GUIDELINE NO.

GD-AP-2304
PAGE 1 OF 6
October 1995

PREFERRED
RELIABILITY
FRACTURE MECHANICS
PRACTICES RELIABILITY

Guideline:

Use a fracture mechanics formulation to estimate the design fatigue life and reliability of metallic
or ceramic structural and mechanical components subject to fluctuating stress. For reusable
spacecraft, update the reliability analysis based on in-service inspection and repair data.

Benefit:

Consideration of fracture mechanics reliability during the design process can assist in the
prevention of failures of structural and mechanical components subject to fluctuating loads in
service. Explicit consideration of the reliability of structural and mechanical components provides
the means to evaluate alternate designs and to ensure that specified risk levels are met.
Probabilistic fracture mechanics analyses may also be applied to life extension of existing
structures, and for problem assessment of in-service fatigue failures.

Potential applications of this method to the Space Shuttle or Space Station include: landing gear,
control surfaces, main engine components, auxiliary power unit components, external tank and
solid rocket booster welds, pressure vessels, propulsion modules, and logistics modules. The
method is also applicable to reusable, Shuttle launched payloads or spacecraft such as Spacelab,
Spacehab, EURECA, SPAS and Spartan.

Stochastic fracture mechanics analysis provides the basis for analysis consistency between
reliability analysis of mechanical systems, such as reliability block diagram analysis, and traditional
deterministic fracture mechanics safe life estimation.

Center to Contact for More Information:

Johnson Space Center (JSC)

Implementation Method:

The method outlined below is a stochastic elastic fracture mechanics approach (for metallic or
ceramic materials) which neglects any crack retardation or acceleration effects. Composites and
other materials with insufficient crack growth data or intractable flaw growth characteristics are
not considered. The detailed development of this approach is essentially the same as that given in
references 1 through 6. The purpose of the simplified approach described herein
is to illustrate some of the advantages and typical results of a stochastic fracture
mechanics analysis. Discussion of technical implementation details, such as the
use of crack growth laws other than the Paris equation (for example, the modified
Forman equation [15]), follows in the technical rationale section.
 GUIDELINE NO. GD-AP-2304
PAGE 2 OF 6
FRACTURE MECHANICS RELIABILITY

For illustration, a Paris crack growth law is assumed, which may be written in the form of a
differential equation. For random applied stress processes, a solution can be written in the form of
a limit state function, M, as:

ac

M ' da
&C <%0 (T&T0)E[S m]
ma Y(a)m( Ba)m (1)
0

in which:
a = crack length
a0 = a(t=0) = initial crack size
ac = critical crack size at which failure occurs
no+T = number of stress cycles in time T at average frequency
no+To = time to crack initiation
C, m = material constants in the Paris equation
Y(a) = geometry function for physical problem under consideration
S = far-field stress range
E[ ] = expected value (mathematical expectation)

Failure is defined to occur when the critical crack length, ac , is exceeded, so that at failure M # 0.
The probability of failure is then the probability that the limit state function is equal to or less than
zero:
Pf = P[M#0] (2)

The first term of Equation 1, the integral, essentially defines the fatigue resistance of the structure
against a crack growing from an initial size, ao, to critical size ac . This integral must be evaluated
numerically in all but the simplest of cases.

Particular forms of the geometry function, Y(a), are available for simple configurations in the
literature or from general purpose fracture mechanics software packages. For unique structural
details, other approaches are available to determine Y(a), such as detailed finite element modeling of
the cracked structure. The second term of Equation 1 defines the accumulated "damage" caused by
the applied stress process. A random stress process is characterized by its power spectral density and
may be described as being narrowband (slowly varying random) or as wideband. In either case closed
form approximations for the second term of Equation 1 are available. If the stress process is
deterministic or if time histories of the stress process are available time domain methods, such as
rainflow cycle identification [10], approximations are available for determining the factors in the
second term of Equation 1. It should be noted that all of the terms in Equation 1 may be treated as
random or uncertain. This enables the modeling of all the sources of uncertainty pertinent to the
problem, such as crack size and location, scatter in crack growth data, etc. Subsequent sensitivity
analyses can be used to determine which variables contribute the most to the fatigue life uncertainty
and require treatment as random, and which variables may be considered as fixed (deterministic).
 GUIDELINE NO. GD-AP-2304
PAGE 3 OF 6
FRACTURE MECHANICS RELIABILITY

