Professional Documents
Culture Documents
2 / Vol. 128, FEBRUARY 2006 Copyright © 2006 by ASME Transactions of the ASME
Downloaded 10 Oct 2007 to 203.78.221.48. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm
Fig. 1 Cycles to reach 0.5 mm crack in plate and specimen by
test
关3兴兲. The accumulated usage factor is then calculated over all the
individual stress-strain cycles of the loading block, following the
Palmgren-Miner rule 共关3兴, Chap. 9兲. The same applies to cases in
which more than one loading block may be applied in a random
sequence, each repeated a specified number of times. In the re-
mainder of the paper, the analyses will be assumed applied to one Fig. 3 Monotonic „lines only… and cyclic „markers… curves for
stress-strain cycle. SA-516 Grade 70 steel. L denotes longitudinal and T denotes
transverse orientation of specimens machined from a plate.
„Reprinted from Fig. 2 of †6‡, Copyright 1984, with permission
4 Stabilized Cycle from Elsevier.…
As part of the design procedure, the local strain approach as-
sumes that a single value of a strain range is used for assessing
fatigue damage for the life of a pressure vessel component. Since Substitution of a monotonic curve for the cyclic curve may
hardening and softening with cycles accompany the initial phase cause problems. For example, Lefebre and Ellyin 关6兴 present
of cycling, during which the strain and stress ranges may change, curve fittings to test data on specimens made of SA-516 Grade 70
the question is: What strain range shall it be? Since for many steel, shown in Fig. 3. Stress amplitude is plotted versus strain
metals the stress and strain ranges tend to stabilize, so that stabi- amplitude for cyclic loading and compared to plots for monotonic
lized hysteresis loops are experienced for the major part of life, loading. The material shows softening with cycles up to strain
the obvious answer is to bypass the hardening and softening with amplitude of about 0.4% and hardening above that level. Prob-
cycles and to accept the strain range of the stabilized cycle as lems may arise within strain amplitudes of 0.1–0.4%, where the
representative of whole life. In some cases, stabilization may be use of the monotonic curve can predict strain ranges in a compo-
difficult to achieve even until failure. Such cases notwithstanding, nent that err on the unconservative side.
the local strain approach assumes that the cyclic action from Of course, an accurate curve fitting to the cyclic data of the
which the cyclic stress-strain curve of the material is derived has material under consideration is preferable, if one is available.
stabilized. That has to be accepted as part of the design procedure. However, for design purposes, approximations could be agreed on
for certain classes of materials, which would parallel those of the
5 Cyclic Material Curve design fatigue curves now used in the ASME B&PV Code 关4兴.
Having established that stabilized action is the target, it follows
that the cyclic-stress-range–strain-range 共or amplitude兲 curve of 6 Applicability
the material provides the information that is needed for the mate- The local strain approach is applicable to cases in which all
rial model used in the analysis. Such cyclic test data are available structural features that affect fatigue damage are defined and can
for many materials. Typical curve fittings to these data can be be modeled with sufficient accuracy. It is not applicable to cases
obtained in terms of three parameters: cyclic elastic modulus E, a in which some structural detail is known to affect fatigue damage
stress parameter, and an exponent. but cannot be modeled, either because its geometry is unknown
共e.g., flaws at the weld toe of an untreated weld兲 or because its
model is unreliable 共e.g., very sharp notch兲. Such cases require
approaches that incorporate the unmodelable details in the test
data, such as, for example, those described by Maddox 关7兴 for
weld joint classes, and more recently by Dong et al. 关8兴.
Limitations on loading are not so clear. Proportional loading
presents no problems, but cases when the principal stress and
strain axes rotate have been shown to pose a problem. Itoh et al.
关9兴 present test data for shear and axial strains that are imposed
nonproportionally to the test section of a thin cylindrical shell,
forcing the principal axes to rotate. The data show unsatisfactory
correlation with predictions using the principal and equivalent
strain ranges that are in the current ASME B&PV Code 关4兴.
