New Combined High and Low-Cycle Fatigue Model to

Estimate Life of Steel Bridges Considering Interaction

of High and Low Amplitudes Loadings

Kamal Karunananda1, Mitao Ohga1,*, Ranjith Dissanayake2,

Sudath Siriwardane1 and Pang-jo Chun1
1Ehime University, Department of Civil & Environmental Engineering, Bunkyo-cho 3, Matsuyama 790-8577, Japan
2University of Peradeniya, Department of Civil Engineering, Peradeniya 20400, Sri Lanka

(Received: 15 March 2010; Received revised form: 7 June 2011; Accepted: 8 June 2011)

Abstract: A new model is proposed in this paper for the estimation of the life of
bridges subject to damage caused by high cycle fatigue combined with low cycle
fatigue taking account of the interaction of high and low amplitude loadings. High
cycle fatigue is caused by normal routine traffic (low amplitude) loading while low
cycle fatigue is caused by extreme (high amplitude) loading situations produced by
such as earthquakes. The model mainly consists of a new damage indicator and a new
strain-life fatigue curve. Total strain is treated as the damage variable. The proposed
model predictions were verified by comparing with fatigue test results for four
materials reported in the literature. The proposed model was then applied to
estimate the fatigue life of a bridge member subject to combined high and low-cycle
fatigue damage caused by normal traffic and by earthquake loadings. The results of
the case study, confirm the importance and applicability of the proposed model.

Key words: high cycle fatigue, low cycle fatigue, combined damage, railway bridge.

1. INTRODUCTION interaction of high and low amplitude loadings may be

Bridges are generally subjected to a regime of high
cycle fatigue (HCF) caused by normal low amplitude
traffic loading (Chen and Lui 1987). Recently, a
number of bridge fatigue failures have been reported
and some of these cannot be explained by HCF alone.
Studies of these failures reveal that overloading
caused by earthquakes is one of the reasons for these
incidents. When a bridge is subjected to earthquake
loading (high amplitude loading), some members may
undergo stresses within the plastic range. These
plastic stresses may cause low cycle fatigue (LCF)
damage during the earthquake while being subjected
simultaneously to in service HCF conditions. This
combined HCF and LCF damage due to the
the cause of a much reduced life (Kondo and Okuya
2007). Therefore, it is important to study life
estimation methods based on combined HCF and LCF
damage caused by high and low amplitude loadings to
prevent unexpected failures of such bridges in the
future.
In the aircraft and mechanical engineering fields,
combined HCF and LCF damage has been studied.
The Coffin-Manson strain-life curve (total strain-
amplitude versus number of reversals to failure) is
used with Miners rule (Miner 1945) as the method
for estimating the life for combined HCF and LCF
damage (Suresh 1998). This method gives better
predictions in situations where HCF damage is less

*Corresponding author. E-mail address: ohga@cee.ehime-u.ac.jp; Fax: +81-89-927-9816; Tel: +81-89-927-9816.

Editor: K.F. Chung.

Advances in Structural Engineering Vol. 15 No. 2 2012 287

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

than LCF damage. Recently, a more accurate method subject to combined HCF and LCF damage caused by
(Kim et al. 2009) has been proposed to estimate the normal service traffic loads and earthquake loadings.
combined HCF and LCF damage where HCF damage
is greater than LCF damage. This approach is based 2. PROPOSED FATIGUE MODEL
on a modified Coffin-Manson curve and Miners This section explains the proposed fatigue model
rule. estimating the life of steel structures for combined HCF
The modification of the Coffin-Manson curve is and LCF damage. Total strain is considered as the
made by changing the slope of the original curve in the damage variable in the proposed model. This is
HCF region to take account of damage interaction composed of elastic and plastic strain components as
effects and damage below the fatigue limit. The factors shown in Eqn 1.
which are used to modify the slope of the curve depend
on the particular materials, but the necessary = el . + pl . (1)
information is not readily available for the majority of
materials. Therefore, obtaining an appropriately
where is the total strain amplitude, el. is the elastic
modified Coffin-Manson curve is difficult for the
strain amplitude and pl. is the plastic strain amplitude.
majority of materials.
Initially, the details relevant to the proposed strain-life
The use of Miners rule is the simplest and the most
fatigue curve are presented below, followed by a clear
widely adopted fatigue life prediction technique. One of
explanation of the proposed damage indicator.
its interesting features is that life calculation is simple
and reliable even when the detailed loading history is
2.1. Strain-Life Fatigue Curve
unknown. However under many variable amplitude
Summation of the independent HCF and LCF damages
loading conditions, Miners rule based life predictions,
does not directly represent the real behaviour of
such as in the case of bridges, have been found to be
combined HCF and LCF damage (Constantinescu
unreliable since the rule cannot capture the loading
2003). To take account of the combined HCF and LCF
sequence effect (Mesmacque et al. 2005; Siriwardane et
damage effect, it is necessary to modify the strain-life
al. 2007, 2008).
fatigue curve in the HCF regime. This modification
These reasons restrict the application of the method
together with a hypothetical ultimate strain
(Kim et al. 2009) to life estimation for combined HCF
requirement in the HCF regime for the proposed
and LCF damage in bridges which are generally
damage indicator (sub section 2.2) leads to the
subjected to variable amplitude loading and subjected to
construction of a different full range strain-life curve
HCF damage which is greater than LCF damage.
for the HCF regime in addition to the strain-life curve
Therefore, a different model is needed, which is
used in LCF regime.
based on commonly available material properties, to
The proposed curve consists of two parts as shown in
estimate bridge life more accurately for combined HCF
Figure 1. The first part of the curve describes the fatigue
and LCF damage taking account also of the interaction
of high and low amplitude loadings.
In this paper, a new model is proposed for estimating Log ( )
the fatigue life for combined HCF and LCF damage as
ULCF
a result of the interaction between high and low
amplitude loadings. The model consists of a new
damage indicator and a new strain-life curve. The new
'f
damage indicator can accurately capture the loading = (2N )b + 'f (2N )c
(i +1)eq E
sequence effect. The strain-life curve used in
conjunction with the new damage indicator enables UHCF
better estimation of combined HCF and LCF damage. y
The model can be applied to a wide range of materials '(i +1)eq
N + Nu
since it depends on commonly available material (i )eq = e
N + Neb'
parameters. i
e
The new fatigue model is first proposed and the
verification of the model is then demonstrated by
N(i +1)R Nu Ny N'(i +1)R (Ni ni) Ni Ne Log (N)
comparing experimentally derived fatigue lives with
theoretical predictions. Finally, the model is applied to Figure 1. Schematic representation of the proposed strain-life
the estimation of the fatigue life of a bridge member fatigue curve

