Sie sind auf Seite 1von 11

Design of Pressure Vessels for

Low-Cycle Fatigue
B. F. LANGER Methods are described for constructing a fatigue curve based on strain-fatigue data
Consulting Engineer, W e s t i n g h o u s e for use in pressure vessel design. When this curve is used, the same fatigue strength-
Electric C o r p o r a t i o n , Pittsburgh, Pa. reduction factor should be used for low-cycle as for high-cycle conditions. When
Fellow A S M E evaluating the effects of combined mean and alternating stress, the fatigue strength-
reduction factor should be applied to both the mean and the alternating component, but
then account must be taken of the reduction in mean stress which can be produced by
yielding.
The complete fatigue evaluation of a pressure vessel can be a major task for the de-
signer, but it can be omitted, or at least drastically reduced, if certain requirements can
be met regarding design details, inspection, and magnitude of transients.
Although the emphasis in this paper is on pressure vessel design, the same principles
could be applied to any structure made of ductile metal and subjected to limited numbers
of load cycles.

I Introduction Ref. [3]) for estimating the fatigue life for combined steady and
alternating stresses.
N THE designing of pressure vessels and piping, it (5) The use of Miner's hypothesis for estimating the cumula-
has not been the usual practice to give serious consideration to tive effect of stress cycles of varying amplitude [4, 5].
the possibility of fatigue failure. This practice has been success- (6) Recognition of the fact that, when a fluctuating stress ex-
ful because pressure equipment, unlike rotating machinery and ceeds the elastic limit of the material, a shift occurs in the mean
aircraft structures, is not usually subjected to large numbers of stress component about which the stress fluctuates [1].
load cycles during its lifetime, and ductile metals can absorb
(7) The use of "fatigue strength reduction factors" Kt, rather
surprisingly large strains for limited numbers of cycles. When,
than "theoretical stress concentration factors" Kt [2].
however, the utmost in reliability and efficient utilization of
The present paper will describe the results of later work on
material is required it becomes necessary to calculate pressure
some of these elements.
stresses and thermal stresses in detail and to determine whether
or not fatigue failure is possible in a few thousand cycles. The
purpose of this paper is to describe methods by which the de- II Construction of a Fatigue Curve
signer can make use of the available data on low-cycle fatigue to
The fatigue curve needed by the designer is one which shows
evaluate the integrity of a component which is to be used at tem-
stress versus cycles and which contains sufficient safety factors to
peratures below the creep range for the material.
give safe allowable design stress for a given number of operating
The basic elements of the proposed design method have been cycles or, conversely, allowable operating cycles for a given value
described in previous publications [1, 2]. 1 They are: of calculated stress. The stress values on the fatigue curves
should be directly comparable to the stress values which the
(1) The use of the maximum shear theory of failure rather
than the maximum stress theory of failure. In order to make designer calculates using his usual methods of analysis for
allowable shear stress values comparable to the more familiar pressure stress, thermal stress, stress concentration, etc.
values of tension and compression, calculated stresses are ex- One obvious way of constructing such curves would be to ob-
pressed in terms of two times the shear stress, which is the largest tain a large amount of low-cycle fatigue data on each material,
algebraic difference between any two of the three principal which would be extremely laborious. Fortunately, Coffin's work
stresses. This quantity is called the "equivalent intensity of on low-cycle strain fatigue has disclosed some fundamental and
combined stress," or more briefly, the "stress intensity." apparently conservative relationships which can be used to esti-
(2) Calculation of stresses using the assumption of elastic be- mate this curve from some of the better-known mechanical
havior. (In situations which involve instabilities, such as plastic properties.
hinges or other forms of nonlinearity, plastic analysis is required.) The terms used in this section are defined as follows, and as
When the elastic limit of the material is exceeded, the elastic cal- shown in Fig. 1.
culation results in a stress value which, although fictitious, is
E modulus of elasticity
=
proportional to strain, which in turn is a good measure of the
N =
cycles to failure
damage being done.
e, =
total strain range
(3) The use of fatigue data from tests in which strain rather
te =
elastic portion of strain range
than stress is the controlled parameter.
e;, =
plastic portion of strain range
(4) The use of the modified Goodman diagram (see Fig. 136 of
c =
a material constant
1 Numbers in brackets designate References at end of paper. tj =
true strain at fracture, commonly known as fracture
Contributed by the Metals Engineering Division and presented at ductility
the Winter Annual Meeting, New York, N. Y „ November 26-Decem-
S = 7 iEet
b e r 1, 1 9 6 1 , o f T H E A M E R I C A N S O C I E T Y O F M E C H A N I C A L ENGINEERS.
Manuscript received at ASME Headquarters. May 31, 1961. Paper A S = stress amplitude
Xo. 61—WA-18. Se = endurance limit