Kalnins 关10兴 has shown that the hysteresis loops for some of the
nonproportional cases 共e.g., case 10 in 关9兴兲 exhibit no elastic un-
loading and the reversal points of different components do not
Fig. 2 Cycles to reach failure in plate and specimen by test coincide, which prevents the use of the methods of Sec. 7. It may
Downloaded 10 Oct 2007 to 203.78.221.48. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm
Fig. 5 Stabilized hysteresis loop
Fig. 4 Cyclic curve and calculated ranges
stabilize in the same way. The software that is used for the cycle-
be that such situations are rare in pressure vessel components. by-cycle method must reflect this behavior. This means that it
Whether a caveat for these cases is or is not needed in design must be able to replicate the cyclic curve for a uniaxial stress state
standards is an open question. in a component. In other words, the calculated stress and plastic
It is assumed for the remainder of the paper that stress and strain ranges must lie on the cyclic-stress-range–plastic-strain-
strain reversal points of all nontrivial components coincide so that range curve that has been input. This requirement will be used as
a multiaxial equivalent stress-strain cycle can be defined 共Sec. a validity check in evaluating the cyclic plasticity models of the
9.2兲. This is ensured for proportional loading. The conditions for software.
which it may also be true for nonproportional loading require To illustrate the validity check, consider a uniaxial stress state
further investigation. cycled in strain control using a cyclic plasticity model that is to be
checked for consistency with the local strain approach. The cyclic
7 Methods and Software curve shown in Fig. 4 共rewritten in amplitudes兲 is input for the
As per Sec. 3, the analysis is applied to each individual stress- monotonic material model. The calculated stabilized hysteresis
strain cycle that is produced by the loading histogram. These loop is shown Fig. 5. It shows a plastic strain range of 0.033 and
cycles must be identified, and the two time points at their stress a stress range of 1330 MPa. Is the cyclic plasticity model consis-
and strain reversals determined. The loading components at the tent with the local strain approach? This is decided by plotting the
reversal points can then be evaluated and the loading range for the coordinates for the calculated ranges, 0.033 and 1330, in Fig. 4. It
cycle determined. is seen that this point lies on the cyclic curve. If the cycling were
At this point, it has been established that the cyclic curve of the done at different strain ranges, the cyclic curve in Fig. 4 would be
material will be used to model the material and the loading will duplicated. Therefore, the cyclic plasticity model used in this
consist of the loading range. Now the question is: What methods analysis is consistent with the local strain approach.
and software will solve the problem in a way that is consistent 7.1.2 Linear Hardening in Cyclic Plasticity. Cyclic plasticity
with the local strain approach? Two basic methods are discussed: models with linear hardening involve two separate components:
cycle-by-cycle and half-cycle methods. For the latter, the twice- isotropic and kinematic. The model with isotropic hardening ex-
yield and Seeger’s methods are included. The cycle-by-cycle pands the yield surface until purely elastic action remains, as
method is discussed only because of its appeal for modeling cyclic shown in Fig. 6. This result does not meet the validity check of
action, but, when used for design purposes, it requires far more Sec. 7.1.1 and is not acceptable. The kinematic component is con-
effort and is less generic to software than the half-cycle methods. sidered next.
The stress and strain ranges obtained by all three methods are the Cyclic plasticity models with linear kinematic hardening have
same. been developed that assume Masing behavior 共Sec. 2.5 of 关14兴兲,
7.1 Cycle-by-Cycle Method. Elastic-plastic finite element according to which magnifying the cyclic stress-strain amplitude
analysis 共FEA兲 is performed over a sufficient number of repeti- curve by a factor of 2 approximates the two branches of a stabi-
tions of a selected cycle until the stress and strain values at the lized hysteresis loop. These models pass the validity check of Sec.
reversal points stabilize. The cyclic-stress-amplitude–strain- 7.1.1. Among the popular finite element programs, ANSYS 关15兴
amplitude curve of the material is used as input for the monotonic linear kinematic hardening model KINH supports a multilinear
uniaxial material model. The loading can be specified as either
between the loading at the reversal points of the cycle or between
plus and minus of the loading amplitude. If in the former case the
hysteresis loop indicates a mean stress, its effect is neglected as
per Sec. 8. Since up- and downloading is performed, a cyclic
plasticity model must be specified to model the unloading and
reloading phases. Two cyclic plasticity models will be considered
for the cycle-by-cycle method. One includes linear and the other
nonlinear hardening. 共For details, see the ABAQUS 关11兴 Standard
User’s Manual II, 11.2.2-2, or Refs. 关12,13兴兲. The question is
which cyclic plasticity model is consistent with the local strain
approach. To answer that question, a validity check is given next.