288 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun

life of plastic strain cycles ( y) which usually relate
to LCF. In this connection, the Coffin-Manson strain-
life curve is utilized as shown below.
'f (2)
= (2 N )b + 'f (2 N )c
E ( )UHCF =
where is the applied strain amplitude, N is the number
of cycles to failure, f is the fatigue strength coefficient,
b is the fatigue strength exponent, f is the fatigue
ductility coefficient, c is the fatigue ductility exponent
and E is the elastic modulus of the material. The first
and second terms on the right hand side of Eqn 2 are
elastic and plastic strain amplitude components of the
total strain amplitude.
The ultimate strain for low cycle fatigue ()ULCF,
which is the total strain amplitude corresponding to
failure in a half reversal (a quarter of a cycle), is
obtained from Eqn 2 as,
( )ULCF = 'f (3)
In Eqn 2, when N equals 0.25 which corresponds to a
' corresponds to the failure life (Ni ni) is defined as ith
quarter of a cycle, the elastic strain amplitude f is
E indicator, Di is stated as,
very small compared to the plastic strain amplitude
(f). The total strain is given by the plastic strain
amplitude as shown in Eqn 3. The second part of the
curve describes the fatigue life for elastic strain cycles
( < y) which usually relate to HCF conditions. This
part of the curve represents a hypothetical fully known
curve (Kohout and Vechet 2001). The shape of the
curve is obtained by directly transforming the
previous fully known stress-life curve (Siriwardane et
al. 2007, 2008) into the elastic strain-life curve shown
below.
b'
N + Nu (4)
= e
N + N e transformed to load level i + 1 as below.
determined from the points (y, Ny) and (e, Ne). ()UHCF
is the ultimate high cycle fatigue strain which is the
elastic strain amplitude corresponding to a half reversal
(a quarter of a cycle) and expressed as,
u (5)
E
where u is the ultimate tensile strength of the material.
()UHCF can be obtained from a monotonic tension test of
the material. It is the true monotonic tension fracture
ductility.
2.2. Damage Indicator
The proposed damage indicator takes account of the
combined HCF and LCF damage due to variable
amplitude loading. The hypothesis behind this fatigue
law is that if the physical nature of the damage is the
same in the two cases, then fatigue life depends only on
the loading condition. Suppose a component is
subjected to a certain strain amplitude ()i for ni cycles
at load level i, and Ni is the fatigue life (number of
cycles to failure) corresponding to ()(i) (Figure 1), then
the reduced life at load level i is obtained as (Ni ni).
The equivalent strain ()(i)eq (Figure 1), which
level damage equivalent strain. Hence, the new damage
( )( i ) eq ( )i
Di = (6)
( ) u ( ) i
where ()u is given by
ULCF ( )i +1 y
( )u = (7)
UHCF ( )i +1 < y
Assuming that at the end of the ith loading level,
damage Di has been accumulated (occurred) as a
consequence of ()i + 1 loading cycles, the damage is

where e is the strain amplitude of the fatigue limit, Ne ( )(' i +1)eq ( )i +1 (8)
is the number of cycles to failure at the strain of e. The Di =
( ) u ( ) i + 1
parameters y and Ny are the yield strain and the
corresponding number of cycles to failure, where Nu is and ()u is expressed as
the number of cycles corresponding to the intersection
of the tangent line of the finite life region and the ULCF ( )i +1 y

horizontal asymptote of the ultimate elastic strain ( )u = (9)
amplitude ()UHCF as shown in Figure 1, and b is the UHFC ( )i +1 < y
slope of the finite life region of the curve which can be

Advances in Structural Engineering Vol. 15 No. 2 2012 289

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

A new damage indicator

Then, ()(i + 1)eq is the damage equivalent strain at
loading level i + 1 and is calculated as,
ni number of cycles ( )i

No
( )(i +1) eq = Di [( )u ( )i +1 ] + ( )i +1 (10)
If ( )i > ( )e ii+1

Ni number of cycles at ( )i
Thus the corresponding equivalent number of cycles
(from fig 1: strain-life curve) to failure N(i + 1)R is obtained from the strain-life curve
as shown in Figure 1. ()i + 1 is the strain at level i + 1
NiR = Ni ni : Residual life and supposing that it is subjected to n(i + 1) cycles, then
the corresponding residual life at load level i + 1, N(i + 1)R
( )(i )eq : Equivalent strain
(from fig 1: strain-life curve)
is calculated as,

If ( )i ( )y
No
( )u = ( )UHCF N ( i +1) R = N (i +1) R n( i +1) (11)

( )u = ( )ULCF Therefore, strain ()(i + 1)eq, which corresponds to

N(i + 1)R at load level i + 1, is obtained from the strain-life
( )(i )eq ( )i
curve as shown in Figure 1. Then the cumulative
Di = damage at the end of load level i + 1 is defined as,
( ) u ( )i

