See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/335805583

Evaluation of the Fatigue Limits of Aluminum Alloys by Thermographic and

eN techniques

Conference Paper · January 2019

DOI: 10.26678/ABCM.MECSOL2019.MSL19-0147

CITATIONS READS

0 105

4 authors, including:

Carlos Bandeira Gabriela Wegmann Lima

Federal Center for Technological Education Celso Suckow da Fonseca Pontifical Catholic University of Rio de Janeiro
6 PUBLICATIONS 10 CITATIONS 2 PUBLICATIONS 1 CITATION

SEE PROFILE SEE PROFILE

Jaime Tupiassú Pinho de Castro

Pontifical Catholic University of Rio de Janeiro
211 PUBLICATIONS 2,481 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jaime Tupiassú Pinho de Castro on 28 October 2021.

The user has requested enhancement of the downloaded file.

Evaluation of Al alloys fatigue limits by thermographic and N techniques

Gabriela Cristina Paiva Martinsa*

Carlos Filipe Cardoso Bandeiraa

Gabriela Wegmann Limaa

Jaime Tupiassú Pinho de Castroa

a
Mechanical Engineering Department, Pontifical Catholic University of Rio de Janeiro (PUC-Rio), Rua Marquês de São Vicente
225, Rio de Janeiro, RJ 22453-900, Brazil. E-mail: [email protected], [email protected], [email protected],
[email protected]
* Corresponding author
Abstract
It is well-known that a material subjected to cyclic loading increases its temperature. Since fatigue
damage is caused by cyclic plasticity, a dissipative process, such temperature increments are much more
pronounced for loadings above the fatigue limit of the material. Based on this fact, La Rosa and Risitano
proposed a thermal approach to measure fatigue limits, associating them to the stress amplitude below
which no significant heat is generated during fatigue loading. This methods works well for steels, but
it would be even more usefull for Al alloys, which reach their fatigue limits at very long lives, tipically
at 5108 cycles. Indeed, it is too laborious to measure such limits using standard mechanical methods,
even when using accelerated procedures like Dixon or Prot, because of the very long time required to
perform the needed tests. This work investigates if the thermal method can reliably evaluate the fatigue
limit of an Al alloy 6351-T6. For comparison purposes, the fatigue limit is also estimated by the
extrapolation of its measured εN curve. Albeit La Rosa and Risitano’s approach is not suitable to
evaluate the fatigue limit of the tested Al alloy, a proposed alternative thermal approach shows potential
to do so.
Keywords
Fatigue, Fatigue Limit, Thermographic Method, N extrapolation

INTRODUCTION
Fatigue is the mechanical failure mechanism that induces crack initiation and/or propagation under cyclic loads, through a
gradual and stable process that can last up to the eventual fracture of the structural component. The fatigue or endurance limit SL is
defined as the stress amplitude below which the crack initiation process is not activated (at least at the surface of the components,
since they may initiate as internal fish eye cracks under much longer gigacycle lives, as reviewed by [1]. Even though SL is a most
important property for structural design purposes, its measurement by classical mechanical procedures is laborious and expensive,
in particular for Al alloys. Indeed, even when using accelerated methods like Prot [2] or Dixon [3], SL measurements require the
test of many specimens (typically at least 10) during a very large number of cycles, to identify if they can last for lives longer than
those associated with them. Such tests typically must last for at least 5106 to 107 cycles for steels, whereas for aluminum alloys they
must last for much longer, usually 5108 cycles or even more. Fast servohydraulic machines, working at say 60Hz, can apply about
5106 cycles per day. Hence, a single fatigue limit test for an Al alloy in such machines requires cycling one specimen for more than
3 months working around the clock, 24 h per day, 7 days per week, certainly a major practical problem. Even very fast rotating
bending machines, working at say 10000rpm or about 167Hz, accumulate less than 15 million cycles per day. They consume much
less energy than the servohydraulic machines, but do not solve the testing time problem. Only resonant ultrasonic testing machines
working around 20kHz can perform such tests in reasonable times, but besides being expensive, they are neither widely available
nor versatile.
Hence, it is no surprise that most mechanical designs must rely on empirical estimations for fatigue limits. According to [4],
fatigue limits SL for steel and Al structural components can be estimated by Eq. (1) and (2), respectively. Such estimates depend on
the ultimate strength SU of the alloy and on empirical factors to quantify the effects of other parameters that can affect S L as well,
such as surface finish ksf; component size ksz, load type klt, working temperature k; reliability kRl, fretting kft, etc. Marin [5] proposed
that, in the absence of more accurate information, the various k i factors (which quantify the effects of the various details that modify
the fatigue limit of the specimen SL in relation to that of the material SL’) should be multiplied, as if they acted independently.
Applying the ki factors only to correct the fatigue limit of the material, implies that the effect of the details is supposed to be
negligible in short lives.

