of the paper. We have some doubt that
thi s condition is satisfied in reality.
As a matter of detail, the right hand side of equation (13) of
the preprint is in need of a factor ( I hr). However, there is
available a more effective formula to evaluate the probability
of the intersection of two failure regions Q, and Qi, see
Ditlevsen(3
9
>.
REFERENCES
38. Kaplan, P.: Identification of uncertainties & evaluation of
their importance. Complementary chapters of Committee
V2 (Applied Design) . Publ. Lloyd's Register of Shipping,
1985, p.6.
39. Ditlevsen, 0. : Uncertainty modelling. McGraw-Hill Int .
Book Co. , 1981, p. 240.
Professor C. Guedes Soares, M.Sc., Dr. lng. (Member): The
authors are to be complimented for an interesting paper on a
topic in which much work still needs to be done.
In addressing the problem of the strength of stiffened hull
plating they have to deal with many issues relevant to ship
primary strength although their conclusions cannot be as
general as that. In fact, collapse of one stiffened plate in the
bottom or deck structure is not necessarily equivalent to hull
girder collapse. The relationship between element and hull
failure must still be established.
Regarding the general organisation of the paper, it is felt that
the authors were generous with their description of the
reliability methods but they were brief regarding the way in
which the methods were applied to the problem at hand,
leaving some questions about what was really done.
One aspect that remained unclear to the writer concerns the
load model adopted in the reliability assessments of Section 5.
Although Section 2.3 refers to a method of load combination
one remains with the doubt as to whether this was applied or
whether equation (14) used design values of each load process.
In this later case it seems more appropriate to include a load
combination factor to account for the non-simultaneous
occurrence of maxima of both loads, as is done in Ref. 40.
In dealing with the individual load processes, it appears that a
normal distribution was used for the maximum SWBM, and
an extreme type I was adopted for the maximum VWBM. This
seems to be somewhat inconsistent because the normal
distribution for the SWBM corresponds to random point in
time occurrences which are represented by the exponential
distribution in the case of VWBM. The use of 'maximum' has
different implications in the two cases. For the SWBM it
means one single occurrence in the region along the ships '
length where the moment reaches the largest valuesfl7l.
However, for VWBM it is the maximum of n independent
outcomes of a random variable which models that stochastic
process.
In studying the influence of the truncation of SWBM, the
authors have concluded that it had almost no influence on the
results. Care should be taken not to generalise that conclusion
to other cases. On one hand, the normal distribution of the
SWBM is much less sensitive to truncation than the
distribution of lifetime maximum SWBM<
40
J. On the other
hand, the importance of the truncation effect is also dependent
on the mean and standard deviation of the SWBM. The larger
these are, the more important truncation becomes, as occurs in
containerships and in ore/oil carriers.
While in the example considered in the paper the truncation
point was about 2 2 standard deviations above the mean. in
containerships that difference can be only ! standard deviacion.
In these cases the- maximum SWBiv1 is more dependent on ;h;:;
truncation effect than on the initial distribution. Furthermore,
the effect of the existence of load instruments cannot be
represented by a truncation of the distribution. It must be
modelled as some sort of filter which avoids most of the
exceedances of the allowed limit. Only then is one able to
represent the maxima that occur in operation, which sometimes
exceed the allowed limit(
40
J
The authors have used the same statistic to describe the
SWBM in both loaded and ballast condition (Table II) which is
in general not correct. The results presented in Ref. 17 for
tankers are only for the loaded condition, as indicated in Table
3 of Ref. 17, in which case the moments are always sagging
ones. In the data of Ref. 17 only bulk carriers experience both
hogging and sagging moments of similar magnitude in the
loaded condition. A further analysis of a more extensive data
set<
40
) indicated that tankers in ballast condition experience
hogging moments of a magnitude about half of the sagging
moments and with a similar standard deviation.
