Академический Документы
Профессиональный Документы
Культура Документы
231
ABSTRACT
This paper reviews and summarizes the development and recent progress of methods of
stochastic fatigue, fracture and damage analysis. Topics covered include structural fatigue,
structural fracture, cumulative damage, maintainability and inspection and structural
damage. Several "'new" methods such as expert
4.1 INTRODUCTION
232
ticular structure. During this process, a generalized mathematical model is used to represent the structural behavior (see report of
Chapter 1). The expected loading and environmental conditions are also represented
mathematically for the computation of structural response. In the case of random and
dynamic load, methods of stochastic structural dynamics are used (see report of Chapter
2). Once the design is implemented and the
structure is constructed, the structural behavior may differ from the analytical results using
the apriori and idealized mathematical model.
In such cases, dynamical measurements can
be made for updating the mathematical model
through the use of system identification techniques (see report of Chapter 5). In the case
of damaged structures, results of inspection
Time
Field Work
J Field
Data
[ Construction
Inspection
1 Workt
Office
Conception
Planning
Field
Desig
._n
Analvsis
Expected
r-
Test Data
I Normal
Repair Maintenance
~l
Demolition
Struciural Identification
r---
I
I
I
k.
"1
~Oamage J.~System
J
ssessment ~
IdentificotionJ
_J
Decision
Analysis
I Math.Nodel
J
Environ,Cond. l for Structure I Response._
L ......
-I
Updated r- - - - ~ ---1
Expeted ~Jpdaled Math.II
Safety Evaluation
D(t), R(t). L(t)
D Damage
R Residual Strength
L Reliability
I Hodel for
Env.Cond, .LSlructure
jII Respons..ee
Econon~c
Con~deration
233
Even with modern science and improved
knowledge, the fatigue phenomenon is not
completely understood. Frequently, it is still
necessary to rely on experimental data in the
case of new materials, new geometrical configuration, and new loading and environmental conditions. Furthermore, relatively large
scatter in test results is expected even for
"identical" laboratory specimens. Consequently, it is necessary and desirable to apply
statistical methods in fatigue studies. Meanwhile, there are cases where it is difficult to
obtain sufficient statistical data because of
the unique design, large size a n d / o r high cost
of the structure(s) involved (e.g., most civil
engineering structures).
During recent years, much progress has
been made in the application of fracture
mechanics to maintainability, lifecycle-cost
analysis and retirement-for-cause life management of aircraft and aerospace structures.
In this report, a comprehensive review of the
available literature is presented.
The failure paths and limit states of large
and complex structural systems are not well
understood, though many methods are now
available for the computation of system reliability (see report of Chapter 3). In recent
years, more experimental data have been obtained on the failure of large and full-scale
structures. However, there are sill discrepancies between mathematical modeling and actual behavior of complex structures (e.g., see
Okamoto et al., 1982). Therefore, there is a
need for further research to establish more
realistic limit states including those for fatigue and fracture damage of structural systems.
The objective of this report is to review and
summarize available stochastic methods of fatigue, fracture, and damage analysis. In addition, several "new" methods such as expert
systems and fuzzy sets and their applications
to damage analysis are briefly introduced and
discussed.
234
fatigue analysis process including the following items:
(a) The shape and size of detectable cracks
may not be clearly defined. Therefore, the
demarcation between crack initiation and
crack propagation may not be established in a
precise manner.
(b) Fatigue test data are subject to enormous statistical scatter with typical coefficients of variation for fatigue lives in many
cases being 30% or higher.
(c) The relationships used to describe fatigue behavior are approximate.
(d) There are m a n y kinds of defects and
discontinuities in welded joints making it difficult to predict initiation and propagation of
fatigue cracks (e.g., see Bowman, Munse, and
Will, 1984).
(e) Environmental and loading processes
causing fatigue behavior may not be well
defined because of incomplete and insufficient records.
(f) The force produced by a given environment or load may not be precisely known.
(g) The state and magnitude of stresses
causing fatigue failure at a joint contain uncertainties resulting from the methods of stress
analysis which are based upon idealized
mathematical models.
(h) Effects of temperature, corrosion, etc.
are not precisely known in most cases. Nevertheless, engineers must make decisions regarding the design and integrity of components to mitigate fatigue failures. Consequently, it is necessary to apply probabilistic
and statistical methods in structural fatigue
studies.
b. Fatigue models
During the defect growth stage, fracture
mechanics can be used to obtain additional
information. The fracture mechanics analysis
should be used in lieu of or in addition to the
S - N type of analysis whereas most available
fatigue data to-date are obtained from con-
= C
(4.2.1)
(4.2.2)
235
the plastic strain range dominates the expression in the high-strain lowcycle region. It was
proposed independently by Coffin (1954) and
Manson (1954). Results of extensive research
during the past two decades show that controlling strain is more meaningful than controlling stress where low-cycle fatigue (say
N < 10 4 cycles) is concerned (Langer 1962;
Yao and Munse 1962).
coalescence, etc., prior to the onset of cycleby-cycle growth. The fracture mechanics approach to fatigue are presented by Rolfe and
Barsom (1977), Hertzberg (1976), and Fuchs
and Stephens (1980).
The basic parameter of fracture analysis is
the stress intensity factor, k, given by the
following relationship.
in which s = a p p l i e d stress, Y ( a ) = f i n i t e
geometry correction factor which may depend
on a, and a = crack depth for a surface flaw
or half-width for a penetration flaw. The stress
intensity factor depends on the crack size,
geometrical configuration, and applied
stresses.
Based on extensive experimental data, the
central region of crack growth rate d a / d n
and the stress intensity factor range AK may
be represented with the following equation
(Paris, 1964):
S = B N -1/2 -4- S e
(4.2.3)
(4.2.4)
K=
Y ( a ) sv/~d
da/dn
= C(AK) m
(4.2.5)
(4.2.6)
(4.2.7)
Good agreement was obtained from experimental results when compared. However, the
single parameter, ASrms, which is assumed to
describe the whole influence of the randomness of the load on the crack growth cannot
236
yield sufficient information about a stochastic
process. It is obvious that identical values of
ASrnl~ can be found for quite different characteristics of stochastic processes. It is difficult to expect that a structure would fail at
the same time when it is subjected to a narrow-band process as when it is subjected to a
wide-band process, even if both processes have
identical ASrms values. The application of this
simple approach requires careful experimental verification using the actual load spectrum. An appropriate choice of AKeq for
some given types of loading and for some
typical structural components can, however,
yield satisfactory assessments of the mean
life-time.
A convenient approximate form for cycles
to failure N can be derived by integrating
eqn. (4.2.6); from an initial crack length a 0 to
a critical crack length a c, and n from 0 to N.
Assuming that a c > > a 0, m > 2 and Y is a
constant, the result of the integration of eqn.
(4.2.6) is
N S m= 1/[a'~/2
'(m/2-
1)CY'~r m/2]
(4.2.8)
da
dn
C(AK)"'
Kc- AK
(1 - R ) K c - A K
for R < 0
for R > 0
(4.2.9)
The time to crack initiation involves a significant statistical variability. It has been considered as a random variable following either
the lognormal or the Weibull distribution (e.g.,
Yang and Chen (1983a, 1983b) Freudenthal
(1965, 1967, 1975), Freudenthal and SchuEller
(1973), Butler (1972), Shinozuka (1968, 1976,
1978), Whittaker and Saunders (1973a,
1973b), Yang (1978), Yang and Trapp (1974)).
The shape parameter ~ for the Weibull
distribution or the standard deviation o~nT for
the lognormal distribution is exclusively related to the coefficient of variation (dispersion) of the tTCI data, and it has been observed to be fairly constant. Thus, o~ and o~n r
are determined from the results of coupon
specimens under laboratory conditions for fatigue-critical components. The scale parameter fl for the Weibull distribution or the
median value of the lognormal distribution is
estimated from a few full scale test articles.
