You are on page 1of 7

Review of fatigue design procedures for pressure vessels

H. Mayer
, H.L. Stark
, S. Ambrose
b, 1
University of New South Wales, Sydney, Australia
Welding Technology Institute of Australia, Australia
When applying detailed fatigue analysis to welds in pressure vessels, designers encounter practical dif®culties with the methods required
by national standards such as ASME BPV Section VIII Div. 2, BS5500 and AS1210.
This paper discusses the main fatigue design stress parameters, being: (a) the stress intensity range and (b) the principal stress range, and
evaluates these for their validity over the scope of fatigue conditions they are required to predict.
It is concluded that a practical and conservative approach for a given weld detail is to calculate geometric stress parameters at the critical
location, evaluate and select the larger of the two parameters above, and to compare this to a fatigue curve based on undressed welded
specimens. It is further concluded that an additional fatigue strength reduction factor of 2 be superimposed when the principal stress direction
changes signi®cantly. q 2001 Published by Elsevier Science Ltd.
Keywords: Fatigue design; Pressure vessels; Stress intensity range; Principal stress range
1. Introduction
Fatigue design of pressure vessels today, is broadly based
on two national standards, those being the ASME Boiler
and Pressure Vessel Code (Section 8 Div. 2, and
Section 3) [1], and BS5500:1997, Annex C [2]. Until
recently the Australian national standard AS 1210
(Supplement 1) [3] has been based on the ASME
Code. Dif®culties in the application in practice of
both the American and British approaches have caused
those responsible for the Australian code to draft a standard
that diverges in important aspects from both the American
and British approaches. It is the purpose of this paper to
raise these differences for discussion.
The ASME approach requires for design, knowledge of
the peak alternating stress intensity, that is the peak local
value (no matter how small the location) of amplitude of
¯uctuation of the Tresca stress intensity (half the
magnitude of ¯uctuation in difference in principal stres-
ses). This is compared with material property data,
based on strain cycling experimental data of plain,
unwelded plate. Of particular note is the requirement
that the ¯uctuations in all three of the principal stresses
need to be known, and those need to be known for the
most localised of peak stress intensity locations. For
machined shapes of known geometry this provides no
insuperable barriers given the advent of ®nite element
analysis (FEA), however for welds, and speci®cally
undressed toes of welds, the designer has little or no
hope of determining such localised peaks given the vari-
able geometry of same. Also for the toes of welds
(which are the most likely of the stress concentrations
to initiate fatigue cracking in practice) there is the further
issue of metallurgical damage coincident with the geometric
stress concentration. Accordingly the material property
approach used in ASME, while readily applicable to de®ned
geometries remote from welds, is not readily applied to weld
By comparison BS5500:1997 Annex C [2] speci®cally
addresses the weld issue and requires knowledge of
the maximum range in principal stress at the weld,
which is then used with fatigue data based on a range
of different weld joint geometries that have been subjected
to uniaxial strain cycling. This weld detail categorisation
approach considers only ¯uctuations in one of the three
principal stresses at the weld. That is, if the largest ¯uctua-
tion in principal stress is across the toe, the principal stress
along the weld is ignored.
Houston [4] applied each of these procedures to a single
example and found these two approaches to differ in fatigue
life estimation by a factor of 50. When Houston's calcula-
tions were repeated using the currently valid rules [1,2], this
factor was reduced to 15.
International Journal of Pressure Vessels and Piping 77 (2000) 775±781
0308-0161/00/$ - see front matter q 2001 Published by Elsevier Science Ltd.
PII: S0308-0161(00)00069-7
* Corresponding author.
Present address: 1A Woods St., North Epping, NSW 2121, Australia.
2. Approaches to component life analysis
2.1. Material property S±N approach
The material property approach has its origins in the early
work carried out by (among others) WoÈhler [5] who inves-
tigated the fatigue behaviour of rail wagon axles. In essence
this approach endeavours to characterise the alternating
stress pattern for comparison with the material property,
fatigue strength.
The material property, fatigue strength, is determined by
testing a number of carefully standardised specimens (typi-
cally small, polished and usually excluding weld metal) to
determine the number of fully alternating (®xed ampli-
tude) cycles of stress (or strain) required to generate
fatigue failure, usually in a speci®ed environment.