𝑆𝐹 (103 ) = 𝑘𝜃 𝑘𝑅𝑙 ∙ 0.76𝑆𝑈 , 𝑆𝑈 ≤ 1400𝑀𝑃𝑎

𝑆𝐹 (103 ) = 𝑘𝜃 𝑘𝑅𝑙 ∙ 0.67𝑆𝑈 , 𝑆𝑈 > 1400𝑀𝑃𝑎
𝑠𝑡𝑒𝑒𝑙𝑠 (1)
𝑆𝐿 (106 ) = 𝑘𝑠𝑓 𝑘𝑠𝑧 𝑘𝑙𝑡 𝑘𝜃 𝑘𝑅𝑙 𝑘𝑓𝑡 ∙ 𝑆𝑈 ⁄2 , 𝑆𝑈 ≤ 1400𝑀𝑃𝑎
6
{𝑆𝐿 (10 ) = 𝑘𝑠𝑓 𝑘𝑠𝑧 𝑘𝑙𝑡 𝑘𝜃 𝑘𝑅𝑙 𝑘𝑓𝑡 ∙ 700𝑀𝑃𝑎, 𝑆𝑈 > 1400𝑀𝑃𝑎
1
 𝑆𝐹 (103 ) = 𝑘𝜃 𝑘𝑅𝑙 ∙ 0.82𝑆𝑈 , 𝑆𝑈 ≤ 325𝑀𝑃𝑎
𝐴𝑙 𝑎𝑙𝑙𝑜𝑦𝑠 { 𝑆𝐿 (5 ∙ 108 ) = 𝑘𝑠𝑓 𝑘𝑠𝑧 𝑘𝑙𝑡 𝑘𝜃 𝑘𝑅𝑙 𝑘𝑓𝑡 ∙ 0.4𝑆𝑈 , 𝑆𝑈 < 325𝑀𝑃𝑎 (2)
𝑆𝐿 (5 ∙ 108 ) = 𝑘𝑠𝑓 𝑘𝑠𝑧 𝑘𝑙𝑡 𝑘𝜃 𝑘𝑅𝑙 𝑘𝑓𝑡 ∙ 130𝑀𝑃𝑎, 𝑆𝑈  325𝑀𝑃𝑎

Even though purely elastic strains cannot cause damage, some authors do not recognize fatigue limits for Al and for some
other non-ferrous alloys, see Figure 1.a. To avoid this problem for structural design purposes, their SN or Wöhler's curves can be
estimated by a two-step bi-parabolic approximation that leads to two straight lines in log-log plots, the second with a smaller slope
starting at lives longer than a very long life NL. Haibach [6] proposed Eq. (3) for steel components, and his idea can be adapted to
describe the SN curves of non-ferrous components, where B and C are, respectively, Wöhler (or Basquin)’s exponent and coefficient.

𝑁𝑆𝐹𝐵 = 𝐶, 𝑁 ≤ 𝑁𝐿
{ 2𝐵−1 (3)
𝑁𝑆𝐹 = 𝐶 ∙ [𝑆𝐿 (𝑁𝐿 )]𝐵−1 , 𝑁 > 𝑁𝐿

(a) (b)
Figure 1 (a) Typical SN curve for steels and Al alloys [7]); (b) SL evaluation from thermal data, Bandeira e al. (2017)

To avoid the need for testing many specimens for a long time, as required by Prot or Dixon mechanical procedures, La Rosa
and Risitano [8, 9] proposed a robust method to measure fatigue limits of steels by a thermal technique using few specimens tested
at much shorter times, which do not even need to be broken. They proposed to correlate load blocks with fixed number of cycles
and incremental stress amplitudes a with the damage-induced surface temperature increments during the crack initiation process.
To evaluate SL, the ratio ∆max/∆N between the maximum temperature variation ∆max induced during a number of cycles ∆N is
correlated to the applied stress amplitude a, as shown in Figure 1.b. This procedure is applied for several a levels until it is possible
to identify a sharp increase in heat generation (damage-no damage transition), characteristic of loads above the fatigue limit. Since
this thermal approach is much faster then the classic mechanical procedures, it can be very useful in practice, and even more if it
could be reliably applied to evaluate the fatigue limits of Al alloys, which are associated with much longer lives than steels.
The purpose of this work is to investigate the use of this thermal technique as a viable tool to estimate the fatigue limits of Al
components, by testing fatigue specimens of the 6351-T6 Al alloy. However, unlike in steels, the number of cycles needed to achieve
temperature stabilization in Al alloys is highly dependent on the stress ranges, making the proposal of a standard technique a much
more challenging task. Hence, after describing the problems identified in the thermal measurements, its estimated fatigue limit is
verified comparing it with literature data and by extrapolating N data measured by standard ASTM procedures. To do so, the stress
amplitude at 5108 cycles, the life usually associated with fatigue limits of Al alloys, is estimated by extrapolating the N curve of
the 6351-T6 Al alloy. This SL estimation method is much more viable than SN-based Dixon or Prot procedures, but albeit the
extrapolation of N data may seem straightforward, it also presents some challenges, as discussed following.