Attempt was also made to account for the
statistical variability of service loads in the
TTCI distribution. Another approach for the
prediction of TTCI is the application of Coff i n - M a n s o n equation in conjunction with the
Neuber's equation. The statistical method for
such an approach is to consider those parameters appearing in eqn. (4.2.2) as random variables as described by Wirsching (1981).
237
airframes, may consist of thousands of
fastener holes in which the crack growth
damage accumulation in each hole has to be
estimated statistically [Yang, 1976]. Since the
applied stress level varies from one location
to another, it is economically impractical to
conduct laboratory tests to obtain the TTCI
distribution at each stress level and possible
load spectra. As a result, an alternate approach, referred to as the method of equivalent initial flaw size, has been suggested and
investigated (e.g., Rudd and Gray (1977),
Yang and Manning (1980), Shinozuka (1979),
Yang (1980), Rudd et al. (1982), Rudd et al.
(1982a, 1982b), Yang et al. (1985), Manning
et al. (1986)].
An equivalent initial flaw size (EIFS) is a
hypothetical crack size assumed to exist in a
structural detail prior to service. The equivalent initial flaw size is obtained by back-extrapolation of observable fractographic data
to time zero for each test specimen using a
suitable crack growth law in the small crack
size region. Such an EIFS would result in an
actual crack size at the actual point in time
when grown forward in service. Once the
distribution function of the equivalent initial
flaw size is established from one set of test
results under one maximum stress level of a
given loading spectrum, it can be grown forward under different stress levels and service
conditions. In other words, the fatigue process is represented by the crack propagation
from EIFS to the critical crack size. Another
advantage of such an approach is that it
allows for a comparison of the initial fatigue
quality for different materials, and manufacturing and assembling processes.
Combination of time to crack initiation and
initial flaw size
238
distribution, but also it is questionable to
what degree the rouge flaws are reproducible
in a distribution form.
239
istence of faults in a specimen. These imperfections may be represented by flaws of idealized (such as elliptical) shapes. Furthermore,
the fracture mechanical concept allows the
calculation of stresses or stress intensities
around a crack tip and the derivation of
fracture criteria for linear elastic or nonlinear
elastic-plastic behavior. Linear Elastic Fracture Mechanics (LEFM) is applicable when
the plastic region around the crack tip is
small in comparison with the crack length.
This is generally the case if the fracture occurs at stresses which are considerably lower
than the yield stress and at plane strain conditions. The application of LEFM is limited
to brittle materials. The stress intensity factor
K, as defined by eqn. (4.2.5), is the governing
parameter for crack growth and fracture. It is
well known that the latter occurs if K exceeds
a critical threshold level Kc, which is called
the fracture toughness. This material parameter is to be determined experimentally for
plane stress or plane strain conditions respectively. Its statistical properties will be discussed later.
Beyond the applicability of LEFM, i.e. for
cases where the size of the plastic region
around the crack tip becomes larger with
respect to the crack length, various methods
of the Elasto-Plastic Fracture Mechanics
(EPFM) are used, e.g. the Crack Opening
Displacement (COD) Method or the J-Integral Method, etc. These concepts are based
on a strain dependent crack opening 8 and by
an energy release defined by a closed contour
(J)-Integral. Again, fracture is assumed to occur if these parameters reach critical values,
i.e. 8 = 6c and J = J~ respectively. Although
these methods are already quite widely in use,
there is still insufficient statistical information on 8~ and J~ available. It is for this and
other practical reasons that for these types of
problems empirical elasto-plastic concepts are
used, e.g. a scheme suggested by Feddersen
(1970) and the so-called Two Criteria Approach (see Harrison et al. (1977)) are utilized.
-.
4.5 ~,,} Z
~3
e~ : e y ( 1 - ~-1
4, oy )2a]
'
il
Oc= Ke/nV/-'~'~
III
oc =
2 Kc
(4.3.1)
~(t)=c(a(t))
oc(ao)
(4.3.2)
240
The two-criteria approach allows also the
consideration of brittle-, plastic collapse- and
mixed-mode failure (see Fig. 4.3.2). The statistical properties of the parameters involved,
i.e. K~., a and Oy (or Or,ij ) cause a non-deterministic failure region as indicated in the
figure. There is a direct relation between these
two methods by simply expressing K r in terms
of f ( o ) a n d o = g ( a ) respectively (e.g., see
Oswald (1983)).
Repeated loads which are lower than those
critical for instable crack extension (fracture)
can cause a crack propagation. As shown in
section 4.2, LEFM, i.e. the stress intensity
factor K proves to be applicable for this case.
This holds also for the COD- and J-Integral
concept respectively for the inelastic region.
Based on the plastic strain A~p, the following
empirical relationship has been suggested by
Solomon (1972):
da/dn
Caa( AEp ) q
(4,3.3)
w.here C a and C 2 are parameters which depend on material constants and the geometry
at the location where the strain is measured.
Equation (4.3.3) shows a formal similarity
with the formulation in the linear case, as
given in eqn. (4.2.6) and also with the strain
dependent low-cycle fatigue formulation of
Coffin-Manson. A detailed discussion concerning the fracture and fatigue formulation
in the linear and nonlinear range is presented
by Oswald and Schu611er (1983). This shows
clearly that the use of the initial crack length
as an input parameter is an advantage of the
fracture concept.
Furthermore, for structures under environmental loading conditions the sequence effect
of the load cycles does not cause acceleration
or deceleration of crack growth in general.
However, it may be important for aircraft
structures, where periodic loading (during
starting and landing) are followed by infrequently occurring overloads such as gust loads
(e.g. Wheeler (1970), Willenborg et al. (1971)).
Theoretical models to consider corrosive el-
KZ
Kr=
/
REGION OF
NO FAILURE
/
/
//
//
REGION OF
FAILURE
//
/.-/"
Sr =
O'failur e
Burdekin/Stone:
Sr
Kr =
Kr =
1 +S r
'n( 1---y~r )
Folias:
Kr = ~ / 1 - S ~
Dugdale:
Kr = ~ c o s ( ~ ' "
Sr)
fects on crack propagation are sill under investigation, particularly the influence of the
respective stress (intensity) levels. To-date,
experimental evidence is mainly used. For
example, Oberparleiter and Schlitz (1981) report that structures under random cyclic loading exposed to salt water environment are
expected to have a reduced fatigue life by an
overall factor of 1.6. Therefore, eqn. (4.2.6)
can be expressed as
da/dn = CT(AK)"
(4.3.4)
241
p ( n ) - (?~l)ne-X'
(4.3.7)
n!
c. Statistical properties of material parameters
f(a) = g(a)h(a)
(4.3.5)
Obviously, this equation must be renormalized. Since h(a) also reveals exponential
characteristics, eqn. (4.3.5) may be expressed
as follows:
f ( a ) = ce c(,-a.)
(4.3.6)
(4.3.8)
n=0
da/dn = F( a,Smax,Smin)
(4.3.9)
242
have been made for the function F(-), e.g.
eqn. (4.2.6). The differential form of the crack
propagation law, however, leads many authors
to apply the methods of stochastic differential
equations to the analysis of fatigue fracture.
In fact, eqn. (4.3.9) may be expressed in the
following form (e.g. Miller and Gallagher
(1981) and Hoeppner and Krupp (1974)):
d a / d t = F[a(t),AK,Kma,R,S ] "
X(t)=
a(to) = A 0
(4.3.10)
1 da
u+(t) dt
- -
(4.3.11)
x R,S)
(4.3.12)
N(t)
Y'. ZkW(t--rk)
(4.3.13)
where N(t) is a homogeneous Poisson counting process with an average rate ~, Z k the
random amplitude of the k th pulse, independent but identically distributed for different
k, the Markovian approximation yields analytical solutions for the statistical moments of
the random time until the crack reaches a
given size. The statistical parameters required
to define the process X(t) are obtained from
simple experiments. These results provide sufficient information to establish the necessary
parameters to solve more complicated problems.
