Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Materials and Design 51 (2013) 916–923

Contents lists available at SciVerse ScienceDirect

Materials and Design


journal homepage: www.elsevier.com/locate/matdes

Technical Report

A thermographic method for remaining fatigue life prediction of welded


joints
P. Williams a, M. Liakat a, M.M. Khonsari a,⇑, O.M. Kabir b
a
Department of Mechanical Engineering, Louisiana State University, Baton Rouge, LA 70803, USA
b
Cameron International, Houston, TX 77041, USA

a r t i c l e i n f o a b s t r a c t

Article history: An in situ technique for predicting the Remaining Fatigue Life (RFL) of metal specimens is used to study
Received 1 March 2013 welded joints. Uniaxial tension–compression tests were carried out using welded and heat treated tubu-
Accepted 27 April 2013 lar specimens of carbon steel 1018. An infrared camera was used to capture thermographic data from the
Available online 9 May 2013
specimens during testing. Short-time excitation (STE) tests were used to determine the slope of temper-
ature increase at several points throughout the fatigue life to characterize the damage. Verification tests
were carried out at different stress levels to study the accuracy of the RFL prediction method on welded
specimens. The presented results show good agreement between the predicted and experimentally mea-
sured fatigue lives. However, tests with welded specimens containing flaws showed significantly shorter
fatigue lives. The prediction method discussed in this paper does not characterize the RFL of flawed
specimens.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction applied load and frequency. It is this change in Rh that allows for a
prediction of consumed fatigue life to be made. To evaluate the Rh,
Welds are commonly used in the manufacture of many compo- a specimen is stopped during a fatigue test, naturally cooled, and a
nents to form the desired shape. However, the process often yields short excitation test (STE) is run to generate a measurable temper-
undesirable material inconsistencies and residual stress in the fin- ature rise captured via an Infra-red (IR) Camera. This process is
ished product which tends to reduce the strength and fatigue life. then repeated throughout the fatigue life at predefined intervals.
Clearly, therefore, a reliable prediction of the fatigue life of welded With the base-line Rh evolution of a specimen characterized, a
components is of great interest to the industry. Yet a review of the short excitation test can be performed on a fatigued specimen to
open literature reveals that there is a general paucity of germane obtain a temperature rise and then comparing the resulting Rh to
information on this subject. the baseline results, the remaining fatigue life can be predicted [1].
It is perhaps appropriate to begin by examining the available Thermography has also been utilized by Naderi and Khonsari
relevant Non Destructive Testing (NDT) techniques for the evalua- [2] to study failure of specimens by entropy developed through fa-
tion of fatigue life of unwelded specimens. Some progress has been tigue. Analyzing this phenomenon allow for the development of a
reported by means of thermographic methods. For example, Amiri health-monitoring system based upon the calculated entropy built
and Khonsari [1], studies the fatigue behavior of Stainless Steel 304 up during fatigue loading. The presented results show promise for
in a rotating-bending apparatus, showed promising results for pre- the use of thermodynamic entropy as a measure of damage accu-
dicting the consumed fatigue life through analysis of the initial mulation and failure prediction materials. The concept of thermo-
slope of temperature rise Rh. The method shown in that paper is ex- dynamics related to material degradation has been investigated
tended in this study in a series of laboratory tests to estimate and developed by Bryant et al. [3], with this studies leading
remaining fatigue life of the welded specimens subjected to uniax- towards the entropic method of fatigue analysis. Further work on
ial tension–compression tests. this topic has also been carried out by Amiri and Khonsari [4]
The variable Rh, discussed by Amiri and Khonsari [1] changes and is discussed in detail by Khonsari and Amiri in ‘Introduction
throughout the fatigue life of a specimen. According to their re- to thermodynamics of fatigue’ [5], reinforcing the promise shown
sults, the value of Rh measured during the fatigue life is greater by an entropic approach to fatigue studies.
than that measured near the beginning of the test under the same Meneghetti [6] utilized thermography and presented an energy
release based parameter for fatigue analysis. The theoretical model
developed shows good correlation with experimental results, and
⇑ Corresponding author. Tel.: +1 2255789192; fax: +1 2255785924. the paper details results for variable loading in fatigue which also
E-mail address: khonsari@me.lsu.edu (M.M. Khonsari).