THE LA ROSA AND RISITANO’S THERMOGRAPHIC METHOD

La Rosa and Risitano proposed to map the surface temperature of fatigue specimens using a high-resolution infrared
thermocamera, claiming that when steel specimens are loaded above their fatigue limits, their temperature increases during the first
part of the test, phase I, until changing its rate  to a much lower or null value  in phase II. This phase II lasts until
the temperature suffers a rapid increase prior to failure in phase III, as schematized following.

2
 Figure 2 Phases of the thermal behavior over the surface of a specimen [10]

La Rosa and Risitano proposed to determine fatigue limits of steels by plotting temperature increments at the end of phase 1
(1 in. Figure 2), or else temperature increase rates ⁄dN1 in phase 1 caused by different stress amplitudes ai = i/2 applied on
a same specimen, which does not even need to be broken. They showed that such Δ1i×σai or else 1i ⁄DN1×σai curves typically
have a bilinear trend with very different slopes, see Figure 1.b again. These two slopes can identify the transition from no fatigue
damage below the fatigue limit to a damage accumulation process above it, and thus measure the S L of the tested specimen. Bandeira
et al. (2017) verified this thermal methodology by coparing the S L of a 1020 steel measured by its procedures with the fatigue limit
measured by reliable mechanical data obtained by testing many specimens following Dixon procedures in a traditional rotating
bending machine.

THE EXTRAPOLATION OF N DATA METHOD

Traditional N design procedures assume fatigue crack initiation is primarily caused by cyclic elastoplastic strain histories
induced by service loads at critical points of structural components. They can estimate fatigue crack initiation lives from the material
N or Coffin-Manson curve, using its Ramberg-Osgood properties to correlate stress ranges at critical points to their correspondent
strain ranges. Such cyclic  curves are usually obtained by joining the tips of several concentric stabilized elastoplastic hysteresis
loops, measured by testing several identical push-pull N specimens at different fixed strain ranges i, all with zero mean stresses
and strains, as schematized in Figure 3.a. As in the monotonic case, cyclic  curves can usually be well described by a Ramberg-
Osgood model. Moreover, many materials have almost symmetric cyclic  behavior under tension or compression, thus their entire
cyclic  curves can be well fitted by:

𝜎 𝜎 1/ℎ𝑐
𝜀 = + 𝑠𝑖𝑔𝑛(𝜎) ∙ | | (4)
𝐸 𝐻𝑐

where the sign() function is needed to correctly describe their compressive parts. H c and hc are cyclic strain-hardening coefficient
and exponent,  is the applied stress,  is the applied strain and E is Young’s modulus.
The number of cycles needed to initiate a fatigue crack under a fixed strain range  is usually measured in push-pull N tests
identical to those used to measure hysteresis loops. The N method assumes that strain ranges correlate with corresponding stress
ranges by Ramberg-Osgood’s equation, and with fatigue lives by Coffin-Manson’s rule:

𝜀𝑎 = Δ𝜀 ⁄2 = Δ𝜀𝑒𝑙 ⁄2 + Δ𝜀𝑝𝑙 ⁄2 = (σc ⁄𝐸 )(2𝑁)𝑏 + 𝜀𝑐 (2𝑁)𝑐 (5)

where el and pl are the elastic and plastic parts of the applied strain range , E is Young’s modulus, c and c are the coefficients
of the elastic and plastic parts of Coffin-Manson’s rule, both power functions of the crack initiation life N, and b and c are their
exponents.
Countless tests confirm that fatigue crack initiation lives can indeed be very well correlated with strain amplitudes a =  ,
as schematized in Figure 3.b. Coffin-Manson’s equation shows that elastic strains are negligible for very short lives, so it tends to
its plastic part in that so-called the low-cycle region, where pl >> el. Likewise, long lives are associate with almost purely elastic
strain ranges, meaning pl << el, and this region of Coffin-Manson’s curves tend to Wöhler’s power function NSB = C, where B
= −1/b and C = (1/2)(c)−1/b.
The extrapolation procedure to estimate fatigue limits from Eq. (5) simply consists of first obtaining the N curve for the Al
alloy and then using its properties evaluate the stress range corresponding to a life of N L = 5108 cycles. Since fatigue limits should
in principle be associated with purely elastic loads, SL can be obtained from the corresponding strain ranges (for Al alloys a(N =
5108 cycles)) directly by Hooke’s law, neglecting the plastic term contribution. Such estimates obviously improve with the number
of long life tests included in the measured N curve.

