Stochastic Model for Damage Accumulation

in Rubble-Mound Breakwaters Based on Compatibility

Conditions and the Central Limit Theorem
Carmen Castillo1; Enrique Castillo2; Alfonso Fernández-Canteli3; Rafael Molina4; and Rebeca Gómez5

Abstract: Motivated by existing damage accumulation modeling difficulties, some suggestions on how to build consistent stochastic models for
damage accumulation in breakwaters are given. The models avoid the selection of easy to use mathematical functions, which are replaced by those
resulting from a set of properties that are expected in the model. It is shown how an inadequate selection of damage progression models leads to
inconsistencies, which are avoided by using the proposed consistent mathematical structures. Dimensional analysis, compatibility conditions, and
the central limit theorem are the bases for building a model aimed at reproducing the stochastic damage progression on breakwater as a result of
general random wave actions. Four elements are identified as essential for a damage accumulation analysis: (1) the statistical distribution of the
initial damage, (2) the mathematical structure of damage growth curves, (3) the damage function, which accounts for how the variables affect dam-
age, and (4) the integral form in which damage accumulates, which permits relating its random character to the normal distribution. These four
elements are combined to derive a general structure of models for determining the damage produced by any wave history, and to obtain the evolution
of densities of cumulative damage with time. Finally, the proposed methods are illustrated by means of a real case example, and one of the models
proposed is shown to be convenient in reproducing damage progression in breakwaters. The proposed method is valid not only for breakwaters but
to other evolutive processes appearing in the maritime and coastal engineering field. DOI: 10.1061/(ASCE)WW.1943-5460.0000146.
© 2012 American Society of Civil Engineers.
CE Database subject headings: Stochastic models; Breakwaters; Coastal engineering; Damage.
Author keywords: Buckingham theorem; Compatibility conditions; Central limit theorem; Random damage growth; Breakwater; Armor.

Introduction 2. The Buckingham theorem, which: (1) permits us to decide

whether a relationship among the selected variables exists, (2)
In this paper, the problem of building models for the stochastic forces us to work with dimensionless ratios, (3) permits us to
analysis of damage accumulation in general from a new point of reduce the initial number of variables involved in the problem,
view is dealt with, and the particular case of breakwaters is con- (4) allows us to determine the minimum number of dimension-
sidered, though the models are applicable to other evolutive pro- less ratios in terms of which the problem can be stated, and (5)
cesses. The selection of arbitrary but simple mathematical functions, facilitates the selection of the appropriate dimensionless ratios.
generally done because of mathematical or numerical convenience, All of these points contain extremely valuable information in
is consciously avoided. On the contrary, the true mathematical struc- model building, which should not be ignored. In addition, if the
ture of functions that arise from the desired properties is discov- selected dimensionless ratios have a clear physical meaning,
ered. To this end, the following four main tools are used: their interpretations will be much more clear and simple.
1. Previous engineering knowledge, which allows identification 3. Functional equations, which allow us to state the expected
of the initial set of variables involved in the problem based on properties in terms of their mathematical representation. In
engineering experience in the field. solving these functional equations, the resulting mathematical
structure of the variable relationships is obtained as a conse-
1
Associate Professor, Dept. of Civil Engineering, Univ. of Castillae quence of such properties. This permits, on one hand, the
La Mancha, 13071 Ciudad Real, Spain (corresponding author). E-mail: identification of the most general functions, which satisfy
[email protected] these properties, and, on the other hand, the knowledge of
2
Professor, Dept. of Applied Mathematics, Univ. of Cantabria, 39001
Santander, Spain. E-mail: [email protected]
which functions lead to the violation of them.
3
Professor, Dept. of Construction and Fabrication Engineering, Univ. 4. The central limit theorem of statistics. This theorem states that
of Oviedo, 33203 Gijón, Spain. E-mail: [email protected] if a random variable is the sum of many random variables,
4
Ph.D. Student, School of Civil Engineering, Polytechnic Univ. of under very general circumstances, it is approximately normal.
Madrid, 28040 Madrid, Spain. E-mail: [email protected] This allows us to certify that a normal distribution is the
5
Ph.D. Student, School of Civil Engineering, Polytechnic Univ. of assumption to be made in one of the steps when the accumu-
Madrid, 28040 Madrid, Spain. E-mail: [email protected] lation of wave effects is what it is looked for.
Note. This manuscript was submitted on April 15, 2011; approved on
February 14, 2012; published online on February 16, 2012. Discussion
period open until April 1, 2013; separate discussions must be submitted for Motivation
individual papers. This paper is part of the Journal of Waterway, Port,
Coastal, and Ocean Engineering, Vol. 138, No. 6, November 1, 2012. This work was motivated by some damage accumulation modeling
©ASCE, ISSN 0733-950X/2012/6-451e463/$25.00. difficulties reported by Melby and Kobayashi (1998). Initially,

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 451
Melby and Kobayashi (1998) suggested for damage accumulation The reader should discover the new ideas presented here and
in breakwaters during the period (tn ; tn11 ), the model not center on the particular cases, which have been used only for

b  illustrative purposes.
5 t 2 tn
b
SðtÞ ¼ Sðtn Þ þ as Ns ; tn # t # tn11 ð1Þ
Tmb Previous Investigations on Damage Progression
in Breakwaters
where the bar refers to mean values, SðtÞ is the mean damage at time
t, as is a constant, Ns is the stability number, b is a constant, and Tm is To do a complete risk analysis for breakwater rehabilitation and
the mean period. However, Melby and Kobayashi (1999) perceived maintenance studies, the probability of failure caused by instability
that Eq. (1) has problems, because predicted values based on Eq. (1) of the armor layer must be analyzed, and therefore knowledge on
are dependent on the prediction time. What really happens, which the deterioration rate of the armor layer is needed. Many authors have
will be subsequently discussed, is that this expression is not invariant dealt with this problem in the past (Burcharth 1994; Burcharth 1997;
with respect to time origin changes. Consequently, they proposed Burcharth 2000; Burcharth 1984; Burcharth and Sorensen 1998;
replacing the model [Eq. (1)] with the following model: Castillo et al. 2004b, 2006; Mínguez et al. 2006). Some of them
assume that the section under constant wave intensity finally reaches
 1=b  1=b  1=b t 2 tn an equilibrium profile, and others state that it is not a real equilibrium
SðtÞ ¼ Sðtn Þ þ as Ns5 ; tn # t # tn11 ð2Þ
Tm but a damage rate reduction.
Breakwater armor stability is highly variable, and this variability is
which solves the aforementioned shortcomings. difficult to quantify, because of the many variables involved in the
Changes in the time origin, which affect variables t and tn , does problem. Breakwater damage is likely to depend on wave height and
change the cumulated damage SðtÞ in Eq. (1), but not in Eq. (2). In wave period, time of operation, structure slope, shape and method of
other words, while Eq. (2) is invariant with respect to time origin armor placement, among other factors. Breakwater stability is stochastic
changes, Eq. (1) is not. in nature, as both the loading (Scotto and Guedes-Soares 2007) and the
Eq. (2) assumes lack of memory. In other words, the damage armor conditions are. There is uncertainty in armor placing, shape, and
produced by waves depends only on the previous damage and the stability, and there is uncertainty in loading by waves and water level.
wave intensity history, but not on how the previous damage has been Some models for damage progression have been described based
reached. on some of the aforementioned variables, but only a few of them
In this paper, functional equations are used to explain the deep provide statistical information on the damage description, basically
reasons for these problems, which rely on the fact that a damage the mean value and standard deviation. However, breakwater
progression law, such as Eq. (1) or (2) cannot be arbitrarily chosen, damage progression with time under wave action should be defined
but must satisfy compatibility conditions. More precisely, the by its associated density function as a function of time, and not only
mathematical structure of the most general damage progression laws by the mean and variance time functions. One of the aims of this
satisfying this condition is derived, and it is shown that the first paper is to provide some ideas on improving damage progression
model [Eq. (1)] does not have this structure, but the second model estimation based on probabilistic assumptions.
[Eq. (2)] does. In fact, the second model [Eq. (2)] is shown to be Both stability and damage progression on rubble mound break-
convenient and simple in reproducing damage progression in waters have been studied by many researchers. Some have de-
breakwaters. veloped empirical models for armor stability as a result of regular
The previous knowledge of Melby and Kobayashi (1998) was waves (Iribarren 1938; Hudson 1958; Hudson 1959; Ahrens and
crucial in selecting the set of variables involved in the problem. They McCartney 1975; Thompson and Shuttler 1975; Van der Meer 1988;
considered that S(t), Ns , t, tn , and Tm were the relevant variables. Losada and Giménez-Curto 1982; Pilarczyk and Den Boer 1983).
An important property of Models 1 [Eq. (1)] and 2[Eq. (2)] is that These models showed the influence of armor characteristics (shape,
they are stated in terms of dimensionless terms, that is, all terms are weight) and positioning (slope) on stability. Some researchers have
dimensionless values. Because this is very relevant, as stated by intended to apply regular wave models for the stability analysis
the Buckingham theorem, using this type of model in engineering under irregular waves (Jensen 1984; Vidal et al. 1991, 1995), 2003),
practice is proposed. 2004, 2006; Jensen et al. 1996, Medina and McDougal 1990), based
The important contribution of Melby and Kobayashi (1998) on a given distribution of wave heights on a sea state and storm
consists of incorporating the statistical behavior of damage by duration or given wave height and period parameters.
providing laws for the mean and standard deviation of damage Other models (Ahrens 1975; Ahrens and McCartney 1975;
progression based on some experimental work, which using some Carver and Wright 1991; SPM84 1984) include the effects of ir-
similarities (dimensional analysis) can be extrapolated to other regular breaking and nonbreaking waves and the wave period,
cases. However, they did not provide the probability density analyze different layer types and armor gradation (SPM84 1984), or
functions (PDFs), which are needed for a more precise analysis of even influence the number of waves (Medina and McDougal 1990).
damage progression in breakwaters. Pfeiffer (1991) compared different models and noted that none of
In this paper, the PDFs are derived too. More precisely, the PDFs, them were valid for predicting extended damage.
because of the central limit theorem, are related to normal distri- Thompson and Shuttler (1975) performed tests on both long-term
bution, and closed expressions for them are given. deterioration and short single-damage storms on riprap armored with
The aim of this paper is not to give a complete model for sto- an impermeable core and concluded the following: (1) the erosion
chastic damage accumulation that was able to deal with the most rate is dependent on the significant wave height, (2) the erosion rate
general case of breakwaters. The paper aims only at giving a general decreases with time, and the damage curves can become nearly
methodology of how the mathematical structure of these models horizontal under some sea states that appear as a stability state, and
should be and what properties should have to avoid compatibility (3) the method of lying the riprap has a significant effect on damage
and dimensionality problems, and to be consistent with the central history. These methods, which are useful to predict damage for
limit theorem (assume normal variables when they need to be a given design storm, do not provide a detailed information of the
normal). damage progression process.

