Fatigue life predictions for riveted lap joints

M. Skorupa, T. Machniewicz , A. Skorupa, A. Korbel

AGH University of Science and Technology, Faculty of Mechanical Engineering and Robotics, al. A. Mickiewicza 30, 30-059 Krakow, Poland

a r t i c l e i n f o a b s t r a c t
Article history: A semi-empirical fatigue life prediction model for riveted lap joints representative of the aircraft fuselage
Received 6 June 2016
skin connections is developed. The effect of the interference fit between the rivet and the hole is taken
Received in revised form 13 September
into account utilizing fatigue test results for 2024-T3 Alclad aluminium alloy coupons with an open
2016
Accepted 15 September 2016
and filled hole. The effect of contact surface friction is considered based on comparisons between the
Available online 15 September 2016 fati- gue lives of lap joints from 2024-T3 Alclad sheets and universal rivets with the Alclad contact
surface and the fatigue lives of similar joints with the Teflon interfoil which eliminates friction between
Keywords:
the sheets. Dependencies between coefficients incorporated into the model to account for the above
Aircraft riveted joints mentioned effects and several quantities related to the amount of rivet squeezing, and thus
Fatigue life prediction representative of the riv- eting process, are provided. The fatigue life is assumed to be controlled by the
Fatigue tests local stress amplitude at the critical location of a joint. Analyses of the fatigue test results for the 2024-
Hole filling effect T3 riveted joints prove that the local stress estimated using the superposition approach can adequately
Faying surface condition represent the combined effect of the applied loading, interference between the rivet and the hole and
faying surface friction condition on the fatigue life. The adequacy of the proposed model is
substantiated by a good agreement between the fatigue lives predicted and observed for lap joints from
D16 aluminium alloy sheets and round head
rivets.
2016 Elsevier Ltd. All rights reserved.

1. Introduction aspect of the fatigue behaviour of riveted joints remains their

The fatigue behaviour of riveted joints is affected by many vari-
ables, often interrelated, associated with the design, production
and the stress level. The installation of a rivet induces rivet hole
expansion and residual clamping between the sheets. Experiments
and analyses indicate a dependency of both hole expansion and
residual clamping not only on the amount of rivet squeeze but also
on the sheet thickness, rivet diameter and rivet type [1(pp. 7387,
134143),2]. Local residual membrane stresses generated due to
hole expansion affect the crack initiation location [1(pp. 188
194),3], crack shape development [1(pp. 199203)], and conse-
quently the fatigue life of a joint [24]. Residual clamping caused
by a difference in the elastic spring back of the rivet and the
sheets gives rise to friction between the sheets under applied
loading. Friction promotes fretting fracture, accounts for some
of load transfer throughout the joint [4,5], and also modifies
the mem- brane stress distribution in the rivet hole vicinity
[6,7]. Though the isolated effect of friction on the fatigue
properties of riveted joints is difficult to experimentally quantify,
frictional load transfer is commonly thought to have a beneficial
influence. A complicated
Corresponding author.
E-mail address: machniew@agh.edu.pl (T. Machniewicz).
response to various common coatings found in aircraft joints, like
corrosion inhibitors and sealants. Available literature data, some
of them being cited in Ref. [3], indicate that the influence of coating
on the fatigue properties can be either detrimental or beneficial.
The design and production of the joint (geometry, magnitude of
clamping, clearance or interference fit between the fastener and
hole), the type of loading (constant or variable amplitude), the load
level, and in the case of variable amplitude loading also spec-
trum severity play a role.
Due to simplifications involved in FE modelling of riveted joints,
such analyses are not always helpful in explaining various experi-
mental trends, even qualitatively. For example, FE stress analyses
in which fretting is simulated by merely assigning a high friction
coefficient value to the contact surface assumed to be flat cannot
be considered a useful tool to investigate frictional load transfer
and draw conclusions about its implications for the fatigue perfor-
mance of a joint [3].
Because of the outlined above, extremely complex determi-
nants of the fatigue behaviour of riveted joints, both the fatigue
life and crack growth predictions are very difficult. The
prediction approaches are strongly empirical to account for
phenomena that are either not well understood or too
complicated to be covered analytically. As aircraft manufacturers
place reliance on damage

