Fatigue Life Estimation of Bellows Based On Elastic-Plastic Calculations
Fatigue Life Estimation of Bellows Based On Elastic-Plastic Calculations
Fatigue Life Estimation of Bellows Based On Elastic-Plastic Calculations
ELASTIC-PLASTIC CALCULATIONS
MASATAKATANAKA
ABSTRACT
A procedure which estimates the low-cycle fatigue life of bellows on the basis of
elastic-plastic calculations is treated. A U-shaped bellows is taken as an example for
the analysis and comparison is made with the measured data. A design formula, in
which Langer's method is modified so that it includes in its expression the strain
concentration factor, is proposed for the fatigue life estimation.
INTRODUCTION
With the aid of high speed electronic computers with huge memories, numerical
methods of solution using the finite difference technique or the finite element
technique have been rapidly developed in recent years as powerful means for
structural analysis. At present refined solution procedures have been established
even for elastic-plastic problems of various structures. Hamada and Tanaka I have
established a numerical method of solution by the finite difference technique for
general axisymmetric shell problems, using the incremental theory of plasticity. The
use of such solution procedures enables us to calculate stress and strain distribu-
tions in the elastic-plastic range with ease and fairly good accuracy and it is worth
using such results in rational design procedures for pressure vessels and other
structural components.
When bellows are used as expansion joints in piping systems it is reoognised
that they are often loaded beyond their elastic limits or that low-cycle fatigue under
tension--compression cycles of axial force is an important design factor. In most
cases, however, bellows are designed with the use of empirical formulae or linear
elastic solutions. As a design procedure of bellows considering the low-cycle fatigue,
an empirical formula has been presented by M. W. Kellogg Co. 2 and an oxperi-
51
Int. J. Pres. Ves. & Piping (2) (1974)-- Applied Science Publishers Ltd, England, 1974
Printed in Great Britain
52 MASATAKA TANAKA
LANGER'S METHOD
For the low-cycle fatigue of completely reversed cycles under uniaxial stress con-
dition, the following relation is valid between the plastic strain range Asp and the
number of cycles to failure Ns:
A~PNf ~ = C (1)
where ~t and C are experimental constants. Equlation (1) is well known as the
Manson-Coffin equation. The constant a is nearly equal to 0.5 regardless of
material properties.
Equation (1) is, however, frequently inconvenient for practical use, because it
includes a plastic strain in its expression. In fact, it has become possible quite
recently to analyse with a good accuracy the plastic strains produced in structures.
Formerly, they have been approximately estimated by the use of linear elastic
solutions. Langer 4 made an attempt to express eqn. (1) in a form easily applicable
to practical design, combining it with elastic stress analysis in the following way.
Let us consider a stress-strain cycle as shown in Fig. 1, then A~p can be expressed
as follows:
Aet' = A~ t - Aee = A~ t - 2Se/E (2)
where E is Young's modulus. Substituting eqn. (2) into eqn. (1), we obtain:
FATIGUE LIFE ESTIMATION OF BELLOws 53
Stress
-I
2S
o/ Strain
L
A,P --,.
-- A _t
s - e a~'12 = c e / ( z u / ) + se (3)
In eqn. (3) S corresponds to the elastic extrapolation stress, as shown in Fig. 1.
Langer proposed that the low-cycle fatigue life of pressure vessels, etc., can be
predicted if the endurance limit is inserted into S and the stress value obtained
from elastic stress analysis into S of eqn. (3~. He considered, 4 for example, that the
constants of eqn. (3), determined as best-fitted to 146 experimental results of
austenitic stainless steel specimens, are as follows:
54 MASATAKA TANAKA
Since the bellows problem is in a multiaxial stress state, the question of how eqn. (I),
which concerns uniaxial stress state, is to be related to such a problem remains
unsolved. Examining a number of experimental results from this aspect, Ohji 7 has
shown that the Manson-Coffin equation is also applicable to the low-cycle fatigue
problems of multiaxial stress state if the following equivalent plastic strain range
Aseqp is used instead ofAe p in eqn. (I), i.e.,
A~eqP = (~)} (h~ijP A~ijP) (7)
F A T I G U E LIFE E S T I M A T I O N O F B E L L O W S 55
In this research we use the equivalent plastic strain range given in eqn. (7). It can
be rewritten for the bellows problem as follows:
2 (Ae~v2 +
Ae,q p = (-~ As~p Aed + Asd2) (8)
Deflection Stress
Elastic I
~c-plastic / ElosticCTension)
e6
0 E.~//IDtrairl II ] I .8 Strain
< 11 / /
W ~
(Compression)
(Compression) (o (b)
tion amplitude w. That is, it shows a procedure in which the maximum strains ca +
and so- in the tension and compression sides are known, respectively, from
Fig. 2(a) when the deflection amplitude w is given, then the residual strains s t and
s c in respective sides are obtained using Fig. 2(b) and, finally, the plastic strain
range As p is determined. The strain value of the point which has the highest prob-
ability of fatigue failure should be used as the strain co-ordinate of the figure. In
what follows, attention will be paid to the point at which yielding occurs for the
first time when loading increases monotonically, i.e., the point which has the maxi-
mum value of the Mises equivalent stress, and it will be shown that satisfactory
results are obtained by such a treatment.