452 / JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012
 Carver and Wright (1991) conducted irregular wave stability The paper is structured as follows. First, the problem is introduced
experiments to try and determine the randomness on damage ini- and the set of variables involved and the different damage concepts
tiation induced by varying armor placement and wave time series are introduced. Next, the model for damage growth, as a result of the
with the same spectral parameters. Font (1968) noted that armor action of deterministic regular waves, assuming a deterministic initial
placement technique is important for determining when initiation of damage of the breakwater, is introduced. Next, the effect of random
damage occurs, but it is not so relevant when damage has evolved. initial damage is considered for the cases of sequences of regular and
Cover depth (minimum remaining depth along the cover layer) was irregular waves, the particular model of the power law function is
shown by Torum et al. (1979) and Davies et al. (1994) to be a useful analyzed, and the model is extended to the case of random waves.
parameter in characterizing the reserve capacity of the armor layer. Later, formulas for the mean, interquartile range, and variance of
Eroded depth (maximum erosion in the cover layer measured normal damage in terms of time are given. Then, how the proposed method
to the slope) gives indication of the progress toward failure (defined can be applied in real practice is explained and applied to an experi-
by exposure of the underlayer). mental example. Finally, some conclusions are given.
Van der Meer (1988) analyzed damage evolution based on
damage profile statistics, basically with mean damage and its
evolution in time based on some experiments. The dimensionless Statement of the Problem and Variables Involved
eroded area S 5 Ae /D2n50 , normalized minimum cover depth C 5
dc /Dn50 , and normalized eroded depth E 5 de /Dn50 were analyzed Assume that a breakwater with an initial random damage (because
through their mean and standard deviation in some experiments and of, for instance, the construction process) is exposed to the action of
compared with normal distributions. These values vary with damage a sequence of waves with random height and period during a given
progression. The standard deviation of damage D increases with its time period. Then, the breakwater damage increases with time. The
mean value: variability of damage increases as the mean damage paper aims at obtaining a model providing the stochastic damage
increases, and higher damage levels may be estimated with smaller progression in terms of time and random wave history. To this end,
relative errors, because the coefficient of variation decreases with the following steps are used:
increasing mean damage. Because the coefficient of variation is the 1. The set of variables playing a role in the problem under study is
quotient between the standard deviation and the mean, it can de- identified;
crease even though the standard deviation increases, if the mean 2. The most general formula relating all these variables in terms
increases more rapidly. Finally, the variability in eroded depth of dimensionless ratios using the ∏ Buckingham theorem is
decreases as damage approaches failure. obtained;
These results provided empirical formulas for breakwater dam- 3. The most general formula giving the damage in terms of
age as a result of depth-limited waves, which depend on stability the number of regular waves is derived using functional
number Ns , mean period Tm , and time t. In addition, damage is a equations;
function of the structure slope, armor gradation, porosity, and stone 4. The deterministic cumulative damage corresponding to se-
shape and method of placement, among other factors. quences of different regular waves and irregular waves is
Van der Meer (1988) also defined the rate of mean damage calculated;
increase for varying conditions of incident waves represented by 5. The PDF of the damage caused by a random initial damage and
significant wave height and mean period. the aforementioned sequences are calculated; and
Medina (1996) proposed an exponential model on individual 6. The PDF of the damage caused by random initial damage and
waves for damage evolution under regular wave attack, which random variation of wave intensity is calculated.
depended on wave height, Iribarren’s number, and the number of Thus, the paper progresses in two stages: first the deterministic
regular waves. The model was later applied to individual incident damage case is analyzed, and next, its stochastic character is studied.
waves (Medina et al. 2003).
Finally, a wide revision on damage progression on rubble-mound
Set of Variables Involved in the Problem
breakwaters was given in Melby (1999).
For the sake of illustration some simplifications are considered, and
it is assumed that the only variables involved in the problem are: (1)
Original Contributions
the mean wave period Tm , (2) the initial damage Dp0 (a dimensionless
Based on the aforementioned considerations, the main contributions variable), (3) the total damage D (a dimensionless variable), (4) the
of this paper are the following: time duration t, (5) the wave height H, (6) the still water depth at the
1. A general methodology to build damage progression models toe of the structure ht , (7) the specific weight of the water gw , (8)
of coastal structures based on dimensional analysis, compat- the armor specific weight g a , (9) the breakwater rubble mound
ibility conditions, and functional equations is presented; slope a (this variable can be replaced by Iribarren’s number), (10)
2. A general damage accumulation curve, which satisfies impor- the nominal diameter Dn50 , and (11) the damage percentile p (this
tant compatibility conditions, is derived and, two particular variable is considered only in the random damage case).
examples, Models 1 [Eq. (1)] and 2[Eq. (2)] reported by Melby The relationships of these variables with the elemental magni-
and Kobayashi (1998), are shown not satisfying and satis- tudes mass, length, and time are shown in Table 1. Because the rank
fying these conditions, respectively; of the associated matrix is 3, from the Buckingham ∏ theorem, any
3. Based on the central limit theorem, a stochastic model related possible existing relationship among these initial 11 variables can
to the normal distribution for damage progression, which has also be written, without loss of generality, in terms of eight di-
general validity, is obtained; and mensionless variables. There are several possible choices, and all
4. The need of this methodology, the consideration of the afore- of them are equivalent in the sense that they are valid to reproduce
mentioned compatibility condition, and the applicability of the the functional relationship of the initial variables.
proposed model are illustrated by its application to the data One example, and though other possibilities exist, the following list
provided by Melby and Kobayashi (1998, 1999), which shows of eight dimensionless variables has been selected: the initial damage
a very good fit of the model in this case. Dp0 , the total damage D, the relative duration (mean number of waves)