Journal of Basic Engineering SEPTEMBER 1 9 6 2 / 3 8 9

Copyright © 1962 by ASME

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


S„ = allowable stress amplitude for a given number of than the strain range is used because this makes S correspond
cycles in magnitude to the endurance limit for complete stress reversal
S u = ultimate tensile strength in ordinary tensile test at the high-cycle end of the curve. (This is the form in which the
= true stress at fracture in tensile test endurance limit is usually reported.) Since
Sy yield strength
limit of elastic behavior (larger of S„ or S J = + eP (4)
Sb
RA per cent reduction of area in tensile test and = 2A S/E (5)
z exponent defined in eq. (12)
M coefficient defined in eq. (13) S - E ( ^ ) = E —2- + E — =
J
F + A S (6)
k exponent defined in eq. (15) V2 2 2
O coefficient defined in eq. (16)
By substituting the values of ep derived from eq. (1) and c from
NOTE: In the discussion of cyclic phenomena the terms "range" eq. (3) into eq. (6), we obtain
and "amplitude" are often confused. In this paper the term
"range" means the total peak-to-peak excursion and the E 100
term "amplitude" means half of the range. S = -In + AS (7)
4 -\/n 100 - RA
Coffin has found [6] that for a wide variety of materials at tem-
peratures below the creep range, the following relationship holds: The fatigue data [10] show that AS is a function of N, so that
eq. (7) is not a satisfactory mathematical model for statistical
VNe D = (1) treatment of the data. It may be observed, however, that, if AS
Furthermore, Coffin has found [7] that the constant c in eq. in eq. (7) were replaced by the endurance limit S„ one necessary
characteristic of the fatigue curve would be satisfied—that is, that
(1) can be conservatively estimated as one half the fracture duc-
S would approach S e when Ar approached infinity. Since A S > S,
tility (true strain at fracture) determined in the ordinary tensile
for finite values of N this introduces some error in the remainder
test. Fracture ductility is defined as
of the curve, but it is on the conservative side. Furthermore,
100 when N is quite small (around 100) A S is a very small part of the
t j = In (2) total value of S. The largest error is found to be around N =
100 - RA
1000.
Therefore c may be calculated from Another possibility would be to make A S equal to the yield
strength, but this would be an unconservative approximation for
_ I 100 (3) constructing a fatigue curve because it results in a curve which
" 100 - RA approaches the yield strength for large numbers of cycles, and the
0.2 per cent offset yield strength is often higher than the endurance
The desired relationship, to which safety factors will later be
limit. Therefore, the best approximation appears to be to take
applied, is one between N, the cycles-to-failure, and a quantity S,
A S = S ( . The proposed equation for plotting the fatigue curve
which is the product of the strain amplitude (half the total strain
therefore becomes
range t,) and the elastic modulus E. The strain amplitude rather
E 100
S = 7=ln + St (8)
4 \/N 100 - RA

A comparison has been made [8] between curves plotted from


eq. (7) and eq. (8) for a material (AISI T y p e 347 stainless steel)
for which both A S and Se were known. Eq. (8) was more con-
servative at all points. The largest difference in values for S
occurred around 1000 cycles and amounted to less than 16
per cent.
A comparison of various curves plotted from eq. (8) with low-
cycle fatigue data has also been made [8]. The comparisons in-
cluded O F H C copper, 2 S Aluminum, 1018 carbon steel, nickel A,
titanium Ti 55-A, 347 stainless steel, 24 ST aluminum alloy, 75
S T aluminum alloy, 422 stainless steel, C R - M o steel, A-201 steel,
and 90 B steel. In most cases the comparison was good, and in all
cases, the calculated curve was conservative as long as the chosen
value of Se was consistent with the fatigue data.
STRAIN Fig. 2 shows the results of a detailed stud}' on austeuitic types
AXIS of stainless steel made from data reported in refs. [8, 9, 10, and
11], N o data were included for notched specimens and no data
were used for temperatures higher than 350 deg C. (As men-
tioned before, the method of analysis being described is limited to
AS
temperatures below the creep range.)
Eq. (8) was used as a mathematical model for statistical evalua-
tion of the data. The "best-fit" curve in Fig. 2 was determined
/
L ^ /
v by applying the method of least squares to the logarithms of the
146 experimental values. The elastic modulus was taken as E =
£ K-
- V - L e 26 X 106 psi and the best-fit was found to occur when RA —
72.6 per cent and S e = 43,500 psi. N o t all of the 146 experimen-
t
tal points are shown in Fig. 2, due to space limitations.
Fig. 1 Stress-strain cycle If no fatigue tests had been available, it would have been

3 9 0 / SEPTEMBER 1962 Transactions of the AS M E

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


reasonable to estimate RA at 50 per cent, which is the minimum failure. (Of course, duplicate tests to improve the confidence
specification value, and Se at 37,500 psi, which is half the specified level are always advisable.) If tests are run at total strain ranges
ultimate tensile strength. The calculated curve for these values e, and e2, with measured stress ranges 2ASi and 2AS 2 , and cycles-
is seen in Fig. 2 to be a safe lower limit. to-failure Ni and N2, the plastic strains per cycle are (ei —
It is the author's opinion that the available evidence is suf- 2ASi/E) and (e2 - 2AS2/E). The plastic strain range is
ficient to allow the use of eq. (8) for the estimation of low-cycle
e„ = MN' (U)
fatigue properties with a high degree of confidence, provided due
conservatism is used in choosing values for Se and RA for ma- where
terials on which low-cycle fatigue tests have not been made. It
is suggested that the lowest expected value of be used except / 2AS,\ , / s 2ASA
when the use of a higher value is backed up by fatigue data. The
log
V (12)
use of high-cycle fatigue data is recommended for the evaluation log N\ - log Ni
of S e , since the errors involved in correlating endurance limit with
ultimate strength and yield strength are well known. For- and
tunately this information is usually available for structural
materials. ( 2AS\ (13)
Manson [12] has proposed another mathematical model for
constructing the fatigue curve from limited amounts of data. He The elastic stress range is
shows that for copper and aluminum the relationship between A S
and N is approximately 2 A S = ON k (14)

where
AS = 2SU[N~0A2 (9)
log 2AS1 - log 2AS 8
where S„, is the true stress at fracture in the tensile test. k = (15)
If eq. (9) is substituted in eq. (7), we obtain the formula log Ni — log N2

