Вы находитесь на странице: 1из 6

Fatigue & Fracture of Engineering Materials & Structures 1998; 21: 535540

FRACTURE MECHANICS AND FATIGUE:


A HISTORICAL PERSPECTIVE
P. C. P
Mechanical Engineering Department/Campus Box 1185, Washington University,
One Brookings Drive, St. Louis, MO 63130, USA

Received in final form July 1997

AbstractIt was just 40 years ago that some of us became aware that the methods of Irwins Fracture
Mechanics could be applied to fatigue crack growth. After the first 10 years of establishing that approach,
many things were known which impact this Symposiums theme of prediction of structural life.
In the subsequent 30 years to the present, the Fracture Mechanics approach applied to life prediction
has often been misused, or has ignored obvious characteristics which effect the precision and relevance
of the predictions. This paper will emphasize some of these areas for improvements, which have been too
often overlooked in practical applications. Some of the overlooked areas are:
(1) the misuse of crack propagation laws in life prediction;
(2) the modelling of crack shape effects in life calculations;
(3) the neglect of significant statistical effects and considerations for particular applications.
In conclusion, it is argued that life prediction methodology should be tailored to the application, rather
than vice versa, which has all too often been the case.
KeywordsCrack growth; Fatigue; Fracture mechanics; Life prediction; Load interaction; Statistical
effects.

INTRODUCTION
Just 40 years ago, George Irwin, at the International Conference on Mechanics in 1956, drew
attention to the elastic stress field surrounding crack tips in loaded bodies and the intensity factor
of that unique field, K, as a tool to analyse abrupt failures. He also clearly indicated the direct
relationship between this intensity factor, K, and the Griffith rate of availability of elastic energy
for the processes of crack extension. Previously, at the Boeing Company, my colleagues had asked
if the GriffithIrwin energy analysis approach could be applied to fatigue crack growth (1955).
The immediate reaction was to see no physical reason for the incremental extension rate of the
fatigue crack to be related to the rate of elastic energy release, because cumulative effects of
plasticity would obviously be involved. Only later, upon understanding the generality of the Irwin
crack tip stress field surrounding the plasticity at a crack tip, did it occur that it was obvious that
the fluctuation of the stress field, as measured by K, should control the rate of crack growth. Early
in 1957, I sent a memo to W. E. Anderson of Boeing [1], explaining this prospect. Unfortunately
at the time, no data on the growth rates of fatigue cracks were available to verify the concept.
It may seem incredible today that in those days of the first modern jet aircraft no one had a
method of predicting how fast a fatigue crack would grow. However, it was well known that small
cracks were frequently observed in all monocock airframes during their lives. Indeed, the British
Comet aircraft of the early 1950s experienced catastrophic failure of their pressure cabins due to
growth to a critical size of such fatigue cracks. (Although I have since obtained a commercial
pilots licence, it is only honest and informative to note that from 1954 until 1965, I declined to
travel by air even on trips for the Boeing Company.)
Prior to this time, Frost and Dugdale [2] of the National Engineering Lab in Scotland and
Head [3] in Australia, had attempted, both empirically and with physical modelling, to determine

1998 Blackwell Science Ltd 535


536 P. C. P

the relationship of variables affecting fatigue crack growth. The lack of a sufficiently comprehensive
set of data against which to test their proposals makes it understandable that they did not come
to completely correct conclusions. Considering the rather unsophisticated testing equipment avail-
able in those days, their efforts were quite good in initiating the search for the controlling
parameters. Subsequently, in 1958, McEvilly along with Illg [4] at N.A.S.A. Langley provided an
extensive set of fatigue crack growth rate data on 2024-T3 and 7075-T6 aluminium alloys. McEvilly
correlated his data making use of a notch analysis parameter, which is somewhat, though obscurely,
related to K. McEvillys contribution has never been sufficiently acknowledged in subsequent
discussions. It was indeed the first successful proposal of such a parameter prior to the Irwin K.
Therefore, after some false starts at obtaining an adequate set of data at the Boeing Company,
it was only in 1959 that sufficient data were obtained to test the validity and range of applicability
of K in correlating fatigue crack growth rate data with a parameter which was easily relatable to
structural configurations. The data from N.A.S.A., and also from S.A.A.B. and the University of
Illinois, on both aluminium alloys were shown [5] to be well correlated using K.