0261-3069/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.matdes.2013.04.094
P. Williams et al. / Materials and Design 51 (2013) 916–923 917

Nomenclature

E elastic modulus (GPa) Rh0 initial slope of temperature rise (°C/s)


F frequency (Hz) r stress amplitude (MPa)
N number of cycles rUTS ultimate tensile strength (MPa)
Nf number of cycles to failure ry yield stress (MPa)
R load ratio
Rh slope of temperature rise (°C/s)

agree well with the model. Dissipated heat energy has also been Spot welded joints have been analyzed using equivalent struc-
shown to characterize fatigue with by Rösner et al. [7] further tural stress as reported by Kang [13], with good agreement be-
emphasizing the utility of thermography in the study of a materials tween the theoretical model and published experimental results.
fatigue behavior. The equivalent structural stress approach is based on Von Mises
Matsumoto [8] presented results of fatigue evaluation using equivalent stress equation. The developed equation was compared
three different Non-destructive methods. They were: Acoustic to experimental results for tensile-shear loading and multiaxial
Impedance, Magnetic Leakage flux and a Thermal method. Acoustic loading, and in both cases showed good correlations with experi-
impedance was measured throughout the fatigue life of the speci- mentally obtained data.
men using an ultrasound sensor, a magnetic sensor was used to Park and Kang [14], also reported on fatigue life prediction of
study the change in magnetic field of the specimen, and an IR cam- spot welded joints utilizing Back propagation Neural Networks
era used for temperature measurements. All three methods of (BNN). This method combines multiple inputs such as load ratio,
analysis show a correlation with the fatigue progression of the nugget size and load angle, and creates a single output which is
specimen, with some slight fluctuations, showing promising re- the predicted fatigue life of the spot welded joints. One of the main
sults of studying fatigue progression in a specimen using a combi- advantages offered by this method is the ability to be trained from
nation of methods. experimental data, to make prediction with multiple input vari-
Various other methods for studying weld behavior under fati- ables. The results obtained from the BNN were compared to exper-
gue loading have been reported. A summary of the commonly used imental data, showing reasonably good correlation for tension and
procedures up to 2005 is given by Crupi et al. [9] The most com- shear loading, and tensile-shear loading.
mon analytical methods include modified Linear Elastic Fracture Recent work carried out by Fan et al. [15] studies the fatigue
Mechanics approach (LEFM), such as Notch Stress Intensity Factor behavior of welded joints using lock-in thermography, and pre-
(NSIF) and Stress Averaging Approach (SAA) as well as the Critical sents reasonable prediction of the weld fatigue lives. The paper
Distance method (CDM). According to Crupi et al. all these methods continues to look into energetic approaches to weld life prediction
are applicable to cruciform specimen with fillet welds, T-joint using with different loading sequences. Overall the results show a
welded specimens and some partial penetration butt-welded spec- good agreement between the thermographic prediction method
imens. The four methods used for predictions are Crack Modeling and the traditional Stress–Number of cycles to failure (S–N) curve
Method (CMM), NSIF, CDM and SAA. The results provided show approach.
reasonable agreement for all methods; generally, however, CMM A thermographic study of welds has also been carried out by
is reported to be the most accurate, and SAA the least accurate Crupi et al. [16] who examined the behavior of flat welded speci-
for most of the different geometries. mens subject to Low-Cycle Fatigue (LCF). The paper discusses the
A further notch based approach is considered by Bilous and heat treatment of the specimen to relieve the stresses from the
Lagoda [10], using geometry and microstructure differences cre- joining process. The same heat treatment procedure was employed
ated by welded joints in specimens to define a stress concentration as a basis for the preparation of tubular 1018 specimens for the fa-
factor. The factor Kf was reported to vary throughout the life of the tigue tests in the present study. The results show a difference in the
steel welded joints, initially increasing and then decreasing closer behavior of the base metal and the weld metal plotted on a tem-
to failure, however in one of the steels tests Kf did continue to rise perature-vs.-cycles graph. Both the welded and base materials
throughout the fatigue test. A difference in the Kf behavior was also showed a three phase behavior to failure in the temperature pro-
discussed and determined to be related to the material, as the file, also seen by Amiri and Khonsari [1] for specimens without a
change in Kf varied between medium and high strength structural weld (or unwelded specimens). This suggests that despite the
steels. weld, the behavior and failure of the specimen should allow for
Finite element approaches have also been considered in the the use of the Rh method for predicting fatigue life. A brief descrip-
study of welded joints and their fatigue behavior. Alam et al. [11] tion of the method follows.
considered both experimental and theoretical weld geometries When subjected to a fatigue load, a specimen generates heat
for laser-welded joints. The experimental and numerical results due to the release of hysteresis energy and there is a relationship
showed the geometry of the weld was significant in determining between material temperature rise and degradation. [1,17–19]
the values and locations of the peak stresses in the joints, and dis- When loaded from a pristine condition there is an initial tempera-
covered geometries that should be avoided to maintain the fatigue ture rise whose slope is Rh. Fig. 1 illustrates how Rh changes with
life of welded joints. Further finite element work has been carried load, with a higher Rh value measured at greater stress levels.
out by Jiang et al. [12] investigating the effects of welding heat in- This parameter has been used to predict the percentage of con-
put and weld layer number on the residual stress in welds. The re- sumed life (PCL) of pristine specimens, with good accuracy. It has
sults show the layers and heat input have a significant impact on not yet been applied to specimens with a stress concentration, or
the strength of repaired welded joints, and emphasizes the compli- material inconsistency such as a weld. Given that with appropriate
cations in the analyzing welded joints due to the large number of heat treatment welds can behave in a similar manner to the base
variables involved. material they have been used to join, this suggests that the method
918 P. Williams et al. / Materials and Design 51 (2013) 916–923