7.1.1 Cyclic Plasticity Validity Check. According to Sec. 4,
the cyclic stress-range–strain-range 共or amplitude兲 curve is used,
which means that the hysteresis loops in the specimens stabilize.
Since the material of the component is supposed to be the same as
that of the test specimens, the loops in the components should also Fig. 6 Stabilized cycle using linear isotropic hardening model
Downloaded 10 Oct 2007 to 203.78.221.48. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm
Fig. 7 Stabilized cycle using linear kinematic hardening model
共curved兲 curve fitting to the cyclic data for the input of the mono- Fig. 8 Cyclic curve and calculated ranges using NLK model
tonic material model, while that of ABAQUS 关11兴 Version 6.3-1
supports only a bilinear curvefit to the cyclic data when the pa-
rameter “hardening= kinematic” is invoked. 共Note that ABAQUS with no unloading and reloading, and do not require cyclic plas-
permits the input of a curved curve fitting to the cyclic data for the ticity models. The advantage is simplicity 共no FEA over cycles兲
input of the monotonic material model when no cyclic plasticity and that they can be performed with any finite element program
with kinematic hardening is specified, which is the case for the that has an incremental plasticity option for static loading. The
half-cycle methods.兲 half-cycle methods give strain and stress ranges that, for practical
For an illustration, consider a single eight-noded brick element, purposes, are the same as those obtained by the cycle-by-cycle
cycled in uniaxial strain control between a strain of 0.03 and method of Sec. 7.1. The two half-cycle methods are considered
−0.01. Figure 7 shows1 the stress-strain response for which the next.
stress range of 1330 MPa is obtained. When cycled with the same
strain range, but fully reversed between 0.02 and −0.02, exactly 7.2.1 Twice-Yield Method. Theoretical support of this method
the same stress range is predicted. In both calculations, the cyclic can be found in the work of Mroz 关16兴. Dowling 关17兴 and Dowl-
curve in Fig. 4 共rewritten in amplitudes兲 is used. The square ing and Wilson 关18兴 applied it to some special cases. More re-
marker in Fig. 4 shows the point with the coordinates of the cal- cently, Kalnins 关19兴 proposed it as a general method for design
culated stress range and plastic strain range. The fact that it lies on and called it the twice-yield method. It is applicable to cyclic
the cyclic curve indicates that the test of Sec. 7.1.1 has been met. primary and nonprimary 共e.g., transient thermal兲 loading; that is,
its applicability is the same as that of the cycle-by-cycle method.
7.1.3 Nonlinear Cyclic Plasticity Models. Nonlinear cyclic The only limitations are stated in Sec. 6.
plasticity models 共e.g., 关11–13,15兴兲, which contain combined From an FEA perspective, the twice-yield method is explained
isotropic/kinematic components, are not designed to receive a ge- by the observation that if in the input the load is specified as the
neric cyclic stress-strain curve of the material as input and calcu- loading range and the cyclic stress-range–strain-range curve is
late the stress and strain ranges that represent a stabilized cycle of used for the material model, then in the output the stress compo-
the same material. The problem is that the input is written for a nents are the stress component ranges and the strain components
specified strain range, which is the end product of the analysis are the strain component ranges. Thus, in one FEA load step, for
and, therefore, unknown before the analysis. For this reason, the which the loading is specified from zero to that of the loading
nonlinear hardening models used in Refs. 关12,13兴 do not meet the range, the output provides the stress and strain ranges that are
validity check of Sec. 7.1.1. needed in the local strain approach.
The following example illustrates the problem. Again a single When coupled with the multiaxial total strain range 共Sec. 9.2兲,
eight-noded brick element is subjected to fully reversed, strain- the twice-yield method is far simpler than the cycle-by-cycle
controlled cycling in one direction, producing a uniaxial stress method. After the reversal points of the loading for the cycle and
state. ABAQUS 关11兴 “data-type= stabilized” parameter is selected, the loading range have been determined, the method is straight-
for which an approximation derived from the cyclic curve shown forward. The quantities that are taken from the output are the
in Fig. 8 is used as input. No isotropic component is used. The multiaxial equivalent stress range, ⌬eq, given by Eq. 共4兲, and the
model includes only the nonlinear kinematic 共NLK兲 component.