In this context it should be mentioned that
extensive investigations have recently been
conducted for various derivatives of the
randomized crack growth rate equation given
by eqn. (4.3.12) recently (e.g. Lin and Yang
(1983, 1985), Yang and Donath (1984), Lin et
al. (1985), Yang and Chen (1985a, 1985b),
Palmberg et al. (1986), Yang et al (1986),
Yang et al. (1982), Yang and Chep (1985c)).
ONe advantage of such a model is that various statistical variabilities and uncertainties,
such as service loading spectra, crack geometry, stress intensity factor, etc., can be
accounted for in X(t) (Yang and Chen (1985a,
1985b, 1985c)).
The modeling of crack size as a Markov
process but discretely valued and discretely
parametered, namely, a Markov chain has
been proposed by Bogdanoff and Kozin (for
a detailed summary of the approach see
Bogdanoff and Kozin (1985a) and Section 4.4
of this report). The method describes the
damage accumulated during each duty cycle.
243
The transition probabilities of the various
crack states are estimated from experimental
data, i.e. a small number of samples of the
material and structural parts respectively for
which the crack sizes have to be calculated.
Based on the relation between the fatigue
concept (linear damage accumulation) and the
fracture mechanics approach, the damage
accumulation concept in context with the
Markov chain model has been modified and
introduced into the fracture mechanics concept by Oswald and SchuEller (1984). The
discretization of the crack length, Aa, is an
arbitrarily chosen fraction of the maximum
possible crack length, i.e. thickness of the
structural part under consideration. The duty
cycle (storm, earthquake, etc.) is defined as
part of a total load history. The transition
probabilities of a crack growing from one
state, i.e. crack interval length, to another
during a particular load event, i.e. duty cycle
may be defined by
<k-i
+aala(j-1)=iaa}
(4.3.14)
where Pi,k represents the transition probability from state i to state k, i.e. due to the load
event j and Zj the crack increment. It should
be pointed out that within this concept the
transition of a crack is allowed not only up to
the next but to any other state. Hence the
transition matrix is represented by an upper
triangular matrix. As the crack growth is a
random variable, eqn. (4.3.14) may be represented as follows
p(k--i+ l/2)Aa
Pk(J) = ~ Pi.kPi(J- 1)
i=1
(4.3.16)
Fl(a )
F2(Smax,Smin)dn
(4.3.17)
4,(a0,ac) = ~
F2(gs,max,Si,min)
(4.3.18)
i=l
Under suitable conditions on the load process, the central limit theorem may be applied
to the right side of eqn. (4.3.18) conditional
on a given (i.e. non-random) value of N(T).
Thus the righthand side is asymptotically
Gaussian for large N(T). Since q~(ao,ac) is a
constant, eqn. (4.3.18) defines a first passage
problem which determines the asymptotic distribution of N(T). In the case of homogeneous cyclic loading the random variable
N(T) is distributed as the inverse Gaussian
distribution asymptotically for large ~b(a0, a~)
(e.g. see Ditlevsen, 1986).
The extensive data set obtained by Virkler
et al. (1978) for fatigue crack growth in 2024-
244
T3 aluminum alloy panels under homogeneous cyclic stressing has recently been the
object of several extensive statistical analyses
(Kozin and Bogdanoff, 1985 c, Ortiz and
Kiremidjian, 1985, Ditlevsen and Olesen,
1985). The sample consists of 163 cycle increment values per test specimen.
The statistical scatter of the experimental
data is in the last two references modeled as
made up of a random "between" specimen
variation and a random "within" specimen
variation, the former being of finite dimensional random vector type and the latter of
random process type.
Ditlevsen and Olesen (1985) suggested a
modified form of the Paris-Erdogan-law, eqn.
(4.2.6)
da
[ AK)"
d--~ = C - - ~
(4.3.19)
245
e. Discussion
The methods presented here do not exhaust the subject of the stochastic approaches
to fracture and crack propagation. However,
they serve the purpose of indicating the main
trends concerning the random characteristics
of loading a n d / o r material properties of these
types of problems.
For stochastic analysis of non-linear, i.e.
elasto-plastic behavior, at present the application of the empirical Two-Criteria Approach
and the Elasto-Plastic Fracture Scheme are
preferred over the theoretical COD- and J-Integral concepts. This is mainly due to lack of
statistical information for the latter approaches.
The dominating role of the initial crack
distribution calls for the continuous need of a
stringent quality control and improved crack
detection methods. Moreover, additional statistical data on fracture toughness are needed.
Probabilistic models for crack propagation
have been developed along the following three
major lines,
(i) equivalent or characteristic loading to
substitute for stochastic amplitudes;
(ii) Markov model which describes the
crack propagation under stochastic loading
a n d / o r in presence of randomness of material
parameters;
(iii) Central limit theorem of probability
which utilizes the fact that the crack length
after n load cycles is a sum of random events.
The first approach and the Markov chain
model provide a good description of experimental results with necessary parameters determined from these results. The ability of
these approaches to predict crack propagation on the basis of experiments that were
made under other conditions still remains to
be investigated.
The Markov chain model may also be
utilized within the fracture concept. For this
purpose the parameter estimates as used in
the crack growth law are based on generic
a. General remarks
246
lifecycle costs. A mathematical model is
needed to describe and analyze data accurately, and predict behavior under conditions
not covered by such data. In addition, it is
desirable to have as few parameters as possible which require evaluation. The model
should also be easy to understand and use in
computations, and possibly be based upon
known physical laws.
The cumulative d a m a g e stems from
material behavior at the atomic or molecular
level. To-date, it appears that material behavior at this level needed in CD is not sufficiently well understood in engineering materials and under laboratory (let alone service)
conditions so that CD models can be based
upon fundamental physical laws. Further, in
view of the magnitude of the effort already
spent and the modest nature of the gains
made, it does not seem probable that much
progress will be achieved in this direction in
the foreseeable future. Therefore, current
models of CD must be phenomenological in
nature and based upon experimental data and
understanding of the phenomena at the macroscopic level. That is, models must be based
upon experimental data.
It is desirable to find the time to reach a
prescribed level of fatigue damage, crack
length, or material lost. From suitable data,
estimates of mean and variance of life time
and approximate confidence intervals can be
obtained using standard statistical methods.
If the coefficient of variation of the lifetime is
consistently very small and the minor fluctuations from the mean are of no consequence, a
deterministic model for mean behavior is appropriate. On the other hand, if the coefficient of variation is consistently large and the
fluctuations from mean behavior are significant, a probabilistic model is appropriate.
Carefully controlled laboratory experiments on life times in CD may produce a
high ( > 15) coefficient of variation in many
cases. In one carefully controlled laboratory
experiment (life of rolling elements of ball
247
(4.4.1)
At time zero assume damage is in state 0,
which corresponds to the satisfactory and
non-failed state. The Weibull distribution is
obtained by finding the distribution for one
event. Thus,
Fw(t ) = 1 - Po(t) = 1 - e -('/s), t > 0
(4.4.2)
The fact that there ar only two damage
states in the Weibull model, satisfactory and
failure, imposes severe limitations. Thus, one
cannot incorporate the properties of the initial damage state distribution, nor can we
trace the evolution of damage accumulation
via N D T methods which is extremely important for inspection and replacement requirements.
The shock model of Esary et al. (1973) is
also based upon a Poisson process approach
in the following manner. It is assumed that a
component is subjected to random shocks
which generate damage, and Pk = Prob
(survives k shocks}, then the failure time
distribution function is
Fw(t)=l-e
-at
k----~-' t = 0
(4.4.3)
k=0
248
(ii) Severity and order of duty cycles:
(iii) State of damage at retirement or
failure:
(iv) Inspection and replacement standards.