While such specimens experience all four stages of fati-
gue, namely crack nucleation, small crack growth, crack
propagation, and complete fracture, because their size is
small the crack propagation stage is usually very short,
and accordingly the data can be regarded as the material
fatigue strength in resisting the formation of cracks. The
relationship is usually presented as a graph of stress ampli-
tude (or stress range) versus number of cycles to failure, the
S±N curve.
To allow application of such data to any given design
situation, allowance must be made for the differences
between the test specimens and the design being consid-
ered. Such factors include difference in Young's Modu-
lus, variable stress amplitudes and ratios, the stress
parameter to be used (e.g. Tresca stress intensity for
multiaxial stresses, or principal stress), mean stress,
overload hardening, size effect, environment, surface
effects, localised peaks in stress intensity due to
geometric stress concentrations and other defects such
as those metallurgical defects associated with welds.
While many of these factors can be incorporated into
the design calculations (using moduli ratio, Miner's rule
etc), the highly variable, localised, and unknown (at the
design stage) stress concentrations, and metallurgical
damage associated with weld toes presents the designer/
stress analyst with severe problems in determining, in prac-
tice, a magnitude of peak stress intensity to take to the S±N
2.2. Weld detail categorisation S±N approach
The dif®culties in application, to fatigue design of welds,
of the material property S±N approach, and the recog-
nition that any location on a pressure vessel may be
welded (e.g. as a result of repairs) have led to the
recent advent of the weld detail categorisation approach
to fatigue design of pressure vessels (BS5500:1997
Annex C).
In essence this approach relies on experimental fatigue
data determined not from polished specimens of parent
metal but from fatigue testing welded joints. Since there
are a great many different geometries of welded joint
(butt, tee, cruciate, toe shape, etc), a family of S±N curves
is required to represent the range of geometries.
Accordingly such S±N curves include the detrimental
effects of weld toe notches and metallurgical damage,
such that the designer need only determine the nominal
stress range in the vicinity of the weld. This has great prac-
tical merit.
Gurney [6] and Richards [8], early workers in the UK,
introduced this approach for steel structures, with
Gurney [9] proposing a set of design rules for welded
steel joints based on a number of typical weld details
categorised into 8 fatigue classes, each having its own
S±N curve. These S±N curves range from a ground
¯ush butt welded joint (highest fatigue strength, almost
equivalent to the material property curve) to a trans-
verse ®llet welded joint (one third of the fatigue
strength of the ground butt weld). Volumes of joint
fatigue data were published by Oliver and Ritter [10].
Other investigators constructed complete or near
complete pressure vessels, to observe their fatigue beha-
viour. The results of a number of these tests are
presented and discussed in Harrison and Maddox [11].
All joints, except the few constructed into pressure
vessels were loaded essentially uniaxially. After
Harrison and Maddox [11] and Maddox [12] adapted
this approach for pressure vessels, BSI eventually replaced
H. Mayer et al. / International Journal of Pressure Vessels and Piping 77 (2000) 775±781 776
geometric stress. The local primary plus
secondary stress intensity at a point which,
while including the concentrating effects of
joint geometry, excludes the highly localised
stress concentration of the weld pro®le itself.
(e.g. for a weld toe this is typically determined
by extrapolating to the location of the toe,
those stresses at least one half of a plate thick-
ness away from the toe).
range of alternating principal stress
amplitude of alternating Tresca stress inten-
2 £ S
t plate thickness
Tresca stress intensity largest difference between the
three principal stresses at a point
In BS5500 [2], S
is the notation used for the maximum range of alter-
nating principal stress, wich is Ds
here. Unfortunately, due to the variety
of notations used by the codes around the globe, it was not possible to select
a set of nomenclature consistent with all the relevant publications. Care
should be taken not to confuse the parameters here, or in the published
the material property approach by the weld detail categor-
isation approach with the publication of BS5500:1997
Annex C.