The response is shown by the curve marked NLK in Fig. 9. It
shows clearly that the stabilized cycle of the cyclic curve that was
input has not been replicated. The stress range given by the NLK
model is 1240 MPa, while the corresponding value on the cyclic
curve is 962 MPa, which is also plotted in Fig. 8. It is clear that
the model does not replicate the cyclic curve that has been input
and fails the validity check of Sec. 7.1.1.
7.2 Half-Cycle Methods. These methods take advantage of
the stabilized form of the hysteresis loop of the cycle. There is no
need to perform the cycle-by-cycle method over a number of
cycles if the two branches of the loop are geometrically similar, as
shown in Fig. 7. FEA over just one branch of the loop gives the
desired stress and strain ranges. That is the basis of the half-cycle
methods. They require only one monotonic FEA of one load step,
1
Jürgen Rudolph of the University of Dortmund, Germany, performed the calcu-
lations for this figure using ANSYS KINH linear kinematic hardening model. Fig. 9 Calculated cycle using NLK model
Downloaded 10 Oct 2007 to 203.78.221.48. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm
equivalent plastic strain range, ⌬ peq, given by Eq. 共5兲. Typical needed here only to identify the consistent stress and strain pa-
finite element programs calculate them automatically. For ex- rameters in the local strain approach. The user of any of the meth-
ample, ABAQUS 关11兴 calls ⌬eq MISES, and ⌬ peq PEMAG. ANSYS ods discussed in Sec. 7 does not have to construct one for an
关15兴 uses similar variable names in the output. A generic output application.
file is scanned for the maximum value of ⌬ peq, and ⌬eq is then In multiaxial situations, hysteresis loops are commonly con-
recorded at the same location. No search of the solution database structed for corresponding stress and strain components sepa-
is required. The total strain range is then obtained from Eq. 共6兲 rately. This does not reveal which strain range is entered on the
using a hand calculator. material strain-life curve. Kalnins et al. 关22兴 developed the con-
cept of a multiaxial equivalent hysteresis loop and showed how to
7.2.2 Seeger’s Method. Seeger gives the general background construct one. This is discussed next.
in 关20兴. Rudolph and Weiss 关21兴 describe the procedure and dis-
cuss its application to weld seams with postweld treatment. It is 9.1 Uniaxial Stress State. To obtain a template for a multi-
applicable to proportional loading; that is, to cases in which all axial stress case, a procedure is outlined first for a uniaxial stress
loading components are multiplied by a single function of time, state in a fatigue test specimen. The calculation is performed with
say, L. the cycle-by-cycle method of Sec. 7.1 using the cyclic plasticity
Seeger’s method performs only one FEA of the component model with linear kinematic hardening.
from L = 0 to the greatest magnitude of L on the histogram and
1. Plot stress versus axial plastic strain over one stabilized
records a selected stress 共兲 and strain 共兲 measure at a number of cycle and obtain a hysteresis loop, which may look like that
L values that is sufficient to permit a curve fitting by an equation in Fig. 5.
共e.g, Ramberg-Osgood兲. The curve fitting between L and is 2. Note that its height is the uniaxial stress range ⌬, and its
called the component yield curve and that between and is width is the axial plastic strain range ⌬ p.
called the local - curve. After the two curvefits are derived, the 3. Calculate the axial total strain range from
unload branch of the hysteresis loop of each cycle is constructed
by assuming Masing behavior 共see 关14兴兲. The stress and strain ⌬
⌬t = + ⌬ p 共1兲
ranges are determined from this branch. For details, see 关20,21兴. E
Regarding the comparison between the two half-cycle methods,
where E is the modulus of the elastic portion of the cyclic
twice-yield method performs an FEA for each stress-strain cycle
curve.
separately, but determines the stress and strain ranges with no
4. Enter ⌬t as ordinate on the strain-life curve to read cycles.
postprocessing. The advantage of Seeger’s method is that the re-
5. Note that ⌬ and ⌬ p lie on the cyclic stress range-plastic
sults of a single FEA can be used for a number of stress-strain
strain range curve of the material, just like the square marker
cycles with different loading amplitudes of the same set of load-
in Fig. 4.
ing.