3
 (a) (b)
Figure 3 (a) Schematic concentric loops, and the cyclic  curve obtained by joining their tips; (b) Scheme of the Coffin-Manson
curve of a metallic alloy (Castro and Meggiolaro, 2016).

MATERIAL AND EXPERIMENTAL PROCEDURES

The thermographic fatigue limit measurements were made in a rotating bending machine under a load radio R = −1 at a
frequency f = 8500 rpm. The hourglass-like fatigue specimens had a large R = 58.8mm constant radius and a polished surface with
measured mean surface roughness Ra = 0.099 m, see Figure 4.a. This is needed to minimize stress concentration and surface finish
effects, and to ensure that their middle section is the critical one, see Figure 4.b. Figure 4.c shows a specimen with its central part
black-painted, to increase its emissivity and thus to improve the infrared camera performance. All specimens were dimensionally
verified and stored in a humidity-controlled environment. Table 1 lists the chemical composition of the 6351-T6 Al alloy used in
this article and Tab. 2 lists the basic tensile properties are yield strength S Y, tensile strength SU and Young’s Modulus E.

Table 1 The Chemical Composition of the 6351-T6 Aluminum alloy

Mn Si Cr Ni Cu Fe Mg Ti Zn Al
0.510 1.080 0.010 0.010 0.040 0.290 0.510 0.020 0.030 97.440

Table 2 The basic tensile properties of the 6351-T6 Aluminum alloy

SY SU E (GPa)
(MPa) (MPa)
324 352 68.2

A compact, bench mounted rotating bending etectromechanical fatigue testing machine RBF 200 was used in all tests. The
major advantages of this veteram equipment are a high loading frequency, up to 10,000rpm, and a low power consumption. The
surface temperature of the fatigue specimen was continuously monitored by an infrared FLIR A320 camera, with resolution 320240
pixels, data acquisition frequency 30Hz, and temperature sensibility 50mK. The temperature data was analyzed using the software
ResearchIR from FLIR. In addition, a black cloth was used to cover the rotating bending machine and the camera to avoid
environmental heat noise during the test. Figure 5 shows the RBF 200 used in this article. See Bandeira et al. (2017) for further
details on thermographic procedures.

4
 (a) (b) (c)
Figure 4 Specimen: (a) geometry and dimensions based based on ASTM E466 [11]; (b) as manufactured; (c) black painted

(a) (b)
Figure 5 RBF 200: (a) overview; (b) lateral view

The  curve was measured following standard ASTM E606 procedures [12], using a uniaxial 100kN Instron servo-hydraulic
testing machine under a load radio R = −1 and at a frequency f = 2Hz, as illustrated in Figure 6.a. All  specimens were
dimensionally verified, and proper care was taken to avoid delayed buckling and other similar problems. Both the clip-gage and the
load cell were properly calibrated against suitable standards before the eN tests. Initially the loading train was aligned within
±0.01mm, to minimize load eccentricity effects. This step is most important to avoid parasitic bending moments during the N tests.
Figure 6.b shows  specimen with the controlling clip-gage mounted on it. Notice in Figure 6.c the very large radius of the 
specimens, to minimize stress concentration effects. Figure 6.d shows a polished 6351-T6 Al alloy specimen.

(a) (b)
5
 (c) (d)
Figure 6 (a) 100 kN Instron servo-controlled testing machine; (b)  specimen with the controlling clip-gage mounted on it; (c)
geometry and dimensions; (d) as manufactured