D = Di ( )( i +1) eq ( )i +1
D( i +1) = (12)
If D < 1
( ) u ( ) i + 1

ni + 1 : number of cycles at i +1 Fatigue failure At the first cycle the equivalent strain ()(i)eq is
No
equal to ()i and the corresponding damage indicator
If ( )i +1 ( )y ( )u = ( )UHCF becomes Di = 0. Similarly at the last cycle, the damage
indicator becomes Di = 1 when ()(i)eq is equal to ()u.
( )u = ( )ULCF Therefore, the damage indicator is normalized to unity
(Di = 1) at the fatigue failure of the material. The
Damage transformation from previous step to next step above procedure is followed until D i = 1. The
( )'(i +1)eq ( )i +1 proposed damage indicator-based algorithm is
Di = D'i = ( )'(i + 1)eq
( )u ( )i +1 summarized in the flow chart shown in Figure 2. Here,
the defined fatigue failure is the time taken for
initiation of a crack at the location of maximum stress
No
If ( )(i +1)eq > ( )e in the structural component and is applicable to a
uniaxial stress state only.
N'(i +1)R = : number of cycles ( )'(i +1)eq
(from fig 1: strain-life curve) 3. VERIFICATIONS OF PROPOSED FATIGUE
MODEL
N(i +1)R = N'(i +1)R ni +1 : Residual life In this section, fatigue test results for four materials and
fatigue lives predicted by a precise LCF model were
( )(i +1)eq : Equivalent strain for N(i +1)R number of compared with fatigue lives predicted by the proposed
cycles (from figure 1: strain-life curve) model. The objective of this section is to verify the
application of the proposed model for HCF and LCF
( )(i +1)eq ( )i +1 combined situations.
D(i +1) =
( )u ( )i +1 The proposed damage indicator has already been
verified by Siriwardane et al. (2008) for a HCF regime.
ii+1
However, it has not been verified for a LCF regime. For
LCF life estimation, Liu et al. (2005) proposed and
Figure 2. Flow chart of the proposed damage indicator verified a precise LCF fatigue model which uses

290 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun

Miners rule with three multiplication factors. Sub (a)

Strain
3a
section 3.1 gives the comparison of the results of the Liu a 1.5a 2a 2.5a
a
et al. (2005) model with the proposed damage indicator
in a LCF regime. Sub section 3.2 explains the a Cycles
1.5a 2a
verification of the proposed model under combined 2.5a
3a
HCF and LCF regimes by comparing with the
experimental test results. (b)

Strain
21a
14a 17a
7a 10a 7a
3.1. Comparison with Predicted Lives of a
Cycles
Precise LCF Model
Liu et al. (2005) proposed a LCF model for more accurate (c)
21a

Strain
life prediction. The model was verified with cyclic loading 17a
10a 14a
7a
test results for A36 solid square steel bars under constant 7a

amplitude loading. Data consisted of low cycle bending Cycles

fatigue results related to various loading histories. The 7a
10a 14a
model consists of a new damage indicator which was 17a
21a
obtained by modifying Miners rule with the three
(d)
multiplication factors: amplitude change factor, power

Strain
21a
13a 18a
14a
coefficient for relative strain and partial cycle factor. This 9a 10a 9a
7a
9a
5a 5a
model also uses a new cycle counting method based on a
4a 2a
half cycles. However, application of this model to other a 3a Cycles
2a
materials, was found to be less appropriate since model 4a
7a
11a 11a 11.5a
parameter determination procedures are lengthy and 13a 12a
18a
difficult. On the other hand, the corresponding cycle 20a 20a

counting technique for random loading is totally different
from the well-known cycle counting techniques used for
such as rainflow or reservoir counting, etc. In this section,
fatigue lives predicted by the proposed model are
compared with the Liu et al. (2005) model in the LCF
regime. The material considered is A36 steel with four
repeating block loading patterns as shown in Figure 3. The
value a represents the amplitude magnification and by
varying the value of a as shown in Table 1, a different
loading block is obtained for each pattern. For example in
pattern I, the value of a was given six different values
(pattern 1 column 2 in Table 1) and six loading blocks
were correspondingly obtained. It was considered that
each loading block repeated until the material failed (i.e.
Di = 1). For each history, the number of cycles to failure
was predicted using the proposed, Liu et al. model as well
as the previous model based on Miners rule. The
comparisons are given in Table 1 and shown in Figure 4.
For all four loading patterns, the previous model
overestimated the fatigue life. These comparisons indicate
that the proposed model correlates more closely with the
Liu et al. model for the LCF regime than do previous
model predictions for all loading patterns (I-IV). Since
there was a difficulty of obtaining A36 steel data which
was used for the verification the Liu et al. model, direct
verification of the proposed model using A36 steel data
was not carried out.
Figure 3. Loading patterns for A36 steel: (a) Pattern I; (b) Pattern
II; (c) Pattern III; (d) Pattern IV
3.2. Comparison with Experimental Lives of
Combined HCF and LCF Tests
The fatigue test results for four materials, available in the
literature (Pereira et al. 2008; Wong et al. 2002; Cook
1982; Colin and Fatemi 2010), were compared with the
fatigue lives predicted by the proposed model. The
materials were P355NL1 steel, 316L stainless steel,
Inconel 718, and S304L stainless steel. P355NL1 is a low
carbon steel which is used in the manufacture of pressure
equipment. 316L stainless steel has many applications in
such as jet engine parts and heat exchangers. Inconel 718
is a nickel-chromium based superalloy which is used in
the manufacture of turbines. 304L stainless steel is used
in many applications such as the cooling systems of
nuclear power plants. The tests were performed both in
the HCF and LCF regimes under different kinds of
variable amplitude loading, such as increasing or
decreasing step loadings, variable amplitude repeating
block loadings and random variable loadings.
3.2.1. Verification for P355NL1 steel
Six fatigue tests of P355NL1 steel were carried out
(Pereira et al. 2008) with two strain range steps of 1%