The authors used 1 month as the typical duration of voyages
for tankers. Results of statistical analysis of voyage duration
indicated mean values of l3 days for small and medium
tankers and 24 days for large tankers<
40
).
Regarding the long-term distribution of VWBM, the authors
have used a simplified approach, as they recognise. This may
be appropriate for comparative and for illustrative purposes.
However, the model uncertainty associated with this method
should be slightly different from the one adopted. The
calculations reported in Ref. 40 for a simplified method
indicated, for tanker hulls, a value of 1 16 for the mean value
and 015 for the standard deviation. This method was more
sophisticated than the authors' in that it used all ship headings.
Thus, the writer estimates that the bias of the authors' method
could be around 1 0. This value includes the bias of the
environmental data of Ref. 25, as reported in Ref. 41.
Furthermore the bias should be the same for sagging and
hogging. Only in ships with fine forms is the non-linearity of
the response significant, leading to different values for hogging
and sagging.
The model uncertainty used by the authors for the strength
model agrees better with the writer's results. In Ref. 42, four
different strength models were considered resulting in mean
values of the model uncertainty between I 11 and 0 98 for
plate induced collapse and between 106 and 090 for stiffener
induced collapse. When all types of collapse were considered
the mean values varied between I 12 and 0 98 for the four
methods. The standard deviation ranged from 7% to 12%.
Incidentally, these values are in the same range as those
obtained recently for plate elements(
43
).
The treatment given by the authors to the effect of corrosion
also deserves some comment. The authors are correct when
stating in Section 4 10 that corrosion rate is unaffected by
plate renewal. However, renewal of the plates affects the hull
section modulus and this has not been accounted for in Section
4.5. In fact Table III shows a steadily decreasing modulus over
a 20 year period.
A more accurate treatment of corrosion that accounts for
replacement was presented in Ref. 43. Defining a mean
lifetime of a plate given by the time necessary for the
corrosion to reach about 25% of the initial plate thickness, it is
possible to derive a distribution for plate renewal.
Accounting for plate renewal means that after an initial period
in which the section modulus decreases, it will reach a steady
state resulting from the existence of a mixture of corroded and
new plates. Thus, at a random point in time, the mean
thickness of the plates is smaller than the initial thickness but it
is always greater than 75% of that thickness.
Finally, it was comforting to notice that although using a
somewhat different reliability analysis than in Ref. 40, the
authors have also obtained values of f3 in the range of 2 and 3.
These are significantly lower values than the range 5 to 7
reported by others, about 10 year ago, using other reliability
models. The values now obtained are much closer to the ones
presented by Faulkner and Sadden'
1
' for warships, showing that
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
the difference between the structural safety of merchant ships
and warships is not as large as was then thought, or as already
noted in Ref. 40.
REFERENCES
40. Guedes Soares, C.: 'Probabilistic Models for Load
Effects in Ship Structures', Report UR-84-38, Division of
Marine Structures, The Norwegian Institute of
Technology, June 1984.
41. Guedes Soares, C.: 'Assessment of the Uncertainty in
Visual Observations of Wave Height', Ocean Engineering,
Vol 12, No. 6, 1985 (in press).
42. Guedes Soares, C. and Soreide, T.H.: 'Behaviour and
Design of Stiffened Plates under Predominantly
Compressive Loads', International Shipbuilding Progress,
Vol. 30, No. 341, 1983, pp 13-27.
43. Guedes Soares, C.: 'Uncertainty Modelling in Plate
Buckling', Proceedings I st. International Symposium on
Ship's Reliability, Varna, Bulgaria, September 1985.
AUTHORS' REPLY
Before answering the points that have been raised the authors
would like to thank the contributors for their interesting and
stimulating discussions.