and
E 100
S = 7= In (10) (16)
G = 2ASiNi~l
WN 100 - RA
in place of eq. (8). The use of this equation reduces the error The desired relationship between S and N is
previously noted in eq. (8) in the region around N = 1000, but
has the disadvantage that S approaches zero when AT becomes S = 1 (Eep + 2 AS) (17)
/L
very large. This is not realistic for materials which have a true
endurance limit. This difficulty, however, is easily overcome by
arbitrarily ending the curve plotted from eq. (9) at the value of N = - (EMN' + GNk)
for which A S = S e and extending it as a horizontal line for larger
values of N . Another disadvantage of eq. (10) is that S u i is not The quantities M, z, G, and k have been defined above in terms
so readily available as S c . of the measured quantities.
Manson's analysis is particularly useful if just a few plastic The fatigue curve described in the foregoing is applicable to
fatigue tests can be made, because it shows how to obtain the cycles of complete stress reversal, where the stress passes through
most information from a small number of tests. Instead of ob- zero and attains equal values of tension and compression. It is
taining the complete curve from a large number of tests, as was well known that at the high-cycle end of the curve, the addition of
done in Fig. 2, it is sufficient to run just two strain-fatigue tests a steady mean stress reduces the endurance limit, and the amount
at well-separated values of strain and observe A S and cycles-to- of reduction can be estimated by means of the Goodman diagram.

10' 1 ! . 1i!11 • ! : ! i1M [ 1 11 1 1 11 1 1 ! ! 1 ; 11

"BEST-FIT' CURVE , RA = 72.6 % , Se = 4 3 500 p5i


:

10° S
RA = 50 % , S E = 37500 psi
oa> 0 /
UJ ^ ^ co era '0 ,
-
C\J ^ ^ ^ Ax

i i f f e „

- ' "*—' ^

DESIGN' CURVE

10 ! 1 II 1 111 ! ' 1 1 11 1 1 ! 11 1 11 1 1 II 1 1 1 1 1 1 1 II 1 I 1
10 10
C Y C L E S TO F A I L U R E

Fig. 2 Fatigue d a t a a n d design curve for austenitic stainless steel

Journal of Basic Engineering SEPTEMBER 1 9 6 2 / 3 9 1

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


It has also been demonstrated [1, 13] that at the low-cycle end along the line CD, but this can be explained by the fact that the
of the curve, where stresses go well up into the plastic range, mean initial test conditions were reported and no consideration was
stress has little or no effect on the fatigue life. I t has been shown given to the real stress cycle which existed at the point of failure
in ref. [1] that this is due to the yielding that occurs during the after the first few test cycles had been completed. In any bend-
first strain cycle, which reduces the effective value of mean stress. ing fatigue test or any fatigue test on a notched specimen, the
When the strain range exceeds twice the yield strain, the mean mean stress can easily shift without the experimenter being aware
stress becomes zero. of it, because the portions of the specimen which remain in the
One of the most troublesome parts of the fatigue analysis elastic range maintain the configuration of the test piece. A
described in ref. [1] is the evaluation of the effect of mean stress. few tests have been reported [14] of axial loading of unnotched
After having gone through the correct procedure, the designer specimens at high values of mean stress, but in these cases the
might well ask whether some residual stress produced by welding, author of ref. [14] has noted that "if the maximum stress in
a preservice test, or an accidental overload may not have actually the range exceeds the tensile static yield strength of the metal, the
produced a mean stress in the member higher than the one he had failure will be by general yielding on the first application of the
calculated. Therefore, it would be much easier, and not unduly load, rather than by fatigue after many applications of the load. - '
conservative, to adjust the fatigue curve downward at the high- This is true for load-controlled tests but not for strain-controlled
cycle end to allow for the maximum possible effect of mean tests.
stress. Keeping the foregoing thoughts in mind, let us find the effect of
The required amount of adjustment can be found by means of mean stress on the amplitude of alternating stress which is re-
the modified Goodman diagram shown in Fig. 3. In this diagram, quired to produce fatigue failure. At zero mean stress, the re-
the mean component of any fluctuating stress is plotted as the quired amplitude is Se. As the mean stress increases along OC
abscissa and the alternating component (half total range) is in Fig. 3, the required amplitude of alternating stress decreases
plotted as the ordinate. Failure is expected for any stress cycle along the line EC. If we try to increase the mean stress beyond
which falls above the line joining S„ endurance limit, to S,„ ulti- C', yielding occurs and the mean stress reverts to C". If we apply
mate tensile strength (line ED in Fig. 3). 2 the mean stress by means of a dead-weight load, we can force it
to hold a value higher than C', but then failure will be produced
more gross yielding than by fatigue, and the fatigue analysis
becomes inapplicable. Therefore C" represents the highest value
of mean stress which has any effect on the fatigue life. The
actual value of mean stress in any specific case might well be
considerably lower, so the assumption that it has the value repre-
sented by point C' is a conservative one. Since Se' is the alternat-
ing stress required to produce failure when the mean stress equals
C', S/ is the value to which the high-cycle end of the fatigue
curve must be reduced if we are to safely ignore the effect of mean
stress. From the geometry of Fig. 3, it can be rh that

S„
for S,<Sb (18)
S.