9.2 Multiaxial Stress State. The five steps in Sec. 9.1 are
8 Mean Stress now retraced for the multiaxial stress case.
What is known about each stabilized stress-strain cycle is only 1. Use again cycle-by-cycle method to calculate all stress 共ij兲
its loading range, which is used to calculate the stress and strain and plastic strain 共pij兲 components at a number of output points
ranges of the stabilized cycle. The loading may begin with asym- over a stabilized cycle that would be sufficient to draw a graph.
metric components at the reversal points, but once the stress-strain Then the following two quantities are calculated at each of the
cycle has stabilized, no information is available regarding the output points:
mean stress of the cycle. Since the only description of cyclic ⬘eq = 1
冑2
冑共1⬘ − 2⬘兲2 + 共2⬘ − 3⬘兲2 + 共3⬘ − 1⬘兲2 + 6共12
⬘2 + 23
⬘2 + 31
⬘2兲
behavior of the material is taken from the cyclic stress-range–
strain-range curve 共Sec. 5兲, which contains no information on 共2兲
mean stress, the magnitude of mean stress, if one is present, is
unknown. ⬘peq =
冑2
3
冑共p11⬘ − p22⬘ 兲2 + 共p22⬘ − p33⬘ 兲2 + 共p33⬘ − p11⬘ 兲2 + 3
⬘ + p23
2 共p12
2
⬘2 + p31
⬘2兲
The lack of knowledge of the mean stress is not a problem 共3兲
when the design fatigue curves of the ASME B&PV Code 关4兴 are
used, in which mean stress is assumed zero when alternating plas- where ij⬘ = 共r兲
ij ⫿ ij,
共r兲
pij = pij ⫿ pij,
i , j = 1 , 2 , 3. The superscript
ticity is present. This is supported by Ellyin 关14兴, who has shown 共r兲, r = 1 , 2, refers to the stress and plastic strain components at the
that mean stress approaches negligible magnitudes when test reversal points, t1 and t2. The minus signs apply to the right-hand
共r兲 共r兲
specimens of SA-516 Grade 70 steel are cycled with various de- downward leg 共see Fig. 5兲 of the hysteresis loop, with ij , pij
grees of mean strain. A reasonable assumption is that the effect of fixed at the upper extreme. The plus signs apply to the left-hand
any mean stress that may actually occur in an individual cycle 共r兲 共r兲
upward leg of the loop, with ij , pij fixed at the lower extreme.
with alternating plasticity in a real component can be neglected.
The resulting curve of ⬘eq versus ⬘peq is the multiaxial equivalent
This is made part of the design procedure considered in this paper.
hysteresis loop of the cycle, which is the counterpart to the
uniaxial hysteresis loop of the fatigue test specimen. It may look
9 Multiaxial Stress and Strain Equivalents like that in Fig. 5.
The objective of this section is to obtain the multiaxial stress 2. Just as in the uniaxial case, this loop identifies the stress
and strain equivalents that are appropriate for the local strain ap- range and strain range that describe the size of the loop. Its height
proach. This will be achieved based on one multiaxial equivalent is the multiaxial equivalent stress range and its width is the mul-
hysteresis loop that represents the cycle as a whole. This loop is tiaxial equivalent plastic strain range, which are now defined by
⌬eq =
1
冑2
冑共⌬1 − ⌬2兲2 + 共⌬2 − ⌬3兲2 + 共⌬3 − ⌬1兲2 + 6共⌬212 + ⌬223 + ⌬231兲 共4兲
Downloaded 10 Oct 2007 to 203.78.221.48. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm
⌬ peq =
冑2
3
冑共⌬p11 − ⌬p22兲2 + 共⌬p22 − ⌬p33兲2 + 共⌬p33 − ⌬p11兲2 + 23 共⌬p212 + ⌬p223 + ⌬p231兲 共5兲
Downloaded 10 Oct 2007 to 203.78.221.48. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm