Escolar Documentos
Profissional Documentos
Cultura Documentos
in Structural Reliability
Robert E. Melchers AssessmentPart 1: Empirical
Centre for Infrastructure
Performance and Reliability,
Department of Civil, Surveying and
Models
Environmental Engineering,
The University of Newcastle,
Reliability theory for the assessment of existing structures for remaining safe life must
Australia 2308
consider the probabilistic deterioration of structural strength as well as multiple applied
e-mail: Rob.Melchers@newcastle.edu.au
loadings. This requires good-quality models of deterioration processes. Models for cor-
rosion, which have been applied in the literature for marine applications, are reviewed. It
is shown that these usually are statistical only, using data pooled from many sources. They
also use little or no theoretical insight. As a result, they provide poor-quality mean-value
information and very high statistical uncertainties. The conclusion is that better modeling
is required and that this involves much better understanding of the factors influencing
marine corrosion. These factors are reviewed briefly, with particular emphasis on marine
immersion corrosion of mild steel. DOI: 10.1115/1.1600467
1 Introduction terms of a constant corrosion rate. For ships, a 10% plate thick-
ness allowance for wastage is typical in Classification Society
Reliability assessment for existing structures is a matter of in-
rules for commercial vessels. For naval vessels, such allowance
creasing importance both for offshore and land-based structural
typically is 0%. These allowances do not provide for localized
systems and for shipping. Since such structures tend to deteriorate corrosion, for example, as might occur at welds or other locations
with time, the reliability assessment must consider the time- of nonuniform materials.
dependent reduction of material strength properties R(t) as well None of these approaches are helpful in estimating the remain-
as it time varying uncertainty as represented by the probability ing safe life of an existing structure. Typically, a much higher
density function f R (r,t). In addition, of course, the reliability as- level of accuracy is required, both for predictions of structural
sessment must consider the applied loading Q and its related un- safety and for prediction of likely future corrosion. This is the
certainty as represented by the probability density function f Q (q). main focus of the present paper, which reviews the attempts that
For the simplest reliability problem, the representation in Fig. 1 is have been made to develop models for marine corrosion. As will
indicative. It is seen that the probability density function of the be seen, the information available to structural engineers about
resistance f R (r,t) moves, with time t, in the direction of lower marine corrosion tends to be anecdotal, not well organized, and of
mean strength and that it becomes flatter i.e., the uncertainty is limited use for simple applications 3. It is also generally insuf-
increased. It is immediately evident that the structural reliability ficient for reliability analyses purposes.
is reduced with time. Of interest in its own right, structural reli- The paper is organized as follows. Section 2 is a short review of
ability assessment is also an important component of risk-based the theory of structural reliability estimation, highlighting the im-
inspection and maintenance planning techniques being developed portance of good-quality deterioration models. Section 3 describes
by a number of authors 1,2. briefly the various forms of corrosion of practical interest and the
In contrast to fatigue, only limited attention has been given to various environments that must be considered. Section 4 reviews
modeling the time-dependent loss of material strength due to cor- statistical data for corrosion of ships and data trends derived from
rosion. This is somewhat surprising since corrosion is important corrosion coupon testing and outlines some statistical models.
for many older structural systems exposed to harsh environmental Section 5 provides a critique of these efforts, reviews the factors,
conditions such as at-sea or along the coast. Measures such as which should be considered and suggests the approach that should
cathodic protection and surface coatings are known to be not al- be used for the development of better-quality probabilistic corro-
ways wholly effective nor sufficiently robust, particularly in the sion models based on understanding of the physiochemical pro-
splash zone and at structural details. cesses. Part 2 of the paper deals with the latter aspect.
For the assessment of structural condition, much attention is
being given to the application of indirect methods to obtain indi-
cations of possible corrosion. Some, such as the measurement of 2 Structural Reliability Estimation
corrosion potentials, provide only the potential for corrosion. Oth-
ers, such as impedance measurements, are very difficult to use in 2.1 General Comments. The theory of structural reliability
field conditions and are of dubious reliability. Moreover, they are was developed originally as an attempt to rationalize the setting of
expensive to use and require access to the region of interest. In factors of safety in structural engineering 4. Because of the con-
addition, there is the problem of extrapolating from highly local- siderable simplifications to the theory, the probability of failure
ized data points to larger structural elements such as ships and most commonly estimated with such procedures may not reflect
offshore structures. reality and is sometimes termed a nominal or formal prob-
For design, a common practical approach is to allow for a sac- ability. This point is important for management decision-making,
rificial amount of materiala corrosion allowance, quoted in such as in maintenance scheduling for existing structural systems,
since for optimal decisions it is desirable to have measures of
Contributed by the OOAE Division for publication in the JOURNAL OF OFF-
probability as realistic as possible 5.