For finite numbers of cycles Sc becomes Sa, as determined from


a fatigue curve such as Fig. 2, and the reduced value Sa' becomes

- S„
for Sa<Sb (19)
MEAN STRESS Sa

Fig. 3 Modified Goodman diagram When the number of cycles decreases to the point where Sa >
Sb, then Sa' = S„ and mean stress has no effect on the allowable
stress amplitude.
Attention is now called to the 45-degree line AB which con-
nects the Sb values on the horizontal and vertical axes. S& is Fig. 2 shows the experimental data from refs. [8, 9, 10, and 11]
called the "limit of elastic behavior" and may be described as the for unnotched austenitic stainless steel. The yield strength of
highest stress amplitude the material can maintain without yield- this material is usually about 30,000 to 40,000 psi. It therefore
ing, even after being cycled several times. It is the yield strength appears that 1 / 2 fie; exceeds the yield strength in the whole range
after strain hardening or strain softening. 3 For practical pur- of interest. Therefore, it may be concluded that this material
poses it can be taken as the higher of the 0.2 per cent offset yield cannot sustain a mean stress at any cyclic stress level which
strength or the endurance limit. Regardless of the conditions would produce failure, and therefore no correction for the effects
under which any given test is started, the true conditions after of mean stress need be made. Experimental verification of this
the application of a few cycles must fall inside the triangle O A B conclusion would be a worth-while research project.
or on the vertical axis above A. The reason for this is that any Fig. 2 also shows a design fatigue curve for austenitic stainless
test which is started under conditions which wrould fall outside of steel based on the best-fit curve and with a safety factor of either
this region would have a maximum stress greater than Sb and 2 on stress or 20 on cycles, whichever is more conservative at
yielding would then reduce the mean stress to a value which would each point. It is believed that these safety factors are sufficient to
make the cycle fall on AB or on the axis above A. It is true cover the effects of size, environment, surface finish, and scatter
that test results have been reported [14] outside this region, of data. Several fatigue tests and simulated service tests on
models of components have confirmed this belief. Service ex-
2 In ref. [1 ] it was recommended that point- D be located midway
perience has also been good, but is probably not lengthy enough
between Sb and <SU, but the author has more recently become con-
to be cited as a strong confirmation of the proposed methods.
vinced that this estimate was too conservative.
3 Coffin [11] has shown that work-hardened materials strain-soften There are materials, however, for which S„ < Sb at and above
under cyclic loading. a finite number of cycles. For such cases, eq. (19) may be used

392 / SEPTEMBER 1962 Transactions of the AS ME

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


Fig. 4 Design curve for m e d i u m - s t r e n g t h pressure vessel steels

to reduce the fatigue curve in the high-cycle region. In choosing The best-known method [3] for expressing the relationship
values for Sb and S„ for use in eq. (19), it must be noted that it is between the fatigue strength-reduction factor Kf and the
conservative to use a high value for Sb and a low value for Sa. theoretical stress concentration factor K, is by means of the
When a single design curve is being constructed to give allowable notch sensitivity q which is defined as
values for a whole group of materials, conservatism requires that
the highest value of Sb and the lowest value of Sa be chosen from
the group. Fig. 4 shows such a curve which can be safely used for
most of the nonheat-treated steels commonly used in pressure
vessel construction, such as SA-201, SA-204, SA-212, SA-302, Notch sensitivity is a property of the material but is also a func-
and the equivalent forging grades. The solid curve has been ad- tion of the notch geometry. Therefore it is not a true material
justed where necessary for the maximum effects of mean stress. property. A somewhat different concept, with wider applicabil-
The dotted curve shows the unadjusted values applicable to zero ity, has been described by Peterson [2, 15, 16], It is best ex-
meau stress. plained by means of a simple example. In Fig. 5(a), a bar with
It may be found that for heat-treated steels, for which the a central hole is subjected to cyclic axial load and the stress dis-
yield strength is not far below the ultimate strength, eq. (19) tribution across the section is as shown. The hypothesis is that
gives an unduly conservative result. When this occurs, it is ad- fatigue failure does not occur when the peak stress reaches the
visable to calculate the mean stress which will actually occur and endurance limit, but rather, as shown in Fig. 5(b), when the en-
not assume that it is always at its maximum possible value. durance limit is reached at some finite distance 8 below the sur-
Take, for example, a heat-treated bolting steel for which Su = face. The rationale for this hypothesis is that, due to the granulai
150,000 psi, Sb = Sy = 130,000 psi, and Sc = 75,000 psi. Assume structure of the material, a finite volume of the material must be
also that Sa = 130,000 psi at 2000 cycles and Sa = 95,000 psi at at a stress equal to the endurance limit in order to cause failure.
10,000 cycles. Mean stress must be considered at any number The dimension 8 might be viewed as a material property, which is
of cycles above 2000, since this is the number of cycles for which smaller for high-strength metals than for low-strength metals.
S„ = Sb. At 10,000 cycles, eq. (19) gives Fig. 6 (reproduced from ref. [16]) shows the approximate rela-
tionship between 8 and tensile strength for steels. Two curves
ri50,000 - 130,0001 are shown, one for the maximum shear theory of failure and one
S ' = 95,000 - = 34,000 psi. for the Mises criterion. Since the values for 8 are empirical, the
L 150,000 - 95,000 J ' 1
answer to any given problem comes out about the same regardless
For this case it would be worth while to calculate the actual of which theory is used.
mean stress rather than accept such a large penalty in reduced
The use of the 8 concept produces several interesting and useful
endurance limit.
results. For one thing, it explains the size effect which has been
noted in fatigue tests of notched specimens. This may be seen
Ill Application ot Stress Concentration Factors by noting what happens in Fig. 5(b) when the size is increased
It is well known that the theoretical stress concentration fac- but 8 is held constant. Another advantage of the 8 concept is
tors which are calculated from the theory of elasticity or deter- that there may be a possibility of extending it to very small notch
mined from photoelastic tests do not always indicate the actual radii, even including tight cracks [2]. Fig. 7 shows the calcu-
weakening which the stress concentration introduces into the lated strength reduction factors for long cracks in steel of 100,000
structure The actual amount of weakening depends not only on psi tensile strength. The abscissa is crack depth, and it may be
the peak stress but also on the stress gradient, the material, and seen that a buried crack of depth 2b is equivalent to a surface
the type of loading. crack of depth 6.