THE INITIAL REACTION TO FRACTURE MECHANICS IN FATIGUE

The initial reaction for this young researcher at that time was that we were on to something
big and important to the safety of aircraft and other structures as well. It was time to tell the
world about it, so a paper [5] was written clearly showing the correlation of data from three
independent sources on two materials. The tools of Fracture Mechanics analysis needed to apply
lab data to structural configurations were also illustrated. Well, that paper was very promptly
rejected by three of the worlds leading journals. All of the reviewers simply stated that no elastic
parameter, e.g. K, could possibly correlate fatigue cracking rates because plasticity was a dominant
feature. They proceeded to somehow totally disregard the facts clearly demonstrated by the data!
Undaunted at the Boeing Company, we proceeded to move ahead with obtaining further data.
All of the companies test equipment was tied-up performing tests for crucial items on production
aircraft. At the time, the rules were that any purchase of an item over $10 000 required approval
of the board of directors of the company. The fact that we ever produced any further results is a
tribute to the patience and persistence of W. E. Anderson, who with the blessing of some upper
level supervisors, home-built some test equipment out of components. By 1961, some initial data
were collected on various materials including steels, titanium and magnesium alloys. The studies
explored the influence of load ratio, frequency and various environments, as compared to lab air.
Although the data taken were predominantly for constant amplitude sinusoidal loading, the effect
of overload causing delays in growth was clear from test interruptions, etc. In addition, a bright
new young man, S. Smith, joined the effort and rose to the task of producing some data under
loadings of Gaussian random processes of various bandwidths. For the early 1960s, in retrospect,
the various effects and variables were in general quite well understood.
Nevertheless, it was not easy to find support for fatigue crack growth studies in that era of
plenty, as compared with today. Many trips to the Wright-Patterson Air Force Base in the early
and mid-1960s were met with polite interest, but no funds were available for such studies. It took
the F-111 failure of 1969 to arouse real Air Force interest. Previously, the Boeing Company had
led the way by providing modest support at Lehigh University prior to obtaining N.S.F. and
N.A.S.A. support there. Moreover, there was a good deal of technical resistance to using Irwins
Fracture Mechanics for anything, including fatigue, before the mid-1960s. Competing approaches
seemed to regard Fracture Mechanics as the enemy. Many materials scientists dismissed the

Fatigue & Fracture of Engineering Materials & Structures, 21, 535540 1998 Blackwell Science Ltd
History of fracture mechanics and fatigue 537

approach as totally lacking, since it was not based on dislocation theory (i.e. simply not a
fundamental approach).
Although much was known in the early-1960s, some key knowledge took until the late-
1960s to be forthcoming. For example, the work of Professor R. P. Wei and students at Lehigh,
e.g. J. D. Landes and R. Bucci, provided a much improved understanding of environmental effects.
Also, early attempts to understand cumulative damage effects of variable loading were quite
hopeless prior to Elbers [6] observation of crack closure effects. R. W. Hertzberg should also be
cited here for this multiplicity of clarifying many effects through careful observation of clever
experimentation. The point has been made that by the end of the 1960s we knew a great deal
about applying Fracture Mechanics to fatigue crack growth phenomena. However, my sincere
apologies to the many great contributors unmentioned here, e.g. J. Schijve and co-workers in the
Netherlands, etc.