Sensitivity analysis can also indicate the parameters for which further data collection could reduce
the overall uncertainty in the fatigue life.

Modern reliability methods, the so-called First-Order Reliability Method (FORM) or Second-Order
Reliability Method (SORM), are available in commercial computer programs to solve Equation 2.
Monte Carlo or more sophisticated simulation techniques are also available [1] [7] [8] [9]. In
particular, PROBAN [1] has been available commercially from Det Norske Veritas (Oslo, Norway)
since 1986 and has been used extensively in the offshore oil industry. PROBAN is available for
UNIX and VAX/VMS based workstations. STRUREL is a PC/Windows based application available
from RCP, GmBH, of Munich, Germany. RELACS is a similar package available from REA, Inc.,
of Golden, Colorado. A NASA-funded application called NESSUS, which runs on UNIX
workstations and mainframes, is available from the Lewis Research Center. The commercial codes
are recommended because of better user-interfaces and better user support. Monte Carlo approaches
generally require direct programming for the solution for the specific problem under study.

For a structure or mechanism in service, the results of inspections may be incorporated into the
analysis and the estimated failure probabilities updated to show the change in reliability based on the
additional information on existing crack size. For each inspection, two outcomes are possible: either
no crack is detected, or a crack is detected and its size or length is measured. Figure 1 is an example
analysis result (from reference [1]) showing reliability as a function of time for which inspections
were assumed at 10 and 20 years, with no crack detected at 10 years, but a 4.0mm crack detected
at 20 years. Note that with the new information gained from inspection at t=10 years, the reliability
is shown to increase as no crack was found. After inspection at t=20 years, reliability is also shown
to increase even though a small crack was detected, but only for a short time. The increase at t=20
years can be attributed to the discovery that the crack length was less than the critical crack length
for the structure, but it should also be noted that the reliability decreases much faster with the
uncertainty tied to the presence of a flaw. Inspection of structures and the new information that is
gained can essentially reset the reliability, and even though a crack may be discovered, this new
information can lead to increased inspections, which can lengthen the life of the item. However, a
crack detection generally decreases the reliability, as in this case after about t=22 years.

Technical Rationale:

A stochastic fracture mechanics approach to fatigue gives an estimate of the reliability of structural
and mechanical components as a function of time in service and allows the reliability estimate to be
updated if the results of in-service inspections and/or in-service load data are available [1] [2]. The
procedure may also be used to optimally schedule inspections, and to compare the adequacy of
different inspection types or quality levels [1] [2]. Type and quality of repair techniques may also be
compared and selected to maintain a desired reliability level. Updating of these analyses as actual
inspections or repairs occur is also possible [2] [3]. The application of in-service reliability estimates
is dependent on the availability of some form of flight load data and accessibility to the structure or
mechanism for inspection. Without such data no updating of the initial design reliability analysis is
possible.
 GUIDELINE NO. GD-AP-2304
PAGE 4 OF 6
FRACTURE MECHANICS RELIABILITY

Figure 1. Reliability vs. Time, Inspections at 10 and 20 Years

The primary intent of this guideline is to make available to the NASA reliability engineering discipline
the engineering mechanics-based methods for estimating the reliability of structural and mechanical
components. Use of such methods would allow for consistency of models and data between the
reliability and structural/mechanical engineering disciplines within NASA. A stochastic fracture
mechanics approach provides a "physics-of-failure" basis for estimating the reliability of components
subject to fatigue and fracture. This enables fault tree or reliability block diagram analyses, or
probabilistic risk assessments, for structural and mechanical components in spacecraft systems to be
performed using the same data and engineering mechanics models as the NASA accepted,
deterministic fracture analysis procedures [14]. For example, the mean time to failure for a pressure
vessel girth weld (as may be needed for a propulsion system reliability block diagram analysis)
estimated using probabilistic fracture mechanics would be rationally consistent with the safe-life
analysis performed to meet current NASA fracture control and safe-life analysis requirements.