While the weld detail categorisation approach relieves
some of the designer's problems, there remains in practice
two areas of dif®culty depending on the stress parameter
used. ASME, and similar standards today, specify the
alternating stress intensity. However, BS5500:1997
considers only the largest range in principal stress. Pres-
sure vessels (unlike typical steel structures) are inevita-
bly subject to stresses that vary multiaxially, not
uniaxially, and so for example on the knuckle of a
pressure vessel end, high circumferential compression
can be combined with high axial tension, causing the
variation in stress intensity at that location to be in the
order of twice the variation of maximum principal
stress. It should be noted that although residual stresses
resulting from welding will cause initial stresses in the
vicinity of the weld to be tensile, the multiaxiality of
the variation in stress can still cause the alternating
stress intensity to be signi®cantly larger than the maxi-
mum alternating principal stress. The complex and
rapidly varying distributions of primary plus secondary
stress intensities typical of the multiaxial stressing in
pressure vessels, particularly at the junctions between
component shells, nozzles etc. cause the stress analyst
dif®culty in isolating a magnitude of stress for use with
the stress category S±N curves. Such stresses need to
be close enough to the weld that it represents the stress
in the vicinity of the weld but excludes the highly loca-
lised stress concentrating effect of the weld toe. While
this can be readily done with typical structural steel
members such as beams, which are predominantly
uniaxially stressed, this can be very dif®cult to isolate
from the full ®eld view of aggregate stresses given by
3. Comparison of approaches
3.1. Inside corner of the nozzle (crotch corner)
Fig. 1 shows this location and it is of a well de®ned
geometry, remote from welds and as such is amenable to
detailed stress analysis using FEA to establish the actual
peak stresses. This particular vessel was required to be
designed for 5000 cycles (n) of pressure ¯uctuation from
0 to 14.8 MPa internal pressure. At the hot spot on the inner
edge of the nozzle the FEA converged to give the three
principal stresses at the internal pressure of 14.8 MPa as
1404, 214.8, and 214.8 MPa. Ignoring the small correc-
tion for Young's Modulus the following fatigue damage is
H. Mayer et al. / International Journal of Pressure Vessels and Piping 77 (2000) 775±781 777
Fig. 1. Stress intensity distribution on pad reinforced nozzle in a cylindrical vessel.
…amplitude of alt: stress intensity†
ˆ …1=2†…404 2…214:8†† ˆ 209:4 MPa
Giving N ˆ 17; 500 from the ASME curve and
damage U ˆ n=N ˆ 5000=17; 500 ˆ 0:29
BS5500:1997 Annex C:
…max principal stress range† ˆ 404 MPa
Giving N ˆ 31; 840 from curve C of BS5500:1997 and
damage U ˆ 5000=31840 ˆ 0:16:
It is of particular note that the ASME approach, using
stress intensity, invokes the extremes in principal stress,
but BS5500 only uses the maximum range in principal
stress. In this example, the difference in the stress range
parameter is only about 4%. However, the difference in
life of around 80% is due to the difference in curves.
Further, on the knuckle of a pressure vessel where the
circumferential stress can be compressive of the same
order of magnitude as the axial tensile stress that differ-
ence could be a factor of 2 (i.e. equivalent to a factor
of approximately 8 on life).
3.2. Edge of the compensation plate
The weld toe at the edge of the compensation plate
presents practical dif®culties for FEA; some particular
small radius has to be assumed for the toe to enable the
FEA to converge. Modelling a small radius of curvature
in a large model is dif®cult. The stress results are extre-
mely sensitive to the magnitude of the radius chosen.
This radius at the weld toe is highly variable in a real
weld. For these reasons it is not practical to attempt to
accurately model weld toes using FEA.
What can be done is to ignore those stress results within
approximately half plate thickness of the toe, and deter-
mine the ªgeometric stressº at the location of the weld
toe by extrapolating from the local stress results. In
practice, when plotting the surface stress approaching
the discontinuity, a clear change in trend is visible as
the effects of peak stress become signi®cant, so
enabling extrapolation to exclude peak stress relatively
simply. (extrapolation techniques are given in Ref.
[2,13]). This renders the FEA within half´t of the
weld toe less critical, although it still has to be suf®-
ciently ®nely meshed to give accurate results outside
that region. Such a ªgeometric stressº represents the
primary plus secondary stresses at the location of the
weld toe, but excludes the highly localised stress
concentration of the weld toe. In this case, two points at a
distance of half´t and 2´t from the weld toe were used and
gave the geometric principal stresses of 1300, 1188 and
0 MPa.
ASME requires a peak stress, and accordingly the
geometric stress has to be converted to a peak stress
using a stress concentration factor; using a factor of
2.5 (Houston [4]) gives the following result
…amplitude of alt: stress intensity†
ˆ …1=2†…404 2…214:8†† ˆ 375 MPa
Giving N ˆ 3; 000 from the ASME curve and damage
U ˆ 5000=3; 000 ˆ 1:67
BS5500:1997 Annex C:
…max principal stress range† ˆ 300 MPa
Giving N ˆ 38; 518; from curve E (C. and
damage U ˆ 5000=38; 518 ˆ 0:13:
Clearly there is signi®cant discrepancy between the
two approaches for the weld toe, which can be traced
to the arbitrary stress concentration factor required to
get a peak stress intensity to take to the ASME curve.