MODELLING OF THE CRACK EXTENSION PROCESS

During those first 10 years, it was abundantly clear that the objective of crack growth rate
studies was to develop methods of predicting life. One approach to attempting to improve such
methods was to formulate physicalmathematical models of the crack growth process upon which
computations could be based at least partially to improve both understanding and the results of
computations. Many models were tried by various people using macro-mechanical to micro-
dislocation approaches without much success. My impression remains that almost all attempts
were made with limited data which did not agree with the main trends of extensive data. Not
meaning to single out any individual or approach, one study claimed a deductive proof that the
crack growth increment must be proportional to the crack tip plastic zone size, i.e. proportional
to K2. However, taking the broad range of crack growth rate data on, e.g. 7075-T6 aluminium
alloy from 10 exp7 to 10 exp2 inches per cycle showed closer to a fourth power of K relationship.
Upon noting that the fact in my own dissertation, the power law, sometimes called the Paris law
is one of the most missused things in making life predictions. Please spare me that notoriety!
Indeed, as a crack grows in size its growth rate normally accelerates rapidly, and if so, life results
exhibit an extreme sensitivity to initial conditions, e.g. the initial growth rate. Therefore, it is
essential to accurately assess the initial local crack growth rates in accurate life computations.
Consequently, laws which fit the broad trends of data, rather than the local initial rates for each
computation, are poor choices for life computations. The broad picture models are only appropriate
for other purposes, e.g. a better understanding of the physics of crack growth.
Now, back in about 1962, Anderson noted that, taking the data for all the base materials and
their various alloys of iron (and steel ), aluminium, titanium, magnesium, zirconium and some
others, then, the measured fatigue crack growth rates (in inert environments) are the same for
each, if the K value is normalized by dividing by E, the elastic modulus. It was only later noted
by Harris [7] that the growth rates in beryllium show that normalizing with E gives this result,
rather than normalizing with density. With every physical model proposed to date, nothing has
shown better results for the comprehensive data than this simple normalization taking DK over
E. Pondering on this fact over the years has led to stating that perhaps this should be explained
with some reasoned physical model before anyone claims a correct model or more detailed effects!
Someone could get famous if they would simply explain that to us!
So it is submitted that physical modelling of fatigue crack growth remains uniquely unsuccessful
for all of these 40 years. But that should not worry us too much, since Fracture Mechanics analysis
methods allow the prediction of growth rates of structural cracks from simple lab test configur-

1998 Blackwell Science Ltd Fatigue & Fracture of Engineering Materials & Structures, 21, 535540
538 P. C. P

ations. It saves us from being required to have an understanding of the detailed micro-physics of
crack growth!
The recent works of many, e.g. Professors S. Sastry, W. Soboyejo and T. Srivatsan, have begun
to explain special microstructural effects, so modelling should not be dismissed with too much
haste. It is at least simply evident that room for further progress remains in this area.
In addition, it seems appropriate to acknowledge fatigue crack growth threshold effects herein,
though discussion of it began only in the late 1960s [8]. The more recent exposition and explanation
of threshold behaviour, including small crack effects, in the works of Professors R. Ritchie [9]
and S. Suresh [10] seem especially relevant. Again, the point is made that initial conditions such
as these are especially dominant in proper structural life estimates.

CRACK SHAPE EFFECTS IN LIFE ESTIMATES

It has already been pointed out that life estimates are most sensitive to initial conditions affecting
the growth rates. This includes a proper assessment of the stress intensity, K, for the real initial
flaw, and its initial growth and changes in shape. This dominance of the initial conditions on the
life originally led C. F. Tiffany (about 1960) to base life estimates only on the initial K level applied
to surface flaws in pressure vessel walls subject to pure tension. For his limited application, the
method worked very well, and for some time became the preferred method for analysis of cyclic
life of space vehicle tankage. It also was suitable for assessing environmental influences on the life
of such flaws. However, the application of Fracture Mechanics methods to more complex structural
situations requires more sophisticated methods. The detailed flaw shape and local structural shape,
as well as loading, must be included in determining the initial K values at various points on the
crack border in order to accurately determine initial growth rates and the progression of initial
changes in shape. All too often, it is seen in practice that an estimate of life is made using formulas
for K for the perfect semi-ellipse where the real initial shape deviates significantly from that shape.
The developers of these erroneous life estimates seem to draw justification from the fact that the
literature solutions for K, both analytical and numerical, are based on the perfect semi-elliptical
shape. Perhaps they may justify this assumption as material scientists by a lack of being provided
solutions for imperfect semi-ellipses. Tada et al. in his Stress Analysis of Cracks Handbook [11],
provides the tools for estimating K values more accurately for imperfect shapes, but few seem to
have discovered that these tools are present. The next edition of that handbook will be published
by A.S.M.E. Press in early 1998, and as his co-author, it seems clear that it would be appropriate
to add a section on how to estimate errors and corrections in K values for various imperfect shapes.
A dramatic example of the need for such a correction occurred in the failure of a helicopter
power transmission gear at a lack of fusion imperfection in the web of the gear at an electron
beam weld. The initial lack of fusion region grew after surface shot peening into a long narrow
surface flaw whose ends doubled back to provide a surface length of only about half of the internal
length. Moreover, the web was subject principally to bending. The observed growth after failure
was a lengthening of only the interior part of the flaw without deepening or increasing its surface
length substantially. An analysis of the K levels at the interior ends of this flaw showed a K level
below the threshold for the ends that grew. However, the residual loading due to shot peening
obviously negated the crack closure effects at these crack ends permitting small crack behaviour
of growth below the threshold. Now this was a case involving other complications, but it does
clearly show the need for special analysis of non-semi-elliptical surface flaws.