The approach outlined herein is a simplified formulation that neglects load interaction effects, such
as retardation, by using the Paris crack growth law. If a Paris equation is properly fit to the basic
crack growth data, the resulting deterministic safe life estimate will be more conservative (shorter)
than the estimated life resulting from a fracture analysis which incorporates retardation. The current
NASA accepted practice in fracture mechanics analysis uses the computer program NASA/FLAGRO,
which includes interaction effects. Load interaction effects may be included in a stochastic fracture
mechanics analysis by using crack growth laws such as the modified Forman equation found in
NASA/FLAGRO. For example, a FORM/SORM formulation which included the modified Forman
equation has been used to study aircraft durability and damage tolerance [14]. A direct Monte Carlo
solution implementing the modified Forman equation may also be found in references 7, 8, and 9.
Issues related to the implementation, applicability, and accuracy of FORM/SORM and Monte Carlo
methods are beyond the scope of this guideline. In general, if a fracture mechanics reliability analysis
 GUIDELINE NO. GD-AP-2304
PAGE 5 OF 6
FRACTURE MECHANICS RELIABILITY

is to be performed and it has been determined from a preliminary deterministic fracture analysis that
the Paris formulation is inadequate, the modified Forman equation (as in NASA/FLAGRO) should
be used following either the approach described in reference 14 or in references 7, 8, and 9.

Two critical parameters in any fracture analysis, deterministic or stochastic, are the size, location and
distribution of initial flaws or cracks, and the ability of nondestructive evaluation (NDE) techniques
to detect a flaw or crack smaller than a certain size. NASA has established standard NDE flaw sizes
for Space Transportation System (STS) payloads [14]. Two recent NASA research projects, one
directed at establishing NDE probability of detection (POD) data, and one directed at gathering initial
flaw distribution data, also may provide additional data for modeling initial flaws and NDE quality.

Deterministic fracture analysis practice uses relatively large safety or scatter factors to account for
the many inherent sources of uncertainty or error, such as analytic model inadequacies, inaccuracy
of stress intensity predictions, and the scatter of experimental crack growth data. Stochastic analysis
methods extend the accepted deterministic methods by allowing (or forcing) the analyst to explicitly
account for these uncertainties by treating them as random variables (or process or fields), requiring
the analyst to consider the likely range and distribution of the parameters. Both the deterministic and
the stochastic analysis will suffer from the same shortcomings of model inadequacy, etc. The
stochastic model has the advantage of addressing the uncertainties specifically using probability and
statistical theory, while the deterministic approach addresses uncertainty in a general manner through
the use of the safety or scatter factor. Use of a stochastic approach and a reliability based design
criteria can be beneficial in avoiding over- or under-conservatism that may result from the use of a
deterministic safety factor approach.

Impact of Nonpractice:

Nonpractice of deterministic fracture mechanics analysis (or test) in a spacecraft or launch vehicle
program would result in the failure to meet NASA design requirements for fracture control [11] [12].
Failure to adequately consider fatigue in the design of structural and mechanical components subject
to fluctuating loads may, at best, result in recurring costs for repair or replacement of components
before their intended design life. At worst, a catastrophic failure, in terms of economic loss and loss
of life, may result.