Fig. 3. Photo of welded specimen, unpainted.

constant throughout a test series, therefore F0 has an effect on all


values of Rh:

Rh ¼ F N rmþ1
a þ Rh0 ð6Þ

with FN = F(N). As excitation test parameters remain constant


Fig. 1. Variation of Rh with change in stress.
throughout a test series, function F varies only with number of fati-
gue cycles.
maybe applicable to welded specimens, if the behavior is relatively
uniform.
3. Material and experiment

2. Theoretical background 3.1. Materials and equipment

Meyendorf et al. [20] presented the following equation to relate The test specimens were manufactured from 1018 carbon steel,
the cyclic temperature rise to the loading conditions on the and were machined into a dog bone shape designed according to
specimen. ASTM: E-466-07. A thread tapping cutter was used to cut the spec-
Rh ¼ F rmþ1 ð1Þ imen circumferentially around the center, creating a 60° v-notch.
a
The specimen was then Titanium Inert Gas (TIG) welded around
where F is a function related to material microstructure, loading the circumference with ER708-2 filler material, with a variable
frequency and temperature, and m is an empirically determined voltage to ensure a good quality connection, before being ma-
material constant. chined again to the final dimensions. Smooth welded joint speci-
F ¼ Fðmicrostructure;loading frequency;load amplitude;temperatureÞ mens have been used in other fatigue studies by Crupi et al. [16]
and Malarvizhi et al. [21]. Re-boring was also carried out, to ensure
The temperature rise experienced by specimens under excita- the removal of any weld bead from inside and the outside of the
tion loading conditions is relatively small, due to the short loading specimen, as weld beads are known to cause early crack formation
time of the excitation test, thus temperature does not have a signif- leading to premature failure [22]. Specimens were heat treated
icant effect on the function F, therefore: after completion of the final machining processes to relieve any
F ffi Fðmicrostructure; loading frequency; load amplitudeÞ ð2Þ residual stresses caused during the manufacturing processes. To
ensure high stress relief, the specimen was heated at 200 °C per
Loading frequency and load amplitude used to obtain Rh are de- hour to a temperature of 650 °C, held for 4 h, and cooled at the
fined as the excitation loading conditions: same rate [13,23]. After the process was completed, the specimen
F ffi Fðmicrostructure; excitation loading conditionsÞ ð3Þ was polished to remove any surface scratches or abnormalities.
The specimen was then sprayed with black paint, to increase the
Amiri and Khonsari [1] described the increase of material deg- thermal emissivity of the surface for thermography analysis.
radation throughout fatigue loading, therefore relating material Fig. 2 shows the dimensions of the welded specimens, and Fig. 3
microstructure to N, the number of fatigue cycles. is an example of the finished 1018 carbon steel specimen.
Prior to fatigue testing, a welded specimen was subjected to a
F ffi FðN; excitation loading conditionsÞ ð4Þ
static test to determine some basic mechanical properties, carried
At the onset of cyclic loading (N = 0) a pristine specimen is ide- out according to ASTM: E8, which are listed in Table 1. The hard-
ally assumed to be defect free. This initial temperature rise is de- ness of the specimen was also measured as it provides a compari-
fined as Rh0. From Eqs. (1) and (4), it can be written as: son point for any further specimens that are made to ensure
similar specimen behavior.
Rh0 ¼ Fðexcitation loading conditionsÞ ð5Þ
Fatigue tests are carried out at a variety of stress levels, at a fre-
where F0 is function obtained when (N = 0). All excitation loading quency, f, of 10 Hz and a load ratio, R (the ratio of minimum stress
parameters (load ratio, frequency and stress amplitude) are kept over maximum stress) of 0.5 The uniaxial tension compression