Fatigue & Fracture of Engineering Materials & Structures, 21, 535540 1998 Blackwell Science Ltd
History of fracture mechanics and fatigue 539

CUMULATIVE DAMAGE EFFECTS IN LIFE ESTIMATES


Being aware of so-called overload delay effects in the early 1960s, early efforts to apply Fracture
Mechanics to life predictions simply assumed that neglecting load interaction effects would result
in conservative estimates of life. It is, however, embarrassing to acknowledge that in my own
dissertation, a crude and naive attempt to discuss a model exists as evidence that we had no idea
on how to properly approach that problem. Previous SN approaches to fatigue were of course
no better off with the Miners Law, which also neglected completely any load interaction effects.
In those days, several researchers showed data indicating damage coefficients differing from one
for variable loading and gave no better than subjective guidelines for guessing how much the
coefficient would differ.
Upon meeting with W. Elber in 1968, and realizing he had made a very important discovery of
crack closure in fatigue, it was really readily apparent why earlier attempts at models of load
interaction effects had failed. On the other hand, it did not become apparent how to formulate
such models. Nevertheless, in this same period, our observations of a fatigue crack growth threshold
were explained and shown to be strongly related to closure by work such as that of Schmidt [12].
Later work at Brown University by Hermann [13] with extremely sensitive clip gauges
(10 exp6 inch) demonstrated that the increments of crack advance are related to the K range for
the crack-opened part of any load excursion independent of past history. However, the prediction
of the level of crack opening loads for variable loading was and is still apparently lacking. It is
obviously a challenging problem and further work is needed in this area.
The F-111 experience with attempting to predict life after proof testing showed clearly the need
to have an accurate model for computational purposes in predicting life. The model, devised
specifically for the F-111 case, showed that semi-empirical models could be devised for particular
circumstances. Subsequent models that have come to my attention all retain that empirical
character, though they are obviously improved for wider applicability. It appears they all need to
be finely tuned to any new application.

STATISTICAL EFFECTS VERSUS BOUNDING IN LIFE ESTIMATES


The blanket statement that statistical effects should be considered is foreign to engineering
training in most disciplines, and is often regarded as unnecessary if a conservative design estimate
will do the job. It is therefore good to have a few examples where statistics can show that the
deterministic approach leads to an incorrect judgement of the situation. Such was the case for the
life consideration and inspectability of the old C-5A wing splice joints.
The original C-5A wing was fabricated from several span-wise planks of aluminium alloy with
overlapping span-wise joints that were fastened by a span-wise row of taperlock bolts. The basic
justification for the Lockhead/Air Force plan to replace these wings was the apparent supposition
that the fatigue life of these joints was too limited. The other justification was lack of inspectability
to anticipate and avert possible fatigue failure. The overlap of the joints was 3/4 of an inch and it
was argued that a crack could develop under the splice and become 3/4 inches long before the
possibility of being detected. Furthermore, it was argued that such a 3/4 inch crack could not
sustain a limit load condition without failure. On the other hand, if a crack just a bit larger were
present, then it could easily be detected by a moving eddy current probe by simply running it
down the splice joint. Now, limit load is a deterministic legislated load as a minimum design load
for the wing. However, if one takes the actual statistical flight loads that the aircraft experienced
in flight, the probability of any aircraft seeing a limit load was less than once in the life of the
whole fleet. So, some years ago, in the early 1980s, J. Gebman of the Rand Corporation [14] took