EXPERIMENTAL RESULTS
Figure 7.a shows max versus number of cycles N for a/SU = 0.7 and some pictures of the temperature distribution on the
specimen surface for the three thermal phases. They show that, as expected, the higher temperatures are localized around the smallest
section of the specimen, where the final fracture occurs. Note that the total temperature variation until failure is very small for this
high a/SU (≈ 2oC), much smaller than observed in steels by Bandeira [13-15]. This can be explained by the much larger thermal
conductivity of Al alloys compared to steels, so that for the same heat generated by the fatigue damage process in both materials,
Al alloys dissipate it inside the material much faster, inducing smaller temperature increments on the specimen surface.
Notice in Figure 7.a that phase 1 is associated to a very small temperature variation with no well-defined gradient, making
difficult to detect the initial fatigue process in Al alloys with this thermal data. In phases 2 and 3 the temperature variations are more
pronounced under almost constant rates ∆/∆N2 and ∆/∆N3, respectively. Figure 7.b suports this conclusion for a/SU = 0.6 and
0.8. Notice as well that, unlike in steels, in this Al alloy phases 1 last for different number of cycles at each applied stress amplitude.
Figure 7.d depicts this Al alloy N1-dependence on the stress amplitude, shwing how the number of cycles of the first thermal
phase varies with a/SU, differently from the steels studied by Bandeira and by Guamán [16This dependence makes the
thermographic method more time consuming for the studied Al alloy than for the previously tested steels.
It is interesting to note that the largest rate d/dN occurs on the third (and last) thermal phase, because of the complete crack
generation and its unstable propagation until final failure. It generates the macro plasticization of the weakest cross section of the
specimen, increasing even more the heat generated. In phase 3 the thermographic camera measurements presents a smaller noise
level in terms of temperature measurement fluctutations, probably because of the fast crack propagation rate associated with it.

(a)

6
 (b) (c)

Phase 1
100000

80000
Number of cycles

60000

40000 y = -3103,1x + 265561

R² = 0,9993
20000

0
0 20 40 60 80 100
a/SU

(d)
Figure 7 (a) Temperature versus cycles of 70% of SU; (b) Temperature versus cycles of 60% of SU, 70% of SU and 80% of SU; (c)
Temperature versus cycles of 60% of SU, 70% of SU and 80% of SU (Zoom in); (d) Adjustment of data from the first phase
To reproduce the La Rosa and Risitano thermography approach, two specimens were tested under stress amplitudes a/SU =
{0.3 → 0.4 → 0.5 → 0.6 → 0.7 → 0.8} for one, and a/SU = {0.2 → 0.3 → 0.4 → 0.5 → 0.6 → 0.7 → 0.8 → 0.85} for the other.
Each a/SU load block lasted the number of cycles needed to characterize the first thermal phase, determined by the straight line
fitted in Figure 7.d. Figure 8.a shows the variation of the maximum temperature on the surface of the first specimen along the
number of cycles, from the smallest to the largest stress amplitude. Note that the more the stress amplitude deacreases, the more
temperature increments 1 deacreases, until it reaches a plateau where there is no significant 1. Then, correlating 1/N with
a/SU is possible to estimate the no damage-damage transition and the fatigue limit for the first specimen (FS) by SL’_LR-R_FS = 0.5SU
= 176MPa, see Figure 8.b.
Likewise, Figure 8.c shows the maximum temperature on the surface of the second specimen. Correlating 1/N with a/SU
is possible to obtain the no damage-damage transition and the fatigue limit for the second specimen (SS) by S L_LR-R_SS = 0.35SU =
123.2MPa, see Figure 8.d. The average for these two fatigue limits is S L’_LR-R_Av = 149.6MPa.
Using Eq. 2, the estimated fatigue limit based on literature recommendations for this Al alloy is S L’_lit = 130MPa, a value 13%
lower than SL’_LR-R_Av, which is a large difference for design purpose. As the thermographic method proposed by La Rosa and
Risitano has some issues and challenges when applied to Al specimens, as discussed above, an optional way to evaluate Al fatigue
limits based on temperature measurements is proposed following.
First notice that there is no significant 1 increase under low a/SU, as shown in Figure 8.b and d. Notice as well that the first
thermal phase is N1-dependent and not well described by thermal data. Then it was decided to perform additional fatigue tests with
higher a/SU = 0.6, 0.7 and 0.8, all until the end of phase 3, i.e, until the specimen breaks.

7
 Q
0,000015

0,00001

Q
y = 6E-07x - 3E-05
R² = 0,9822
0,000005

0
0 50 100
a/Su

(a) (b)

Q
1,00E-05
8,00E-06 y = 2E-07x - 7E-06
R² = 0,6009
6,00E-06

Q
4,00E-06
2,00E-06
0,00E+00
-2,00E-06 0 50 100
a/SU

(c) (d)
Figure 8 (a) max versus cycles under increasing load steps for the first specimen; (b) S L’ evaluation from the thermal data for the
first specimen; (c) max versus cycles under increasing load steps for the second specimen; (d) S L’ evaluation from the thermal
data for the second specimen