The bellows taken up for comparison between analytical and experimental results
is U-shaped and has four convolutions, as shown in Fig. 3. Using the notation and
I--
!~ dI
d2 - -
t. b
o
. . . . .J; 176~
-I
(Unit : ram)
Bet lows a b rl=2 h I' dp d~ I1
No.I 64.1 56.1 448 0-825 130 99~ 106 35
NO.2 61.7 55.5 4"43 1.050 1,50 106 40
NO.:5 52"4 44'0 4"56 0"/86 I~) 68 78 40
the corresponding values given in the figure, the configuration parameters are as
follows. For No. 1 bellows:
b 0"839 r~ r2 h
. . . . . . 0"0870 - = 0"0151 0 = - (11)
a a a a 2
Among these, No. 1 bellows is that used in the low-cycle fatigue test by Hamada and
Takezono. 3 Nos. 2 and 3 bellows are tested in this research by using the same
apparatus as mentioned above. All the bellows are made of SUS 27 stainless steel
and solution-treated after forming to remove residual stresses. The stress-strain
relation of this material in uniaxial tension test is shown in Fig. 4. In elastic-plastic
analysis a linear hardening material is assumed, as shown in the figure by the solid
lines. Using the following material constants:
30 ~
E
20 J
R E/Y:910
H'=O.O2E
E =2-1x 104 I~/tn,.n2
,I
o 05 I0 1'5 2'0 %
b
E
- - = 910 H ' = 0"02E v -- 0"3 E = 2"1 x 104 kg/mm 2 (12)
Y
elastic-plastic analysis of the processes of axial tension and compression is carried
out. Here Y, H ' and v denote the yield stress, the plastic tangent modulus and
Poisson's ratio, respectively. A non-linear elastic solution is necessary to obtain
residual strains. It can be furnished, however, if Hooke's stress-strain relations are
always used in the solution procedure for elastic-plastic problems. The linear
elastic solution can also be obtained by such a treatment.
_l
- ] C OA
A
lh
v v
A B
Numerical analysis of Nos. 1 to 3 bellows reveals that for all the bellows the
Mises equivalent stress:
Figure 6 shows the relations between Aeeq~ and the non-dimentional deflection
w* at point A of Fig. 5 which are obtained from the elastic-plastic analysis. Here
w* is a non-dimensional expression of the deflection w divided by the elastic limit
deflection w t, i.e.,
w w/h if,
w* . . . . . --=- (14)
w, .'dh wl
~ No. 2
No. I
No. 3
<3
I
2 3 4 5 6 7 8 9 I0
~= w/~i
Fig. 6. Theoretical relations between non-dimensional deflection w* and equivalent plastic strain
range AeqP
The values of ff~ = wJh determined from the linear elastic solution are as follows:
~P; = 0.270 No. 1 bellows
wt = 0.136 No. 2 bellows
rp~ = 0.330 No. 3 bellows (15)
Using eqn. (15) and Fig. 6, the value of Ae=qp at an arbitrary deflection amplitude
can be obtained for a given bellows unit.
Next let us compare the analytical with the experimental results. The latter were
obtained in such a way that the number of cycles accumulated until a fatigue crack
60 MASATAKA TANAKA
penetrating the shell thickness was regarded as Ny. The same device as that of
Hamada and Takezono 3 was employed, in which the fatigue testing machine was
automatically stopped when the air that had been filled in the test bellows with
about 0.2 kg/cm 2 before the experiment decreased in pressure to about 0.02 kg/cm 2
as a fatigue crack propagated through the shell thickness. The fatigue crack
penetrating the shell thickness was, in most cases, found in the bottom portions of
bellows convolution.