1998 Blackwell Science Ltd Fatigue & Fracture of Engineering Materials & Structures, 21, 535540
540 P. C. P

the actual statistical flight spectrum along with the established rates of growth of the span-wise
splice cracks and simply calculated the probable life and size of cracks that would actually occur
at failure. It turned out that the 50/50 probable size of cracks at failure were over 3 inches long.
Cracks that were 2 inches long had a survival probability of more than 95%, and for those of
1.5 inches more than 99%. Moreover, the flight time to grow from an inspectable size to beyond
1.5 inches was several hundreds of flight hours. That means that, if there was a flight safety
problem, the span-wise splices were easily inspectable at conveniently spaced intervals considering
other inspection requirements to detect any possible cracks. The lesson here is that the deterministic
analysis did not just give a vague answer to the problem, it gave the wrong answer! On the other
hand, the Air Force never did those inspections because at the time they were presenting arguments
to the congress that the C-5A needed new wings and the inspections would have weakened their
arguments. So you and I bought them new wings at only about $10.00 per citizen (1980 dollars).
The above example is one which clearly shows that some statistics applied to the right problems
can often do more than just clarify the answers to our engineering calculations. We should try to
make use of the statistical approach in any case which has wide scatter such as the growth rates
of fatigue cracks. However, when calculating crack growth life, the deterministic components of
the calculations also have errors, e.g. those in stress intensity, K formulas, discussed earlier, and
for those components of the calculations techniques where bounding the errors may be more
fruitful than trying to regard them as statistical. Furthermore, on some occasions the statistical
approach may get so complex that extending the bounding approach to some statistical variables,
though simplistic, may clarify the results. Let us keep an open mind to such matters in order to
tailor our life calculation methods to suit the particular problem at hand. It is hoped that the
examples provided in this discussion have demonstrated that differing approaches are advantageous
for the particular circumstances of interest.

REFERENCES
1. P. C. Paris (1957) The mechanics of fracture propagation and solutions to fracture arrester problems.
The Boeing Company Document D-2-2195.
2. N. E. Frost and D. S. Dugdale (1958) The propagation of fatigue cracks in sheet specimens. J. Mech.
Phys. Solids 6, 92110.
3. A. K. Head (1956) The propagation of fatigue cracks. J. Appl. Mech., ASME 23, 406.
4. A. J. McEvilly and W. Illg (1958) The rate of fatigue-crack propagation in two aluminum alloys. NACA
TN 4394.
5. P. C. Paris, M. P. Gomez and W. E. Anderson (1961) A rational analytic theory of fatigue. T he T rend
in Engineering at the University of Washington, 13, 1, 914.
6. W. Elber (1971) The significance of crack closure. In: Damage T olerance in Aircraft Structures, AST M-
ST P 486, 230242.
7. P. C. Paris and D. O. Harris (1970) An engineering evaluation of the status of utilization of beryllium
from the viewpoint of fracture mechanics. In: Proceedings of the Beryllium Conference: A Report of the
National Materials Advisory Board, Vol. 1, NMAB-272, NAS, Washington, D.C. 537568.
8. P. C. Paris (1970) Testing for very slow growth of fatigue cracks. Closed L oop Magazine (MTS Systems
Corp.) 2(5).
9. R. O. Ritchie (1982) Environmental effects on near threshold fatigue crack propagation. In: Fatigue
T hresholds, EMAS, Vol. 1.
10. S. Suresh and R. Ritchie (1984) Propagation of short fatigue cracks. Int. Metals Rev. 29, 445476.
11. H. Tada, P. C. Paris and G. R. Irwin (1973) T he Stress Analysis of Cracks Handbook, Del Research
Corporation, 2nd Edn 1985 (3rd Edn to be published by A.S.M.E. Press, 1998).
12. R. A. Schmidt and P. C. Paris (1973) Threshold for fatigue crack propagation and the effects of load
ratio and frequency. AST M-ST P 536.
13. P. C. Paris and L. Hermann (1982) Fatigue T hresholds, EMAS.
14. J. Gebman and P. C. Paris (1978) Probability that a damaged element will not cause structural failure:
basic method and example. The Rand Corp., R-2238-RC.

Fatigue & Fracture of Engineering Materials & Structures, 21, 535540 1998 Blackwell Science Ltd

Вам также может понравиться