Mr Viner has asked why we have expressed our results in
terms of annual rather than cumulative reliabilities. Both
annual and cumulative reliability indices and failure
probabilities can play a role in rational decision-making and
rule development. Only if the structure being analysed is
subject to some form of non-stationary influence, for example
corrosion, does the task of evaluating the reliability over
different time intervals need detailed consideration. Although
we have developed a method (Ref. 13) which allows us to
evaluate the cumulative reliability of a structure undergoing
uniform corrosion, it tends in its present form to result in
rather wide bounds on the reliability index at the reliability
levels reported in the paper. In addition, we are currently
unable to obtain cumulative results which reflect the effect of
the renewal of plating due to excessive corrosion diminution.
The fact that we have concentrated on the failure of a stiffened
panel as opposed to hull girder failure does not mean that we
think that hull girder failure modes should in general be
ignored. However, if the structural elements comprising the
hull girder all have adequate levels of reliability, then the
reliability of the hull girder will be assured provided that ship
service and structural arrangements follow traditional practice.
When and if the structural arrangement of the hull girder
differs radically from traditional forms then a reliability
analysis of the hull girder should be performed. An exercise of
this type is described in Ref. 44, and includes analyses in
intact and damaged conditions.
Mr Viner suggests that the values we propose for the yield
stress may be a little high. On this point the authors concur
with Dr Frieze that the value to be taken should lie between
the static and dynamic values, as is the case in the paper.
However, in port a ship can be subjected to very low strain
rates, and therefore it may well be more appropriate during
loading and unloading conditions to take the static yield stress.
Within the paper these conditions have not been analysed. The
authors agree with the need to systematically investigate the
relationship between mill test and zero or low strain rate
values of yield stress. The findings would be useful for both
deterministic and probabilistic calculations. Dr Frieze has asked
about the strain rates at which the model tests were carried
out. Although material yield strengths were available for the
model tests, no information was given with respect to strain
rates. Therefore it is not possible to identify whether the
results apply to static or dynamic conditions. Concerning Dr
Frieze's comments on the difference between the tensile test
strain rates of the USA and other countries, it should be noted
that ANSI/ ASTM E 8 specifies that the applied elas1ic stress
rate shall not exceed 690 MPa/min (II 5 N/mm' per second).
Lloyd's Register'-; Rules comply with the lACS unified
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
requirement which states that yield stress is to be determined
with an elastic stress rate of l-3 kgf/mm
2
per second ( 10-30
N/mm
2
per second).
Dr Frieze suggests that the section modulus of the hull girder
should have been reduced to take account of the reduced
effective width of the stiffened panels under extreme
compressive loads. An approach of this type is described in
Ref. 36. Because we use the ultimate compressive load
carrymg capacity and the gross cross-sectional area in
eva!uating the collapse stress of the stiffened panel, it would
be mcorrect to use the effective hull girder section modulus.
This, in reply to Mr Chalmers, is why the standard deviation
of the structural response model uncertainty is relatively small.
Dr Frieze has raised several points relating expressly to the
strength formulation rather than to the reliability analysis; the
first of these concerns the importance of stiffener distortion
and our use of tolerance values rather than measured
imperfections. We believe that initial stiffener distortions are
important and can see no justification for ignoring
the1r .. The quantity of published test data giving full
details of tmtlal deformation measurements is very limited.
Nommal values were therefore used to correlate with test data
in order to be consistent.
The effect of shakedown has not been ignored. It is because of
the anticipated effects of shakedown that we have adopted a
res1dual stress level that 1s equal to only 10% of the yield
stress. Dr Frieze is correct when he suggests that a slenderness
(LhrR) .J(i]E (where R=radius of gyration) in excess of 10
ts outside tbe practical range for merchant ships; this is why
we have not considered tensile failure of the outstand. The
ultimate strength formulation is intended for application only
where lateral pressure is not the dominant loading component.
It would therefore be unrealistic to investigate the accuracy of
results over the full lateral pressure range. The correction
factor about which Dr Frieze asks has been introduced because
the theory otherwise overestimates the effect of initial
deformation for low slenderness beam-columns. The factor was
therefore applied to force a full yield solution for stocky
members.