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 453
Table 1. Variable Dimensions Written in Terms of Fundamental Magni- discussed first, and next an extension to the sequences of different
tudes Length, Mass, and Time regular and irregular waves is provided.
Dimension Tm Dp0 D t H ht gw ga a Dn50 p
Length 0 0 0 0 1 1 22 22 0 1 0 Damage Caused by Regular Waves
Mass 0 0 0 0 0 0 1 1 0 0 0
In this section, a model for damage accumulation for the case of
22 22
Time 1 0 0 1 0 0 0 0 0
regular waves is constructed. Because deriving a formula for Dp 5
hðtp; Dp0 ; Nsp ; hpt ; Dp; apÞ, as explained in Eq. (3), is too complicated,
following Melby and Kobayashi (1998), fixed values of hpt , D, and
t* 5 t/Tm , the relative still water depth at the toe of the structure a are assumed, that is, hpt 5 hpt0 ; Dp 5 Dp0 , and ap 5 ap0 , and then the
hpt 5 ht =Dn50 , the relative excess specific weight D  5 ðga 2 gw Þ=gw , aim of this section consists of deriving a formula
the stability number Nsp 5 H=ðD  Dn50 Þ, the rubble mound slope a,    
and the probability p, that is, the total damage D can be written as Dp ¼ h t p; Dp0 ; Nsp ; hpt0 ; Dp0 ; ap0 ¼ Dp tp; Dp0 ; Nsp ð8Þ
 
Dp ¼ h t p; Dp0 ; Nsp ; hpt ; Dp; ap ð3Þ giving the total damage D in terms of variables tp; Dp0 , and Nsp , all of
them were assumed to be deterministic. More precisely, the general
for the deterministic case, and as form of the damage growth function in terms of time t* and initial
  damage Dp0 for a given stability number Nsp , assuming that all other
pp ¼ r tp; Dp0 ; Dp; Nsp ; hpt ; Dp; ap ð4Þ variables are fixed, is derived. This implies that the Dpðtp; Dp0 ; Nsp Þ
function is valid only for selected ratios of hpt , D, and a, that is,
for the random case. valid for any combination of values of the initial variables that keep
The asterisks used throughout refer to dimensionless variables. those ratio values fixed.
Because of the similarities of the cyclic wave action and cyclic
fatigue loads, fatigue models (Castillo et al. 1999, 2008a, b; Castillo
Damage Measures and Hadi 1995) can be extended to include the case of rubble-mound
As indicated, D is the damage and an assumed dimensionless breakwaters.
variable (with asterisk). In this section, several possibilities for D As indicated in the Introduction, the function Dpðtp; Dp0 ; Nsp Þ
are discussed. Following Melby and Kobayashi (1998), three cannot be arbitrarily chosen. When modeling, researchers usually
options are considered. choose easy to deal with functions, such as polynomial, expo-
1. The eroded depth nential, or Fourier (sine and cosine) type functions, but this has
some theoretical and engineering disadvantages. Thus, the function
Dp1 [ E p ¼
de
ð5Þ Dpðtp; Dp0 ; Nsp Þ is selected in order to satisfy some conditions that go
Dn50 well above of its obvious increasing character on t*, Dp0 , and N0p , as
will be subsequently discussed.
where de 5 maximum difference between undamaged and To obtain the form of the function Dpðtp; Dp0 ; Nsp Þ, a functional
damaged profiles measured normal to the slope. equation was used considering the following compatibility condi-
2. The dimensionless eroded area tion that this function must satisfy:

Ae If the breakwater with initial damage Dp0 is subject to the wave action Nsp
Dp2 [ Sp ¼ ð6Þ during a time tp 5 t1p 1 t2p , the resulting final damage, according to the
D2n50
definition of the function D*(), will be Dpðt1p 1 t2p ; Dp0 ; Nsp Þ. However, this
value D can also be obtained in two steps: first the breakwater is subject to
where Ae 5 cross-sectional eroded area. the action of waves during a time t1p . Therefore, the damage will be
3. The dimensionless cover depth Dpðt1p ; Dp0 ; Nsp Þ, and next, the resulting breakwater, with initial damage
Dpðt1p ; Dp0 ; Nsp Þ is subject to an extra period of duration t2p . Thus, using the
Dp3 [ Cp ¼
dc
ð7Þ D() function definition again, the final damage Dp½t2p ; Dpðt1p ; Dp0 ; Nsp Þ; Nsp 
Dn50 is obtained. Because both values must provide the same result, no matter the
values of t1p and t2p , one gets the functional equation
where dc 5 minimum depth of armor remaining.
While E  and S are direct measures of damage, because they   h   i
increase with damage, C is an indirect measure, because it Dp t1p þ t2p ; Dp0 ; Nsp ¼ Dp t2p ; Dp t1p ; Dp0 ; Nsp ; Nsp ð9Þ
decreases with damage. Therefore, it should not be called damage.
To deal with the case of random waves, the case of deterministic It is obvious that not any function D() satisfies Eq. (9) and that this
waves first needs to be analyzed, discussed, and understood. Once equation is useful to identify the functions Dpðtp; Dp0 ; Nsp Þ that
a model for deterministic damage progression has been constructed, satisfy this compatibility condition and those that fail to satisfy it.
the random case can be considered. The following two sections deal Fortunately, Eq. (9) is the well-known translation equation
with the deterministic and the random cases, respectively. (Aczel 1966; Castillo et al. 2004a; Castillo and Ruiz Cobo 1992),
whose general solution is

Damage Accumulation Caused by Deterministic      

Dp tp; Dp0 ; Nsp ¼ g g21 Dp0 ; Nsp þ tp; Nsp ; " Dp0 ; tp
Waves and Initial Damage
ð10Þ
In this section it is assumed that the breakwater or structure is subject
to deterministic waves, and a model for deterministic damage ac- where g(×,×) is an arbitrary invertible function with respect to its first
cumulation is developed. The case of a sequence of regular waves is argument.

454 / JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012
 From a practical point of view, this is an interesting result stating  
Dp ¼ g tp; Nsp ¼ as Nsp5 tp þ g
b
ð13Þ
that the function of three arguments D(×,×,×) cannot be chosen ar-
bitrarily to represent the problem, and that the only degree of
freedom is a two argument function g(×,×), invertible with respect to where as and g are constants and t  is the dimensionless time as-
its first argument. sociated with damage D, we have
In summary, Eq. (10) provides the most general formula satis-
!1=b
fying Eq. (9) for the deterministic damage D progression in terms   Dp 2 g
of the initial damage Dp0 and time t*, for constant Nsp . tp ¼ g 21
Dp; Nsp ¼ ð14Þ
To illustrate the practical interest of Eqs. (9) and (10), they are as Nsp5
applied to the Melby and Kobayashi models.
and replacing Eqs. (13) and (14) into Eq. (10), we obtain the
Example 1 (Melby and Kobayashi Damage Curve) model
The Melby and Kobayashi damage curve [Eq. (1)], that in di- 2 !1=b 3b
    p
mensionless terms can be written as D 2 g
Dp t p; Dp0 ; Nsp ¼ as Nsp 4 þ t p5 þ g
5 0
ð15Þ
  as Nsp5
Dp t p; Dp0 ; Nsp ¼ Dp0 þ as Nsp5 tp
b
ð11Þ

does not satisfy the compatibility condition [Eq. (9)], because the which can be written as

1=b h  p 5 i1=b
two values
   1=b  p
    Dp tp; Dp0 ; Nsp 2 g ¼ D0 2 g þ as Ns tp ð16Þ
Dp t1p þ t2p ; Dp0 ; Ns p
¼ Dp0 þ as Ns p5
t1p þ t2
p b

The particular case of g 5 0 provides the Melby and Kobayashi

and (1999) alternative proposal [Eq. (12)] for the damage progression
      curve, showing that the model [Eq. (16)] is an extension of the Melby
Dp t2p ; Dp t1p ; Dp0 ; Nsp ; Nsp ¼ Dp0 þ as Nsp5 t1pb þ t2pb and Kobayashi model [Eq. (12)].

are different, unless b 5 1.