Experimental data must be available so that
these parameters can be evaluated inspecific
applications. Therefore, the parameters that
define the model should be incorporated in a
fashion that will allow relatively simple estimation methods for the determination of their
magnitudes. The model should be coherent in
the sense that all parts of the model that
describe various facets of the C D process fit
together in a unified structure.
The purpose of this section is to describe a
phenomenological probabilistic model that is
comprehensive, coherent, and robust. Applications of the model to many sets of real
experimental data, with the excellent fits and
predictions as well as diagnostic results obtained, indicate that the model is useful for
the study of fatigue life, crack growth and
wear processes (See Bogdanoff et al. 1978,
1980, 1981 and 1985).
b. Stationary MC-model
Po = {Th,.-., % } , where
-Pl
ql
...
P2
q2
...
0
0
0
0
0
0
0
0
......
......
Pt, 1
0
q~- l
1
cr,=P{Do=j},
A duty cycle (DC) is a repetitive period of
operation in the life of a component, the
damage of which accumulates during this
period. The precise mechanism of the accumulation of this non-negative increment of
damage during a D C is not known. It is
assumed that the increment of damage at the
end of a D C depends in a probabilistic manner
only on the amount of damage present at the
start of the DC, on that D C itself, and is
independent of how damage was accumulated
up to the start of that DC. In other words, the
damage accumulation is treated as an imbedded Markov process where damage is considered only at the end of a DC. The time
x = 0,1,2 . . . . is measured in numbers of DC's
irrespective of their duration. Moreover,
damage is indexed by a discrete set of states
y = 1,2 . . . . . b, where state b denotes retire-
j=l
..... b
(4.4.5)
I2, j= 1
j= 1
(4.4.6)
Fw(x;b ) = p x ( b )
(4.4.7)
249
we can write
b-1
F w ( x ; b ) = y" ~rjFw(x;j,b )
(4.4.8)
j=l
Fw(x;b ) = Fw(x;1,b )
(4.4.9)
E{WI,b} = Y'~ (1 + r j ) - r n w
(4.4.10)
j=l
b-I
2
P~')= ~ ~Px,(J)
(4.4.13)
j=l
j= 1
b-1
p,4(Wl,b) = 3o 4
b-I
j=l
b--1
+ E r/(1 +rg)
j=l
where rg =pj/qj. Moments of Wg,b are found
in a similar manner.
Failure may occur before state b is reached. Let the (1 b) row vector
P= {P,,...,Pb}
(4.4.11)
(4.4.12)
j=2
25{)
including cdf's, where appropriate, which is
most useful in describing and analyzing life
d a t a ( B o g d a n o f f a n d Krieger. 1978;
Bogdanoff and Kozin, 1985).
c. Non-stationary MC-model
In the non-stationary case, let P~ define the
severity of the j t h DC; this includes variation
in loading, environmental changes, and
changes in material properties all as a function of time. Then we replace eqn. (4.4.6) with
X
~x = fi0~rPi
(4.4.15)
(4.4.16)
251
In this section we are concerned with fitting MC-models to real fatigue, crack growth
and wear data. This require estimation of the
parameters of suitable MC-models. A detailed discussion of the parameter estimation
problems is beyond the scope of this report.
The reader should have sufficient experience
with many examples of fitting MC-models to
CD data so that a visual inspection of an edf
points to the general nature of the appropriate model. Then, the method of moments or least squares can be used to obtain
acceptable parameters.
A few examples are given to illustrate the
diagnostic, predictive and cdf fit capabilities
of the MC-model approach, with few parameters.
Saunders- Birnbaum Data (1969)
Figure 4.4.1 shows the edf (empirical distribution function) of the life data of 6061-T6
aluminum at zero mean stress with amplitude
of 31,000 psi. Inspection of the cdf either
visually or by means of the time-transformation technique reveals that a constant severity
DC model with ~r1 = 1 and Pb = 1 will produce an acceptable cdf. The probability transition matrix P has only unit steps with p j =
constant. We take one DC to represent 103
1.000
.g00
.800
.700
.600
/~
FIT
.500
.400
EDF
.300
BIRNBAUM SAUNDERS
.200
WEIBULL
FIT ~
DATA
/
.100
0.000
0.0
Fig, 4.4.1.
21'.2
42.2
63.6
84.5
105.0
127.2
252
1.0
B--MODEL
FIT
0.0
0
Fig. 4.4.2.
90
253
/Fv-
Fw
NONSTATIONARY
B--MODELFIT
/~'x
/'-"~ EDFDATA
.5
t-
a. General remarks
ic
./
x
J
400
800
Fig. 4.4.3.
Each of these possible explanations or a
combination of them occurs. The time-transformation-condensation technique was developed later. The cdf shown in Fig. 4.4.3 was
obtained from a nonstationary model using
this technique. Seven parameters are required
with ~r1 = 1 and Pb 1; four parameters describe the stationary model which produces
F w , and three parameters describe the time
transformation
=
y = x + 1.12x 2 + 67.18x 4
(4.4.17)
254
(1983)] for screening one bad part and certifying the remaining 99.9% for additional safe
utilizations of the life capacities of engine
disks.
The retirement-for-cause (RFC) life management is based on the applications of fracture mechanics and nondestructive evaluation
(NDE), under which each rotor disk is inspected periodically. When a crack is detected
during inspection, the disk is retired; otherwise, it is returned to service until the next
inspection maintenance. This procedure could
be repeated until the disk has incurred detectable damage (crack), at which time it is
retired for that reason (cause). Retirementfor-cause (RFC) is, then, a methodology under which an engine disk would be retired
from service when it has incurred quantifiable
damage, rather than because an analytically
determined minimum design life had been
exceeded.
To execute the retirement-for-cause procedvre, an optimal inspection interval should be
determined such that the life-cycle-cost of
engine components is minimum and a high
level of the component reliability should be
maintained. Hence the RFC analysis methodology should be capable of estimating the
probability of component failure, the percentage of replacement during each inspection maintenance, the percentage of components to be inspected by different NDE systems and the life-cycle-cost model as well as
various costs associated with it.
Various segments in the fatigue reliability
analysis including maintainability, such as
scheduled inspection and repair/replacement
maintenances or scheduled proof test maintenance for practical applications are reviewed
in the following.
b. Failure criteria
As the crack size increases in service, the
residual strength of the cracked component
reduces. For non-redundant components, it
suffices to determine the crack size distribution in service, and failure occurs as the crack
size exceeds the critical crack size. Theoretically, the critical crack size is a random variable. However, since the crack growth rate is
very high near the critical crack size, the
effect of the critical crack size variability on
the fatigue failure is rather small compared to
other random variables involved in the entire
design system. For the redundant structures,
such as fail-safe or multiple-load-path components, cracks may be arrested by crack
stoppers or other redundant members. In this
case, the distribution of the residual strength
should be computed from that of the crack
size. Then, failure occurs as the applied load
exceeds the residual strength [Yang (1974),
Shinozuka (1976)]. Relations between the
crack size and the residual strength of a component are available for some types of components based on fracture mechanics computations (see many references in Yang (1974,
1981), Oswald and SchuEller (1984)].
Nondestructive evaluation (NDE) system
Current nondestructive evaluation (NDE)
techniques are not capable of repeatedly producing correct indications when applied to
flaws of the same length. The chance of detecting a given crack length depends on many
factors, such as the location, orientation and
shape of the flaw, materials, inspectors, inspection environments, ect. As a result, the
probability of detection (POD) for all cracks
of a given length has been used in the literature to define the capability of a particular
NDE system in a given environment. Many
POD curves can be found, for instance, in
ASCE (1982), Packman et al. (1976), Yee et
al. (1976), Berens and Hovey (1981) Yang
and Donath (1982, 1983) and references
therein.
