v. Scheepsbouwkunde
IN TiTUE OF
rtWehnische Hogeschool
Li -; V .Yos
Delft
SHIP STRUCTURAL DESlisi FOR EXTREME. LOADS
by
Edward V. LeT,jJ
Prepared for Special Issue
of Schiffstechnik in Honor of the
Birthday of Professor O. Grim
7 September 1976
U i., . L,/
/
/
,., r"/
1 7 .r,..
/
,A
1 6 NOV. 1976ARCHIEF
-E,E4E3A
Introduction
Dr.
Grim has always been interested in the extreme behavior of ships atsea and this interest has led him into valuable investigations of non-linear
responses. In addition to such important studies the full understanding of
.extreme behavior requires the application of tools
obtained from probability
theory.
It is the latter subject that will be discussed briefly here.The idea persists in some quarters that the problem of extreme conditions
is a deterministic one. If we knew the worst wave a ship or structure will
ever encounter, we could calculate the re:,:imum loads and design accordingly. But in reality the "worst wave" may not give the worst load. Furthermore, we cannot ever claim to know the worst wave or the worst load without some
prob-abilistic qualification -- such as the worst in 100 years (i.e.,
"the
100-yearwave"). Finally, the design of the structure cannot be so
precise
that thestrength just barely exceeds the maximum load. Variability in the strength of
the material, unfairness and misalignment of structure, material flaws and weld defects, inadequacies in the design theories, etc., all lead to variabil-ity in the strength of the structure.
The above difficulties can be overcome either by including a generous em-pirical factor of safety or by utilizing a probabilistic approach, i.e. the
use of reliability theory. Let us limit
ourselves to
the problem of designinga ship with adequate but not excessive longitudinal hull strength. This
prob-lem takes on
an entirely different aspect when we ask simply, has theproba-bility of
failure been reducedto
an acceptable value? To answer it we need to know the distribution functions of load (or demand) and of strength(capa-bility)
expressed in the same units. Then we can determine the probability offailure by integration. Finally, we must decide
what is
an acceptable valuefor this failure probability. Probability of Failure
(
If y is demand and z is capability, the probability of failure may be
de-fined as the probability that the ratio zly is greater than 1.0. Since :riy is
an instantaneous
value
of safety factor, we can also say that the failureprobability is the probability that the safety factor always exceeds unity.
It can be shown that
the
cumulative distribution of the safety factor is givenby (1) p(z I [1 - Q(y)]r(z) dz Y J.
--Where Q(y) is the cumulative distribution of demnvd, r(z) is the density function of capability.
The integral
can be
evaluated numerically to determine its value atz0/y. =
1.0, which
is the probability of failure per cycle of loading.NordenstrOm has investigated the probability of failure when both demand and capability can be expressed as specific probability functions, such as
normal, Weibull, and Frechet (2). Mansour has also investigated this problem
(3)
Loads
The development of the cumulative probability of demand or loads has been developed elsewhere on the basis of considering the entire population of indi-vidual peak-to-trough or peak-to-mean amplitudes of stress or bending moment
in a ship's lifetime (4). It can be estimated on the basis of a summation of
.many Rayleigh distributions (5). It need not be expressed as a discrete
dis-tribution function, although it has been shown that it can be represented by a
so-called Weibull distribution with suitable parameters (2). A log-normal or
exponential distribution is not acceptable, however (6).
The significance of such a long-term distribution as that shown in Fig. 1 is that it shows the probability, P. (per cycle of load) of exceeding
differ-ent levels of stress or bending momdiffer-ent. For design purposes it is convenient
to interpret the curve in terms of a large data sample of N cycles, where N =
1/P. Hence, an alternate scale of N is shown In the figure. For example,
-probability of exceedance of 10-6. means that in a data sample of 106 cycles of
load we would expect that one value would exceed the indicated level. Of
par-ticular interest is the bending moment at 108 which corresponds to a typical oceangoing Ship's lifetime of 25 years.