9.52mm
r=37mm

16mm
= 6.27mm
38mm 28mm

Fig. 2. Schematic of the welded specimen.


P. Williams et al. / Materials and Design 51 (2013) 916–923 919

Table 1
Material properties of 1018 steel.

Yield stress, Tensile strength, Elastic Hardness,


ry (MPa) rUTS (MPa) modulus, E HRB
(GPa)
Pristine 305 575 140.8 71
Welded 295 429.3 193.5 60.9

tests were carried out using a TestResources axial-torsion, servo-


Fig. 5. Diagram of experimental loading conditions.
hydraulic fatigue tester rated at 50 kN axial load and 2 kN torsional
load. The temperature evolution was captured using an infra-red
(IR) camera, Mikron 7500, with a temperature range of 0–500 °C,
320  240 pixel resolution and a sensitivity/Noise Equivalent Tem-
perature Difference (NETD) 0.08 °C at 30 °C, and an accuracy
of ±2%. A data acquisition rate of 1 frame per second was used to
capture the surface temperature during the test.

3.2. Experimental procedure

Fig. 4 shows the equipment that was previously described in


Section 3.1 in the set up used to collect experimental data. Thermo-
Fig. 6. Loading sequence for test specimens.
graphic data is collected from the specimen via IR camera through-
out the fatigue loading and STE tests.
The experimental procedure is based on two different cyclic
loading scenarios. First there are the STE tests—used to find the Initially the specimens are subjected to an excitation load for
initial slope of temperature evolution at predefined intervals approximately 15–20 s to determine the initial slope of tempera-
throughout the entire fatigue life—and the second loading is the ture rise. After this step, the specimen is then cycled at the defined
fatigue loading, following ASTM: E466-07. Figs. 5 and 6 show a fatigue load until the operator chooses to stop the experiment. The
diagram of the two different loading conditions and the loading stopping point in the fatigue test can, of course, vary depending on
sequence applied. the stress level and the expected length of the fatigue test. Once
For the STE tests, a constant stress level of 273 MPa was applied, the test is stopped, the specimen is allowed to cool until it returns
at a load ratio of 0.545. These are determined by the operator to to ambient temperature, to ensure during the subsequent excita-
create a measurable temperature rise. The fatigue tests had a con- tion test the true temperature rise is measured. This process is re-
stant load ratio of 0.5, and the stress levels applied varied from peated until the specimen fails. The entire fatigue life of a
320 MPa to 240 MPa for different tests to cover both low- and specimen interrupted at several intervals during the test is as-
high-cycle fatigue. sumed to remain unchanged as if it was continuously fatigued
without interruption. This is proved by the verification tests pre-
sented in this paper.
At the conclusion of the experiment, the temperature rise deter-
mined from the data and IR-camera images captured and saved in
the computer by the data acquisition system. In previous tests with
pristine specimens, the area of highest temperature occurred
approximately in the center of the specimen, allowing for simple
measurement in post processing [24]. The technique for obtaining
the temperature rise at the hotspot on the specimen was to apply a
short line, approximately 5 mm in length to average the tempera-
ture along this short line. Fig. 7 shows an illustration of a typical IR
camera image and the data averaging line.
The welded specimen behaved slightly differently compared to
the unwelded specimens. The hotspot tended to appear in the base
material shown in Fig. 8, either above or below the weld and mov-
ing slightly during the excitation tests; see Fig. 9. Due to this phe-
nomenon it was not possible to use the same data averaging point
and obtain good results.
The method employed involves selecting a large rectangular
area that covers the majority of the gage section of the specimen,
thus allowing for capture of the all the temperature variation
regardless of the position. Fig. 10 shows the rectangle data averag-
ing point on the gage section of a test specimen, significantly larger
than the previous point shown in Fig. 5. This provided better re-
sults, and was used to collect all the Rh data used in the results.
Ummenhofer and Medgenberg [25] used thermography to study
Fig. 4. Photo of experimental set up. fatigue damage processes in welds, and showed IR camera images
920 P. Williams et al. / Materials and Design 51 (2013) 916–923