Advances in Structural Engineering Vol. 15 No. 2 2012 291

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

Table 1. Comparison of predicted lives by previous and proposed models with Lius life (Liu et al. 2005) for
A36 steel

Predicted fatigue life (cycles)

Pattern Amplitude magnification factor Liu et al. model Previous model* Proposed model
I 0.03 40.5 49 44
0.0275 54.5 64 58
0.025 75.5 84 79
0.0225 108.0 119 111
0.02 159.0 173 162
0.015 399.0 427 403
II 0.0045 33.0 43 38
0.004 49.5 61 56
0.0035 80.0 93 86
0.003 136.5 153 142
0.0025 251.0 271 254
0.002 523.0 548 517
III 0.0055 31.5 51 43
0.005 43.0 74 59
0.0045 63.0 101 82
0.004 92.5 140 119
0.0035 141.5 209 182
0.003 232.5 342 298
0.0025 414.5 608 532
IV 0.0055 35.5 70 58
0.005 52.5 89 76
0.004 113.0 178 154
0.0035 179.0 268 236
0.003 299.5 434 384
0.0025 542.5 764 684

*Coffin-Manson curve with Miners rule

and 0.5% according to the ASTM E606-92 standard.
The specimens were tested under two loading
sequences; increasing (pattern A) and decreasing
(pattern B) as shown in Figure 5. Three similar types
of specimen were tested for each loading sequence.
The tests were carried out in such a way that the first
block was applied for a specified number of cycles,
not causing material failure. The second block of
loading was then applied until failure was observed.
Further, all tests were strain controlled and conducted
in null strain ratio. As shown in Figure 5, two strain
amplitudes are higher than the yield strain of the
material. Therefore, the entire loading sequence of the
tests was in LCF regime. There, only the damage due
to LCF was verified. Damage due to combined HCF
and LCF was not verified on the P355NL1 material. A
detailed summary of the tests is given in Table 2.
The number of cycles to failure for these tests was
predicted by the proposed model. In addition, the
previous model based on Miners rule was used to
predict the number of cycles to failure. The results
obtained are given in Table 2 and shown in Figure 6.
Both previous and proposed models give close
estimates of fatigue life for increasing loadings
(pattern A). But predicted lives by the previous model
differ significantly from experimental lives observed
in the case of decreasing loading (pattern B). The
correlation coefficient for the proposed model
predictions with experimental results was estimated as
0.97. The correlation coefficient for the previous
model predictions with experimental results was
estimated as 0.64. The illustrations of Figure 6 and the
difference in the correlation coefficients is clear
evidence that the proposed method is the better
predictor.
3.2.2. Verification for 316L stainless steel
Twelve fatigue tests on 316L stainless steel were carried
out (Wong et al. 2002) under two loading patterns
(patterns A and B) under zero mean strain conditions as
shown in Figure 7. Initially, a percentage of LCF life
(corresponding to strain ranges of 0.8% and 1.6% for
patterns A and B, respectively) was applied followed by
HCF loading (strain range of 0.6%) until failure. That is,

292 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun

(a) 500 (b) 600

Previous model (miner's rule) Previous model (miner's rule)
Proposed model Proposed model

Predicted life (cycles)

Predicted life (cycles)

100
100

30 30
30 100 500 30 100 600
Liu's life (cycles) Liu's life (cycles)

(c) 700 (d) 800

Previous model (miner's rule) Previous model (miner's rule)
Proposed model Proposed model
Predicted life (cycles)

Predicted life (cycles)

100 100

30 30
30 100 700 30 100 800
Liu's life (cycles) Liu's life (cycles)

Figure 4. Comparison of predicted lives versus Lius lives (Liu et al. 2005) for: (a) loading pattern I; (b) loading pattern II; (c) loading
pattern III; (d) loading pattern IV

(a)
is given in Table 3. The number of cycles to failure in
Strain

these tests was predicted by the proposed model. In

0.01 addition, the previous model based on Miners rule was
used to predict the number of cycles to failure. The
0.005 results obtained are given in Table 3 and shown in
y = 0.002 Figure 8. The proposed model gives a closer prediction
of experimental fatigue life than the previous model for
Cycles both patterns A and B. The correlation coefficient for
the proposed model predictions with experimental
(b) results was estimated as 0.88. Similarly, the correlation
Strain

coefficient for the previous model predictions with

experimental results was estimated as 0.65. Therefore,
0.01
the illustrations of Figure 8 and the difference in the
correlation coefficients show that the proposed model is
0.005 the better predictor.
y = 0.002

3.2.3. Verification for Inconel 718 alloy

Cycles
Figure 5. Loading patterns used for P355NL1 steel: (a) pattern A
(increasing loading); (b) pattern B (decreasing loading)
tests were conducted under a combined HCF and LCF
regime. A detailed summary of the experimental results
Six fatigue tests of Inconel 718 were carried out (Cook
1982) both under increasing step loading (pattern A) and
variable amplitude repeating block loading (pattern B) as
shown in Figure 9. Four specimens were tested under
increasing loading with different numbers of strain range
steps. Two specimens were tested under variable