On the contrary, the Melby and Kobayashi damage curve Location-Scale Families
[Eq. (2)], that in dimensionless terms can be written as
In this section, the authors try to simplify the general model
  h   1=b ib [Eq. (10)] without imposing very strong conditions. An important
Dp tp; Dp0 ; Nsp ¼ Dp0 þ as Nsp5 tp
1=b
ð12Þ simplification was obtained if the damage gðtp; Nsp Þ function is
assumed to belong to a location-scale family, that is, to a common
satisfies the compatibility condition Eq. (9), because the two values parametric family for all values of Nsp with its parameter values
dependent on Nsp
  h   1=b  p ib
Dp t1p þ t2p ; Dp0 ; Nsp ¼ Dp0 þ as Nsp5 t1 þ t2p
1=b
      
Dp ¼ g tp; Nsp ¼ q a Nsp tp 2 b Nsp ð17Þ

and where q is any invertible function. Eq. (17) means that the effect of
    changing Nsp is a change in the scale aðNsp Þ and a change in the
Dp t2p ; Dp t1p ; Dp0 ; Nsp ; Nsp
location bðNsp Þ parameters.

h b
 p 1=b  1=b p ib 1=b  1=b p A family F is said to be a location-scale family if it is stable with
¼ D0 þ as Nsp5 t1 þ as Nsp5 t2 respect to location and scale changes of the variables. In other
h words, X 2 F implies that aX 1 b 2 F , where a and b are
1=b  1=b  p ib
¼ Dp0 þ as Nsp5 t1 þ t2p constants. One example of a location-scale family is the normal
family, because if X is normal Nðm; s2 Þ, then Y 5 aX 1 b is normal
Nðam; 1 b; a2 s2 Þ. One example of a nonlocation-scale family is
are identical, showing that, contrary to their previous model the beta bðm; nÞ family, because if X is a b(m, n) random variable,
[Eq. (11)], this Melby and Kobayashi new proposal [Eq. (12)] then aX 1 b is not beta. In particular, the beta has range (0, 1) and
satisfies the aforementioned compatibility condition. aX 1 b has range (b, a 1 b).
This illustrates the practical importance of the compatibility A family is a location-scale family if and only if it is invariant
condition [Eq. (9)] and, more important, the relevance of Eq. (10). with respect to changes of scale (units of measurement) and changes
The initial problems experienced by Melby and Kobayashi (1998) of origin. These families are unavoidable when the variables in-
with the damage progression law [Eq. (11)] could have been avoided volved permit changes of scale (e.g., length, speed, time) and
if damage progression curves were selected using Eq. (10). For- changes of origin (e.g., time).
tunately, they detected the problem and solved it by replacing With this weak assumption from Eq. (17), t can be obtained in
Eq. (11) with Eq. (12). terms of D, that is
 
  q21 ðDpÞ þ b Nsp
Example 2 (an Extension of the Melby and Kobayashi p
t ¼ g21 p p
D ; Ns ¼   ð18Þ
Damage Curve) a Nsp
Because Eq. (10) provides the most general form of models satisfying
Eq. (9), any invertible function gðtp; Nsp Þ can be used. If we choose and then Eq. (10) becomes

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 455
         h     i
Dp tp; Dp0 ; Nsp ¼ q q21 Dp0 þ a Nsp tp ð19Þ Dp2 ¼ q q21 Dp1 þ a Nsp2 t2p
h n h     io   i
Eq. (19) provides the total damage, including the initial damage ¼ q q21 q q21 Dp0 þ a Nsp1 t1p þ a Nsp2 t2p
Dps and the damage associated with a sequence of duration t  of h
waves of intensity Nsp . Because of the presence of q() and q21 () in       i
¼ q q21 Dp0 þ a Nsp1 t1p þ a Nsp2 t2p ð24Þ
Eq. (19), D is independent of bðNsp Þ, which implies that only scale
families are needed, that is, bðNsp Þ [ 0, to obtain the damage
progression model [Eq. (19)]. Therefore, the solution Eq. (10) of Notice that the initial damage for the t2p extra duration is the previous
functional Eq. (9) helps discover that scale families are sufficient for damage Dp1 .
the models. Eq. (19) is simpler than Eq. (10), because the latter Repeating the process for time t3p of extra waves of intensity
depends on function gðDps ; Nsp Þ of two arguments, while the first Nsp3 ,. . ., and time t‘p of extra waves of intensity Nsp‘ , results in
depends on two functions q21 ðDps Þ and aðNsp Þ of a single argument
(invariance with respect to change in units (scale). This simplifi-   ‘ 
P 
Dp‘ ¼ q q21 Dp0 þ a Nspi tip ð25Þ
cation is the result of assuming a common scale family for all values i¼1
of Nsp , as done in Eq. (17), and not because of an arbitrary
assumption. that is, a formula providing the damage Dp‘ , resulting after [ se-
In the following example, it is shown that Melby and Kobayashi quences of regular wave trains, each sequence different from the
use this assumption. This can be considered an argument in favor of rest.
using scale families, which arise from physical considerations, as The sum in Eq. (25) is not imposed by any assumption arbitrary
demonstrated by these authors. or not, but arises naturally from the structure of model [Eq. (19)].

Example 3 (Melby and Kobayashi Damage Curve) Damage Caused by Irregular Waves
Melby and Kobayashi (1999) use a particular case of the location
family The most real situation is the case of irregular waves, which is
subsequently considered. If the load history is given by the contin-
uous function Nsp 5 Nsp ðtpÞ, that is, the wave intensity changes with
Dp ¼ qðt pÞ ¼ t p
b
ð20Þ time (number of cycles), we get
8 9
and the scale parameter < ð
tp
=
   
   1=b DpðtpÞ ¼ q q21 Dp0 þ a Nsp ðtÞ dt ð26Þ
: ;
a Nsp ¼ as Nsp5 ð21Þ 0

Eq. (26) is very important and clearly shows the role played by the
In fact, replacing Eqs. (20) and (21) into Eq. (19), obtains the fol- different variables and elements in the damage accumulation pro-
lowing model: cess as follows:
1. The q function provides the shape of the damage curve and
2 !1=b 3b
depends on the material properties of the breakwater;
Dp0 2. The term q21 ðDp0 Þ is the contribution of the initial damage to
DpðtpÞ ¼ as Nsp5 4 þ tp5 ð22Þ
as Nsp5 the total damage;
3. The aðNsp Þ damage function says how the wave intensity Nsp
affects damage; and
Ð tp
which is the model proposed by Melby and Kobayashi. This proves 4. The integral term 0 a½Nsp ðtÞdt is the contribution of the wave
that it is a particular case of the general model [Eq. (19)]. Note that sequence to the total damage and, as indicated, arises from
replacing Eq. (20) into Eq. (17) obtains Eq. (13). the structure of Eq. (19).
Consequently, in the subsequent sections, scale families of the With these interpretations of the model’s [Eq. (26)] terms, we
form [Eq. (17)] with bðNsp Þ [ 0 are assumed. have an important physical view of our model [Eq. (26)].