In practical applications, a nondestructive
inspection limit, aNDE, is usually specified,
which is a crack size corresponding to a high
detection probability and confidence level.
255
There are two possible errors that can occur
in any inspection system: failure to give a
positive indication in the presence of a crack
size greater than aNDE, referred to a Type I
error, and the giving of a positive indication
when the crack size is smaller than aNDE,
referred to as Type II error. It is possible to
reduce either Type I or Type II error using
multiple N D E systems, see Yang and Donath
(1982, 1983) and Yang and Chen (1985a,
1985b). Furthermore, various N D E systems
have been reviewed in ASCE (1982).
Scheduled maintenance procedures
Two types of scheduled maintenance procedures will be considered; (i) scheduled inspection and repair or replacement maintenance, and (ii) scheduled proof tests maintenance. For most of the fatigue-critical components, the scheduled maintenance is a necessity to assure their structural integrity [Barrois
(1973), Coffin and Tiffany (1976)]. The role
of maintenance becomes more important as
the service life of the component increases.
Currently, scheduled maintenance procedures
are determined from the deterministic calculation results, with the possible exception of
the retirement-for-cause for gas turbine engine components (e.g., Harris et al. (1980),
King (1983), Yang and Chen (1985a, 1985b)].
Scheduled inspection and repair or replacement maintenance The purpose of
nondestructive inspection is to detect the
crack if it exists. When a detectable crack
exists at a critical location, the residual fatigue life is significantly reduced. If a crack is
detected in a detail, either during production
or service inspection, the cracked detail may
be repaired, if the crack size is small enough
to be repairable; otherwise other repair procedures, such as retrofit or other procedures
may be needed. For some fatigue-critical
components, however, the repair is not feasible so that the entire component should be
replaced. For instance, a gas turbine engine
disk consists of many critical locations, such
256
Life-cycle-cost optimization
While the maintenance programs have been
demonstrated to improve the structural reliability in service, they may be expensive. Although in some situations the maintenance
program is a necessity, the cost should be
balanced by its benefit [Yang (1976b)]. Hence
the economical consideration for maintenance procedures is important, and it can be
handled more appropriately by the statistical
approach. In this connection, the life-cyclecost optimization using the return-of-investment (ROI) as an objective function should
be mentioned. Any objective function used
for the optimization purpose should take into
account the cost of failure, the cost of repair
or replacement, the cost of inspections and
the associated cost of maintaining the N D E
systems, the initial cost, the cost of down
time, etc. While preliminary investigations in
this regard have been attempted, further research is needed,
Finally, the subject problems reviewed
above apply essentially to metal fatigue. For
composite materials that have become very
important in industrial applications, the applicability of fracture mechanics has not been
proven to be promising. This is because the
materials are inhomogeneous and anisotropic,
and the fatigue damage includes different
Structures deform under the action of loadings. When the loading conditions are severe
the deformation may be permanent and,
hence, damage occurs. The extent and nature
of structural damage necessarily depends on
(1) the material used (steel, reinforced concrete, masonry or wood); (2) structural configuration and construction (frame, shear wall,
etc.); and (3) loading environments (static,
dynamic). As many buildings and structures
are uniquely constructed and most loadings
are variable and stochastic in nature, it is not
surprising that there is no simple commonly
accepted procedure to describe structural
damage. While fatigue and fracture are generally referring to the process of damage at a
particular location of structural components,
the damage of nonstructural components such
as claddings, partitions, and contents of the
structure can be also very important. Therefore, in general, local as well as global measures of the state of the structure are required
in describing structural damage.
Until recently, definitions of structural
damage have been mostly descriptive; the
Modified Mercalli Intensity scale which corresponds to various degrees of damage caused
by earthquake is a good example. Other
gradations of structural damage state such as
minor, moderate, severe, major and partial
257
1 [~. oairliS i
8c = ~
[~
i=1
xiri
(4.6.1)
in which wi = importance factor (consideration such as life hazard and cost), ~/, = service
history influence coefficient for demand, S, =
response (demand) in i th element, x i = service
history influence coefficient for capacity, and
r, = resistance (capacity) in i th element. The
cumulative effect of the load and degradation
of resistance are included through the two
influence coefficients. However, determination of ~, 71 and x is not elementary and
remains to be studied.
258
Based on the linear cumulative damage
function to predict low cycle fatigue failure in
metal specimen by Yao and Munse (1962),
Kasiraj and Yao (1969) evaluated the seismic
damage of a simple structure for deterministic
loading conditions. Tang and Yao (1972) extended the work for random ground excitations. Recently, Stephens and Yao (1985) proposed a similar damage function as follows:
ASp,
i=l
"'I
)
(4.6.2)
in which Ym= 1 - 0.86r, r = relative deflection ratio, i.e., the ratio of the compression
change to tension change in plastic deflection
in cycle i, ASpt= tension change, ASpf-tension change in a one cycle test to failure
conducted at the relative deformation ratio of
cycle i. D will range from 0 (no damage) to
1.0 (critically damaged). This model has been
used to assess the damage of several reinforced concrete structures, including models
tested in laboratory and real structures
damaged during actual earthquakes.
It is pointed out that the determination of
the foregoing damage functions requires time
histories of the structural response, therefore,
it is more suitable to assess damage of structures for which such records are available. To
assess uncertainties and predict reliability,
obviously a large number of such responses
are required or analytical methods based on
random vibration need to be developed.
For structures that suffer cumulative
damage, a good measure of the degree of such
damage is the total hysteretic energy dissipation, since it depends on the inelastic response amplitude as well as the number of
cycles of stress reversals. Measuring of damage
based on such energy consideration has been
advocated by Akiyama (1985) based on studies of a large number of steel buildings under
earthquake excitation. Whereas Banon and
Veneziano (1982) studied laboratory test results of reinforced concrete members and
(~max
8~ + ~
f dE
(4.6.3)
259
found that the variability of the member
capacity in terms of limiting value (collapse)
of D c can be approximately described by a
lognormal variate with a mean of 1.0 and
standard deviation of 0.54. An index based
on a nonlinear function of these two response
parameters does not show significant improvement in terms of reduction of uncertainty. To predict damage of structures under
future loadings, a recently developed hysteretic restoring force model and random vibration method (Pires, et al. 1983; Wen, 1980)
can be used to calculate the statistics of 6ma~
and f d E given the intensity and power spectral density function of the excitation. As a
result, the statistics of the damage index due
to future excitation, D 1, can be easily
evaluated. The risk of the limiting value of
damage being reached can be expressed as
P [ D I l D c > 1] in which the uncertainties in
both loading and resistance have been incorporated. A method for synthesis of member
damages to obtain the damage index of the
structure has been proposed. The calibration
of the proposed damage index with damage
observed in a member of R.C. buildings during several recent earthquakes in the U.S. and
Japan indicated that a value of less than 0.4
corresponds to reparable damage and a value
of more than 1.0 corresponds to collapse
(Park, Ang, and Wen, 1985).
c. Other approaches
It is important for structural engineers to
estimate the damage state at the time of inspection as well as the up-date reliability of
an existing structure. Yao (1979), and Liu and
Yao (1978) reviewed several damage functions and discussed the general problem of
structural identification. Nevertheless, it is
difficult to clearly define the degree of damage
of a complex structural system in the real
world. The theory of fuzzy sets (Zadeh, 1965,
1973) may be useful in the solution of such a
problem.
260
processing techniques, the selection of which
may require subjective evaluation of certain
experts (Stephens and Yao, 1985).
As another alternative, Toussi and Yao
(1982-83) applied the theory of evidence to
make an analysis of damage. The proposed
algorithm was applied to assess the damage
state of the scaled laboratory model of a
10-story reinforced concrete frame subjected
to repeated earthquake-like excitations.