Advances in Structural Engineering Vol. 15 No. 2 2012 293

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

Table 2. Comparison of predicted lives with experimental lives for P355NL1 steel

First strain Second strain Experimental fatigue Predicted fatigue

step step life (cycles) life (cycles)
Strain Number of Strain Number of Test result Average Previous Proposed
Specimen range (%) cycles (n 1 ) range (%) cycles (n2) (n 1 + n2) life model* model
D201 1.0 337 0.5 2525 2862
D202 1.0 337 0.5 3600 3937
3399 9024 4577
D203 1.0 673 0.5 977 1650
D204 1.0 673 0.5 2727 3400
2525 5837 2007
D205 1.0 1010 0.5 1790 2800
D206 1.0 1010 0.5 1404 2414
2607 2641 1805
D301 0.5 3620 1.0 922 4542
D302 0.5 3620 1.0 1092 4712
4627 4440 4607
D303 0.5 7240 1.0 1507 8747
D304 0.5 7240 1.0 1044 8284
8515 7715 7972
D305 0.5 9377 1.0 1528 10905
D306 0.5 9377 1.0 1693 11070
10987 9648 9832
*Coffin-Manson curve with Miners rule

(a)
Strain

104

0.004
y = 0.0032
0.003
Predicted life (cycles)

Cycles

(b)
Strain

0.008
Pattern A-previous model
Pattern A-proposed model y = 0.0032
Pattern B-previous model 0.003
Pattern B-proposed model
Cycles
103
103 104
Experimental life (cycles)

Figure 6. Comparison of predicted lives versus experimental lives

for P355NL1 steel Figure 7. Loading patterns used for 316L stainless steel:
(a) pattern A; (b) pattern B

amplitude repeating block loadings. The maximum
strain amplitude of each block was increased in four
stages. These tests were conducted in combined HCF
and LCF regimes. A detailed summary of the tests is
given in Tables 4 and 5 for increasing step loading and
variable amplitude repeating block loading, respectively.
The number of cycles to failure for these tests was
predicted by the proposed model. In addition, the
previous model based on Miners rule was used to
predict the number of cycles to failure. The obtained
results are given in Tables 4, 5 and shown in Figure 10.
The correlation coefficient with the experimental results

294 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun

Table 3. Comparison of predicted lives with experimental lives for 316L stainless steel

First strain Second strain Predicted fatigue

step (LCF) step (HCF) life (cycles)
Specimen Strain Number of Strain Number of Experimental fatigue Previous Proposed
range (%) cycles (n 1 ) range (%) cycles (n2) life (cycles) model* model
A-1 0.8 1097 0.6 25477 26574 34204 31634
A-2 0.8 2742 0.6 20918 23660 31028 26991
A-3 0.8 4387 0.6 15597 19984 27852 22037
A-4 0.8 6581 0.6 8188 14769 23617 15996
A-5 0.8 8774 0.6 873 9647 19381 13448
A-6 0.8 10968 0.6 10968 15144 11520
B-1 1.6 125 0.6 22714 22839 33001 24843
B-2 1.6 313 0.6 20831 21144 28006 14444
B-3 1.6 501 0.6 17925 18426 23038 8150
B-4 1.6 751 0.6 9923 10674 16369 4338
B-5 1.6 1002 0.6 2130 3132 9726 2175
B-6 1.6 1252 0.6 1252 3057 1366
*Coffin-Manson curve with Miners rule

(a)

Strain
Pattern A-previous model
Pattern A-proposed model
Pattern B-previous model
Pattern B-proposed model
y = 0.0055
Predicted life (cycles)

104
Cycles

(b)
Strain

y = 0.0055

103
103 104 Cycles
Experimental life (cycles)
Figure 9. Loading patterns used for Inconel 718: (a) pattern A
Figure 8. Comparison of predicted lives versus experimental lives
(increasing type step loading); (b) pattern B (variable amplitude
for 316L stainless steel
repeating block loading)

for the proposed model predictions was estimated as
0.99. Similarly, the correlation coefficient for the
previous model predictions was estimated as 0.76. The
illustrations of Figure 10 and the difference between the
correlation coefficients is evidence that the proposed
method gives predictions which correlate better with
experimental results than does the previous model.
3.2.4. Verification for S304L stainless steel
Three fatigue tests on S304L stainless steel were carried
out (Colin and Fatemi 2010) under random loading.
These test results were used to verify the proposed
model. All tests were strain controlled and a summary of
the tests is given in Table 6. Experimental results were
compared with the lives predicted by the proposed
fatigue model as well as the previous model based on
Miners rule. The comparisons obtained are given in
Table 6.
In Table 6, the first two tests were in LCF regime
while other four were combined HCF and LCF tests. In
relation to the experimental results, the estimated
predictions of the proposed model show a percentage

Advances in Structural Engineering Vol. 15 No. 2 2012 295

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

Table 4.Comparison of predicted lives with experimental lives of Inconel 718 for increasing type step loading

Predicted fatigue
life (cycles)
Stress range Strain Ni Experimental fatigue Previous Proposed
Specimen (GPa) range (%) (cycles) life (cycles) model* model
N-7 1.02 0.5 7600
1.22 0.6 3400
1.44 0.7 1680
1.63 0.8 1000
1.78 0.9 640
1.89 1.0 460
1.94 1.1 260
15040 14102 18959
L-6 1.91 1.1 510
1.94 1.2 360
1.97 1.3 290
1.98 1.4 220
2.00 1.5 170
2.03 1.6 140
2.02 1.7 110
1800 759 2276
N-8 1.63 0.8 2360
1.86 1.0 1065
1.95 1.2 560
1.98 1.4 325
4310 2588 5501
T-7 1.41 0.7 4000
1.71 0.9 1830
1.84 1.1 800
1.85 1.2 600
7230 4779 9443
*Coffin-Manson curve with Miners rule

Table 5. Comparison of predicted lives with experimental lives of Inconel 718 for variable amplitude
repeating block loading