However, the above lifetime bending moment is only an average value. If
there were 100 similar ships operating in the same service for which data were
available, we would expect a
variation in
this bendingmoment-to-be-exceeded-once. Karst has shown on the basis of a binomial model (approximated by the
Poisson density function) that one ship out of 100 would be expected to exceed
a stress corresponding to N = 302 x 108 = 1010 (7). This -- or the value
cor-responding to 1000 ships (1011) -- could be used as a design load. However,
as shown above, if we proceed to the concept of probability of failure the distribution itself will be utilized, rather than a single value.
Structural Response
Less is known about the distribution of structural response or
capabili-ty. Perhaps the best treatment is that of Mausour and Faulkner (8), who
con-sider the effects of a great many uncertainties. They assume that the
proba-bility density function is normal, with
a
mean value for each potential modeof failure determined by standard structural analysis techniques. The
stand-ard deviation is determined by estimating the coefficients of variation (ra-tios of standard deviation to mean value) for each factor of uncertainty and
then combining them. Although a great deal more research is needed to
estab-lish reliable values of all coefficients, the paper gives a good idea of the order of magnitude of the variability of capability.
Objective uncertainties, based on Ang's subdivision (9), are random
var-iations that can be measured experimentally. They include material
proper-ties, scantlings and imperfections in construction. Under the first category
-- material properties Young's modulus (E) and yield strength were
consid 2 consid
-
-ered, the first being assumed to be small enough to be ignored. The coeffi-'cient of variation for yield strength was believed to be 6-82: and a value of
87 was assumed for a tanker structure in either tension or compression.
Scantling variations include both variability in plete thickness and in areas
and inertias of stiffeners. An overall value of was selected on the basis
of an ISSC study (10). Under imperfections
in construction,
cracks and flawswere arbitrarily excluded on the grounds that they represented non-ductile
be-havior,
but they should obviously be studied further. Shape imperfections maybe favorable or unfavorable, and they depend on mode of failure. It was
as-sumed that they can be included in the subjective uncertainties. Finally
re-sidual welding stresses were taken to be a type of imperfection that must be allowed for directly in the case of compression, and a value of 12% was as-sumed for a typical tanker (compression only).
Subjective uncertainties (9) are those that have not been measured and
therefore
must be estimated on the basis of engineering judgement. They de-..d on mode of failure -- such as deck or bottom yielding, strut-panel com-i,ression failure, grillage instability, etc. -- excluding fatigue.Estimated
mean
values and coefficients of variation given in (8) for aanker are summarized in Table 1.
Table 1
3
-The above figures are quoted as examples of variability in structural
capability, but are not endorsed for design use. Such data for ship
struc-tures are extremely scarce, and it Is felt that a great deal more research
needs to be done on the subject. Mention may be made of the fact that a
pro-ject in this area is expected to be undertaken soon by
the
Ship StructureCommittee.