Damage Caused by Sequences of Regular Waves Damage Accumulation Caused by Random Initial
In this section, the assumption of constant Nsp
is abandoned and Damage
a time-dependent Nsp ðtÞ is assumed. The case of sequences of regular
waves is discussed first, and then the general case of irregular waves The random behavior of damage in breakwaters can be the result of
is analyzed. two main causes: the initial damage, and the random character
If the breakwater has initial damage Dp0 and it is subject during of waves. In this section it is considered first that the initial damage
time t1p to random waves of intensity Nsp1 , the cumulated damage Dp1 of the breakwater is random and next that the waves are random.
according to Eq. (19) becomes
h   i Damage Caused by Regular Waves
Dp1 ¼ q q21 Dp0 þ a Nsp1 t1p ð23Þ
In this section, it is assumed that a breakwater of slope a is subject to
regular waves (constant Nsp ) and that Dp0 is the only random variable,
If the breakwater is the next subject, during time t2p , to extra waves which, according to Eq. (10), implies that D is random too. It is also
of intensity Nsp2 , the cumulated damage Dp2 becomes assumed that the function g is known.

456 / JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012
 Let fDp0 ðd0p Þ be the PDF of Dp0 , FDp0 ðd0p Þ the corresponding illustrated, as the areas of the shadowed regions, associated with
cumulative distribution function (CDF), and p a probability asso- t* 5 40,000 waves and t 5 50,000 waves.
ciated with a given percentile xp , where capital letters have been used Alternatively, one can assume a different initial density fDp0 ðd0p Þ,
to refer to random variables and lower case letters to the particular as say the one corresponding to t 5 40,000 waves (Fig. 1), that leads
values they take. to a nonnull initial probability of failure. Therefore, the proposed
The aim of this section consists of determining the PDF of D model includes both cases.
in terms of the PDF of Dp0 for given t* and Nsp . Because D and Dp0 Eq. (27) has the inconvenience that both the g and FDp0 functions
are related by Eq. (10), to determine the PDF of D* in terms of the must be assumed or determined experimentally.
PDF of Dp0 , a change of variable needs to be performed. Thus, If we assume the scale families in Eq. (17) for gðtp; Nsp Þ from
the CDF of D after t  cycles becomes Eq. (27), we obtain
      
p ¼ FDpðtpÞ dp; tp; Nsp ¼ Prob½DpðtpÞ # d p pp ¼ FDpðtpÞ ðd pÞ ¼ FDp0 q q21 ðdpÞ 2 a Nsp t p ð28Þ
     
¼ Prob g g21 Dp0 ; Nsp þ tp; Nsp # d p
    
¼ Prob Dp0 # g g21 dp; Nsp 2 tp; Nsp Damage Caused by Irregular Waves
    
¼ FDp0 g g21 dp; Nsp 2 tp; Nsp ð27Þ In the previous section, it was assumed that Nsp was constant with
time. In this section, Nsp is a function of time Nsp ðtpÞ, and Dp0 is
which is the looked-after function [Eq. (4)]. random, with CDF FDp0 ðd0p Þ and scale families are assumed. Then,
the following CDF of the damage D(t) can be obtained by a sim-
Taking derivatives in Eq. (27), the PDF of D* becomes ple change of variable [Eq. (26)]
 
fDp x; t p; Nsp 0 8 tðp
91
         <   =
fDp g g21 x; Nsp 2 tp; Nsp g9 g21 x; Nsp 2 tp; Nsp FDpðtpÞ ðtpÞ ¼ FDp0 @q q21 ½DpðtpÞ 2 a Nsp ðtÞ dt A
¼ 0    : ;
g9 g21 x; Nsp 0
ð29Þ
Example 4 (Beta Density for Initial Damage)
For example, assume that the density function of the initial damage which is the CDF of the resulting damage D(t) after time t subject
fDp0 ðd0p ; 0; Nsp Þ is a beta density b(2, 2) and that the damage to random waves of intensity Nsp ðtpÞ.
growth curve is gðtp; Nsp Þ 5 tp2 . Then, the densities of the damage Eq. (29) has practical interest, because it provides the CDF of the
levels for t 5 0, 10,000, 20,000,. . ., 50,000 waves are shown in cumulated damage for any given history of regular or irregular
Fig. 1. waves in terms of the CDF of the initial damage and the wave history
It is interesting to analyze Fig. 1 after including the critical Nsp ðt pÞ.
damage dcp in it. Initially and in the first stages t # 30,000 waves, the
possible values of the damage levels are below the critical damage
dcp 5 0:6, the one producing failure, and then, the probability of Power Law Damage Function
failure is null. This explains how for this example at the first stages of
the loading process none of the samples fail. However, afterward, In this section our models are further simplified by making some
that is, when t . 30,000 waves, the probability of failure is positive, additional assumptions. A particular case of interest can be obtained
and failure can occur. In Fig. 1, the probabilities of failure have been by assuming the power law damage function

Fig. 1. Evolution of the probability-density function fDp ðd p; t pÞ with the number of waves t  and illustration of the critical damage dcp (all in di-
mensionless terms)

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 457
 qðt pÞ ¼ tp þ g
b
ð30Þ and

ð Þ
8 tðp
9b
where b and g 5 constants and bðNsp Þ 5 0, leading to [Eq. (17)] <   =
FDpðtpÞ ðd pÞ ¼ FDp0 ðdp 2 gÞ1=b 2 a Nsp ðtÞ dt þg
     b : ;
Dp ¼ g t p; Nsp ¼ a Nsp tp þ g ð31Þ 0

that has the flexibility of positive and negative curvatures for b . 1 ð35Þ
and b , 1, respectively, and leads to the inverse damage function
Example 5 (Illustrative Example)
  ðDp 2 gÞ1=b
tp ¼ g21 Dp; Nsp ¼   ð32Þ Fig. 2 illustrates one example of damage growth curves for four
a Nsp different initial damages (Dp0 5 0; 1=3; 2=3; 1), corresponding to
the case
The power law function appears very frequently in engineering 8 9b
tðp
problems and has been given some important justifications by <   p  =
Bazant (2006) and Bazant and Pang (2007), who even demonstrate Dpðt pÞ ¼ p 1=b
D0 þ a Ns ðtÞ t dt
that g must be null. In these cases, the power law assumption is not : ;
0
arbitrary but deeply justified.
Note that Eq. (31) is separable in Nsp and t , that is, the g function with a½Nsp ðtÞ1=b 5 0:038 and b 5 1/2.
can be expressed as the product of two different functions: cðNsp Þ, In the model, the damage d does not stabilize asymptotically
including the effect of Nsp and tp , the effect of t  on damage
b
with t  → ‘.
progression.
Once function g is available; taking into account Eq. (10), we
can obtain the cumulated damage D(t) caused by the action Example 6 (Melby and Kobayashi Damage Curve)
of t waves of intensity Nsp when starting from an initial damage Choosing g 5 0, we obtain the model [Eq. (2)], that is, the second
Dp0 , as Melby and Kobayashi damage curve, which shows that it is a par-
h ticular case of Eq. (33). More precisely, Melby and Kobayashi
1=b   ib
DpðtpÞ ¼ Dp0 2 g þ a Nsp t p þ g5½Dpðt pÞ 2 g1=b assume the power function in Eq. (30).
 1=b  
¼ Dp0 2 g þ a Nsp tp ð33Þ
Damage Caused by Random Waves
In this case, Eqs. (26) and (29) become Until now, the variability of the cumulated damage because of the
8 tðp
9b variability of the initial damage has been considered. However, it
<   = is more important to consider the effects on the stochastic process
DpðtpÞ ¼ ðDp0 2 gÞ1=b þ a Nsp ðtÞ dt D(t) caused by random waves Nsp ðt pÞ. To this end, the normal
: ;
0 approximation is crucial.
tðp
 
þ g5½DpðtpÞ 2 g1=b ¼ ðDp0 2 gÞ1=b þ a Ns ðt  Þ dt Normal Approximation
0 If the period of action of waves is long enough, and this happens
ð34Þ practically in all cases of breakwater design, based on the central

Fig. 2. Damage progression curves (d  versus t ) for four different initial damage values