Journal of Basic Engineering SEPTEMBER 1 9 6 2 / 3 9 3

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


t t i
300

2 5 0

, — M A X I rfUM SHEAR
200
2
UJ
cr
t- I 5 0

MISES—/

I 00

50 1 1 1 1 1 1 V X 1
.0001 .0002 .0005 .001 .002 .005
S (INCHES)
S e = FATIGUE LIMIT
Fig. 6 Material constant 5 versus tensile strength for steel (from Peterson,
Ref. [16])

1?
cr. j
! i

O
\ - 1 1 •
C )
<1
LL 10
y
7?.
O 9
Lb) 1-
|
MAX STRESS
( >
8
SURFACE
0 JUJJJJL, / CRACK
cr 7 HURItU^
CRACK
T
1- B
CD
V B y
III 5
RR
/ w
I -
</>
/
LLI
z> 3
/ i
/
Fig. 5 Stress distribution in bar w i t h circular hole o 1 i

1- !
2
<R
LI_
A relationship of the kind shown in Fig. 7 can also be used to 1
i•

estimate the required sensitivity of a method of nondestructive 0 0.2 0.4 0.6 0.8 1.0 1.2
testing, such as radiography. When the radiographer uses a DEFECT SIZE PARAMETER,2b (inches)
penetrometer having a thickness of, say, 2 per cent of that of the
section he is inspecting, he must realize that defects smaller than Fig. 7 Fatigue strength-reduction factors for cracks
2 per cent will be undetected. If sensitivity is expressed in per
cent of the section thickness, Fig. 8 shows the strength-reduction basis for the commonly accepted practice of using lower values
factor which may be produced by hidden defects in various thick- of Kf for lower values of N. In (C), however, although nominal
nesses of material. The quality of a vessel design can be charac- stress is the controlled parameter, the material in the root of the
terized by a factor K, which is the ratio between the highest peak notch is really being strain cycled, because the surrounding
stress at a discontinuity, hole, fillet, etc., to the average stress material is at a lower stress and behaves elastically. Therefore,
intensity in the shell. If the desired vessel quality corresponds it should be expected that the ratio between curves (.4) and (C)
to K = 3, then we see from Fig. 8 that 2 per cent sensitivity is should be independent of N and equal to Kt. The experimental
adequate up to almost 3-inch thickness and 1 per cent sensitivity verification of this relationship is not adequate, but some does
is adequate up to 5-inch thickness because the hidden defects exist [2]. The use of a Kf independent of N together with strain-
will then not be the controlling factor. If a lower quality vessel cycling fatigue data appears to the author to be both logical and
is being built, say, K = 5, then 2 per cent sensitivity is adequate conservative, even though it is contrary to the most commonly
up to 9-inch thickness. accepted practices.
In the field of low-cycle fatigue, it has been common practice In applying stress concentration factors to the case of fluctuat-
to use lower stress concentration factors for small numbers of ing stress—that is, a cycle which has both a mean and an alternat-
cycles than for large numbers of cycles. This is reasonable when ing component—it has been the common practice to apply K s to
the allowable stresses are based on stress-fatigue data, but is not only the alternating component. A careful studj' of what hap-
advisable when strain-fatigue data are used. Fig. 9 shows a pens to fluctuating stresses when they enter the plastic range re-
typical relationship between S versus N curves from ( A ) strain- sults in a better method of using the stress concentration factor.
cycling tests on unnotched specimens, (B) stress-cycling tests on Take, for example, the case of a material with 40,000 psi yield
unnotched specimens, and (C) stress-cycling tests on notched strength and 30 X 10® psi modulus made into a notched bar which
specimens. The ratio between the ordinates of curves (B) and has a Kf — 3. The bar is cycled between nominal tensile stress
(C) decreases with decreasing cycles-to-failure, and this is the values of 0 and 20,000 psi. Common practice would call the

394 / SEPTEMBER 1962 Transactions of the AS M E

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


I
ir
o

z
o
H
o
ZD

I 2 3 4 5 6 7 8 9 10 II 12 13 14 15 16
SECTION THICKNESS (INCHES)

Fig. 8 Fatigue strength-reduction factors for protection f r o m h i d d e n defects

10" T 1 I I I I 1 I I I I I

- A : S T R A I N - C O N T R O L L E D T E S T S , UNNOTCHED (ORDINATE I S - g E 6 , )

- B : S T R E S S - C O N T R O L L E D T E S T S , UNNOTCHED (ORDINATE IS STRESS)

" C : S T R E S S - C O N T R O L L E D TESTS , NOTCHED (ORDINATE IS N O M I N A L S T R E S S )

10

10" 10' 10" 10

Fig. 9 Typical relationships b e t w e e n stress, strain, a n d cycles-to-failure

mean stress 10,000 psi and the alternating component (V2) X 3 X example, if the yield strength had been 50,000 psi, the adjusted
20,000 = 30,000 psi. The stress-strain history of the material at value of mean stress would have been 20,000 psi, and the common
the root of the notch would be, in idealized form, as shown in Fig. practice would have given an unconservative result.
10. The calculated value of Et/2, which would be compared
with strain-fatigue data, is 30,000 psi. The basic value of mean IV Determination of Need for Fatigue Evaluation
stress is also 30,000 psi, but yielding causes an adjustment down Since most pressure vessels are subjected to limited numbers of
to 10,000 psi. Therefore the yielding during the first cycle is pressure and temperature cycles during their lifetime, considerable
seen to be the justification for the common practice of ignoring design effort could be saved by defining the conditions which do
K f when calculating the mean stress component. It so happens or do not require that a fatigue evaluation be made. In order
that, for the case chosen, the common practice gives exactly the to define these conditions, several factors must be considered.
same result as the proposed method described here. The common For pressure cycling only, we need to know:
practice, however, would have given the same result regardless (1) The expected range of pressure cycles, which may be con-
of the yield strength of the material, whereas the proposed method veniently stated as a fraction of the design pressure. Let us call
gives different mean stresses for different yield strengths. For this fraction F.