There is no practical alternative route to get a peak
stress intensity for use with the ASME curve using
FEA. Accordingly it is the authors' view that the
BS5500 approach is the better method to assess fatigue
performance at a weld toe.
4. Characterisation of governing stress
Each of the two approaches considered above asks for a
characteristic ¯uctuating stress. The material property
approach requires the maximum alternating stress intensity
amplitude, S
, while the weld detail characterisation
approach in BS5500:1997 uses the maximum principal
stress range, Ds
The typical fatigue test specimen is cycled using uniaxial
stress, or uniaxial strain of ®xed amplitude. The loads on a
real component may well subject that component to: (i)
varying mean stress; (ii) varying amplitude stress; (iii)
multiaxial stress; and (iv) out-of-phase loading, (including
rotating principal axes).
4.1. Tensile alternating stress with lateral compression
The literature contains considerable evidence to the effect
that virtually all fatigue and fracture properties of metals
and components are signi®cantly affected by multiaxial
loading [7,14,15]. The most damaging case being when
tensile stresses normal to a crack plane act in conjunction
with compressive stresses parallel to the crack plane. In such
a multiaxially stressed case the crack opening due to the
H. Mayer et al. / International Journal of Pressure Vessels and Piping 77 (2000) 775±781 778
tensile stress is assisted by the compressive stress, which
may or may not itself be cycling. An example of such multi-
axial stressing is on the internal surface of a pressure vessel
knuckle, where, particularly if the knuckle radius is small in
comparison with the crown radius, a substantial circumfer-
ential compressive stress accompanies the tensile axial
The range of alternating principal stress, Ds
(BS5500:1997), effectively ignores that lateral stresses are
present, and, in the case of the knuckle, this parameter
would effectively ignore the smaller of the axial tensile
stress or the circumferential compressive stress, and so it
does not account for the fact that fatigue life is reduced by
their combined presence. On the other hand, the alternating
stress intensity S
, (ASME, AS1210), increases because of
the presence of the lateral stresses of the opposite sign.
Though it is not entirely clear from the literature, it appears
that S
is probably conservative in the case of multiaxial
loading. However, it is clear that Ds
, is unconservative
in its treatment of multiaxial stresses of mixed sign.
4.2. Biaxial tension
Here, the same mechanism allows the crack growth to be
inhibited by the presence of the lateral tension. But, because
the stresses on the third principal axis are zero, as is gener-
ally the case on the outside surface of a pressure vessel, the
two parameters are equal to each other, both ignoring the
restraining effects of the lateral tension. Hence, both are
4.3. Triaxial compression
In the extreme case where an element is subjected to
¯uctuating hydrostatic pressure, it can be readily theorised
that, given three equal principal stresses, the alternating
stress intensity, S
, is zero and, hence, this parameter
predicts no crack growth or associated fatigue damage to
the element. On the other hand, the parameter Ds
, the
principal stress range, will equal the range in one of the
principal stresses regardless of the other two being of
equal magnitude, and so it will indicate fatigue damage.
The literature indicates that the use of Ds
is probably
conservative in that it ignores the restraining effects of
triaxial compression.
4.4. Triaxial tension
As a direct extension from the biaxial tension case, it is
clear that tension parallel to the crack plane in two direc-
tions will serve to reduce crack opening even more than
tension parallel to the crack plane in only one direction.
Here, the Ds
parameter retains the value of the range of
the largest principal stress, irrespective of the magnitude of
the other two principal stresses and, hence, predicts this case
even more conservatively than the biaxial tension case.
The value of the S
parameter is less than in the biaxial
tension case, because the shear stress on the element is
reduced. In the limiting case, when all principal stress
components are equal, and proportional throughout the
cycle, S
is zero at all times and predicts no crack growth.
However, the tension still opens the crack and, therefore, the
crack will grow, thus making the parameter unconservative
in this limiting case. Since it is not known at which point S
becomes conservative, it must be said to be unconservative
in any triaxial tension case.