458 / JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012
 p
limit
Ð tp theorem, we can assume that theÐ random Ð variable Uðt Þ 5 In addition, from Eq. (37), we have
a½N p
ðtÞ dt for large t  is normal Nð t kdt; t rdtÞ 5 Nðktp; rtpÞ,   b
DpðtpÞ ∼ N m0 þ ktp þ g; s20 þ rtp
0 s 0 0
where k and r are constants. ð42Þ
The central limit theorem states that under very general conditions,
if we sum (integrate) many random variables, the sum is approxi- and then, we can approximate the variance of D as
mately normal, no matter what distribution the summands have. Thus, h i2  
we need not to be concerned about the statistical distribution of var½Dpðt pÞ  bðm0 þ kt p þ gÞb 2 1 s20 þ rt p
summands, because their sum will be approximately normal.
D(t ) is the damage associated with time t , and Dp0 5 Dpð0Þ is s20 þ rtp
¼ b2 ð43Þ
the initial damage (t 5 0). Then, if we assume that both [D  ðtÞ 2 ðm0 þ ktpÞ222b
g]1/b and ðDp0 2gÞ1=b belong to a given location-scale family F ,
because U(t ) has been assumed normal with a deep justification, for which for t  → ‘ gives
the compatibility of Eq. (34), one possibility for this family F is the 8
normal family. Consequently, because of compatibility conditions, >
> b , 1=2
>
< 0r if
we can also assume that
lim var½DpðtpÞ ¼ if b ¼ 1=2 ð44Þ
 p 1=b   tp → ‘ >
> 4k
D0 2 g ∼ N m0 ; s20 ð36Þ >
:‘ if b . 1=2

and then which shows that for b 5 1/2 and t → ‘, the dispersion tends to
  a constant, and otherwise it tends either to zero or ‘. Because it is
½DpðtpÞ 2 g 1=b
∼ N m0 þ kt p; s20 þ rtp ð37Þ difficult to believe that the variance tends to zero, assuming b , 1/2
can be on the unsafe side. Thus, the use of this assumption must be
that is well justified by experiments if not rejected as being physically invalid.
FDpðtpÞ ðxÞ ¼ Pr½DpðtpÞ # x
n 1=b o Practical Application of the Proposed Method
¼ Pr Dps ðtpÞ 2 g # ðx 2 gÞ1=b
2 3 In this section, how the proposed method can be applied in practice
1=b
6ðx 2 gÞ 2 m0 2 kt 7 p is summarized. Because the proposed method is contained in the
¼ F4 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 5 ð38Þ CDF [Eq. (38)], application to a practical case requires estimation
s20 þ rtp of the following parameters: mp0 and sp0 (initial damage parameters),
g and b (breakwater typology and characteristics), and k  and
where F(x) 5 CDF of the normal N(0,1) random variable. Values of r  (future wave action). To this end we can proceed as follows.
b , 1 imply negative skewness, and values of b . 1 imply positive
skewness. Melby and Kobayashi (1999) obtain both cases of Step 1: Estimate Parameters mp0 and sp0
skewness, which will be subsequently described.
The interesting result is that the CDF of the cumulated damage This can be done based on previous experience on breakwaters
D(t) is given by Eq. (38), and that this result, generally valid, is noting that mp0 and sp0 are the mean and standard deviation of the
a consequence only of the central limit theorem and a compatibility initial damage. Normally the contribution of the initial damage to
condition. In other words, Eq. (38) is not arbitrarily chosen, but final damage is small compared with that of wave action.
a consequence of these two important properties that must be satisfied.
Once we know the CDF of D(t), the median of D(t ) can be Step 2: Estimate Parameters g and b
calculated
Because these parameters are breakwater dependent, they can be
median½DpðtpÞ ¼ ðm0 þ ktpÞb þ g ð39Þ estimated based on laboratory tests, similar to those of Melby and
Kobayashi (1999), or provided by experienced engineers based on
and the interquartile range similar tests.

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffib
Q0:75 2 Q0:25 ¼ m0 þ kt þ 0:675 s20 þ rt p
p
Step 3: Choose a aðNsp Þ Function

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffib For example, aðNsp Þ 5 as Nspb and fit as and b based on the experi-
2 m0 þ ktp 2 0:675 s20 þ rt p ð40Þ ments suggested in Step 2. In any case, it is convenient to analyze
the sensitivity of the predicted damages to these parameters and
adopt a conservative behavior.
which for t  → ‘ gives
8
>
> b , 1=2 Step 4: Estimate Parameters k* and r*
>
>
0
pffiffi
if
<
0:675 r
lim Q0:75 2 Q0:25 ¼ pffiffiffi if b ¼ 1=2 ð41Þ Based on the definition of these parameters, we can use past data of
tp → ‘ >
> k p
tdata years and use the estimates
>
>
: ‘ if b . 1=2 p
Ðt data  p 
a N0 ðtpÞ dt p
showing that for b 5 1/2 and t  → ‘, the variance tends to a constant, ^k p ¼ 0
p ð45Þ
otherwise it tends either to zero or ‘. tdata

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 459
 ( )2 h  1=b ib
p
Ð p1=b
tP iþ1
Dpðt pÞ ¼ D0 þ as Nspb tp ð47Þ
a½Ns ðtpÞdt p 2 k p
data

i¼0 i
^rp ¼ p ð46Þ
tdata to the experimental data of Series A9. The parameter estimation has
been completed using least squares (LS), that is, minimizing the sum
Step 5: Use Eq. (38) of squares between the observed and the predicted values. The
This allows us to obtain the percentile curves, in particular the following cases have been analyzed:
median curve, and to evaluate the evolution of the breakwater failure
probability with time. Case 1
First, Eq. (1) of Melby and Kobayashi has been considered, which
Example of Application led to the estimates indicated in Table 2, where LS 5 20.91. The
fitted model is shown in Fig. 3(a), where the fit is not very good in
In this section and only as a particular example of the model, the the central part.
proposed methods and concepts are illustrated using some experi-
ments reported by Melby and Kobayashi (1998, 1999). These
authors conducted three series of tests (A9, B9, and C9) on rubble
mound structures at the U.S. Army Engineers Waterways Ex-
periment Station. The sequence and duration of waves were dif-
ferent for the three series, and several layer profiles were measured;
therefore, the evolution of the mean and standard deviation of
damage with the number of waves was determined. The damage
used refers to D  [ S, that is, to the eroded area.
In the example, Eq. (33) is applied with cðNsp Þ 5 as Nspb and g 5 0,
that is,

Table 2. Parameter Estimates and Goodness of Fit (Least-Squares Errors)

for Different Cases
Series A9

Least
Case Model b b as Dp0 squares
Melby and Eq. (1) 1/4 5.000a 0.02500 0.0000 20.91
Kobayashi
Melby and Eq. (47) 1/4a 5.000a 0.01100 0.0000a 17.02
Kobayashi
A3 Eq. (47) 0.544 2.809 0.00316 1.8080 9.990
A4 Eq. (47) 1/2a 3.000a 0.00437 1.7400 10.070
A5 Eq. (47) 0.264 5.000a 0.00949 1.8760 13.100
A6 Eq. (47) 1/2a 3.110 0.00398 1.7990 10.030
A7 Eq. (47) 0.520 3.000a 0.00352 1.8160 10.000
A8 Eq. (47) 1/4a 3.000a 0.05908 0.6100 39.030
Series B9

Least
Case Model b b as Dp0 squares
B3 Eq. (47) 0.269 3.485 0.03280 1.1760 1.280
B4 Eq. (47) 1/2a 3.000a 0.00550 1.6230 1.630
B5 Eq. (47) 0.4878 5.000a 0.00120 2.3360 3.620
B6 Eq. (47) 1/2a 2.424 0.00870 1.3430 1.520
B7 Eq. (47) 0.347 3.000a 0.02370 1.1240 1.350
B8 Eq. (47) 1/4a 3.000a 0.05736 0.0185 1.770
Series C9

Least
Case Model b b as Dp0 squares
C3 Eq. (47) 0.321 4.402 0.00909 1.51600 0.560
C4 Eq. (47) 1/2a 3.000a 0.00583 1.34000 2.330
C5 Eq. (47) 0.301 5.000a 0.00638 1.64000 0.715
C6 Eq. (47) 1/2a 4.381 0.00179 1.77600 1.850
Fig. 3. Proposed models fitted to the Melby and Kobayashi data (dots in
C7 Eq. (47) 0.379 3.000a 0.01810 0.89200 1.330
the figures) for Series A9; mean and the mean 6 1 SD curves corre-
C8 Eq. (47) 1/4a 3.000a 0.05767 0.02140 3.060
a
sponding to Cases 1, 2, and 3 [from (a) to (b)]
Fixed values.