The fatigue lives predicted by the present method, in which the constants of
eqn. (1) are assumed as:
= 0.5 C = 0.65 (16)
so that best-fitted estimation anay be given to the uniaxial fatigue data, and those
by eqns. (5) and (6), are compared with the measured data of bellows in Figs. 7(a)
to (c). The fatigue life estimated by the present method is in good agreement with
the measured one. The maximum principal stress value Sp is used as the elastic
extrapolation stress S in Langer's method. The relations between Sp and ~ = w/h
obtained from linear elastic solution are as follows:
Sn = 86-0~ (kg/mm 2) No. 1 bellows
Sp = 164.9~ (kg/mm 2) No. 2 bellows
Sp = 79.6}~(kg/mm 2) No. 3 bellows (17)
(:~o \ o : Experimental
1.5
\\ - : Longer, eqn. ( 5 )
\\ -:Takezono, ecln (6)
' ~ 0 \\\\ -- :Present theory
% ~ \\\\\ :'
~ . 1.0
'G
0.5
<
0 t ~ I i I I I
i0 3 104 IO5 I06
CyclestofailureNf
(a)
FATIGUE LIFE ESTIMATION OF BELLOWS 61
15 -
~ \\ 0 :Experimentol
------ :Longer,eqn.(5)
c~'~ \\ : Takezono,eqn.( 6 )
~ , - :Presenttheory
I 0 -
.-=_
.B --..
~ o5
,....,
-
t I i I I I
0102 io3 K)4 ~)s
Cycles to failure Nf
(b)
:L~O
0 :Experimental
.... Longer, ecln.(5)
~ ~'\\ --~ :Takezono,eqn (6)
\ ~ :Presenttheory
~2.0
"8
~1.0
8
0 I i I i I I I
io3 ~ ~ ~G
Cyclas to failure Nf
(c)
Fig. 7. Comparison between estimated lifetimes and measured ones of bellows: (a) No. 1 bellows;
(b) No. 2 bellows; (c) No. 3 bellows.
62 MASATAKATANAKA
\
\ \\ :Linger, era.(5 )
\ \ . :TQl~zono, ecp. {6)
7
29 i~,. 3 \\\
~6
~5
.-=_ 47 6 "%.'~.\ XXxxx\
4
.5
t I I I J I I
1102
Cycles to failure Nf
TABLE 1
CONFIGURATION PA.R)tdvl][n~R~OF BELLOWS ANALYSED
of Table 1. It is apparent from Fig. 8 that while there are many cases where both
the estimated fatigue lives agree well with each other, there are still other cases
where the Hamada-Takezono equation (eqn. (6)) gives an unsafe fatigue life
estimation and further that the Langer equation (eqn. (5)) is always unsafe.
In order to experimentally investigate the above point which can be concluded
from a theoretical standpoint, another low-cycle fatigue test was carried out for
No. 5 bellows. The configuration parameters of the bellows used in this experiment
were not identical with those of the No. 5 bellows because of difficulty in its forming.
\ , Io :Experimental
------:Lar~er, eq~ (5)
6
' \
\\\\ ----- :Takezon~ecln.(6)
:Preserd theory
~4
2 -
I -
oe i I
103
i I
104
i
I05
I I
I06
Nt
Fig. 9. Additional experimental support for fatigue life estimation by present theory.
64 MASATAKA TANAKA
The following proportionality between the total strain amplitude 5o and the
deflection amplitude w can be derived from linear elastic solution:
e = k,w (22)
where k, is a constant. If Kc is a strain concentration factor, then the actual strain
range Aet of the elastic-plastic deformation state can be expressed as follows:
0 :Experimental
1"5 \ -- :Longer,eqn (5)
\\ .... Talezono, eqrk(6)
-- :ModifiedL~'~ger,eqn(25)
\ ~ \\\\
1.0
05
o" I
io 3
1 I
io4
i I
~:~
-i t
~s
(a)
FATIGUE LIFE ESTIMATION OF BELLOWS 65
I-5-
J'O --
<,
0.5-
I I J I I I
Nf
(b)
3.0
0 : Ex~tol
\ .... L , x ~ r , eqn.(5 )
. -----: Tokezono,eqn.(6 )
:Modified Lor~er, eqn.(25)
2.0
1.0
0 / 3 i I i I i I
io 4 ~ ~e
Nf
(c)
Fig. 10. Comparison of estimated lifetimes by eqns. (5), (6) and (25) with measured ones:
(a) No. I bellows; (b) No. 2 bellows; (c) No. 3 bellows.