Fig. 7. The data averaging point used for a pristine specimen.


Fig. 9. Hotspots converging on the center of the specimen.

Fig. 10. Schematic of the rectangle data averaging point used in post processing of
Fig. 8. Two hotspots visible during excitation test. the welded specimens’ results.

with higher temperatures above and below the weld around the
edge of the Heat Affected Zone, in similar locations to the welded
specimen shown in Figs. 8 and 9. It should also be noted that dur-
ing the fatigue loading of the specimen the hotspot appeared to
gradually move to the center of the specimen, however this is most
likely due to conduction of heat from the more ductile base mate-
rial around the weld.

4. Results

The results presented in Fig. 11 shows the temperature evolu-


tion for a series of tests with different stress levels for the welded
specimens. Fig. 11 clearly shows approximate linear progression
for the Rh evolution for all the different stress levels. Using these re-
sults, the individual curves for each stress level can be used to pre-
dict the fatigue life of a fatigue specimen, by a single excitation
test. Fig. 12 shows the Rh evolution compared to normalized fatigue
Fig. 11. Plot of Rh vs. number of cycles, N.
life, allowing for direct comparison between the curves for the dif-
ferent stress levels tested. The trend shown for each stress level is
relatively linear, with a decreasing gradient as stress values de-
crease, relating to the amount of thermal energy released under experienced cyclic fatigue using Fig. 11 and then perform an exper-
the different stress levels. Similar trends were reported for pristine, imental fatigue test to verify the predictions. The procedure for the
unwelded specimens by Amiri and Khonsari [1], showing both a verification tests begins with applying a fatigue load to the speci-
linear temperature evolution, and decreasing gradients for lower men, until the operator chooses a stopping time (or number of
stress levels, suggesting the heat generation from the material is cycle), this is determined by the expected fatigue life and the spec-
relatively linear regardless of the addition of a welded joint. imens behavior. Once the test has been stopped, the specimen is
Several verification tests were carried at known stress levels, to left to cool to ambient temperature, and then a STE test is carried
first predict the remaining life of a specimen that had already out, using the same excitation loads, and stress ratio, as in the
P. Williams et al. / Materials and Design 51 (2013) 916–923 921

Fig. 14. Rh vs. PCL plot for r = 267 MPa, without Rh0.
Fig. 12. Plot of Rh vs. normalized number of cycles, N/Nf.