Due to the large temperature dispersion under equal a/Su, the temperature average and its increasing rate for thermal phases
1, 2 and 3 is included in this analysis. Albeit phase 3 represents an already initiated fatigue crack, whose damage process is described
by fracture mechanics concepts, it is also included in this evaluation. Figure 9 illustrates average temperatures  =  − 0, where
0 is the ambient temperature; and average temperature rates /.
Notice that temperature variations are more precise for phases 2 and 3, since their standard deviations are smaller, mainly for
phase 2 where R2  1. For temperature increasing rates, the best result were obtained for phases 1 and 2, but mainly for phase 2
where R2 = 0,91. In this way, only fatigue limits based on thermal phase 2 are considered for  and , as shown in Table
3. Notice in this table that temperature variations  yield fatigue limits closer to literature recommendations, only 9,2% lower in
a conservative way. Temperature rates , on the other hand, yield much higher fatigue limits. Hence, although this data set
is quite limited, it can be tentatively recommended that the thermal estimation of the fatigue limits of Al alloys should be based on
phase 2 temperature variations.
2,50 3,00
2,00 2,50
y = 3,4436x - 1,1551
2,00 R² = 1
1,50
Average (∆θ2)

y = 4,6486x - 2,8853
Average (∆θ1)

R² = 0,8863 1,50
1,00
1,00
0,50 0,50
0,00 0,00
0 0,2 0,4 0,6 0,8 1 0 0,2 0,4 0,6 0,8 1
σa/Su σa/Su

(a) (b)

8
 10,00 1,40E-04
1,20E-04
8,00
1,00E-04
y = 9,1173x - 1,9468 y = 0,0003x - 0,0002
Average (∆θ3)

6,00 8,00E-05

Average (∆θ/∆N1)
R² = 0,982 R² = 0,844
4,00 6,00E-05
4,00E-05
2,00 2,00E-05
0,00 0,00E+00
0 0,2 0,4 0,6 0,8 1 -2,00E-05 0 0,2 0,4 0,6 0,8 1
σa/Su
σa/Su

(c) (d)
6,00E-05 2,00E-04
5,00E-05
1,50E-04
4,00E-05
y = 0,0002x - 6E-05

Average (∆θ/∆N3)
Average (∆θ/∆N2)

y = 0,0001x - 6E-05 1,00E-04

3,00E-05 R² = 0,4591
R² = 0,9176
2,00E-05
5,00E-05
1,00E-05
0,00E+00 0,00E+00
0 0,2 0,4 0,6 0,8 1 0 0,2 0,4 0,6 0,8 1
σa/Su σa/Su

(e) (f)
Figure 9 (a) Average  versus cycles under a/SU with linear adjustment; (b) Average  versus cycles under a/SU with
linear adjustment; (c) Average  versus cycles under a/SU with linear adjustment; (d) Average  versus cycles under
a/SU with linear adjustment; (e) Average  versus cycles under a/SU with linear adjustment; (f) Average  versus
cycles under a/SU with linear adjustment.

Table 3 Fatigue limit

Thermal data
SL’ (MPa)
[Phase 2]
 118,1
 211,2

Since due to time limitations mechanical tests were unpractical in this case, to verify this thermally-estimated SL for the 6351-
T6 Al alloy, initially 11 traditional N tests were made, see Figure 10. At least two points at each  were used, as required by the
ASTM E606 standard. Figure 10 shows the data points and the N curve fitted to this first set of data, el/2 = 0.0051(2N)-0.034 and
pl/2 = 1.2263(2N)-0.893. Notice how well the elastic and plastic components (obtained under fixed  strain-range control) can be
fitted by the two power terms of Coffin-Manson’s equation.

9
 Figure 10 Coffin-Manson curve fitted to the first set of data measured by testing 6351-T6 Al alloy

Then, to improve this N curve fit in longer lives, 6 more tests with smaller strain ranges (0.35%, 0.32%, and 0.3%) were
made. This new data is added to the previous ones, resulting in the graph shown in Figure 11. The new values of Coffin-Manson’s
elastic coefficient c, plastic coefficient c, elastic exponent b and plastic exponent c are, respectively, 443.3MPa, 2.49, -0.067, and
-0.99.
Finally, 11 more N tests were made, under small (0.28% and 0.3%) and intermediate (0.45%, 0.55%, and 0.65%) strain
ranges. Figure 12 shows the resulting 6351-T6 Al alloy  curve fitted to all the 28 tests. The final Coffin-Manson properties of
this Al alloy are listed in Table 4.