Journal of Basic Engineering SEPTEMBER 1962 / 3 9 5

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


(2) The expected number of pressure cycles in the required discussion, however, let us use Fig. 4, which is more conservative
lifetime of the vessel. and applicable to a variety of medium-strength pressure vessel
(3) The average stress intensity in the vessel wall produced by steels. This curve includes all necessary corrections for the effect
the design pressure. This will normally be about the same as the of mean stress on the endurance limit.
allowable stress value for the design temperature tabulated in (5) The ratio of the highest calculated (or estimated) local
the A S M E Code for the material being used. For the purposes stress intensity in the vessel to Sm. Let us call this ratio K.
of the present discussion, call this stress value Sm. If thermal cycling as well as pressure cycling is involved we
(4) The applicable design fatigue curve for the material being also need
used. Fig. 2, developed in the preceding section, shows such a (6) The maximum expected temperature difference between
curve, applicable to austenitic stainless steel. For the present two adjacent points. (The term "adjacent points" means that
the points must be spaced not more than about two times the
thickness of the member under consideration. For example,
the difference between the temperature of a nozzle and that of
the shell to which it is attached is significant. The temperature
range across the thickness of a shell plate or flange is also sig-
nificant. On the other hand, the temperature difference between
YIELD STRENGTH the top and the bottom of a long vessel is not significant because
there is sufficient flexibility between these two points to accom-
modate the differential expansion.)
/ — BASIC VALUE OF (7) The expected number of thermal cycles during the required
/ MEAN STRESS
lifetime of the vessel. A thermal cycle is the establishment and
removal of the temperature difference described in (6) above.
Let us first consider the case of pressure cycling alone. The
amplitude (half range) of the highest local stress intensity will be
ADJUSTED VALUE
OF MEAN STRESS
S„it = - KSm F (21)

/Sait must be held within the limits of Sa on the fatigue curve,


Fig. 4. It is therefore possible to construct families of curves
.002 showing pressure range versus allowable cycles for various values
STRAIN of K and Sm. For example, if the design details and inspection
methods are chosen with sufficient care to keep K down to a
value of 3, Fig. 11 shows the relationship between pressure range
and cycles for various values of S,„. A similar fa mil}' is shown
in Fig. 12 for various K values at <S,„ = 20,000 psi. Fig. 12 is of
particular value to the designer because it tells him how low he
must keep K, and K is a measure of the quality of the inspection
and the design details such as fillet radii, discontinuities, and
Fig. 10 Idealized stress versus strain history

IOO

10* 10"
ALLOWABLE CYCLES

Fig. 11 Allowable pressure transients for K = 3 and various values of Sm

396 / SEPTEMBER 1962 Transactions of the AS M E

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


nozzle attachments. If Fig. 12 tells him that he can afford a AT = change in T during a cycle
high K value, he can save money on inspection and design de- H = a factor depending on geometry
tails. Sth = thermal stress intensity
Any combination of pressure range and expected cycles which
The factor H has a magnitude of the order of unity but usually
falls below the limits shown in Fig. 11 or Fig. 12 does not re-
lies between 0.5 and 2. Let us assume, conservatively, that
quire further fatigue evaluation. A combination which falls
H = 2. The thermal stress intensity amplitude is then
above the applicable curve is not necessarily unsatisfactory, but
requires more detailed study.
A similar generalization can be derived for the case of fluctuat- Salt = - X K X 2 X EaAT = KEaAT
ing thermal stress. Thermal stresses are characterized by the
expression
Salt
AT = (23)
St h = HEaT (22) KEa
Here, again, S a it must be kept within the Sa limits of the fatigue
where
E — modulus of elasticity, psi curve, Fig. 4. For the case of a carbon steel vessel with good de-
a = coefficient of thermal expansion, in/in/deg F sign details Ea = 180 psi/deg F and K = 3. Therefore
T = max temperature difference between adjacent points, S„ (24)
AT =
deg F 540

ALLOWABLE PRESSURE TRANSIENTS

sm= 20,000 psi

tn U-
co O

ALLOWABLE CYCLES
0I i i i i mil i mil i i i i mil i lllllli
IO3 IO4 IO5 IO6 10
Fig. 12 Allowable pressure transients for Sm = 2 0 , 0 0 0 psi and various values of K