Acceptable Failure Probability
With probability data on both structural demand and capability available,
the probability of failure can be calculated, as previously shown. The final
step in a rational approach to designing a ship for extreme loads is then the
determination of an acceptable level of probability. J.F. Dalzell, in an
in-formal memo to the Ship Research Committee (12 May 1970), gave a valuable analysis of Some published Lloyd's Register data on merchant ship losses (11), which covered 13 years (1949-1966) and 390,000 ship-years of service
experi-Estimated Uncertainties for
, nip
77;
ree.ngth
/IL
c, Displacement, Mean Stress Tons/in.2 Tanker (8)n1 ,nn
LAJV Coefficients Objective of Variability, Subjective % Total n,/Jo.eee
a,
Deck yieldBottom beam-column 16.5 15.0 3.6 8.3 3.0 7.0 9.1 11.3 (middle tank empty)
e,:;iae Bottom yield 16.5 8.6 3.0 9.1
Deck strut-panel 15.4 8.8 7.0 11.3 Deck grillage instability 16.5 8.6 7.0 11.2
--tonsee. He assumed that the losses desienated "Foundered" (31% of all losses)
Te. cases
of
complete structural fnauie, althoughthere were no doubt
nu-Jous exceptions. Assuming a 25-ycer average ship life, he concluded that a
figure of
about 0.006would be a
rease:elbie value foo theprobability of
failure that has been tacitly accepted over the past 20 years for large
oceangoing ships. In proposing a specific figure for a new design criterion,
however, we
feel that a moreconservative figure should
be adopted, and Dal-.-11's suggestion of 0.001 is tentativcAy proposed (12).A basic approach to determining the probability level to be used in a design criterion is that of "expected loss", which has been summarized in conventient form by Freudenthal for application to maritime structures in
general (13). It is based on the principle that the best design is the one
that
minimizes
the expected total cost, where the latter consists of the sumof initial cost and failure cost, as is explained below. It is very
diffi-cult, if not impossible, to assign a dollar value to passengers and crew, but it will be assumed -- perhaps over-optimistically -- that in this day of
ef-. ient communication and life-saving technology a ship may founder without
the crew being lost.
Expressed as an equation, the total:expected cost to
be minimized is,
L ==. I P
1
I = initial cost of the ship (or structure)
P1 = probability
of
failure (in a lifetime)F = anticipated total cost of failure (replacement cost + cargo loss ±
temporary charter of replacement ship + loss of business from cus-tomer reactions + cost of pollution or other environmental effects, etc.)
Some approximate calculations (14) confirm that the above tentative
fig-tire of 0.001 for lifetime failure probability, is reasonable. However, more
work is needed on this subject, based on extensive, routine reliability data
on actual ship hull
girder
failures.Numerical Example
A very interesting numerical example of a reliability analysis, leading to an estimate of lifetime failure probability was given by Mansour and
Faulkner (8) for a frigate. Of greater interest for the present purpose
are
results given in less complete form for the 85,000-ton deadweight tanker. In
both examples the demand (or load) probabilities were determined in a
some-whnt :simplified manner. Hence, a recalculation of the tanker load
probabil-ity distribution has been made, in order to permit actual results of a com-plete failure probability calculation to be presented in numerical and graph-ical form. The starting point is the tabulation of "Expected bending moment" vs. significant wave height given in Table IV (8), obtained by means of the MIT Seakeeping program and Pierson-Moskowitz spectra assuming long-crested
irregular head seas. Expected bending moment values were converted to MS of
.record by dividing by 1.25 and to stress by dividing by the section modulus
(deck) of 152,600 in.2-ft. Resulting values were corrected approximately for
the effect of short-crestedness and for an equal distribution of all ship-wave headings using a factor of 0.70 obtained from previous calculations. Data
were plotted
and read off to correspond to our usual wave height groups,-4--+ where
Standard deviations of rms.stresses were. c7!atimated on the basis of previous
calculations,
and
the Walden distribution of wave heights in the NorthAtlan-tic was assumed. (15).
See
'Table 2 below.Table 2
Assumptions for Long-Term 'toad Calculations
_.
'-c&
...-k, .)
The
procedure for obtaininga
long-term stress curve for each weathetgroup and then integrating to obtain a single long-term curve is described in
detail elsewhere (5) (16). Results are presented graphically in Fig. 2,
where the cumulative distribution of demand as calculated is shown on the same .scale as the density function of capability, as given by Mansour and
Faulkner (8) for the critical case of deck strpt-pauel buckling in the
sag-ging condition (mean stress = 15.4 tons per sq. in.,
standard deviation =
15.4 x 0.113 = 1.74 tons per sq. in.) and also for the case of deck grillage
instability. See Table 1,
-Numerical solution of the previously given equation for probability of
failure by deck strut-panel
buckling gives
a value of p = 1.2 x 10-12 for thefull load condition if still water bending moment is zero. This would be en-.
tirely acceptable, for it
implies
that only one ship out of 10,000 similarships (with lifetimes. of 108 cycles) would fail.