Predicted fatigue
life (cycles)
Stress Strain N Experimental fatigue Previous Proposed
Specimen range (GPa) range blocks life (cycles) model* model
M-6 1.70 0.9 217
1.88 1.0 150
1.98 1.2 100
1.97 1.4 400
16473 7441 22639
X-7 1.62 0.8 150
1.93 1.1 150
1.89 1.2 150
1.89 1.3 400
16150 5956 24559
*Coffin-Manson curve with Miners rule

variation of 15.4%, whereas for the previous model the 4. CASE STUDY: FATIGUE LIFE
figure is 26.6%. Therefore, fatigue lives predicted by the ESTIMATION OF A BRIDGE MEMBER
proposed model are more accurate than previous model Fatigue life estimation of a bridge member is discussed
predictions. in this section. The evaluations are especially based on

296 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun
Pattern A-previous model
Pattern A-proposed model
Pattern B-previous model
104 Pattern B-proposed model
Predicted life (cycles)
103
102 2
10 103
Experimental life (cycles)
Figure 10. Comparison of predicted lives versus experimental lives
for Inconel 718
secondary stresses and strains, which are generated in
the region of the riveted member connection due to
stress concentration effects caused by normal traffic and
by earthquake loadings.
4.1. Considered Member
One of the bracing members of the longest railway
bridge in Sri Lanka (Siriwardane et al. 2007, 2008) was
selected for fatigue life estimation. The selected
member is shown in Figure 11(b). Its thickness is small
compared to its other dimensions. Previous time-history
analysis of the global structure of this bridge reveals that
this member is also one of the highly stressed members
when subject to earthquake loading.
Table 6. Comparison of predicted lives with experimental lives for S304L stainless steel
Maximum Minimum
Test strain strain
1 0.01
2 0.01
3 0.005
4 0.005
5 0.003
6 0.003
*Coffin-Manson curve with Miners rule
Advances in Structural Engineering Vol. 15 No. 2 2012
4.2. Stress-Strain Analysis
The combined HCF and LCF damage was evaluated
based on the state of strain when a release of contact
(tightness) of rivets occurs and all the riveted locations
have no clamping force. The clamping force is
generally defined as the compressive force in the plates
which is induced by the residual tensile force in the
rivet. The residual force in a rivet arises when the rivet
becomes shortened in length due to cooling, after the
hot rivet is inserted into the holes in the plates in order
to connect them, and a second head is formed from the
protruding shank. This clamping force generates a
triaxial state of stress in the connected plate
(Siriwardane et al. 2007, 2008). Since this study
assumes that the riveted locations have no clamping
force (value of clamping force is zero), the connected
104
members are considered to be subject to a biaxial stress
state. Therefore, a critical member without rivets can be
analyzed assuming a biaxial state of stress using a 2D
finite element analysis. Allowing for the symmetry of
geometry, boundary and loading, one half (hatched
area) of the member was subjected to FE analysis as
indicated in Figure 11(c). Nine node isoperimetric shell
elements were used for the FE analysis as shown in
Figure 12(a). The actual air gap restraint conditions
were considered in the model to represent unilateral
contact (ghe) between the rivet and the plate. Roller
supports were provided along the boundary (bc) to
allow for longitudinal elongation of the plate. The
interaction between holes was considered by the
assigned boundary condition. To achieve continuity of
the stress field between the global structure and sub
structure, it is necessary to consider any interface
between the two structures at every iterative step. In
this model, therefore, the primary tensile stress history
(Figure 13) of the member, which had been obtained
Experimental life (blocks) Predicted life (blocks)
Previous Proposed
Test life Average life model* model
0.01 211
0.01 196
203 259 239
0.005 1601
0.005 1740
1671 2789 2118
0.003 14463
0.003 15666
15065 20122 10075
297
 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

(a) (c)
30 mm 60 mm 30 mm

Rivet hole
60 mm diameter 20 mm

(b)
120 mm

Uniform cyclic loading

Considered member Considered connection Note: Hatched area is subjected to FE analysis

Figure 11. Considered member: (a) general view of the bridge; (b) close view of considered connection and member; (c) geometric details
of connection

(a) (b) (c)

d c
Highly
stressed
location
during
earthquake
loading
(454.8 MPa)
h
e g
f

Highly
stressed
location
during the
service
loading
(314.3 MPa)

a b

Figure 12. Stress analysis: (a) FEM mesh; (b) maximum von Misses stress distribution during service loading; (c) maximum von Misses
stress distribution during earthquake loading

from measured strain histories at the mid span of the
bridge under normal routine traffic loading was applied
to the bottom face (ab) as uniform pressure. The
position of the ab boundary of the sub model was
determined considering the distribution of the far field
primary stress in the member. The maximum stress
contours obtained are shown in Figure 12(b). This
shows that operating stresses are below the yield limit
of the material (therefore are in the HCF regime) and
the highly stressed locations are subjected to uniaxial
stress states. Assuming that a single day traffic
sequence is repeated every day, a one day time history

298 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun

60
0.0015
Primary stress (MPa)

Secondary strain
50

40 0.001

30
0.0005
20

10
0 100 200 300 400 500 600 700
0 100 200 300 400 500 600 700 Traffic sequence
Traffic sequence
Figure 15. Secondary strain variation of the critical location with
Figure 13. Primary stress variation of the member with traffic traffic sequence per single day
sequence per single day