Table 4 Coffin-Manson properties

c (MPa) c b c
504.68 1.89 -0.082 -0.96

Although the ASTM E606 standard accepts the use of only 10 specimens to measure a suitable N curve, notice how large is
the variation in Coffin-Manson’s properties as the number of specimens used in the data-fitting process increases. In table 5, these
values are listed in sequence, to facilitate their comparation. This maybe unexpected result certainly justifies a debate about the
number of specimens really needed to generate a reliable  curve, since it indicates that it is possible to question the validity of N
curves that are based on a small number of specimens, even they if measured strictly following standard procedures.
Figure 13 can contribute to this debate. They show the variation of Coffin-Manson properties as the number of specimens
increases. The plot with five specimens includes the first five tests carried out in different strain ranges (0.4%, 0.5%, 0.6%, 0.7%,
and 0.8%). Next, in the case of 10 specimens, one more test was added in each strain range, satisfying the minimum requirements
of ASTM E606. It includes almost all the data shown in Figure 10, except for a third specimen with 0.4% strain range. From them
on, the specimens were added almost in the sequence in which they were done.

10
Figure 11 Coffin-Manson curve fitted to data measured by testing 6351-T6 Al alloy N test specimens for second set of data

Figure 12 Coffin-Manson curve fitted to data measured by testing 6351-T6 Al alloy N test specimens for all tests

Table 5 Coffin-Manson properties comparation

 17 28
specimens specimens specimens
c
347.82 443.30 504.68
(MPa)
c 1.23 2.49 1.89
b -0.034 -0.067 -0.082
c -0.89 -0.99 -0.96

11
 (a) (b)
Figure 13 (a) the variation of elastic Coffin-Manson properties as the number of specimens increases; (b) the variation of plastic
Coffin-Manson properties as the number of specimens increases.

This analysis shows a stabilization of Coffin-Manson’s elastic properties just over 20 specimens. However, the same behavior
was not observed in Coffin-Manson’s plastic properties. Although there is an amortization of these properties’ oscillation, it cannot
be said that there was a convergence. This can be explained by the fact that there are relatively few specimens with high strain
ranges, since the purpose of these experiments was to estimate fatigue limits extrapolating the elastic curve to very large lives.
The stable hysteresis loops generated during the tests were quite symmetrical, indicating that this 6351-T6 Al alloy is Masing
material and confirming that they can be used to obtain its Ramberg-Osgood properties, see Figure 14. The cyclic strain-hardening
coefficient Hc and exponent hc, obtained by properly fitting the tips of all measured  stable elastoplastic hysteresis loops, are
786.63 MPa and 0.17.

Figure 14 Cyclic  curve of 6351-T6 Al obtained by fitting the tip of stable  loops

Since stresses and strains should be purely elastic at “infinite lives”, it is possible to estimate the fatigue limit from the elastic
part of Coffin-Manson’s equation, using Hooke’s law to correlate the stresses and strains. Thus, strictly speaking, Eq. (6) does not
need to use Ramberg-Osgood cyclic properties, eliminating a possible source of error. Assuming the infinite life for Al alloy at N L
= 5108 cycles, using the measured Young’s Modulus from Tab.2 and Table 4 properties, it is trivial to use Eq. (6) to extrapolate the
stress amplitude (𝜎𝑎 ) for the required “infinite live” value.

2𝜎𝑎 𝜎
( ) = 2 ( 𝑐 ) (2𝑁𝐿 )𝑏 (6)
𝐸 𝐸

This a value is a suitable N-based estimate for the fatigue limit SL of the tested Al alloy. However, since the N specimens
were tested under push-pull loads whereas the thermal estimated SL was obtained under rotatory bending, it may be necessary to
use a load type fatigue strength modifying factor, as shown in Eq. (2), to properly compare them. As recommended by Juvinall [17],
klt = 0.9 was used when comparing the axial push-pull and the bending loads. So, in Table 6 it is possible to list the fatigue limits of
this aluminum alloy estimated by the  extrapolation, using the different properties listed in Table 5. Once again, notice the large
variation in the estimated values due to the increase in the number of specimens used to fit the  curve, see Figure 15.

Table 6 Estimates for the fatigue limit of the 6351-T6 Al ally obtained by extrapolating  data.
 17 28
specimens specimens specimens

12
 SL’ (MPa) 191 123 102

Figure 15 Fatigue limit of material versus number of specimens

Like in Figure 13, there is a stabilization in the fatigue limit estimated by the  extrapolation just over 20 specimens, an
expected behavior since the fatigue limit depends exclusively on Coffin-Manson’s elastic properties. These results emphasize the
importance of using redundant long life tests to obtain reliable extrapolations using the  curve.
Finally, Table 7 compares the various fatigue limit estimates for the tested 6351-T6 Al alloy obtained by all methods used in
this work. Since the variation in such estimates is non-negligible, it indicates that, unlike for the steels, the tested methods still
cannot be used as reliable substitutes for the much more laborious mechanical tests for Al alloys.