SHORE MECHANICS AND ARCTIC ENGINEERING. Manuscript received, January 9, In addition, it is important to note that the theory of structural
2002; final revision, January 2003. Associate Editor: A. Naess. reliability as it currently exists can be used only in part to describe
and human error usually are not considered. This may be impor-
tant in management decisions. where XX(t) is a vector process and ( ) denotes the positive
component only. The term E(X n Xx)x n n(t).x(t)0 repre-
2.2 Basic Structural Reliability Problem. There are two sents the outward normal component of the vector process at the
key elements in analyzing structural reliabilitythe loads acting domain boundary. This result can be extended readily to allow for
on the system and the resistance of the system to support them. gradual deterioration of structural strength with time with the re-
For offshore platforms, ships at sea, etc., the loads principally will sult that v and p f (0,t L ) both become time dependent. However, as
be forces generated due to wave action and ship motion. Such the structural resistance becomes lower the Poisson approximation
forces will be generated globally, such as those leading to hull used for Equation 1 becomes gradually less applicable. It follows
girder bending, and locally, such as those at side and bottom pan- that accurate representation of deterioration is important in assess-
els, at hatches, etc. The forces will be a function of the wave ing the validity of application of Equation 2.
height and, hence, sea-state, and can arise also from wind condi- Since the above formulation is not always easy to apply in
tions. For ships, bow-area loads generated under high-speed practice simplified and asymptotic methods such as the FOSM/
operationsthe so-called slamming forcesmay need to be con- FORM time-invariant techniques are often applied. Except for
sidered. There may be effects also from differential temperatures. certain special problems, these estimate the probability of failure
In general, there will be several possible loading systems acting p T approximately. The estimate will approach p T as the models
on the structure; most of these will be functions of time and, for technical reliabilityincluding those for loads, resistances,
hence, will be best described by a random process. A typical re- and structural dimensionsbecome more refined. It can be esti-
alization of one such process is shown in Fig. 2. mated from 4:
Various forms of deterioration such as fatigue and corrosion
will cause the strength limit state function G Q,R(t) to exhibit
deterioration with time. In general, there will be uncertainty both
about the precise strength and the rate of strength loss with time,
p T
G(x)0
f X x dx (3)
Journal of Offshore Mechanics and Arctic Engineering NOVEMBER 2003, Vol. 125 265
Journal of Offshore Mechanics and Arctic Engineering NOVEMBER 2003, Vol. 125 267
immersion corrosion data to 1994 for mild-steel coupons pooled shown in Fig. 8. It is evident that there is still a large degree of
from all available sources. The data showed a large degree of scatter in the data. This is considered to be due to: not all envi-
scatter. After detailed examination of the data, this was attributed ronmental and material variables having been properly isolated; a
mainly to seawater temperature differences. An approximate tem- preponderance of short-term data, often from laboratory trials us-
perature correction was applied 27, using estimates for mean ing artificial seawater; and a lack of an appropriate theoretical
seawater temperatures at each site. This produced the points model on which to base data comparisons 28.
The data of Fig. 8 can be used to construct an empirical statis-
tical model using the power law format commonly used in atmo-
spheric corrosion modeling work. After calibration, this estimated
the mean corrosion depth as 27:
m d 0.084t 0.823 (4)
with a standard deviation function given by,
s d 0.056t 0.823 (5)
A log-normal probability distribution model for the corrosion
depth was found to be consistent with the data. However, the
coefficients and the exponents of Equations 4 and 5 were ex-
tremely sensitive to data points at longer exposure times. This
suggests that fundamentally the model is not a good one.
4.2.2 Splash-Zone Marine Corrosion. There are very few
studies of splash-zone corrosion reported in the literature. Figure 9
shows a summary of pooled data from a variety of sources in
terms of equivalent metal loss versus duration of exposure. As for
immersion corrosion, there has been rather poor control and re-
porting of environmental conditions.
4.2.3 Atmospheric Marine Corrosion. In contrast to the ear-
lier cases, there is a considerable body of data and modeling ef-
forts for atmospheric corrosion, although rather little is specifi-
Fig. 7 Localized corrosion at weld heat-affected zone HAZ cally for marine corrosion. Most effort has gone into industrial
between stiffener and plate atmospheres 28,29,30. The data and modeling specifically for
Table 3 Typical corrosion rates for oil tankers
marine corrosion have been summarized recently 31,32.
4.2.4 Pitting and Crevice Marine Corrosion. As noted, pit-
Segregated Ballast Tank Corrosion Rates
ting corrosion is not particularly critical for the strength of mild-
Corrosion and low-alloy structural members. However, it may be critical for
Location Rate mm/yr Remarks containment. There is an extensive literature on pitting corrosion
Deck plating 0.100.50 Uncoated in general. A summary of data and models for marine corrosion is
Deck longitudinals 0.251.00 Uncoated available 33,34. For stainless steels exposed long term to immer-
Side shell plating 0.06 0.10 Uncoated sion conditions, crevice corrosion can be a significant problem.