Moses and Yao (1984) suggested the use of
the following reliability function:
(4.6.4)
tr
h~)(t)
Lv(t~)
h~)(t)
Design Information
Statistical Data
Expected Loading and Enviromental Condition
J Classical Th.of
Structurat
Reliability
Objective
Safet 7 Measure. Bayesian Statistics
System Identification
Fuzzy Sets
Rule--Based System
LT (t i)
y
es
\
J
Random Processes
First Passage
Probability
Random Vibration
Possible
Forms
.of Hazard
Functions
hT
/
/
(t)
No
Take
Apropiote
Action
T
Fig. 4.6.1.A suggestedprocess for safetyevaluation of existing structures (after Mosesand Yao (1984)).
261
262
Brown, C.B. (1979). A fuzzy safety constructed probabilities. J. Eng. Mech. Div. ASCE, 105 (EM5).
Brown, C.B. (1980). Entropy constructed probabilities.
J. Eng. Mech. Div., ASCE, 106 (EM4).
Brown, C.B., Furuta, H., Shiraishi, N. and Yao, J.T.P.
(1984). Civil Engineering Applications of Fuzzy Sets,
Proceedings, First International Conference on Fuzzy
Information Processing, Kauai, Hawaii, 22-25 July
1984, Edited by J.C. Bezdek (in Press).
Brown, C.B., and Leonard, R.S., (197l). Subjective uncertainty analysis. Preprint No. 1388, ASCE National Structural Engineering Meeting, Baltimore,
Maryland, 19-23 April 1971.
Brown, C.B. and Yao, J.T.P. (1982). Instructions and
opinions: the use of fuzzy subjectivity. Architectural
Science Review, 1982.
Brown, C.B. and Yao, J.T.P. (1983). Fuzzy set's in
structural engineering. J. Struct. Eng. ASCE, 109 (5):
1211-1225.
Butler, J.P. (1972). Reliability analysis in the estimation
of transport-type aircraft fatigue performance. In:
Proc. Int. Conf. Structural Safety and Reliability,
Edited by A.M. Freudenthal, Pergamon.
Coffin, L.F. Jr. (1954). A study of the effects of cyclic
thermal stresses in a ductile material. ASME Trans.,
16: 931-950.
Coffin, M.D. and Tiffany, C.F. (1976). New air force
requirements for structural safety, durability, and
life management. J. Aircraft AIAA, 13 (2): 93-98.
Criteria of the ASME Boiler and Pressure Vessel Code
for Design by Analysis in Section III and VIII
(1969). Division 2, ASME.
Culver, C.G., Lew, H.S., Hart, G.C. and Pinkham, C.W.
(1975). Natural Hazard Evaluation of Existing Buildings. U.S. department of Commerce, National Bureau
of Standards, Building Science Series, Washington,
D.C.
Davidson, J.R. (1975). Reliability after inspection. Fatigue of Composite Materials, ASTM-STP 569, pp.
323-334.
Ditlevsen, O. (1981a). Reliability against defect generated fracture, J. Struct. Mech., 9 (2): 115-137.
Ditlevsen, O. (1981b). Reliability against fracture of
butt welds as inferred from inspection data. Eng.
Fracture Mech., 14 (4): 713-724.
Ditlevsen, O. (1986), "Random fatigue crack growth--a
first passage problem. Eng. Fracture Mech., 23 (2):
-467 477.
Ditlevsen, O. and Olesen, R. (1985). Statistical analysis
of the Virkler data on fatigue crack growth. To be
published in Engineering Fracture Mechanics (Preprint: DCAMM Report No. 314, Technical University of Denmark, Lyngby, Denmark).
Ditlevsen, O. and Sobczyk, K. (1985). Random fatigue
crack growth with retardation. Accepted for publication in Engineering Fracture Mechanics (preprint:
DCAMM Report No. 309, Technical University of
Denmark, Lyngby, Denmark).
Dubois, D. and Prade, H. (1980). Fuzzy Sets and Systems: Theory and Applications. New York, Academic
Press, 1980.
Eggwertz, S. (1972). Investigation of fatigue life and
residual strength of wing panel for reliability purposes. Probabilistic Aspects of fatigue, ASTM-STP
511, 1972, pp. 75 105.
Esary, J.D., Marshall, A.W., and Proschan, F. (1973).
Shock models and wear processes. Annals of Prob., 1
(4): 627-649.
Evans, A.G. and Wiederhorn, S.M. (1974). Proof testing
of ceramic materials--an analytical basis for failure
prediction. Int. J. Fracture Mech. 10 (3): 375-392.
Fatigue Design Handbook. AE-4, SAE,.
Feddersen, C.E. (1970). Evaluation and Prediction of
the Residual Strength of Center Cracked Tension
Panels. ASTM STP 486, pp. 50-78.
Freudenthal, A.M. (1947), Safety of structures. Trans.
ASCE, 112:125 180.
Freudenthal, A.M. (1965). The Expected Time to First
Failure, Air Force Materials Laboratory, AFMLTR-66-37, WPAFB.
Freudenthal, A.M. (1967). Reliability analysis based on
time to first failure, Conference Paper presented at
the Fifth I.C.A.F. Symposium, Melbourne, Australia.
Freudenthal, A.M. (1975). Reliability Assessment of
Aircraft Structures Based on Probabilistic Interpretation of the Scatter Factor. Air Force Materials
Lab., AFML-TR-74-198, WPAFB.
Freudenthal, A.M. and Gumbel, E.J. (1953). On the
statistical interpretation of fatigue tests, Proc. Royal
Soc. London, Series A, 216: 309-322.
Freudenthal, A.M. and Gumbel, E.J. (1956). Physical
and statistical aspects of fatigue. Adv. Appl. Mech.,
4.
Freudenthal, A.M. and Schu611er, G.1. (1973). Scatter
Factors and Reliability of Aircraft Structures. NASA
Grant NGR-09-010-058, The George Washington
University, Washington, D.C.
Fu, K.S. and Yao, J.T.P. (1979). Pattern recognition
and damage assessment, presented at the Third ASCE
EMD Specialty Conference, University of Texas,
Austin, TX.
Fuchs, H.O. and Stephens, R.K. (1980), Metal Fatigue
in Engineering, Wiley.
Fujino, Y. and Lind, N.C. (1977). Proof-load factors
and reliability. J. Struct. Div. ASCE, 103 (ST4):
853-870.
Gumbel, E.J. (1958). Statistics of Extremes. Columbia
University Press.
263
Gumbel, E.J. (1963). Parameters in the distribution of
fatigue life. J. Eng. Mech. Div. ASCE, 89, (EM5).
Harris, J.A., et al. (1980). Engine Component Retirement for Cause. Contract F33615-80-C-5160, Pratt a
Whitney Aircraft Group, Government Product Division, West Palm Beach, FL.
Harrison, R.P., Milne, J. and Loosemore, K. (1977).
Assessment of the Integrity of Structures Containing
Defects. CEGB Rep. R / H / R 6 Rev. 1.
Heer, E. and Yang, J.N. (1971). Structural optimization
based on fracture mechanics and reliability criteria.
AIAA J., 9 (5): 621-628.
Hertzberg, R.W. (1976), Deformation and Fracture
Mechanics of Engineering Materials. John Wiley.
Hoeppner, D.W. and Krupp, W.E. (1974). Prediction of
component life by application of fatigue crack
knowledge. Eng. Fract. Mech., 6: 47-70.
Housner, G.W. and Jennings, P.C. (1977). Earthquake
Design Criteria for Structures. EERC, Report No.
77-06, California Institute of Technology, Pasadena.
Ishizuka, M., Fu, K.S. and Yao, J.T.P. (1983). Rulebased damage assessment system for existing structures. Solid Mech. Arch., 8: 99-118.
Kasiraj, I. and Yao, J.T.P. (1969). Fatigue damage in
seismic structures, J. Struct. Div. ASCE, 95 (ST8):
1673-1692.