460 / JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012
Case 2 Cases A3 and A4 are very similar in quality from the point of
view of LS errors and better than the Melby and Kobayashi models.
Eqs. (2) or (47) of Melby and Kobayashi have been considered,
This can be seen by comparing the LS values (17.02 and 9.99 or
and the parameters are shown in Table 2. The associated LS 5
10.07, respectively), or by comparing Figs. 3(a and c), but Case A4
17.02, and the fitted model is shown in Fig. 3(b). Fig. 3(b) shows
has the advantage of an integer value for the Nsp exponent, which was
that the damage progression law using Nsp5 is too conservative.
3, and not 5, as Melby and Kobayashi suggest. This means that the
In fact, the curvatures in the figure suggest an exponent smaller
value of the exponent requires further investigation if it is to be
than 5.
extrapolated outside the ranges of the experimental values.
Using the normal approximation, the densities of cumulated
Case 3 damage for t 5 10,000e50,000 waves have been calculated, which
are plotted in Fig. 4, which shows the skewness of these densities.
Eq. (47) has been considered and the parameters b, b, as , and Dp0
This information is more complete than simple means and variances
without any constraints by LS have been estimated, obtaining the
as a function of time.
parameter estimates in Table 2 with LS 5 9.99, which is a substantial
The same methods have been analyzed for Series B9 and C9, and
improvement with respect to Case 1, as seen in Fig. 3(c).
similar results have been obtained (Figs. 5 and 6).

Case 4
Eq. (47) was used and the parameters b, b, as , and Dp0 were estimated Conclusions
by LS, subject to b 5 1/2 and b 5 3, obtaining the parameter
estimates in Table 2 with LS 5 10.07, which also is a substantial The main conclusions are the following:
improvement with respect to Case 1. The motivation to fix b 5 1/2 1. Selection of a damage accumulation method for breakwaters
was that the observation of the experimental results of Melby and must be done considering dimensional analysis and some
Kobayashi suggest a possible constant variance for large time t , consistency rules to get simple models and to avoid dimen-
which implies b 5 1/2. sional and consistency problems.
2. The damage model can be defined by supplying four main
elements: (1) a damage progression law (the g function)
Case 5
corresponding to fixed breakwater slope a, still water depth
Eq. (47) was used and parameters b, b, as , and Dp0 were estimated by at the toe of the structure ht , and armor specific weight g a ; (2)
LS, subject to b 5 5, obtaining the parameter estimates in Table 2 the distribution function of the initial damage of the break-
with LS 5 13.10, which is worse than Cases 2 and 3. The motivation water; (3) the wave history Nsp ðtpÞ ; and (4) the function, taking
to fix b 5 5 was to be consistent with the Melby and Kobayashi into account the effect of wave intensity on damage.
exponent for Nsp . 3. The proposed model reproduces not only the damage accu-
mulation but its statistical distribution, that is, contrary to
other existing models that are deterministic, the proposed
Cases 6--8
model is a stochastic model. The densities of the cumulative
Eq. (47) was used and the parameters b, b, as , and Dp0 were estimated damage are shown to be related to the normal distribution
by LS, subject to b 5 1/2, b 5 3, or both, obtaining the parameter because of the central limit theorem, and closed formulas for
estimates in Table 2 with LS 5 10.03, LS 5 10.00, and LS 5 39.03, them are given.
respectively. 4. The proposed model for cumulative damage can also be
The corresponding results for Series B and C are given in applied to evaluate the damage produced by known existing
Table 2. storms, that is, to assess the actual state of a breakwater after

Fig. 4. Proposed model [Eq. (47)] fitted by least squares to the Melby and Kobayashi data (dots in the figures) for Series A9; mean and the mean 6 1
standard deviation curves resulting from the experimental data and associated densities

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 461
 Fig. 5. Proposed model fitted to the Melby and Kobayashi data (dots in the figures) for Series B9; mean and the mean 6 1 SD curves

Fig. 6. Proposed model fitted to the Melby and Kobayashi data (dots in the figures) for Series C9; mean and the mean 6 1 SD curves

suffering severe storms. This can be done using Eqs. (25) or 8. The proposed models [Eqs. (19) and (22)] extend the Melby
(26), depending on the available wave data (sequence of and Kobayashi model [Eq. (2)] and guarantee the satisfaction
individual waves or continuous wave data). of the time origin change invariance, which is crucial to avoid
5. Based on the central limit theorem, a closed formula, not only model inconsistencies, such as those previously discussed. In
for the mean of the cumulated damage in terms of time, but for addition, Eq. (38) provides a formula for the time evolution of
the dispersion, has been given. This formula indicates that the cumulative distribution function of damage written in
there was a threshold value of the exponent b, which leads to terms of the normal distribution, which is of general applica-
constant variance with time. Above this value, the variance bility because it is based on the central limit theorem.
increases, and below it decreases. 9. From a practical point of view, the proposed model [Eq. (22)]
6. The model fits very well the experimental data provided by can be applied in the same way as the Melby and Kobayashi
Melby and Kobayashi (1998, 1999), not only in the mean model [Eq. (2)], with the only difference being that the
trends but in the variances too. parameter g needs to be estimated.
7. The Melby and Kobayashi model [Eq. (2)] satisfies the
compatibility condition [Eq. (9)], and uses the power law
scale family for damage accumulation, which is what makes Acknowledgments
it a valid model. In addition, this model can be comple-
mented by using the CDF for damage accumulation given in We are indebted to the Spanish Ministry of Science and Technology
Eq. (38). (Projects BIA2009-10483 and TRA2010-17818) for partial support.