Side shell 0.100.25 Uncoated
longitudinals
Bottom shell plating 0.04 0.10 Uncoated 5 Development of Better Models
Transverse bulkhead 0.300.50 Uncoated
plating
The above shows that attempts, to date, at developing models
Transverse bulkhead 0.200.60 Uncoated for corrosion have been based largely on empirical relationships
vertical stiffeners developed from pooling of data from a variety of sources. As
demonstrated clearly in Figs. 8 and 9, simply collecting data and
pooling it without detailed analysis results in statistical models no longer condoned. Hence, the historical, statistical information,
with very large degrees of uncertainty and with consequent low while interesting, is unlikely to be applicable to new situations nor
predictive capability. This should not be unexpected, yet it is the is it capable of generalization.
approach most commonly adopted. Unfortunately, such models To improve this situation, it is considered that the approach that
are unlikely to be of much practical use except for the most gen- must be followed is to develop models based firmly on established
eral, broad-brush decisions. physiochemical concepts for corrosion. Such models will need to
Moreover, as these pooled data statistical models are based on be calibrated to actual field data. For this, two possibilities exist
historical data, they may be inapplicable to more modern applica- subjective supplementation of the existing information or com-
tions with different operational and maintenance practices. For missioning of new tests, under better control conditions. The latter
example, much of the ship data for bulk carriers applies to vessels is an expensive and long-term process as illustrated in the release
that did not have coatings for their ballast tanks, a practice that is of the complete five-year test results of the 1980s ASTM
Fig. 9 Collected data for splash-zone corrosion data corrected for temperature with 5 and 95 percentile bands 26
Journal of Offshore Mechanics and Arctic Engineering NOVEMBER 2003, Vol. 125 269
round-robin worldwide test program for copper-bearing mild possible to relate these findings back to the original structure, see
steel, 90/10 copper-nickel, and aluminum 35. This work makes Fig. 10. An example of this for the corrosion of holds in bulk-
an important contribution to the data available for marine immer- carrier ships is available 37.
sion corrosion. It provides information about uniform corrosion
and about pitting and has relatively good information about criti- 6 Discussion
cal environmental factors. Although the discussion here has been confined largely to cor-
Importantly, laboratory data are unlikely to be very helpful for rosion as an independent deterioration mechanism, in practice,
the calibration of models although it may be important for scien- corrosion is not an independent issue for remaining life assess-
tific understanding. Typically, such tests have used either saline ments of aging systems. Corrosion interacts with applied stresses,
water or artificial seawater. Importantly, these mixtures do not fatigue, mechanical damage, and, most importantly, with protec-
replicate truly seawater conditions because one of the most sig- tive systems such as cathodic protection, paint coatings, and man-
nificant ingredients, bacteriological and biological matter, is not agement practices. The interaction with each of these phenomena
present. or materials is a matter that cannot be ignored in practice, even
The initial framework for the development of marine corrosion though the interactions are not fully understood in all cases. There
models has been outlined 36. This requires: 1 detailed under- is considerable ongoing research of a fundamental nature into the
standing of factors influencing corrosion and 2 reference back to mechanics of impressed current cathodic protection and its influ-
fundamental corrosion mechanics. Table 4 summarizes the factors ence on corrosion losses under a variety of marine conditions
that normally are considered to influence marine immersion 38. Eventually, this will need to be translated into engineering
corrosion. design rules and guides for the protection of aging infrastructure,
Because of the fundamental nature of these requirements, it has including probabilistic models. However, the first priority from a
been necessary to move away from corrosion observations on ac- probabilistic modeling viewpoint must be to develop better mod-
tual structures and the concomitant large uncertainties in the data els for the seemingly most elementary problemimmersion cor-
to a study of the behavior of corrosion coupons. It should then be rosion under at-sea conditions.
It should be clear that the development of such probabilistic
corrosion models must follow the development of deterministic,
physiochemical corrosion models. These must reflect a reasonable
degree of physical reality if they are to have predictive power
beyond the data from which they were calibrated. In this sense,
they might be viewed as engineering models rather than scien-
tific ones, or phenomenological models, conveying the essence
of behavior, if not its complete details. The background and ap-
proach for the development of such models is discussed in Part 2
of this paper.
7 Conclusion
The structural reliability theory necessary for the assessment of
ageing infrastructure systems and for their maintenance planning
was outlined and it was argued that there is an urgent need for
Fig. 10 Relationship between structural and coupon speci- better-quality models to represent adequately the deterioration
men corrosion. mechanism of corrosion. Attention was given to marine corrosion
Journal of Offshore Mechanics and Arctic Engineering NOVEMBER 2003, Vol. 125 271