66 MASATAKA TANAKA
Equation (25) is to be called the modified Langer equation which takes account of
the strain concentration factor. If we assume as Kc = 1 in this equation, we obtain:
(CE/2) t/~,
N f = (S - Se) l/at (26)
whose form is identical with that of eqn. (5) or eqn. (6).Therefore, if the results
obtained by eqns. (25) and (26) are compared with the measured data, the influence
of strain concentration becomes dear.
In Figs. 10(a) to (c) the fatigue lives predicted by eqns. (5), (6) and (25) are
compared with the measured data of Nos. I to 3 bellows, when Kc is determined
from elastic-plasticanalysis and the constants of eqns. (25) and (26) are assumed
as in eqn. (4).The strain concentration factor is calculated only for the maximum
principal strain ,. The fatigue life given by eqn. (25) agrees well with that of
eqn. (6) and also with the measured one. From this fact it may be concluded that
eqn. (5),which gives Ix~st-fittedestimation to uniax/al stressstate,must be modified
as in eqn. (6) for the bellows problem, mostly because Langer's method does not
take account of any strain concentration phenomenon which occurs in elastic-
plastic deformations.
DISCUSSION
It has been made clear that the two procedures based on elastic-plastic calculations
suggested in this paper are effective for the low-cycle fatigue life esti/nation of
bellows subjected to completely reversed loading. If the relations between deflec-
tion amplitude and the equivalent plastic strain range or the strain concentration
factor are given in design charts, then we can design bellows with regard for their
low-cycle fatigue failure. This procedure seems to be more reasonable than the
other, i.e., eqn. (25), based on a principal strain concentration factor, and, in
addition, is applicable to low-cycle fatigue problems of other more complicated
FATIGUE LIFE ESTIMATION OF BELLOWS 67
structural components. The writer concludes that much effort should be made in
the future to express the relations between the deflection amplitude w and Aecqp or
Kc in design charts so that reliable lifetime estimates can be made.
CONCLUSIONS
A procedure has been discussed whereby estimates of the low-cycle fatigue life of
bellows subjected to completely reversed loading can be made on the basis of
elastic-plastic calculations. The analytical results obtained by this procedure for
bellows subjected to completely reversed-deflection cycling of axial force were
compared with the measured data, and the effectiveness oftbe procedure ascertained.
Further considerations on the bellows problem revealed the following facts:
(I) While there are many cases where the Hamada-Takezono equation (eqn. (6))
is effective, there are still some cases where eqn. (6) gives an unsafe fatigue life
estimation if the bellows configurations are changed. This fact was also ascertained
from the experimental point of view.
(2) It is because Langer's equation (eqn. (5)) does not take account of any
strain concentration phenomenon occurring in the elastic-plastic deformation of
actual structures that it must be modified as in eqn. (6) for the bellows problem.
(3) If Langer's method is modified so that it includes in its expression the strain
concentration factor as shown in eqn. (25), then such a modified Langer equation
gives a good fatigue life estimation to the bellows under consideration. The use of
eqn. (25) may be recommended for the design of bellows because of the simplicity
of its expression.
In this paper the problem to be dealt with was restricted to that of bellows
subjected to completely reversed loading and a case of completely-reversed-
deflection cycling of axial force was analysed. It should be possible, however, to
apply the present method to other low-cycle fatigue problems of shell structures
other than bellows, since refined solution procedures for their elastic-plastic
problems have been already established and are readily available.
ACKNOWLEDGEMENTS
The writer wishes to express his cordial thanks to Professor M. Hamada, Osaka
University, for his encouragement and useful discussion. In addition the writer
expresses his gratitude to Professor R. K. Penny, University of Liverpool, for his
critical reading of the manuscript and helpful suggestions. He is also grateful to
Mr T. Ohiwa and Mr K. Iyoshi, formerly Students of Osaka University, for their
aid in performing experimental work.
68 MASATAKATANAKA
REFERENCES