Table 7 Estimates for fatigue limit of the Al 6351 Al alloy

ThermographyRisitano ThermographyHigher a/SU
Literature (MPa)  extrapolation (MPa)
(MPa) (MPa)
130 149.6 118.1 102

CONCLUSION
This work tests the viability of using of thermographic and of N extrapolation methodologies as rapid experimental estimate
techniques for obtaining the fatigue limit of the 6351-T6 Al alloy, which is too laborious to measure by mechanical tests since it is
associated to very long lives, normally assumed as 5108 cycles. This is a most important problem for practical applications, since
the reliability of most structural designs, which must avoid fatigue crack initiation issues, can only be achieved by using very high
safety factors when based on estimated instead of measured material properties.
The thermographic approach was previously qualified to measure fatigue limits of steels, and it can very significantly reduce
the fatigue testing costs by much reducing the testing time and the quantity of specimens required. However, the difficult in
stabilizing the specimen temperature during the thermographic tests, which can be associated to the high thermal conductivity of
the Al alloys, still is a challenging obstacle in such tests. A proposed methodology reduces but does not eliminate this problem.
The extrapolation of N curves measured by standard procedures at much lower lives, although not as fast as the thermograpic
option, still can be an attractive option when compared to fatigue limit measurements based on standard SN tests, which require too
much testing time to be completed. However, the data presented here clearly shows that the small numbers of specimens accepted
by the ASTM E606 standard clearly is not enough to obtain reliable S L estimates.

REFERENCES

[1] Castro J. T. P., Meggiolaro M. A., 2016. “Fatigue Design Techniques, volume 2: Low-Cycle and Multiaxial
Fatigue.” CreateSpace.
[2] Prot, E. Marcel, “Fatigue testing under progressive loading, a new technique for testing materials”, Revue de
Metallurgie, Vol. XLV No.12, 1948, p. 481.
[3] Dixon, WJ. The up-and-down method for small samples. Am Stat Assoc J 60(312):967-978, 1965.

13
 [4] Castro J. T. P., Meggiolaro M. A., 2016. “Fatigue Design Techniques, volume 1: High-Cycle Fatigue.” CreateSpace.
[5] Marin, J. Mechanical Behavior of Engineering Materials, Prentice-Hall 1962.
[6] Haibach, E “Modified linear damage accumulation hypothesis accounting for a decreasing fatigue strength during
increasing fatigue damage”, LBF TM Nr.50, Darmstadt, Germany, 1970.
[7] Madhukarϯ, Samatham et al. A Study on Improvement of Fatigue Life of materials by Surface Coatings. 2018.
[8] La Rosa G, Risitano A., 1999. Thermographic methodology for the rapid determination of the fatigue limit of
materials and mechanical components. In J Fatigue Mater Struct Components 1999; Vol 22, p. 65-73.
[9] Fargione G., Gerarci A., La Rosa G., Risitano, 2002. A rapid determination of the fatigue curve by the
thermographic method. International Journal of Fatigue, Vol. 24, p. 11-19.
[10] Lipski, A., 2016, “Accelerated determination of the fatigue limit and the S-N curve by means of the thermographic
method for X5CrNi18-10 steel”, Acta Mechanica et Automatica, vol. 10, 22-27.
[11] ASTM E466, 2015. “Standard practice for conducting force controlled constant amplitude axial fatigue tests of
metallic materials.” Chapter 5.
[12] ASTM E606, 2012 “Standard test method for strain-controlled fatigue testing”.
[13] Bandeira, C.F.B., Kenedi, P.P. and Castro, J.T.P, 2018, “On the use of thermographic method to measure fatigue
limits”, Latin American Journal of Solids and Structures, 2018, 15(10 Thematic Section), e60.
[14] Bandeira, C.F.B., Kenedi, P.P. and Castro, J.T.P, 2017, “Thermography – a faster method to obtain the fatigue or
endurance limit of materials”, International Symposium on Solid Mechanics, 6th Edition.
[15] Bandeira C. F. C., Kenedi P. P.; Castro J. T. P., Meggiolaro M. A., 2017. “On the use of the thermographic
technique to determine the fatigue limit of a cold drawn carbon steel”. Seventh International Conference on Very
High Cycle Fatigue, p. 329-334, U. Siegen. ISBN 978-3-00-056960-9.
[16] Guamán A., Castro J. T. P., Vieira R. B., Paiva V. E. L., Freire J. L. F., 2017. “Fatigue Limit Assessment of a Low
Carbon Steel using Dixon’s Up-And-Down and Infrared Thermography Methods” International Conference on
Advances in Experimental Mechanics 29th – 31st August 2017, University of Sheffield, UK.
[17] Juvinall, RC. Stress, Strain & Strength, McGraw-Hill 1967.

14

View publication stats