It is standard JSC engineering practice to specify fracture control requirements on structural and
mechanical components that are classified as fracture critical. As a part of these requirements, it is
required that a component be shown by analysis or test to be able to survive a deterministic minimum
number of service lifetimes (usually 4) without the largest possible undetectable flaw growing to
failure [11] [12]. While this process has proven to be adequate to ensure against failure, design
practices which do not take advantage of stochastic methods for fracture analysis may produce over-
conservative and uneconomical structures and mechanisms.

In mechanical or structural system reliability analyses, failure to use a stochastic fatigue or fracture
mechanics approach to the estimation of reliability will result in life estimates that are incompatible
 GUIDELINE NO. GD-AP-2304
PAGE 6 OF 6
FRACTURE MECHANICS RELIABILITY

and inconsistent with the results of the standard NASA deterministic safe-life analyses required by
the Space Shuttle and Space Station Programs [11] [12].

Related Guidelines:

Guideline GD-AP-2303: Spectral Fatigue Reliability

References:

[1] PROBAN-2: Example Manual, Report No. 89-2025, A.S Veritas Research, Hovik, Norway, August, 1989
[2] Madsen, H.O., Skjong, R.K., and Kirkemo, F., "Probabilistic Fatigue Analysis of Offshore Structures - Reliability
Updating Through Inspection Results", A.S Veritas Research, Hovik, Norway
[3] Madsen, H.O., Skjong, R.K., Tallin, A.G., and Kirkemo, F., "Probabilistic Fatigue Crack Growth Analysis of Offshore
Structures, with Reliability Updating Through Inspection", Society of Naval Architects and Marine Engineers, Proceedings,
Marine Structural Reliability Symposium, Arlington, VA, October 5-6, 1987
[4] Wirsching, P.H., Torng, T.Y., and Martin, W.S., "Advanced Fatigue Reliability Analysis", International Journal of
Fatigue, Vol. 13, No. 5, 1991, pp. 389-394
[5] Wu, Y.-T., Burnside, O.H., and Dominguez, J., "Efficient Probabilistic Fracture Mechanics Analysis", Numerical
Methods in Fracture Mechanics: Proceedings of the Fourth International Conference, Luxmoore, A.R., Owen, D.R.J.,
Rajapakse, Y.P.S., and Kanninen, M.F., Eds., Pineridge Press, Swansea, U.K., 1987
[6] Veers, P.S., Winterstein, S.R., Nelson, D.V., and Cornell, C.A., "Variable-Amplitude Load Models for Fatigue Damage
and Crack Growth", Development of Fatigue Loading Spectra, STP 1006, American Society for Testing and Materials, 1989
[7] Sutharshana, S., et al., A Probabilistic Fracture Mechanics Approach for Structural Reliability Assessment of Space
Flight Systems, Advances in Fatigue Lifetime Predictive Techniques, STP 1122, American Society for Testing and
Materials, 1991
[8] Sutharshana, S., et al., Computational Methods for Probabilistic Flaw Propagation Analyses, Proceedings of the ASCE
Structures Congress 91, April, 1991
[9] Moore, N. R., et al., "An Improved Approach for Flight Readiness Certification - Probabilistic Models for Flaw
Propagation and Turbine Blade Failure", JPL Publication 92-32, December, 1992.
[10] Matsuishi, M., and Endo, T., "Fatigue of Metals Subjected to Varying Stress", Presented to the Japan Society of
Mechanical Engineers, Fukuoka, Japan, March, 1968.
[11] Fracture Control Requirements for Payloads Using the National Space Transportation System, NHB 8071.1A
[12] Fracture Control Requirements for Space Station, SSP 30558
[13] Sigurdsson, G., et al., "Probabilistic Methods for Durability and Damage Tolerance Analysis", presented at the 1992
USAF Aircraft Structural Integrity Program Conference (ASIP), San Antonio, Texas, December 1-3, 1992.
[14] Fatigue Crack Growth Computer Program NASA/FLAGRO Version 2.0, JSC-22267A (Draft), January, 1993.