King, T.T. (1983). USAF Engine Structural Durability
and Damage Tolerance. AIAA Professional Study
Series, June 30-July 1, 1983, Seattle, Washington.
Kozin, F. and Bogdanoff, J.L. (1981a). A critical analysis of some probabilistic models of fatigue crack
growth, Eng. Fract. Mech., 14: 59-89.
Kozin, F. and Bogdanoff, J.L. (1983a). An approach to
accelerated testing, Trans. Canad. Aero. Space Inst.,
29 (1): 60-76.
Kozin, F. and Bogdanoff, J.L. (1983b). On life behavior
under spectrum loading. Eng. Fract. Mech., 18 (2):
271-293.
Kozin, F. and Bogdanoff, J.L. (1985a). Adaptiveupdating of fatigue model for reliability and maintenance
based upon service information. 40th MFPG (NBS),
Proc., Gaithersburg.
Kozin, F. and Bogdanoff, J.L. (1985b). On the statistic
modeling of cumulative damage in composites.
ICOSSAR 85, Proceedings, Vol. 1, Kobe, Japan.
Kozin, F., and Bogdanoff, J.L. (1985c). On probabilistic
modeling of fatigue crack growth. In: I. Konishi,
A.H.-S. Ang and M. Shinozuka (Eds.), Structural
Safety and Reliability, Vol. II, IASSAR (Proceedings
of ICOSSAR'85, Kobe, Japan, 1985), pp. 331-340.
Kozin, F. and Bogdanoff, J.L. (1986). A model of
prediction of failure of long chains. To appear in
ASCE J. Eng. Mech.
Krawinkler, H. et al. (1983). Recommendations for Ex-
264
Miller, M.S. and Gallagher, G.P. (1981). An analysis of
several fatigue crack growth rate descriptions. Measurement and Data Analysis, ASTM-STP 738, pp.
205-251.
Miner, M.A. (1945). Cumulative damage in fatigue.
ASME J. Appl. Mech., 12: 159-164.
Moses, F. and Yao, J.T.P. (1984). Safety evaluation of
buildings and bridges. In: D. Faulkner, M.
Shinozuka, R.R. Fiebrandt, and I.C. Franck (Eds.),
The Role of Design, Inspection, and Redundancy in
Marine Structural Reliability, Committee on Marine
Structures, National Research Council, Washington,
D.C., pp. 349-385.
Newmark, N.M. and Rosenblueth, E. (1971). Fundamentals of Earthquake Engineering. Prentice-Hall.
Oberparleiter, W. and Schlitz, W. (1981). Fatigue life
prediction in a corrosive environment in: D.
Faulkner, M.J. Cowling and P.A. Frieze (Eds.), Integrity of Offshore Structures, The University of
Glasgow, Scotland.
Ogawa, H., Fu, K.S., Hanson, J.M. and Yao, J.T.P.
(1983). An Extension of Structural Damage Assessment System: SPERIL-I. Technical Report, School
of Civil Engineering, Purdue University, West
Lafayette, IN.
Okamoto, S., Nakata, S. Kitagawa, Y., Yoshimura, M.
and Kaminosono, T. (1982). A Progress Report on
the Full-Scale Seismic Experiment of a Seven Story
Reinforced Concrete Building-Part of the US-Japan
Cooperative Program. Building Research Council,
Ministry of Construction, Japan.
Oliveria, C.S. (1975). Seismic Risk Analysis for a Site
and a Metropolitan Area. EERC, Report No. 75-3,
University of California, Berkeley, CA.
Ortiz, K. and Kiremidjian, A.S. (1985). Time Series
Analysis of Fatigue Crack Growth Rate Data.
Manuscript (first author: Aerospace and Mechanical
Engrg. Dept., Univ. of Arizona, Tucson, Arizona
85721; second author: Dept. of Civil Engrg., Stanford Univ., Stanford, California 94305).
Oswald, G.F. (1983). A Contribution to the Reliability
of Structures Taking into Account Time Variant
Systems Properties (in German). Diss., Techn. Univ.
Munich.
Oswald, G.F. and SchuEller, G.I. (1983). On the reliability of deteriorating structures. In: G. Augusti et al.
(Eds.), Proc., 4th Int. Conf. on Appl. of Statistics
-and Prob. in Soil & Struct. Engr. (ICASP-4), Pitagora
Editrice, Bologna, pp. 597-608.
Oswald, G.F. and SchuEller, G.I. (1984). Reliability of
deteriorating structures, Eng. Fract. Mechs., 20 (3):
479-488.
Packman, P.F. et al. (1976). Reliability of flaw detection
by nondestructive inspection, ASM Metal Handbook,
Vol. 11, 8th Edition, Metals Park, Ohio, pp. 214-224.
Palmberg, B., Blon, A.D. and Eggwertz, S. (1986). Probabilistic damage tolerance analysis of aircraft structures. To appear in Probabilistic Fracture Mechanics
and Reliability, Chapter Ill, American Society for
Testing and Materials, Special Technical Publication, ASTM-STP, Jan. 1986.
Palmgren, A. (1924), Die Lebensdauer von Kugellagern.
Zeitschrift des Vereines Deutscher lngenieure, 68
(14): 339-341.
Paris, P.C. (1964). The fracture mechanics approach to
fatigue. In: J.J. Burke, N.L. Reed and V. Weiss
(Eds.), Fatigue, An Interdisciplinary Approach,
Syracuse University Press, pp. 107-132.
Park, Y-J. and Ang, A. H-S. (1985). Mechanistic seismic
damage model for reinforced concrete. J. Struct.
Eng. ASCE, 111 (4): 722-739.
Park, Y.-J., ANg, A. H-S. and Wen, Y.K. (1985). Seismic
damage analysis of reinforced concrete buildings. J.
Struct. Eng. ASCE, 111 (4): 740-757.
Parker, R.J., Zaretsky, E.V. and Dietrich, M.W. (1971).
Rolling-Element Fatigue Life of Four M-Series
Steels. NASA TN D-7033, p. 14.
Parzen, E. (1959). On Models for the Probability of
Fatigue Failure of a Structure. NATO 245.
Pires, J.E.A., Wen, Y.K. and Ang, A. H-S. (1983).
Stochastic Analysis of Liquefaction Under Earthquake Loadings. SRS Report No. 504, University of
Illinois at Urbana-Champaign.
Rolfe, S.T. and Barson, J.M. (1977). Fracture and Fatigue Control in Structures, Prentice-Hall.
Roufaiel, M.S.K. and Meyer, C. (1981), Analysis of
Damaged Concrete Frame Buildings. University of
Columbia, N.Y., Techn. Report No. MSF-CEE-8121359-1.
Rudd, J.L. and Gray, T.D. (1977). Quantification of
fastener hole quality. Proc. 18th A I A A / A S M E / S A E
Structures, Structural Dynamics and Materials Conference.
Rudd, J.L., Yang, J.N., Manning, S.D. and Garver,
W.R. (1982). Durability design requirements and
analysis for metallic airframe. Design of Fatigue and
Fracture Resistant Structures, ASTM STP 761. 1982,
pp. 133-151.
Rudd, J.L., Yang, J.N., Manning, S.D. and Yee, B.G.W.
(1982a). Damage assessment of Mechanically
fastened joints in the small crack size range. Proc. of
the Ninth U.S. National Congress of Applied Mechanics.
Rudd, J.L., Yang, J.N., Manning, S.D. and Yee, G.B.W.
(1982b). Probabilistic fracture mechanics analysis
methods for structural durability. Proc. of A G A R D
Meeting on Behavior of Short Cracks in Airframe
Components, 1982.
Saunders, S.C. and Birnbaum, Z.W. (1969). Estimation
265
for a family of life distributions with application to
fatigue. J. Appl. Probability, 6 (2): 338-347.
Schneider, R.R. and Dickey, W.L. (1980). Reinforced
Masonry Design. Prentice Hall.