However, When the constant still water bending Moment of 671,000 ft.-'tons (sag) is included, the probability decreases to P1 = 3.7 x 10-8,
com-pared to 1.3 x 10-8 in (8). The reason
for this
moderate discrepancy isbe-lieved to be differences in the methods of calculation of demand. If a ship
encounters 108 cycles of stresS in.a
normal
lifetime, either figure implies'that the ship would be expected to fail by deck strut-panel buckling during
its lifetime. This is obviously unacceptable. Assuming
that
the ballast'loading would give
a much lower probability of failure,the
combinedproba-bility for loaded and ballast would be slightly reduced, but would not change
the
situation significantly.- If it is assumed that, the design of the deck is modified to increase its.
load carrying ability to equal or exceed the grillage strength,
then
theea-__ 5 Sig. Wave .Height Range Ft. Frequency of Occurrence % Mean RAS Stress Tons/in .2 Standard Deviation. -RMS Stress
Tonslin.2
_
_ 0 - .3 8.75 0.0229 0.0020 3 - 6 23.75 0.0275 0.0033 6:- 9 30.70 . 0.0760 . 0.0066 91 ,-- 12 20.35 0.2294 0.1075 12 - ify 6.90. 0.3670 0.1396. 1i6 - 21 4.95 0.6147 0,1841 21_.,. 27 3.35 0.8716 0,2176 27 - 34 1.06 i.0689 0.2353 .. 34. - 42' 0.17 1.2916 0.2392 > 47.nj19
.1,crilA
0.2556pability is governed by the latter. (Mean.
Strength = 16,5
tonsiin,2 andstandard deviation = 11,2%). On this basis the probability of failure
be-comes,
PI = 7.5 x 10-9. On the basis of a binomial model, assuming a ship 'lifetime of 109 cycles, this means, that there is a lifetime failureprobabil-ity of 0.528. Or 53 ships out of a, fleet
of
100 Would be expected to fail.This Is also unacceptable, but the result, must be considered tentative for a
number of reasons:
North Atlantic sea
conditions have been assumed, which are generally
more severe than most tanker routes 7- In spite of the famous Cape
of Good Hope rollers.
Assumed standard deviations of structural capability (8)
may
be toosevere.
3, No allowance was made for lateral
longitudinal
bending in oblique.seas, which tends to increase deck edge. stresses (14).
4. Since the calculation of failure probability depends on the
uncer-tain "tails" of the distributions of demand and capability, there is a possibility for appreciable error in the calculated probabilities. The still water bending moment has been considered to be
determinis-tic. However, in a more complete treatment, the variability of
still
water moffient should be taken into consideration (14).Nevertheless, results given in this simple example are close enough to
reasonable figures to encourage further work on all aspects of this interest- 5,
ing approach to ship structural design. CONCLUSION
It is concluded that the basic calculation procedures are now available
for rational
ship longitudinalstrength
design on the basis of reliabilityprinciples. The greatest need appears to be for data On the probability
as-pects of hull strength. Second in importance is the need for statistical
da-ta on heavy weather damage to longitudinal strength members related to number
of ships "at risk." A third area for further work is the collection of wave spectra for unusually severe situations, such as the approaches to the Eng-lish Channel (continental shelf) and the waters off South Africa.
,
ACKNOWLEDGEMENTS
This paper
would not have been possible
withoutthe basic
wOrk ofMan-sour and Faulkner (8) on structural capability, nor without the continuing
support of the American
Bureau of Shipping and Ship
Structure Committee forresearch. at Webb Institute. The author wishes to thank various members of
a
the Webb staff for their assistance, including Professor Robert Zubaly,
Pro-fessor Bruce Stephan and Student Assistant R.P. Dallinga (of Delft Technical University).