240

200
A L L O W A B L E THERMAL
180
TRANSIENTS
O
1 160
U:
U.
Q 140
a:
S
£ 120
2~ 100
UJ
«
1 80
x

io2 io3 IO4 io5


ALLOWABLE CYCLES

Fig. 13 Allowable thermal transients for carbon steel vessels

Journal of Basic Engineering SEPTEMBER 19 6 2 / 3 9 7

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


Fig. 13 shows the allowable thermal transients for carbon Nu, A',3, . . ., N,„. The usage factors npi/Npi, np2/Np2, etc., and
steel vessels with K = 3 and K = 5. If another material had nn/Nlh nt2/Nt2, etc., can then all be calculated. Each group of
the same fatigue curve and the same modulus but a higher co- usage factors which produces additive stresses must be summed
efficient of thermal expansion, the curves would be lower. Any up and, if the sum of any group exceeds unity, a fatigue study is
case which falls below the applicable curve does not require needed.
fatigue analysis. In using Fig. 13 it should be noted that the The procedure can be illustrated by an example: An autoclave
ordinate is AT, the change in T during the given cycle, which is with a flanged closure is to be used in a process requiring 2000 psi
not always the same as T itself. For the startup-shutdown cycle, and 600 deg F. The operation is continuous during the week,
AT = T. During operation, however, there may be many with shutdown over the week-ends. Ten batches of material per
cycles which do not bring the temperature differences to zero, so week are produced. The reloading operation between batches in-
that the most damaging transient might be one for which AT < T volves reduction of pressure to 1000 psi and reduction of tempera-
but the number of cycles is large. ture to 300 deg F. Thermocouple readings taken on similar
Even simpler sets of rules can be derived if the type of service vessels have shown that, during normal steady-state operation,
allows additional assumptions to be made. Suppose, for ex- the flange temperature runs lower than the body temperature by
ample, carbon steel vessels are being designed to a maximum K as much as 100 deg F, but during reloading the temperature dif-
value of 3 and a maximum nominal stress (S m ) less than 30,000 ference is reduced to 50 deg F. The material used in the vessel
psi. The service is such that startup and shutdown will not is ferritic steel and the design stress is 20,000 psi. The design de-
occur more than 5000 times and significant thermal transients tails have been chosen to keep the ratio of peak stress to nominal
during operation will not occur more than 100,000 times during stress down to a value not greater than 3. The required life of the
the lifetime of the vessel. Since pressure can change rapidly, no vessel is 20 years at 50 weeks of operation per year.
assumption is made regarding the maximum number of pressure This operation violates the general requirements stated above
cycles. in that (1) the pressure fluctuates during normal operation
From Fig. 11 we see that an infinite number of pressure cycles through more than 25 per cent of the design pressure, (2) the
up to 25 per cent of design pressure can be imposed without danger maximum temperature difference exceeds 80 deg F, (3) the
of fatigue damage. The 5000 expected startup cycles at 100 per change in temperature difference during normal operation ex-
cent of design pressure also give a point below the applicable ceeds 40 deg F, and (4) the pressure and thermal stresses are
curve. Going now to Fig. 13 we see that the 5000 start-up cycles probably additive at the flange-to-shell junction. Therefore, a
should not be allowed to involve a temperature difference of more more detailed study is needed.
than 80 deg F and the 100,000 thermal transients should not in- The cycles to be considered are:
volve a change in temperature-difference of more than 40 deg F.
The rule for these operating conditions can therefore be stated (1) 100 per cent pressure fluctuation from startup and shut-
as follows: down. For this cycle, Fx = 1, npi = 20 X 50 = 1000, Npl (from
A fatigue evaluation is required if, during normal operation, Fig. 11) = 20,000, n p l / N p l = 0.050.
(2) 50 per cent pressure fluctuation from reloading. For this
(1) The pressure fluctuates through a range exceeding 25 per
cycle F2 = 0.5, np2 = 20 X 50 X 10 = 10,000, Np2 = 450,000,
cent of the design pressure, or
np2/Np2 = 0.022.
(2) The maximum temperature difference between any two
adjacent points of the vessel exceeds 80 deg F, or (3) 100 deg F temperature difference during startup. For
(3) The temperature difference between any two adjacent this cycle na = 1000, Ar,i (from Fig. 13) = 2 5 0 0 , « „ / A f n = 0.400.
points of the vessel changes by more than 40 deg F. (4) 50 deg F change in temperature difference during reloading-
For this cycle nl2 = 10,000, N,2 = 35,000, n,2/N,2 = 0.286.
It may be noted that in the foregoing discussion no considera-
tion has been given to the cumulative effect of various cycles, nor The sum of all the usage factors is 0.758, which is safely below
to the additive effect of pressure and thermal cycles. There- unity. Therefore no detailed fatigue evaluation is required. The
fore, the curves and rules given above imply that example given is, however, a borderline case and it would not
(1) The pressure cj'cles and the thermal cj'cles do not produce have taken much increase in the severity of the thermal transients
their highest stresses at the same point in the vessel and therefore to make a fatigue evaluation necessary. For example, if the
need not be added. startup were too rapid and produced a transient temperature dif-
ference of 150 deg F, N a would have been 550 cycles and ria/Na
(2) All significant pressure cycles can be conservatively repre-
would have been 1.82. In that case it would have been necessary
sented by a single value of pressure range F applied for a given
to calculate the thermal stress at the flange-to-shell junction to
number of cycles.
see if there really was any danger of fatigue failure. It may also
(3) All significant thermal cycles can be conservatively repre-
be observed that poor design details, such as sharp fillets, cannot
sented by a single value of AT applied for a given number of
be tolerated for service of this type. Fig. 13 shows that if K — 5,
cycles.
a AT of 100 deg F can only be allowed for a total of 350 cycles
If these conditions are not all met, it is still possible to avoid without detailed analysis.
the complete fatigue analysis by means of a simplified cumulative
damage evaluation. The most general case is that of a vessel
which will be subjected to pressure transients Fi, F2, F3, . . . ., F„ V Summary
for cycles numbering npi, np2, nD3, . . . ., npm respectively, and Justification is given for the use of strain-fatigue data rather
thermal transients AJTI, AT2, AT3, . . .., ATn for cycles numbering than stress-fatigue data in the designing of components for limited
na, Ha, tii3, . . . ., lit,,, respectively. The pressure transients will numbers of load cycles, for service at temperatures below the
always produce stresses which are additive among themselves, creep range. It is shown that the strain-fatigue curve can be
but the thermal stresses may not be additive either to the pres- estimated with fair accuracjr and good conservatism when only
sure stresses or among themselves. The allowable cycles for each the elastic modulus, the endurance limit, and the per cent reduc-
value of F and AT can be read directty from the applicable curves tion of area are known.
of Figs. 11 or 12 and 13, to give Arp:, No2, Arp3, . . . ., Npn and N, , The stress concentration factors used to evaluate the cyclic life