Schu~ller, G.I., (1985). A consistent reliability concept
utilizing fracture mechanics, In: S. Eggwertz and
N.C. Lind (Eds.), Proc., IUTAM Symposium
Stockholm, Springer Verlag, Berlin, Heidelberg, pp.
145-156.
Schu~ller, G.I. and Freudenthal, A.M., (1972). Scatter
Factor and Reliability of Aircraft Structures. NASA,
Techn. Rep. CR-2100, Washington, D.C., 20546,
Nov., 1972.
Sendeckyj, F.P. (1981). Fitting models to composite
materials fatigue data. Test Methods and Design
Allowables for Fibrous Composites, ASTM-STP 734,
American Society for Testing and Materials, pp.
245-260.
Shibata, A. and Sozen, M.A. (1976). Substitute structure
method for seismic design in R/C. J. Struct. Div.
Proc. ASCE.
Shinozuka, M. (1968). Structural Reliability Under
Condition of Fatigue and Ultimate Load Failure.
Air Force Materials Lab., AFML-TR-68-234.
Shinozuka, M. (1976). Development of Reliability-Based
Aircraft Safety Criteria: An Impact Analysis,
AFFDL-TR-76-36, Vol. 1, WPAFB, 1976.
Shinozuka, M. (1978). Reliability-Based Scatter Factors,
Vol. 1: Theoretical and Empirical Results, Air Force
Flight Dynamics Lab., AFFDL-TR-78-17.
Shinozuka, M. (1979). Durability Methods Development, Volume IV--Initial Quality Representation.
AFFDL-TR-79-3118, September 1979.
Shinozuka, M. and Yang, J.N. (1969). Optimum structural design based on reliability and proof-load test.
Ann. Assurance Science, Proc. of the 8th Reliability
and Maintainability Conf., Vol. 8, pp. 375-391.
Shinozuka, M., Yang, J.N. and Heer, E. (1969). Optimal
structural design based on reliability analysis, Proc.
8th International Symp. on Space Technology and
Science, Tokyo, Japan, pp. 245-258.
Solomon, H.D. (1972). Low cycle fatigue crack propagation in 1018 steel. J. Materials, JMLSA, 7 (3):
299-306.
Sozen, M.A. (1981). Review of earthquake response of
reinforced concrete buildings with a view to drift
control. State of the Art in Earthquake Engineering.
Turkish National Committee on Earthquake Engineering, Istabul, Turkey.
Stephens, J.E. and Yao, J.T.P. (1985). Data Processing
of Earthquake Acceleration Records, Technical Report No. CE-STR-85-5, School of Civil Engineering,
Purdue University, W. Lafayette, IN.
Stephens, J.E. and Yao, J.T.P. (1985). Estimation of
266
Whitney, J.M. (1981). Fatigue characterization of composite materials. Fatigue of Fibrous Composite
Materials, ASTM-STP 723, American Society for
Testing and Materials, pp. 133-151.
Whittaker, I.C. and Besuner, P.M. (1969). A Reliability
Analysis Approach to Fatigue Life Variability of
Aircraft Structures, AFMLTR-69-65.
Whittaker, I.C. and Saunders, S.C. (1973a). Exploratory
Development on Application of Reliability Analysis
to Aircraft Structures Considering Interaction of Fatigue Crack Growth and Periodic Inspection. Air
Force Materials Lab., AFML-TR-72-283.
Whittaker, I.C. and Saunders, S.C. (1973b). Application
of Reliability Analysis to Aircraft Structures Subjected to Fatigue Crack Growth and Periodic Inspection. Air Force Materials Lab., AFML-TR-73-92.
Wiggins, J.H., Jr. and Moran, D.F. (1971). Earthquake
Safety in the City of Long Beach Based on the
Concept of Balanced Risk. J.H. Wiggins Company,
Redondo Beach, C.A.
Willenborg, J.R.M., Engle, R.M. and Wood, H.A. (1971).
A Crack Growth Retardation Model using an Effective Stress Concept. AFFDL-TM-FBR-71-1.
Wirsching, P.H. (1981). The Application of Probabilistic Design Theory to High Temperature Low Cycle
Fatigue. NASA CR-165488.
Yang, J.N. (1974). Statistics of random loading relevant
to fatigue. J. Eng. Mech. Div. ASCE, 100 (EM3):
469-475.
Yang, J.N. (1976a). Reliability analysis of structures
under periodic proof test in service, AIAA J., 14 (9):
1225-1234.
Yang, J,N. (1976b). Statistical estimation of service
cracks and maintenance cost for aircraft structures.
J. Aircraft AIAA, 13 (12): 929-937.
Yang, J.N. (1977). Optimal periodic proof test based on
cost-effective reliability criteria. AIAA J., 15 (3):
402-409.
Yang, J.N. (1977). Reliability prediction for composites
under periodic proof test in service, American Society
for testing and materials, ASTM-STP 617, Composite Materials, Testing and Design, pp. 272-295.
Yang, J.N. (1978a). Fatigue and residual strength degradation for graphite/epoxy composites under tension-compression cyclic loading. J. Composite
Materials, 12: 19-29.
Yang, J.N. (1978b). Statistical approach to fatigue and
fracture including maintenance procedures. In: Perrone et al. (Eds.), Fracture Mechanics, University
Press of Virginia, Charlottesville, 1978, pp. 559-577;
Proc. Tenth Symposium on Naval Structural Mechanics, Washington, D.C.
Yang, J.N. (1980). Statistical estimation of economic
life for aircraft structures. J. Aircraft AIAA, 17 (7):
528-535.
267
sequence on statistical fatigue of composites. AIAA
J., 18 (12): 1525-1531.
Yang, J.N. and Jones, D.L. (1982). Fatigue of
graphite/epoxy [ 0 / 9 0 / + 4 5 / - 45] laminates under
dual stress levels. Composite Technol. Rev., 4 (3):
63-70.
Yang, J.N. and Liu, M.D. (1977). Residual strength
degradation model and theory of periodic proof tests
for graphite/epoxy laminates. J. Composite
Materials, 11: 176-203.
Yang, J.N., Hsi, W.H., Manning, S.D. and Rudd, J.L.
(1986). Stochastic crack growth models for application to aircraft structures. To appear in Probabilistic
Fracture Mechanics and Reliability, Chapter IV,
American Society for Testing and Materials, Special
Technical Publication, ASTM-STP, Jan. 1986.
Yang, J.N. and Manning, S.D. (1980). Distribution of
equivalent initial flaw size. 1980 Proceedings Annual
Reliability and Maintainability Symp., San
Francisco, CA, Jan. 22-24, 1980, pp. 112-120.
Yang, J.N., Manning, S.D. and Rudd, J.L. (1985a).
Evaluation of a stochastic initial fatigue quality
model for fastener holes, Presented at ASTM Conference, March 18-19, 1985, Charleston, S.C., to
appear in Fatigue in Mechanically Fastened Composite and Metallic Joints, ASTM Special Publication.
Yang, J.N., Manning, S.D., Rudd, J.L. and Hsi, W.H.
(1985b). Stochastic crack propagation in fastener
holes, Proc. A I A A / A S M E / A S C E / A H S 26th Structures, Structural Dynamics and Materials Conf.,
April 15-17, 1985, Orlando, FL, pp. 225-233, AIAA
Paper No. 85-0666; also: J. Aircraft AIAA, 22 (9):
810-817.
Yang, J.N., Salivar, G.C. and Annis, C.G. (1982) Statistics of Crack Growth in Engine Materials--Vol. 1:
Constant Amplitude Fatigue Crack Growth at
Elevated Temperatures. Air Force Wright Aeronautical Lab., Technical Report AFWA1-TR-82-4040,
WPAFB.
Yang, J.N., Salivar, G.C. and Annis, C.G. (1983). Statis-