Dr Frieze asks about the way in which we have non-
dil!lensionalised the strength model uncertainty. Clearly, it is
ummportant whether we non-dimensionalise the error between
the predicted and the test values with respect to the test value
or s1mply non-dimensionalise the predicted value with respect
to the test value. The former quantity is equal to the latter
mmus l 0. The significant point is therefore whether it would
have t>et:n better to have non-dimensionalised with respect to
the values rather than the test values. In practice, it
1s not poss1ble to state categorically that a particular course of
action is better than another; for example, the error between
test and predicted values might be almost constant over the
wh<_lle range of possible values, in which case it would clearly
be to non-dimensionalise with respect to either test or
predicted values. The appropriate course of action must be
on the basis of an examination of the data. As reported
m the paper, non-dimensionalising the predicted values with
respect to the test results gives a mean value of 0 9991 with a
standard deviation of 00963; when the data are tested against
a normal distribution having the same mean and standard
deviation the x
2
statistic (for 7 degrees of freedom) is 490. If
the results are non-dimensionalised with respect to the
predicted values, then the mean and standard deviation
obta!ned are 10062 and 00975 respectively. The x
2
statistic
i.n case is 545 (again for 7 degrees of freedom).
Neither d1stnbutwn would be rejected as being non-normal at
the 50% level. The first of the distributions was chosen
because it has a slightly lower x
2
statistic and is (visually) less
skew. It is, of course, only because the mean values are close
to I 0 that both the non-dimensional ising methods result in
sets of data which are nearly normal; it is trivial to show that
if the means had not been close to 10 only one of the sets of
non-domensionalised data could have been nearly normal.
Dr Frieze is correct when he states that we have ignored any
correlatiOn between the hydrodynamic pressure and the vertical
wave bending moment. In Section 5.1 we give the dynamic
pressure m a normal ballast condition as 00485 N/mm
2
this
308
is the value which has an exceedance probability of w-a in
each wave encounter. The hydrostatic pressure acting on an
empty tank this condition is 005 N/mm
2
By comparison,
the pressure acting on the bottom of an emoty
ballast tank m a deep draught condition is 023 N/mm
2
while
the 10-
3
value of the dynamic pressure in this condition is
00348 N/mm
2
The largest total pressures are therefore
experienced. in the loaded condition, and are predominantly due
to hydrostatic pressure. We have concluded from this that the
dynamic pressure has only a small influence on the strength of
the bott?m and c?nsequently that we could ignore its
correlatiOn with the bendmg moment without significantly
reducing the accuracy of our results.
Dr final point concerns the possibility of reducing the
negative tolerance on plate thickness in order to reduce the
am?unt of plate produced. A draft requirement
which the average thickness of rolled plate to not less
than the nommal thickness 1s due to be submitted for lACS
Council consideration.
With reference to Mr Chalmers' point on corrosion and ship's
age, an oil tanker with 'cc' notation will only suffer from
corrosion where there is local breakdown of steelwork
protective coatings. As regards merchant ships in general, the
reader IS referred to Ref. 33 for a more detailed dtscusswn on
the subject of corrosion.
The authors agree with the points Dr Goodman has made on
the specialist nature of reliability analysis. Dr Goodman's
points reflect the reservations that several discussers have
about the practicability of reliability-based design. It is likely
for the foreseeable future that level 2 procedures will be used
for the. design of only novel or high-risk projects
for which there IS little or no relevant service experience.
However: the authors foresee that a major application of level
2 reliability of the type presented in the paper will be
to c.ahbrate traditiOnal level l requirements represented by the
Soc1ety's current rules and direct calculations.
Professor Faulkner's belief that there is a considerable
in the structural safety levels of existing ships as
determmed method presented in the paper is one that
we IS an area .subject to on-going investigation by
Lloyd s Register and requires the consideration of many more
aspects of ship service, design, and construction than have
been dealt with in the paper.