original series of tests. This produces an Rh value, which can then Table 2
Verification test results on welded specimens.
be compared to a plot of Rh vs. N/Nf the normalized failure cycles
for the specimen. Using the Rh value from the STE a value of N/Nf r Rh (°C/sec) N Nf Predicted Experimental Absolute
or the percentage of consumed life can be determined. The speci- (MPa) Consumed Consumed Error (%)
Life, PCL Life, ECL (%)
men is then loaded with the fatigue load until failure, so an exper-
(%)
imental N/Nf value can be determined. This experimental value is
310 0.47 3947 5101 68.4 77.37 8.97
compared with the predicted value from the STE test.
267 0.4594 33,005 49,218 74.2 74 0.2
In verification Test 1, a stress of 267 MPa was applied to the 310 0.4667 3512 4741 68.0 74.07 6.07
specimen. The test was stopped after 33,005 cycles and the excita- 285 0.4194 10,022 15,035 63.5 66.65 3.15
tion test was performed. The fatigue loading was then continued
and the specimen failed at 49,218 cycles. From both the equation
displayed in Fig. 13 and the plot itself, the experimental value for ship. The resulting equation for prediction of the consumed life
Rh from verification Test 1 of 0.4594 predicts a PCL of 73.7%. The now reads:
experimental value was found to be 74%.
PCL ¼ 2:012Rh  0:1824
The agreement between the prediction and experimental val-
ues is close. Amiri and Khonsari [1] also used verification tests The results of the verification tests with the first term removed
on pristine specimens and also reported small errors between are shown in Fig. 14.
the predictions and the experimental values. However, examina- From this improved plot, the value of PCL predicted is now
tion of the first set of verification test results show that the initial 74.2%, much closer to the experimental value of 74%. This process
temperature rise Rh0 measured at 0 fatigue cycles (N0 = 0), tended was completed for all four verification tests, and the accuracy of
to bias the fit of the curves. The situation is different when deal- the all predictions was increased. Table 2 shows the predictions
ing with welded specimens due to the residual stresses in the made during the verification tests, and the comparison to the
material after the welding process. Recall from Section 3.1 that, experimental values.
heat treatment was applied to the specimen to relieve the From the results in Table 2, there is clearly close agreement be-
residual stress experienced during welding. However, it is unli- tween the predicted values of PCL and the experimental values.
kely that heat treatment can return the material to completely The removal of the initial point Rh0 increased the accuracy further,
pristine stress free condition, and thus, effectively, the welded with the maximum error shown in the verification tests being less
specimens cannot be considered pristine. Hence, it was decided than 9%.
to remove the first data point and derive a new curve-fit relation- Fig. 15 shows clearly the effect of a welded joint on the fatigue
life of a specimen, where to obtain comparable fatigue lives of
welded specimens, a significant reduction in applied stress is re-
quired. Results for pristine 1018 were collected within a range of
stress ratios between R = 0.53 and R = 0.62. The results shown
in Fig. 15 compare well with the Strain–Number of cycles to failure
(e–N) curves reported by Crupi et al. [16], suggesting similar spec-
imen behavior, with the cited and our experiments using stress re-
lieved specimens this reinforces our experimental results.

5. Discussion

Most test specimens failed in the base material, either above or


below the weld, with the weld maintaining its shape throughout
the fatigue loading. Fig. 16 shows a failed specimen, where the
welded section of the specimen, in the center appears to bulge
slightly as a result of considerable necking clearly visible in the
base material above and below the weld. The heat treatment ap-
Fig. 13. Rh vs. PCL plot for r = 267 MPa. plied to relieve the stress in the weld appears to have made the
922 P. Williams et al. / Materials and Design 51 (2013) 916–923