4.5. Out-of-phase loading
In the case of rotating principal axis directions, that is
when shear and normal stress alternate out-of-phase, crack
growth rates in steel have been observed to increase by a
factor of up to ,2.5 under high cycle fatigue conditions
[16,17]. This represents a reduction factor of ,1.3 on
strength. The effect is much more severe under low cycle
fatigue conditions. Both the ASME and BS5500 approaches
consider only the largest changes in alternating stress to
which an element is subjected, whether or not the loads
H. Mayer et al. / International Journal of Pressure Vessels and Piping 77 (2000) 775±781 779
Table 1
Performance of design parameters under various loading conditions
are alternating out-of-phase. Hence, this effect is not consid-
ered and both codes may be unconservative in such cases.
4.6. Load state discussion
The conclusions above are collated in Table 1. The table
reveals immediately that both parameters can be
unconservative in some load states. This implies that each
of the two approaches, ASME BPV-VIII-2 and
BS5500:1997 Annex C, can lead to unconservative esti-
mates of fatigue life. Of particular concern is the
BS5500:1997 use of the range of principal stress, Ds
where tension normal to a crack plane acts in conjunction
with compression parallel to that plane, such as in the
knuckle regions of a vessel.
It is also evident from Table 1, that either one or the other
parameter is conservative in all states of in-phase cyclic
loading, which implies that between the two parameters
conservative estimation of fatigue life can nevertheless be
It is a simple exercise to calculate both parameters once
the stresses at the point of interest are known. Hence, to be
conservative, the designer may simply apply the larger (and
more conservative) of the two parameters.
In consideration of out-of-phase loading, a fatigue
strength reduction factor of 2 is proposed.
5. A new approach in AS1210 supplement 1-2000
A revised version of AS1210 was being drafted at the
time of completion of this paper. Among other modi®ca-
tions, a new approach to detailed fatigue analysis is
proposed. This new approach, is based on the arguments
presented in this paper, and is subject to public comment.
Consequently, the published version of AS1210 may vary in
detail from this approach, which is summarised as follows.
There are now two ®lters aimed at removing the need for
fatigue analysis. These two ®lters are much simpli®ed from
the previous code and so minimise design costs for those
cases clearly not in need of detailed fatigue analysis. Where
there is need for detailed fatigue analysis the following
If there is no structural singularity such as a weld toe, but
simply rolled plate or forging without repair welds, the three
principal stresses are established using elastic analysis at the
location of the peak stressing. Both the maximum range of
principal stress, Ds
, and the maximum range of stress
intensity, S
, are established for that point. The larger of
these two parameters is then chosen as the design parameter,
and corrected for Young's Modulus (for both material and
temperature). The permissible number of cycles, N, is then
established from a design curve, which is equivalent to
curve C from BS5500 Annex C, so allowing the calculation
of the fatigue damage, U.
If the peak stress region is a weld toe, then the three
principal stresses are established at the location of the
weld toe by extrapolation from the surrounding local stress
analysis results (linear elastic FEA or equivalent) beyond a
perimeter that is half plate thickness from the weld toe. The
procedure is then as before, excepting that a second design
curve, which is equivalent to Curve E from BS5500 Annex
C, is used to determine N.
Curves C and E of BS5500:1997 are the ones being
considered at the time of writing. Slightly different curves
may be used in the published document (to cater for ASME
parent metal).
As an alternative, consideration is being given to present-
ing the fatigue curves in a strain form. The advantages of
such a presentation being that the curves can be presented as
common for all metals and temperatures. Further, such a
strain presentation of the curves enables the practitioner to
use non-linear FEA, and the S±N data can be represented
with two curves only, being one for welded material &
another for non-welded material.
If the detail is poorer in fatigue performance than a weld
toe, such as for example the root of a partial penetration
®llet weld, then the procedure is to use the average stress
determined across the weld throat further applying a stress
concentration factor of 2 to the design stress parameter
before going to curve E.
In those cases where the principal stress directions vary,
that is where the principal stresses ¯uctuate signi®cantly out
of phase, then a further fatigue strength reduction factor of 2
is applied.
6. Conclusion
Practical dif®culties arise for the designer in the fatigue
analysis of welds in pressure vessels. For example, the
geometry of weld toes are effectively singularities for the
purposes of elastic ®nite element stress analysis. Yet, most
national standards, such as ASME, require an elastically
derived peak stress intensity at the highly localised peak
stress location. Other standards, such as BS5500, allow cate-
gorisation of the weld detail, but rely on the range of maxi-
mum principal stress, without apparently taking account of
multiaxial stresses. When calculating the fatigue life of
some components, the two codes differed by a factor on
life of greater than 10.