Professor Faulkner suggests that the target safety level should
be at the lower end of the range of safety levels implicit in
current rule designs. We are of the opinion that it is in fact
desirable that the target safety level should be chosen in a
more formal manner than this perhaps implies. Only in the
Simplest cases is it possible to formulate a design procedure
that Ie.ads to completely uniform levels of safety. It is therefore
essential that .the method. w.e employ in determining the target
level be consistent; IS 1t should lead to the same target
safety level when apphed after a rule has been calibrated as it
be.fore. If this condition is not satisfied, repeated
calibrations of the same rule will result in a different level of
safety each time. We suggest tentatively that the target safety
level shou.ld be set t.o co.rrespond to the average probability of
failure, wtth the caltbratton method chosen to give the same
average probability of failure after calibration as before.
Because of the non-linear relationship between reliability index
and probability of failure, this corresponds to a target level
near the lower end of the range of reliability indices.
In reply to Dr Lackenb, our intention is to include fatigue
and behavwur m future work on the reliability analysis
of ship As far as the fatigue limit is concerned,
typical welded JOIOts have fatigue limits equivalent to a stress
range which Is generally below 50 N/mm
2
The applied stress
range can clearly exceed this value.
Mr Sole has asked that we comment on the sensitivity of the
results to our and in particular to our assumptions
concermng the verttcal wave bending moment. The results
presented in the paper are sensitive to the assumptions made,
that much is incontrovertible. Moreover, because of the very
nature <;>f reliability analysis (i.e. it attempts to quantify
uncertamty) we are forced to make assumptions in areas that
would be completely ignored in a deterministic analysis.
However, we have not attempted to provide results that are in
some way actuarial. To do so would have been both unrealistic
and unnecessary. Our objective was to quantify in a notional
or comparative sense the level of safety associated with the
stiffened panels we have analysed. By doing this, we are able
to compare the levels of safety implicit in existing rules with
the levels of safety that are implicit in new or revised rules.
The outcome of this is that although the numerical results of
the reliability may be sensitive to the assumptions we
make, the deciSions that are made on the basis of those results
will not be. This does not of course mean that the assumptions
we made are of no consequence. It does however mean that
we are consistent about our assumptions, and
provided that the assumptions correctly identify and rank the
major sources of uncertainty relating to the structure being
analysed, then we will be able to use reliability analysis as a
decision-making tool.
As far as the assumptions relating specifically to the vertical
wave bending moment are concerned, they have been
described as being too severe by Professor Lehmann and Mr
Ostergaard, and not sufficiently severe by Professor Soares.
The conditions described for the strip theory analysis are
somewhat pessimistic in that the ship is assumed to be in head
and travelling at its service speed throughout its life. This
IS why the mean values of the model uncertainties on the
bending moments were taken as being significantly less than
10, so as to represent generally unrestricted seagoing service.
Professor Caldwell is certainly right in presuming that the
reliability of a whole deck may differ significantly from the
reliability of a single element of it. The extent to which it will
differ depends largely on the degree to which the strengths of
the elements that make up the deck are correlated. If all the
are (as opposed to simply being nominally
1denttcal), the reltab1hty of the deck will be the same as that of
the individual elements. Unfortunately, the situation is very
much more complex if the elements are not identical; it is then
necessary to investigate the way in which loads are
redistributed from failed elements to the surviving elements,
and also the extent to which failure of one element can
precipitate failure in adjacent elements. As we said in response
to Mr the combination of adequate element safety levels
and trad1ttonal structural arrangements is generally sufficient to
ensure adequate hull girder safety.
Dr Pittaluga has raised the point about correlation between
dynamic pressure and vertical wave bending moment that was
previously raised by Dr Frieze. We are in agreement with Dr
Pittaluga's conclusion that it is unimportant in this case. The
method proposed in Ref. 37 does indeed provide a means for
taking these effects into account; however, we are not at
present in a position to endorse this approach. It is not clear
that the additional complexity involved will be rewarded by
improved accuracy.