300
was taken as the loading block in this study. The 200

Primary stress (MPa)

obtained secondary stresses and strain histories for
100
normal traffic loading are shown in Figures 14 and 15,
respectively. 0
In this case study, fatigue life was calculated for 100
damage caused by the normal traffic (HCF) combined
200
with earthquake (LCF) loadings. The elasto-plastic
analysis was then conducted by applying (to the ab 300
0 10 20 30 40 50
interface) the primary stress history (Figure 16) which Time (sec)
was obtained by time history analysis of the global
structure under earthquake loading. The maximum Figure 16. Primary stress history of the member during the
stress contours obtained are shown in Figure 12(c) and earthquake
it was decided that the state of stress is uniaxial. The
secondary stresses obtained and strain histories for 450
Secondary stress (MPa)

earthquake loading are shown in Figures 17 and 18, 300

respectively. 150
The secondary strain variations obtained (Figures 15 0
and 18) are complex and also of irregular shape. These
150
strains should be reduced to a series of equivalent strain
cycles at zero mean strain. In order to achieve this 300
objective, initially the famous rainflow cycle counting 450
technique (Dowling and Socie 1982) was used to 0 10 20 30 40 50
identify the strain ranges and sequences of closed strain Time (sec)
cycles. The modified Goodman relation was then used Figure 17. Secondary stress history of the critical location during
the earthquake

300
to transfer these counted cycles to equivalent mean
Secondary stress (MPa)

250 strain zero stabilized cycles.

200

150
4.3. Fatigue Curves
The mechanical and cyclic properties of the bridge
100
material are given in Table 7. From these properties,
50 ULCF, UHCF, y, e, Ny and Ne are estimated as 0.95,
0.00234, 0.00163, 0.00117, 1.076 105 and 5.83 105 ,
0 100 200 300 400 500 600 700
respectively as explained in sub section 2.1. Then, b and
Traffic sequence
Nu were estimated as 0.19 and 8426, respectively (sub
Figure 14. Secondary stress variation of the critical location with section 2.1). From these parameters, the corresponding
traffic sequence per single day stress-life curve for HCF and the strain-life curve for

Advances in Structural Engineering Vol. 15 No. 2 2012 299

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

0.1 (a)
Log ( )
0.075
Secondary strain

0.05
0.025
0 475
0.025
0.11
0.05 N + 1069
= 235
N + 5.83 105
0.075
0.1
0 10 20 30 40 50
Time (sec)
235
Figure 18. Secondary strain history of the critical location during
the earthquake
1069 5.83 105 Log (N)
Table 7. Mechanical and cyclic properties of bridge
(b)
material Log ( )

Property Value 0. 95000

Youngs modulus (GPa) 203 986.5 (2N )0.11 + 0.95(2N ) 0.64

=
Yield strength (MPa) 330 203 103
Ultimate tensile strength (MPa) 475
Fatigue strength (MPa) 235
Fatigue strength coefficient (MPa) 986.5 0.00234 N + 8426 0.19
Fatigue strength exponent 0.11 0.00163 = 0.00117
N + 5.83 105
Fatigue ductility coefficient 0.95
Fatigue ductility exponent 0.64

0.00117

combined HCF and LCF loading for the bridge material
under consideration were constructed as shown in
Figure 19.
4.4. Fatigue Life Estimation
Until an earthquake occurrence, the member was
subjected only to HCF damage caused by normal traffic
loads. The damage due to usual traffic loads (HCF) was
therefore evaluated using a previously proposed HCF
model (Siriwardane et al. 2008) with the stress-life
curve shown in Figure 19(a) since accurately captures
the HCF loading sequence effect. In the case of the
proposed combined HCF and LCF model, the gradient
in the elastic region of the proposed strain life curve (b)
is higher than that of the strain life curve (b) used for
HCF life prediction (Siriwardane et al. 2008). For this
reason, lives predicted by the proposed model for
purely HCF conditions exceed actual fatigue lives.
Therefore, the proposed model cannot be applied when
purely HCF conditions apply. During an earthquake,
however, the member is subjected to LCF damage.
When the bridge is again used after the earthquake, the
damage due to normal traffic is less than the damage
caused by normal traffic when it had acted alone. For
this reason, the previously proposed model cannot be
used for damage estimation of usual traffic after the
8426 1.076 105 5.83 105 Log (N)
Figure 19. Fatigue curves for considered bridge material for the
use: (a) before the earthquake; (b) during and after the earthquake
earthquake. Therefore, during and after an earthquake,
fatigue damage has to be evaluated using the proposed
combined HCF and LCF model (section 2). The
updated damage (damage accumulation) was calculated
by following the flow chart given in Figure 2. The
fatigue life was estimated when the damage indicator
reached unity.
An earthquake was assumed to occur at different times
in the bridge life (5, 10, 20, 30, 40, 50 and 54 years). It
was assumed that the normal traffic load resumed
following the earthquake. The fatigue life of the member
was estimated using three approaches: (1) proposed
model; (2) previous model (Coffin-Manson relationship
with Miners rule); (3) Miners rule with the proposed
strain-life curve. The obtained results are given in Table
8. The results indicate that LCF damage by earthquake
loading produces an appreciable reduction of bridge life.
For the proposed model, percentage reduction of life was
higher if the earthquake occurs at the beginning of the
bridge life when compared to those occurring at later
times. Comparison of fatigue life reveals that the

300 Advances in Structural Engineering Vol. 15 No. 2 2012

Kamal Karunananda, Mitao Ohga, Ranjith Dissanayake, Sudath Siriwardane and Pang-jo Chun

Table 8. Comparison of calculated fatigue lives of the riveted connection

Previous model Proposed curve with

(Miners rule) Miners rule Proposed model
Time of Fatigue Percentage Fatigue Percentage Fatigue Percentage
earthquake* life reduction life reduction life reduction
(years) (years) of life (%) (years) of life (%) (years) of life (%)
5 26.5 35.4 34.0 17.0 22.0 63.3
10 26.5 35.4 34.0 17.0 24.0 60.0
20 26.5 35.4 34.0 17.0 29.0 51.7
30 30.0 26.8 34.0 17.0 34.0 43.3
40 40.0 2.4 41.0 31.7
50 50.5 15.8
54 54.0 10.0
Without 41.0 41.0 60.0
earthquake