462 / JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012
References Iribarren, R. (1938). “Una fórmula para el cálculo de los diques en escollera
[A formula for the calculation of rock-fill dikes].” Technical Rep.
HE-116-295, Revista de Obras Públicas, Madrid, Spain (in Spanish).
Aczel, J. (1966). Lectures in functional equations and their applications,
Jensen, O. J. (1984). A monograph on rubble-mound breakwaters, Danish
Vol. 19, Academic Press, New York.
Hydraulic Institute, Hørsholm, Denmark.
Ahrens, J. P. (1975). “Large wave tank tests of riprap stability.” Technical
Jensen, T., Andersen, H., Grønbeck, J., Mansard, E. P. D., and Davies, M. H.
Memorandum 51, U.S. Army Engineer Waterways Experiment Station,
(1996). “Breakwater stability under regular and irregular wave attack.”
Coastal Engineering Research Center, Vicksburg, MS.
Ahrens, J. P., and McCartney, B. L. (1975). “Wave period effect on the Proc., 25th Coast. Eng. Conf., Vol. 2, ASCE, Reston, VA, 1679e1692.
Losada, M. A., and Giménez-Curto, L. (1982). “Mound breakwaters under
stability of riprap.” Proc., Civ. Eng. Oceans III, ASCE, Reston, VA,
1019e1034. oblique wave attack.” Coast. Eng., 6(1), 83e92.
Bazant, Z. P. (2006). “Mechanics-based statistics of failure risk of quasi- Medina, J. M., and McDougal, W. G. (1990). “Discussion on “Deterministic
brittle structures and size effect on safety factors.” Proc. Natl. Acad. Sci. and probabilistic design of breakwater armor layers”. J. Waterw. Port
USA, 103(25), 9434e9439. Coast. Ocean Eng., 116(4), 508e510.
Bazant, Z. P., and Pang, S. D. (2007). “Activation energy based extreme Medina, J. R. (1996). “Wave climate simulation and breakwater stability.”
value statistics and size effect in brittle and quasibrittle fracture.” Proc., 25th Coast. Eng. Conf., ASCE, Reston, VA, 1789e1802.
J. Mech. Phys. Solids, 55(1), 91e131. Medina, J. R., Garrido, J., Gómez-Martín, M. E., and Vidal, C. (2003).
Burcharth, H. F. (1984). “Fatigue in breakwater armour units.” Proc., 19th “Armor damage analysis using neural networks.” Proc., Coast. Struct.,
Int. Conf. Coastal Eng., Vol. 3, ASCE, Reston, VA, 2592e2607. ASCE, Reston, VA, 236e248.
Burcharth, H. F. (1994). “Reliability evaluation of a structure at sea.” Proc., Melby, A., and Kobayashi, N. (1999). “Damage progression and variability
Int. Workshop Wave Barr. Deep Water, Port and Harbour Research on breakwater trunks.” Coastal structures, 1999, Balkema, Rotterdam,
Institute, Yokosuka, Japan. Netherlands, 309e316.
Burcharth, H. F. (1997). “Reliability based design of coastal structures.” Melby, J. A. (1999). “Damage progression on rubbleemound breakwaters.”
Advances in coastal and ocean engineering, Vol. 3, WS Publishing, Technical Rep. CHLe99e17, U.S. Army Corps of Engineers, Water-
Singapore, 145e214. ways Experiment Station, Vicksburg, MS.
Burcharth, H. F. (2000). “Reliability based design of coastal structures.” Melby, J. A., and Kobayashi, N. (1998). “Progression and variability of
Chapter 6, Coastal Engineering Manual Part VI (CEM), Coastal En- damage on rubblemound breakwaters.” J. Waterw. Port Coast. Ocean
gineering Research Center, Vicksburg, MS, VI.6.1eVI.6.47. Eng., 124(6), 286e294.
Burcharth, H. F., and Sorensen, J. D. (1998). “Design of vertical wall caisson Mínguez, R., Castillo, E., Castillo, C., and Losada, M. (2006). “Optimal cost
breakwaters using partial safety factors.” Proc., 25th Int. Coastal Eng., design with sensitivity analysis using decomposition techniques.
Vol. 2, ASCE, Reston, VA, 2138e2151. application to composite breakwaters.” Struct. Saf., 28(4), 321e340.
Carver, R. D., and Wright, B. L. (1991). “An investigation of random Pfeiffer, H. L. (1991). “Comparison of four rubble-mound stability equa-
variations in the stability response of stone armored, rubble mound tions with prototype data from burns harbor breakwater.” M.S. thesis,
breakwaters.” Technical Rep. CERC-91-17, U.S. Army Engineer Civil Engineering Dept., Texas A&M Univ., College Station, TX.
Waterways Experiment Station, Vicksburg, MS. Pilarczyk, K. W., and Den Boer, K. (1983). “Stability and profile de-
Castillo, C., Mínguez, R., Castillo, E., and Losada, M. (2006). “An optimal velopment of coarse materials and their application in coastal engi-
engineering design method with failure rate constraints and sensitivity neering.” Proc., Int. Conf. Coast. Port Eng. Devel. Countr., World
analysis. Example application to composite breakwaters.” Coast. Eng., Association for Waterborne Transport Infrastructure (PIANC). Brussels,
53(1), 1e25. Belgium, 43e61.
Castillo, E., Fernández-Canteli, A., and Hadi, A. S. (1999). “On fitting Scotto, M. G., and Guedes-Soares, C. (2007). “Bayesian inference for
a fatigue model to data.” Int. J. Fatigue, 21(1), 97e106. long-term prediction of significant wave height.” Coast. Eng., 54(5),
Castillo, E., Fernández-Canteli, A., Pinto, H., and Ruiz-Ripoli, M. L. 393e340.
(2008a). “A statistical model for crack growth based on tension and SPM84 (1984). Shore protection manual. U.S. Army Engineer Waterways
compression Wöhler fields.” Eng. Fract. Mech., 75(15), 4439e4449. Experiment Station, U.S. Government Printing Office, Washington, DC.
Castillo, E., Fernández-Canteli, A., and Ruiz-Ripoli, M. L. (2008b). “A Thompson, D. M., and Shuttler, R. M. (1975). “Riprap design for wind wave
general model for fatigue damage due to any stress history.” Int.
attack. A laboratory study in random waves.” Rep. EX707, H. R.
J. Fatigue, 30(1), 150e164.
Wallingford, Wallingford, U.K.
Castillo, E., and Hadi, A. S. (1995). “Modeling lifetime data with application
Torum, A., Mathiesen, B., and Escutia, R. (1979). “Reliability of break-
to fatigue models.” J. Am. Stat. Assoc., 90(431), 1041e1054.
water model tests.” Proc., Coast. Struct., ASCE, Reston, VA, 454e469.
Castillo, E., Iglesias, A., and Ruiz-Cobo, R. (2004a). Functional equations
Van der Meer, J. W. (1988). “Rock slopes and gravel beaches under wave
in applied sciences. Mathematics in science and engineering, Elsevier,
attack.” Ph.D. dissertation, Delft Hydraulics Laboratory, Emmeloord,
New York.
Netherlands.
Castillo, E., Losada, M., Mínguez, R., Castillo, C., and Baquerizo, A.
Vidal, C., Losada, M. A., and Mansard, E. P. D. (1995). “Suitable wave-
(2004b). “Optimal engineering design method that combines safety
factors and failure probabilities: Application to rubble-mound break- height parameter for characterizing breakwater stability.” J. Waterw.
waters.” J. Waterw. Port Coast. Ocean Eng., 130(2), 77e88. Port Coast. Ocean Eng., 121(2), 88e97.
Castillo, E., and Ruiz Cobo, M. R.(1992). Functional equations and Vidal, C., Losada, M. A., and Medina, R. (1991). “Stability of mound
modeling in science and engineering, Vol. 161, Marcel Dekker, New breakwater’s head and trunk.” J. Waterw. Port Coast. Ocean Eng.,
York. 117(6), 570e587.
Davies, M. H., Mansard, E. P. D., and Cornett, A. M. (1994). “Damage Vidal, C., Martín, F. L., and Migoya, L. (2003). “Parámetros de oleaje para
analysis for rubble mound breakwaters.” Proc., 24th Coast. Eng. Conf., la descripción de la evolución del daño en diques rompeolas.” Actas de
Vol. 1, ASCE, Reston, VA, 1001e1015. las VII Jornadas Españolas de Costas y Puertos, Balkema, Rotterdam,
Font, J. B. (1968). “The effect of storm duration on rubble mound break- Netherlands, 1e10 (in Spanish).
water stability.” Proc., 11th Coast. Eng. Conf., Vol. 1, ASCE, Reston, Vidal, C., Medina, R., and Lomónaco, P. (2006). “Wave height parameter
VA, 779e786. for damage description of rubble-mound breakwaters.” Coast. Eng.,
Hudson, R. Y. (1958). “Design of quarry-stone cover layers for rubble- 53(9), 711e722.
mound breakwaters.” Research Rep. 2-2, U.S. Army Engineer Water- Vidal, C., Medina, R., Martín, F. L., and Migoya, L. (2004). “Wave height
ways Experiment Station, Vicksburg, MS. and period parameters for damage description on rubble-mound
Hudson, R. Y. (1959). “Laboratory investigation of rubble-mound break- breakwaters.” Proc., 29th Int. Conf. Coast. Eng., Jane McKee Smith
waters.” J. Wtrwy. Harb. Div., 85(WW3), 93e121. (ed.), Vol. 4, World Scientific, Hackensack, NJ, 3688e3700.

JOURNAL OF WATERWAY, PORT, COASTAL, AND OCEAN ENGINEERING © ASCE / NOVEMBER/DECEMBER 2012 / 463
Copyright of Journal of Waterway, Port, Coastal & Ocean Engineering is the property of American Society of
Civil Engineers and its content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email articles for
individual use.