reduced, the heat treated base metal was the weakest part of the
specimen.
Fig. 17 shows one of only two specimens that failed in a location
that was not in the base metal. Both specimens were loaded with
stress of 250 MPa, expected to be in the High-Cycle Fatigue (HCF)
regime. Initially the specimens behaved in the same manner as
the previously tested welded specimens, a small temperature
change was measured, and the profile then stabilized, as shown
in Fig. 1. However, very little deformation occurred in the base me-
tal, and the typical bulge at the center of the specimen due to the
weld was not visible, suggesting the specimen was not approach-
ing failure. After a reasonable length of time under fatigue loading
conditions, 45,000 and 70,000 cycles respectively, the specimens
showed a sudden large temperature rise, followed by failure at
the weld, approximately in the center of the specimen. Unlike
the usual cup-and-cone fracture site visible in a ductile material,
Fig. 15. S–N curves for pristine 1018 carbon steel and welded and heat treated a relatively flat failure site with a visible flaw became noticeable.
specimens.
Examination of the fracture surface in Fig. 17 shows the existence
of a flaw at the top of the specimen cross section. Both specimens
that failed in this manner had similar flaws in the weld in similar
welded section the strongest part of the specimen and greatly locations. The only difference between the specimens was the flaw
increased the ductility of the base material. This explains why size, the specimen with the larger flaw failed earlier, at 45,000 cy-
the hotspots occur there, i.e. due to the higher deformation under cles, and the smaller flaw had a longer fatigue life at approximately
the loading eventually leading to failure at those locations. 75,000 cycles.
In the test series results shown in Fig. 10, all the specimens To confirm whether HCF leads to a ‘‘flaw-based’’ failure as op-
failed in the base metal, suggesting that the welds in these speci- posed to ‘‘plastic-deformation’’ based failure in the base metal as
mens were high quality and had no inherent flaws. The specimens seen previously, a further test was carried out at an even lower
were not checked prior to testing, so the detection of flaws in a stress level of 240 MPa. The results from this loading can be seen
specimen was solely based on specimen behavior and failure. The in Fig. 11, with a fatigue life of 153,000 cycles. The failure of this
behavior of the specimens was very consistent, despite the varia- specimen was similar as the previous results described failure in
tions that are possible with welds specimens, such as differences the base metal, with a slight bulge shown at the weld location,
in material microstructure and weld quality. Verification tests showing the specimen behavior and failure is similar in both
showed good agreement with the original series of fatigue tests low- and high-cycle fatigue.
in number of cycles to failure, and showed that results obtained These unexpected failures do not truly assess the strength of
in Fig. 10 could be reproduced accurately. Failure in the welded the weld, and the prediction method does not account for flaws
specimens occurred in the base metal, either above or below the in the specimen, and the significant reduction in fatigue life that
welded joint. The position of fracture for each specimen was re- they cause. The processed data from the early failing specimens
corded, with the results showing slightly more failures below the showed good agreement with the results in Fig. 11, a good linear
weld. Previous experiments by Ummenhofer and Medgenberg progression between the curves for 261 MPa and 240 MPa, sug-
[25] and Crupi et al. [26] have shown welded specimens failing clo- gesting that until the flaw in the specimen became the dominating
ser to weld in the weld toe region. However those specimens had factor in the specimen failure, the Rh evolution was in agreement
not been stress relieved, or ground flat, suggesting that fracture with the previous specimens. Initially under the loading the base
outside of the weld is not an uncommon phenomenon. Due to material deformed more than the weld, making the base metal
the variation in fracture location being relatively even, it seems un- the source of heat generation. However during this cycling the
likely that the failure location in the base metal is due to loading or flaw in the weld continued to grow, until its size allowed it to
stress concentration in the material. overtake the base metal deformation at the heat generation source
Fig. 14 shows a comparative S–N for pristine 1018 carbon steel and failure location.
and the specimens containing a welded joint that had been stress
relieved. Both sets of results in the plot show a good linear corre-
lation, the crucial difference between the results is the reduction
in the stress level the welded specimen can withstand. A pristine
specimen showed a fatigue life of 9091 cycles with a maximum
stress of 445 MPa, however a similar fatigue life for a welded spec-
imen was obtained at a lower value of 298 MPa. Despite the ap-
plied stress relief, the overall strength of the specimen is

Fig. 16. Photo of failed welded specimen. Fig. 17. Photo of fracture surface of failed flawed specimen.
P. Williams et al. / Materials and Design 51 (2013) 916–923 923