3 9 8 / SEPTEMBER 1 9 6 2 Transactions of the AS M E

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms


should be the estimated fatigue strength reduction factors rather Weldino Joltrlwl Research SUP11lement, vol. 32, no . 1, 1953, pp. 23s-
than the theoretical stress concentration factors. l'vl ethods are 30s.
14 James O. Smith, "The Effect of Range of Stress on the x"utiguc
available for estimating these fnctors for most notch geometries Strength of Metals," University of Illinois, Engineering Experiment
and even for defects such us cracks. When allowable cyclic Station Bulletin Series No. 334, voL 39, 110. 26, February 17, 1942.
stresses are based on stmin-ratigue data, the sume fnctor should 15 H. E. Peterson, "A :M ethod of Estimating the Fa.tigue Strength
be used for low-cycle as for high-cycle conditions. When calculat- of a Member Having a Small Ellipsoidal Cavity," IME-ASME,
International Conference all Fa.tigue of Metals, Loudon, 1956.
ing the effects of combined mean and alternating stress, the 16 R. E. Pcterson, "Analytical Approach to Stress Concent.ration
fatigue strength reduction factor should be applied to both the Effect in Fatigue of Aircraft Materials," Westinghouse Research
llleaD ami the alternating component, but then uccount must be Labornt.ories, Scientific Paper 1O~0509-0 - P2, Sept.ember, 1959. Also
taken of the reduction in Ulean stress which can be produced by published in Proceedings of W ADC Symposium, Fat.igue of Aircraft
Structures, August, 1!J59, WADC-TR 59~507, p. 273.
yielding.
The complete fatigue evalu ation of a pressure vessel, iucluding
ca lculation of discontinuity stresses, st.reugth reduction factors
nnd thermal stresses, and study of cumulative damage from all
possible pressu re and temperature cycles, cnn be a major task
for the designer. It can be shown, howevcr, that under conditions
which include a large number of applications, the complete fa-
tigue eva luation can be omitted provided certain requirements
are met regarding design details, inspection, and magnitude of
transien ta.
Although the emphasis in this paper is on pressure vessel de-
sign, the saine principles could be applied to any structure made
of ductile metal and subjected to limited numbers of loud cycles.

VI Acknowledgments
T he author is indebted to several of his associates for sug-
gestions and contl'ibutiotls to the development of the design
methods proposed in this paper. Particular mention sho uld be
made of rvJr. R. E. Peterson, ,1I,'estinghouse Research Laboratories,
Dr. L. F. Coffin, General Electric R esearch Laboratories, D ..
W. E. Cooper, General Electric Knolls Atomic Power Laborr..·
tory, tlnd Mr. J. L. ~dershon, Bureau of Ships. The development
of eq. (8) resulted from discussions held by n. Task Gmup of t.he
ASME Boiler and Pressure Vessel Committee under the chair-
manship of D r. 'V. T. Lank ford , U. S. Stccl Corporation. The
statis tical study of t.he dat.a in F ig. 2 was made by Mr. T. Shiroa-
moto, Westinghouse Bettis Atomic Power Laboratory. The
assistance of Mr. W. J. O'Donnell , of Bet.tis, is also appreciated .

References
1 B. F. Langer, "Design Vnlues for Thermal Strel>S in Du ctile
.Mn.terials," Weldi1l(J Journal Research Supplement, vol. 37, September ,
1958, p. 411s.
2 B. F. Langcr, "Application of Stress Concent.mt.ion Factors,"
Bellis Teclmical Review, WAPD.B'J'-IS, April. 19GO, p. I.
3 R. E. Peterson, "Stress Concentrat.ion Design Factors," John
Wiley & Sons, Inc., New York, N. Y., 1953.
4 B. F. Langer, "Fatigue Fail ure From Strcss Cycles of Varying
Amplitude," Journal of Ap1Jlied llfecha7l1'c8, vol. 4, TUANS. ASME,
vol. 59, 1937.
5 l\'l ilton A. :iVIiner, "Cumulative Damage in Fatigue," JOlLrllal
of Al'l;Heil Mecham'C8, vol. 12, TRANS. ASME, vol. 61, 1945, p. A-1S9.
G L. F. Coffin, Jr. , " A Study of the Effects of Cyclic Thermal
Stresses all a Ductile IVIetnl," TI(ANS, ASME, vol. 76, 1954, p. 93l.
1 J. F. Tnvcrllelli lind L. F. Coffin, Jr. , "A Compilntion a nd
Interpretation of Cyclic Strain Fatigue Tests on Metals," Trans.
ASM, \'01. 51. 1959, p, 438.
8 J. 1". Tavcrnclli and L. F. Coffin, Jr., "Experimental Support
for Gencl'lllized Equation Predicting Low-Cycle Fatigue," General
Electric Research Labs. i\Jcmo Report No. :ME-S3,September, 1959.
9 A. Jolmnnson, "}'ntigue of Steels at Constnnt Strain Amplitude
and Elevated Tempcmtllre ," Proeeedillgs Colloquium on Fat.igue,
Stockholm , Spri nger, Berlin, 1955.
10 E. E. Bnldwin, G. J . So kol, and L. F. Coffin, Jr., "Cyclic
Strnin Fatigue Studies all AIS I~347 Stainless Steel," AS7'M Proceed-
i1l0s, vol. 57,1957, p. 561.
11 L. F. Coffin, Jr., and J. F. Tavernelli, "The Cyclic Straining
and Fatigue of Metals," 'J'rans. MetallurGical Society, AlME, vol.
215, October, 1959, pp. 791-807.
12 S. S. IVlunson, "Thermnl Stresses in Design-Part 19. Cyclic
Life of Ductile IVIaterials," Machine Desigl~, July 1, 19GO, p. 139.
13 J. II. Gross, S. TSBlld, Bnd R. D. Stout, "Factors Affecting
Resist·ance of Pressure Vessel Steels to Repented Overloading,"

Journal 01 Basic Engineering SE P T E MBER 1962 399

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 05/05/2015 Terms of Use: http://asme.org/terms

Das könnte Ihnen auch gefallen