The major fatigue design stress parameters: (i) the stress
intensity range, S
, and (ii) the principal stress range, Ds
were evaluated for their performance over the scope of fati-
gue conditions they were required to predict, and for their
ability to be calculated.
It was concluded that a practical and conservative fatigue
analysis approach for a given weld detail is to use for the
stress parameter the larger of S
and Ds
, and to avoid the
dif®culty with the singularity at the weld toe by extrapolat-
ing from the surrounding values to give a geometric stress at
the singularity. The alternating stress parameter so evalu-
ated is compared to one of two fatigue curves. A curve based
H. Mayer et al. / International Journal of Pressure Vessels and Piping 77 (2000) 775±781 780
on parent material property data is used when the location in
question is free of any welded material, and another curve
based on butt welded, unground specimens, is used when
welded joints may be present at the location at any time in
the vessels life. These details may vary slightly in the ®nal
published AS1210.
It is further proposed, that an additional fatigue strength
reduction factor of 2 be applied where the component
stresses in a multiaxially stressed location ¯uctuate out of
phase signi®cantly.
[1] ASME BPV-VIII-2. Boiler and pressure vessel code: section VIII Ð
Rules for construction of pressure vessels: division 2 Ð alternative
design rules. Appendix 5. New York: The American Society for
Mechanical Engineers, 1995.
[2] BS5500:1997. British standard speci®cation for un®red fusion welded
pressure vessels. Annex C. UK: British Standards Institution.
[3] AS1210 Supplement 1-1990. Un®red pressure vessels Ð Advanced
design and construction (supplement to AS1210-1989). Appendix C.
Sydney: Standards Australia.
[4] Houston R. British standards institution boiler and pressure vessel
design criteria. In: Nichols RW, editor. Pressure vessel codes and
standards. Amsterdam: Elsevier, 1987 (chap. 5).
[5] WoÈhler, A. Versuche uÈber die Festigkeit der Eisenbahnwagenachsen.
Zeitshrift fuÈr Bauwesen, 1860. In: ASM handbook, vol. 19: fatigue
and fracture. OH: ASM International, 1996. p. 15.
[6] Gurney TR. Fatigue of welded structures. Cambridge: Cambridge
University Press, 1968±1979.
[7] Beaver PW. Biaxial fatigue and fracture of metals: a review. Metals
Forum, Australia, vol. 8, no. 1, Summer, 1985.
[8] Richards KG. Fatigue strength of welded structures. Cambridge:
Abington Hall, 1969.
[9] Gurney TR. Fatigue design rules for welded joints. The Welding
Institute Research Bulletin 1976;May:115±24.
[10] Oliver, Ritter. Catalogue of S±N curves of welded joints in structural
steel. DuÈsseldorf: Deutscher Verlag fur Schweisstechnick, 1979.
[11] Harrison JD, Maddox SJ. Derivation of design rules for pressure
vessels. The International Institute of Welding Document, XIII-941-
80, 1980.
[12] Maddox SJ. Fatigue design rules for pressure vessels. The Interna-
tional Institute of Welding Document, XIII-1215-86, 1986.
[13] IIW Group XIII-XV. Fatigue design of welded joints and compo-
nents. The International Institute of Welding Document, XIII-1539-
96/XV-845-96, 1996.
[14] Leevers PS, Radon JC, Culver LE, (1977). In: Beaver PW, editor.
Biaxial fatigue and fracture of metals: a review. Metals Forum,
Australia, vol. 8, no. 1, 1985.
[15] Charat IMH, Garrat GG, (1980). J Test Eval JTEVA, 8,9. In: Beaver
PW, editor. Biaxial fatigue and fracture of metals: a review. Metals
Forum, Australia, vol. 8, no. 1, Summer, 1985.
[16] Kanazawa K, Miller KJ, Brown MW, (1977). J Engng Mater Technol
(Trans. ASME, Series H), 99, 222. In: Beaver PW, editor. Biaxial
fatigue and fracture of metals: a review. Metals Forum, Australia, vol.
8, no. 1, Summer, 1985.
[17] Kitagawa H, Yuuki R, Tohgo K., (1979). Fatigue Engng Mater Struct,
2, 195. In: Beaver PW, editor. Biaxial fatigue and fracture of metals: a
review. Metals Forum, Australia, vol. 8, no. 1, Summer, 1985.
H. Mayer et al. / International Journal of Pressure Vessels and Piping 77 (2000) 775±781 781