*After construction

proposed model predictions differ from the previous ACKNOWLEDGEMENTS

model predictions. For the previous model, the reduction
of service life is constant irrespective of the time of the
earthquake occurrence. If Miners rule is used with the
proposed strain-life curve, fatigue life predictions are
increased over the previous model predictions. This
verifies that the proposed strain-life curve better
represents the combined HCF and LCF behaviour than
the Coffin-Manson relationship. The differences in the
case study results confirm the importance of an accurate
combined HCF and LCF model when estimating the
fatigue life of existing connections.
5. CONCLUSIONS
A new fatigue life prediction model for combined
HCF and LCF damage is proposed. The model was
verified by comparing predicted lives with
experimental lives for four materials subjected to
variable amplitude loading. It was shown that the
proposed fatigue model gives a realistic fatigue life
for combined HCF and LCF damage for variable
amplitude loading situations where detailed stress
histories are known. The proposed fatigue model was
utilized to estimate the fatigue life of a bridge
member. The case study recognized the need to take
account of earthquake induced LCF damage in
addition to the HCF damage caused by the normal
routine traffic loading in steel bridges. The
importance of accurate life prediction under
conditions of combined HCF and LCF damage was
also confirmed. Since the proposed model only relates
to uniaxial stress states, it is recommended that these
models be extended to take account of multiaxial
stress states in the future.
The authors wish to express their sincere gratitude to
Professor M.P. Ranaweera and the team of experts who
worked on the Sri Lankan Railway Bridge project, for
their expert advice in carrying out this research. The
kind support given by the Sri Lanka Railways (SLR) is
also appreciated.
REFERENCES
Chen, W.F. and Lui, E.M. (1987). Handbook of Structural
Engineering, CRC Press, New York, USA.
Constantinescu, A., Van, K.D. and Maitournam, M.H. (2003). A
unified approach for high and low cycle fatigue based on
shakedown concepts, Fatigue Fracture of Engineering
Materials and Structures, Vol. 26, No. 6, pp. 561568.
Cook, T.S. (1982). Stress-strain behavior of Inconel 718 during low
cycle fatigue, Journal of Engineering Materials and
Technology, Vol. 104, No. 3, pp. 186191.
Dowling, S.D. and Socie, D.F. (1982). Simple rainflow counting
algorithms, International Journal of Fatigue, Vol. 4, No. 1, pp.
3140.
Kim, Y.H., Song, J.H. and Park, J.H. (2009). Expert system for
fatigue life prediction under variable loading, Expert System
with Applications, Vol. 36, No. 3, pp. 49965008.
Kohout, J. and Vechet, S. (2003). A new function for fatigue curves
characterization and its multiple merits, International Journal of
Fatigue, Vol. 23, No. 2, pp. 175183.
Kondo, Y. and Okuya, K. (2007). The effect of seismic loading on
the fatigue strength of welded joints, Material Science and
Engineering A, Vol. 468470, pp. 223229.
Liu, W., Liang, Z. and Lee, G. (2005). Low cycle bending fatigue
of steel bars under random excitation. Part II: design
considerations, Journal of Structural Engineering, ASCE, Vol.
131, No. 6, pp. 919923.

Advances in Structural Engineering Vol. 15 No. 2 2012 301

 New Combined High and Low-Cycle Fatigue Model to Estimate Life of Steel Bridges Considering Interaction of
High and Low Amplitudes Loadings

Mesmacque, G., Garcia, S., Amrouche, A. and Rubio-Gonzalez, C. NOTATION

(2005). Sequential law in multiaxial fatigue, a new damage total strain amplitude
indicator, International Journal of Fatigue, Vol. 27, No. 4, pp. el. elastic strain amplitude
461467. pl. plastic strain amplitude
Miner, M.A. (1945). Cumulative damage in fatigue, Journal of e strain amplitude of the fatigue limit
Applied Mechanics, Vol. 12, pp. 159164. f fatigue ductility coefficient
Pereira, H.F.S.G., de Jesus, A.M.P., Fernandes, A.A. and Ribeiro, y yield strain amplitude
A.S. (2008). Analysis of fatigue damage under block loading in UHCF ultimate strain of high cycle fatigue
a low carbon steel, Strain, Vol. 44, No. 6, pp. 429439. ULCF ultimate strain of low cycle fatigue
Siriwardane, S.C., Ohga, M., Dissanayake, R. and Taniwaki, K. f fatigue strength coefficient
(2007). Different approaches for remaining fatigue life u ultimate tensile strength
estimation of critical members in railway bridges, International Nu number of cycles corresponding to the
Journal of Steel Structures, Vol. 7, No. 4, pp. 263276. intersection of the tangent line of the finite life
Siriwardane, S., Ohga, M., Dissanayake, R. and Taniwaki, K. region and the horizontal asymptote of ultimate
(2008). Application of new damage indicator-based sequential strain amplitude
law for remaining fatigue life estimation of railway bridges, Ne number of cycles corresponding to the fatigue
Journal of Constructional Steel Research, Vol. 64, No. 2, pp. limit
228237. Ni fatigue life corresponding to i
Suresh, S. (1998). Fatigue of Materials, Cambridge University Ny number of cycles corresponding to yield strain
Press, Cambridge, UK. ni applied number of cycles of i
Wong, Y.K., Hu, X.Z. and Norton, M.P. (2002). Plastically Di damage indicator after ith loading
elastically dominant fatigue interaction in 316L stainless steel E elastic modulus
and 6061-T6 aluminium alloy, Fatigue Fracture of b fatigue strength exponent
Engineering Materials and Structures, Vol. 25, No. 2, pp. c fatigue ductility exponent
201213. b slope of the finite life region

302 Advances in Structural Engineering Vol. 15 No. 2 2012