-6
Pie
-References
.1. E. Abrahamsen
and
N. NordenstrOm, "Designand
Reliability of ShipStruc-tures",
Trans, Spring Meeting, SNAtiE. 1970,?, Nordenstrft, "Methods for Predicting Long-Term Distributions of Wave
Loads and Probability of Failure for
shills",
Appendix VIII. Det NorskeVeritas Research Report No. 69-23e-S, 1969.
3, A.E. Mansour, "Probabilistic
Design
Concepts in Ship Structural Safetyand Reliability", Trans. SNAME,
1972-4; E.G.U. Band, "Analysis; of Ship Data to Predict Long-Term Trends of Hull
Bending Moments", ABS Report, November 1966.
5-,
E:V. Lewis,
'Predicting Long-Term Distributions of Wave-Induced BendingMoment on Ship Hulls", SNAZIE Spring Meeting, 1967.
6. E.V. Lewis, "Long-Term Applications of
St. Denis/Pierson
Technique toShip Design", SNAME Seakeeping Symposium, October 1973.
1.0.J. Karst,
"Statistical Techniques", Appendix A to Reference 16.A.E. Mansour and D. Faulkner, "On Applying the Statistical Approach to
Extreme
Sea Loads and Ship HullStrength",
Transaction of the RoyalIn-stitution of Naval Architects, 1972.
Ang, "Elements of Structural Reliability and its Implications in
Design", and, 'Extended Reliability Basis for Formulation of Design
Crite-ria", Proc. ASCE Engineering Mechanics Division Conferences,
"Probabil-istic Concepts and Methods in -Engineering", and "Probability and Random
Processes in Engineering", Purdue University, Lafayette, Indiana,
No-Vember 1969,
.Report of Committee 10 on "Design Procedure", Proc. of the Fourth ISSC
Tokyo, 1970, published by SNAJ. February 1971.
114, 14.1 Beer, "Analysis
of
World Merchant Ship Losses', Trans. RINA, 1969_. 12. R.S. Little, F.V. Lewis, and F.C. Bailey, "A Statistical Study of
Wave-Induced Bending Moments. on Large Oceangoing Tankers and Bulk Carriers",
Trans. SNAME, 1971.
13 A-M. Freudeuthal, and W.S. Gaither,. "Probabilistic Approach tO Economic
Design of Maritime Structures", XXIInd International Navigation Congress
Section II,
Subject 5, Paris,
1969.
14. E..V. Lewis, D. Hoffman, W.M. Maclean, R. van Hoff, and R. Zubalye "Load
Criteria for Ship Structural Design", Ship Structure Committee Report
SSC-240, 1973...
15, IL. Walden,_ "Die Eigenschaften der Meereswellen im Nordatl
Ozean", Deutscher Wetterdienst Seewetteramt Publ. No, 41,
16: D. Hoffman, 3. Williamson, and
E.V.
Lewis, "Analysis andof Full-Scale Data on
Midship
Bending Stress of .Dry CargoStructure Committee Report, SSC-196, June 1969.
--t antischen. 1964 Interpretation Ships", Ship N. A.H-S.
-7-i!,3-' 10-5
0-5 r.-T4 VO-3(X>Xj) TOTAL PROBABILITY OF EXCEEDINGXj
Fig- 1.
:Typical Long-Term Probability Curves, Extrapolated
from Full-scale Data and Predicted from Model Tests.
S.S- Wolverine State in the North Atlantic
(5)
4g 11/11/1111 . LONG
,MODEL
7 j-FULL-SCALE TERM PROBABILITY DATA --k-
CURVES 22 , DATA [2pi i I ..1 1--,.. I Ic
t-tr) C.1 I.
, 1 INUM3F_R OF CYCLES 109 .r.19 ;o7 IV 13,5 t 0.4 i0,3W IV
0.13)tuu,E,