Pittaluga's second point is certainly an important one. This
IS a matter about which decisions have to be made on other
th_an purely technical grounds. In the paper we have treated
st1ll water as a stochastic process. In doing so
we do not between a truly random loading process
(wave bendtng moment), and a process that is uncertain and is
subject to random influences but is nonetheless wholly
controllable. Moreover, if we are to treat still water loads as
stochastic processes then these processes will have to be
modelled differently for each ship type. This will make the
task of formulating consistent longitudinal strength
requirements for different types of ships very complex. In
some of the work we have carried out since the paper was
read, we have treated still water bending moment as a
deterministic quantity. This is perhaps more in line with the
concept of permissible still water bending moments in the
assessment of longitudinal strength.
Professor Betts suggests that in view of some of the
assumptions and approximations made in the paper the
calculated failure rates should be lower than observed failure
rates. There are, however, two significant approximations
STRUCTURAL RELIABILITY ANALYSIS OF STIFFENED PANELS
which cause the results to overestimate the failure frequency.
Firstly, the results presented in the paper take no account of
structural redundancy. Secondly, the reliability analysis
assumes that both the still water and wave bending moment
spatial distributions have their peak values at the same location
as the stiffened panel being analysed.
Lehmann and Mr Ostergaard have suggested that
bucklmg IS not m fact the correct failure mode for the panel
discussed in the paper. An investigation of stresses at the
support indicates that a mid-span failure is indeed the critical
mode. Although applied stresses at the support are in excess of
those at the capacity is also greater. The strength at
mid-span Is based on the extreme fibre stress which is
influenced by lateral pressure, deformations and residual stress.
In contrast, full yield can generally be sustained at the
supports. Although in some cases extreme fibre yield may
occur at the supports in advance of a mid-span failure, this
condition is unlikely to constitute an ultimate limit state.
Pr<_Jfess.or Lehmann and Mr Ostergaard are surprised that the
rehabi11ty of the panels is decreased by corrosion. Corrosion
does not of course significantly reduce the critical buckling
stress of a stocky panel, but the reduction of the hull girder
modulus (also caused by corrosion) effectively
mcreases the loads applied to the panels. This is responsible
for the reduction in reliability.
Professor Lehmann and Mr Ostergaard have suggested that the
uncertainties relating to the vertical wave bending moment
have been significantly underestimated. We suspect that this is
due to a misconception about the way in which we have
modelled the distribution of vertical wave bending moment. It
must be appreciated that the model uncertainty applied to the
vertical wave bending moment is not the only uncertain
quantity that contributes to the modelling. Within a single
wave cycle the wave bending moment is modelled by an
exponential distribution that is fitted to the strip theory results
at the 10-B probability of exceedance level. For any large
number of wave cycles, the distribution of the largest bending
moment will therefore be a Fisher-Tippett Type I Extreme
distribution, i.e. a Gumbel distribution<4S). Fig. 5 shows the
probability density functions both for the parent exponential
dist.ribution, and for the Gumbel distributions corresponding to
penods of I hour, I year, and 20 years. The coefficients of
variation of these distributions are respectively 0 454, 0 196,
and 0165. The Gumbel distribution is therefore the major
source of uncertainty in our modelling of the wave bending
moment.
We agree that the expression given by Ditlevsen for the
pro?ability content of the intersection of two failure regions is
easier to evalua.te than the expression given in equation (13);
however, equatton (13) is perhaps more immediately
comprehensible.
In reply to Professor Soares, the distributions used to model
the still water and wave bending moments are not inconsistent.
The maximum wave bending moment during a one hour
mterval has been modelled by a Gumbel distribution. The
maximum still water bending moment during a single voyage
has been modelled by a Gaussian distribution. As is stated in
Section 4.2, the still water bending moment has been assumed
to remain constant throughout a voyage. Clearly, the maximum
:-vater moment during a year is not normally
d1stnbuted, but th1s ts accounted for in the load combination
procedure used in the paper, see equation (10).