6. Conclusions [5] Khonsari MM, Amiri M. Introduction to thermodynamics of mechanical


fatigue. CRC Press, Taylor and Francis Group; 2013.
[6] Meneghetti G. Analysis of the fatigue strength of a stainless steel based on the
The Rh method for RFL prediction uses short time excitation energy dissipation. Int J Fatigue 2007;29:81–94.
tests to generate a small temperature rise in a metal specimen. This [7] Rösner H, Sathish S, Meyendorf N. Thermographic characterization of fatigue.
Rev Progr Quant Nondestruct Eval 2001;20:1702–9.
temperature rise is then compared with previously collected
[8] Matsumoto E. Non destructive evaluation of fatigue by thermal, acoustic and
experimental data to provide a predicted value of the consumed fa- electromagnetic techniques. Proc Eng 2011;10:3656–61.
tigue life. The results of a series of experimental tests with tubular [9] Crupi G, Crupi V, Guglielmino E, Taylor D. Fatigue assessment of welded joints
using critical distance and other methods. Eng Failure Anal 2005;12:129–42.
1018 carbon steels how that the Rh method for predicting PCL holds
[10] Bilous P, Lagoda T. Structural notch effect in steel welded joints. Mater Des
promise with welded specimens, with relatively small errors be- 2009;30:4562–4.
tween the predicted values and the experimental values. [11] Alam MM, Karlsson J, Kaplan AFH. Generalising fatigue stress analysis of
For the experiments conducted, provided that a welded speci- different laser weld geometries. Mater Des 2011;32:1814–43.
[12] Jiang WC, Wang BY, Gong JM, Tu ST. Finite element analysis of the effect of
men has a high quality weld and appropriate heat treatment, the welding heat input and layer number on residual stress in repair welds for a
weld is likely to be strong, possibly the strongest part of the spec- stainless steel clad plate. Mater Des 2011;32:2851–7.
imen. In the experiments the base metal showed higher tempera- [13] Kang HT. Fatigue prediction of spot welded joints using equivalent structural
stress. Mater Des 2007;28:837–43.
ture generation that the weld initially, greater deformation, and [14] Park JM, Kang HT. Prediction of fatigue life for spot welds using back-
necking throughout the fatigue loading. Two specimens suffered propagation neural networks. Mater Des 2007;28:2577–84.
failure in the weld, unlike the previous specimens, and both of [15] Fan J, Guo X, Wu C, Ma G. Rapid measurement of fatigue behavior of welded
joints using the lock-in infrared thermography. In: 11th International
these were discovered to have visible flaws on the fracture surface. conference on quantitative infrared thermography, 11–14 June, Naples, Italy,
The base metal had not deformed significantly as seen before in 2012.
other welded specimens and failure occurred suddenly, signifi- [16] Crupi V, Chiofalo G, Guglielmino E. Using infrared thermography in low-cycle
fatigue studies of welded joints. Welding J 2010;89:195–200.
cantly before the expected fatigue life. Despite the heat treatment,
[17] Naderi M, Amiri M, Khonsari M. On the thermodynamic entropy of fatigue.
the flaws in the welded section made it weaker than the base Proc R Soc A 2010;466:423–38.
metal. [18] Amiri M, Naderi M, Khonsari M. An experimental approach to evaluate the
critical damage. Int J Damage Mech 2011;20:89–112.
Prediction of the consumed fatigue life is however dependent
[19] Naderi M, Khonsari MM. A comprehensive fatigue failure criterion based on
on the quality of the welds in the specimens. Rh evolution does thermodynamic approach. J Compos Mater 2012;46:427–47.
not account for flaws in the welds, and specimens containing flaws [20] Meyendorf N, Rösner H, Kramb V, Sathish S. Thermo-acoustic fatigue
are not suitable for collecting data. NDT methods such as ultra- characterization. Ultrasonics 2002;40:427–34.
[21] Malarvizhi S, Raghukandan K, Viswanathan N. Fatigue behaviour of post weld
sound examination are already in use to locate crack and other heat treated electron beam welded AA2219 aluminium alloy joints. Mater Des
flaws in welds, and components containing these should be 2008;29:1562–7.
avoided in the field. [22] Radaj D. Design and analysis of fatigue resistant welded structures. Arlington
Publishing; 1990. p. 100–101.
[23] Masubuchi K. Residual stresses and distortion, Welding, brazing, and
References soldering, ASM Handbook, vol. 6. ASM International; 1993. p. 1094–102.
[24] Amiri M, Khonsari MM. Rapid determination of fatigue failure based on
[1] Amiri M, Khonsari MM. Non-destructive estimation of remaining fatigue life: temperature evolution: fully reversed bending load. Int J Fatigue
thermography technique. J Failure Anal Prevention 2012;12:683–8. 2010;32:382–9.
[2] Naderi M, Khonsari M. Real-time fatigue life monitoring based on [25] Ummenhofer T, Medgenberg J. On the use of infrared thermography for the
thermodynamic entropy. Struct Health Monitor 2011;10:189–97. analysis of fatigue damage processes in welded joints. Int J Fatigue
[3] Bryant MD, Khonsari MM, Ling FF. On the thermodynamics of degradation. 2009;31:130–7.
Proc R Soc A 2008;464:2001–14. [26] Crupi V, Guglielmino E, Maestro M, Marinò M. Fatigue analysis of butt welded
[4] Amiri M, Khonsari MM. On the role of entropy generation in processes AH36 steel joints: thermographic method and design S–N curve. Mar Struct
involving fatigue. Entropy 2012;14:24–31. 2009;22:373–86.

You might also like