Soares draws attention to the importance of the way
m wh1ch the truncatiOn of the still water bending moment is
treated. He is of course right in stating that the distribution of
the lifetime maximum still water bending moment is more
sensitive to truncation than is the single voyage distribution.
Although we have only considered annual rather than lifetime
reliabilities, our methodology does handle this aspect of the
modelling correctly.
Professor .Soares has raised a r<;>int concerning the correct way
of modellmg the effects of loadmg instruments in limiting still
water bending moments. It is true that stitl water bending
moments may on occasion exceed the permissible limit. When
thi;, occurs it will be as the result of a gross error. Such events
STRUCTURAL REUABILITY ANALYSIS OF STIFFENED PANELS
do not fall within the intended scope of our paper, and it was
therefore appropriate that the still water bending moment was
truncated at the permissible limit rather than being filtered.
The statistical results presented in Ref. 40 show that tankers in
a ballast condition experience hogging moments that are over
twice the magnitude of the sagging moments experienced in the
loaded condition. However, these results cover all tankers in
the sample range of 22,700 to 423,700 tonnes deadweight and
results for larger tankers of the size considered in the paper
are not presented separately. SWBM values compiled by Mano
et aJ(I9) substantiate the mean values given in Table II of the
paper. It should be noted that the values in the table are not
applicable to either tankers of medium size or segregated
ballast designs.
The results presented in the paper are based on the assumption
that the ship will experience 12 voyages a year. Making an
allowance for time spent loading and discharging, this is in
close agreement with Professor Soares' estimate of a typical
voyage duration for large tankers of 24 days. As regards the
bias on hogging and sagging VWBM, more recent work
suggests that for ships of the size considered in the paper the
difference in the bias values is somewhat overstated.
Professor Soares has raised an interesting point concerning the
effect of corrosion on hull girder section modulus. Provided
that steelwork is replaced as soon as it reaches its corrosion
limit, then the average material thickness will tend towards a
steady state value equal to the mean of the initial thickness and
the replacement thickness. This is in fact implicit in equation
(22), where the mean of the distribution tends towards v/2 as n
increases. As Professor Soares points out, the values given in
Table III do not take replacement into account in the
calculation of section modulus. If immediate replacement is
}l(i
assumed, the section moduli reduce less quickly than is
indicated by the table. In the case of the 'cc' notation ship the
difference is not significant. For the non-'cc' notation ship
taking account of immediate replacement gives deck and keel
section moduli after 20 years of 90 10 and 91 84% of the
nominal values respectively. The corresponding steady state
values are 85 45 and 86 17% of nominal for deck and keel
respectively. Table III gives values of 86 62 and 89 87% of
nominal for deck and keel respectively after 20 years.
Replacement is in fact likely to occur only after a survey has
been carried out, so the section moduli will decrease linearly
between the times at which material is replaced and increase
effectively instantaneously at those times. This is an area in
which further investigation may prove necessary, but for the
present time we feel that our use of the values given in Table
III is justified.
Professor Soares somewhat erroneously compares results in
Refs. I and 40 with those in our paper. Both Refs. I and 40
present reliability indices for hull girder failure over a 20 year
(lifetime) period based on a yield capacity factored by a
strength model error which is determined subjectively. In our
study, stiffened panel element failure is considered explicitly
and the reliability indices are calculated on an annual basis.
REFERENCES
44. Tomiyasu, 0., Hori, T., Tateishi, M., Rashed, S. M. H.,
and Miwa, S.: 'Strength Evaluation of Novel
Unidirectional-Girder-System Product Oil Carrier by
Reliability Analysis', SNAME, November 1985.
45. Gumbel, E. J.: 'Statistics of Extremes', Columbia
University Press, 1958.

Potrebbero piacerti anche