http://dx.doi.org/10.1016/j.ijfatigue.2016.09.007
0142-1123/ 2016 Elsevier Ltd. All rights reserved.
tolerance designs, available prediction models mainly concentrate r S aBP 1 RTR K aBR aFR RTR K a b kb K
f ;BP f ;BR f ;b
on crack growth. The concept of equivalent initial flaw size (EIFS), 1
which is the size of a hypothetical crack assumed to be present at
the rivet hole prior to service, has been conceived to allow
correla- tion of the entire crack growth using the stress
intensity factor (SIF) [8]. The EIFS must be derived by the back
extrapolation of the crack size versus number of cycles (N) data
known from fracto- graphic measurements to N of zero.
Investigations indicate that EIFS values for riveted joints are
affected by manufacturing, design solution, loading conditions and
a crack growth prediction methodology used. An ample
overview of approaches applicable to cracks in riveted lap joints
[1(pp. 241268)] leads to a conclu- sion that currently available
crack growth prediction techniques are not capable of providing
fully reliable results, especially in the case of multiple site
damage (MSD).
Fatigue life predictions models for riveted joints employ con-
cepts commonly applied to the crack initiation life analysis for
notched structural members. This type predictions are method-
ologically easier than crack growth predictions in that the
quantity assumed to control the fatigue life, usually a function of
the stress and/or strain at a notch, is associated with the
uncracked geometry and, in contrast to the SIF, it does not vary
with crack growth. It should be emphasized that reliable
predictions on the fatigue life can indirectly bring some
information on the crack development stage as fatigue tests of
fuselage panels with riveted splice connec- tions indicate that
cracks visible on the outer surface are detected after 5075% of
the panel total fatigue life [9,10].
In models presented in Refs. [1113] it is assumed that the fati-
gue life is governed by the amplitude of the local stress at the crit-
ical location of a riveted joint. For a lap joint with multiple rivet
rows considered in the present work, the critical location is the
rivet hole edge in one of the outer rows. Like the SIF in crack
growth predictions for riveted joints, this stress is obtained by
superposition of solutions for the bypass load, transfer load and
secondary bending obtained for a plate with a single hole. Because
the local stress is determined taking only the geometry of the
joint into account, the applicability of the models referred to
above is limited to cases when the actual joint, for which the
predictions are made, and the reference joint, for which the
model has been tuned, differ solely in geometry. It is required that
the riveting pro- cess should be similar for the actual and
reference case, but the similarity criterion cannot be formulated
in a precise way. In Ref. [2] the present authors proposed to
explicitly account in the fati- gue life prediction model for the
influence of riveting. This was achieved utilizing a unique
relationship between rivet hole expan- sion on the sheet interface
and the multiplier in the Basquin equa- tion fit to the local stress
amplitude versus fatigue life data points. Produced in Ref. [2]
comparisons between the fatigue lives com- puted and observed
in over 80 fatigue tests on aluminium alloy lap joint specimens
involving various combinations of the produc- tion variables
indicated a significant improvement in the predic- tion accuracy
compared to a model which disregards the effect of riveting.
The literature evidence, e.g. [4,7], as well as experiments by the
present authors [3] indicate that riveted joints characterized by
the same rivet hole expansion but different friction conditions on
the faying surface may show very different fatigue properties.
Such results imply that transferability of the model will increase
if the effect of the interference between the rivet and the hole
and the effect of frictional load transfer on the fatigue life will be
consid- ered individually.
In this paper, the implementation of the above conception in
the fatigue life prediction model is presented. Accordingly, the
stress at the critical location is expressed as
where S is the applied stress level, Kf,BP, Kf,BR, and Kf,b are the fatigue The deviations from the nominal rivet hole diameter (do) ranged
notch factors for a finite width plate with open hole under remote from
tension, pin loading, and pure in-plane bending respectively, RTR = 0 to 0.03 mm. The rivet pitch in row (s) was the same as the rivet
TTR/P is the transfer load ratio (TTR transfer load, P applied load), row pitch (p) and approximately equal to 5d. The 2024-T3 Alclad
kb = Sb/S is the bending factor (Sb nominal stress due to secondary
bending), aBP, aBR and ab are the rivet hole expansion dependent
coefficients, and the aFR coefficient accounts for the contribution
of frictional forces to load transmission.
The derivation of the RTR ratio, the kb factor, and the fatigue
notch factors in Eq. (1) was addressed in previous works
[3,14,15]. This paper starts with the presentation of experimental
investigations required to determine the aBP, aBR, ab and aFR coef-
ficients and to relate them to riveting process characteristics. The
experiments included measurements of several quantities corre-
sponding to the rivet squeeze and a study of the evolution of fric-
tion on the faying surface of a riveted joint during cyclic loading.
Fatigue tests on riveted lap joints as well as on coupons with open
and filled holes were also carried out. The three-row riveted lap
joint specimens of 2024-T3 Alclad sheets were intended to be a
simple representation of longitudinal connections of the skin in
the pressurized aircraft fuselage. The generated experimental
results are subsequently used to validate the superposition
approach involved in Eq. (1). The predictive capability of the model
is evaluated based on comparisons between the fatigue lives com-
puted and observed in a large number of fatigue tests on various
riveted lap joints of D16 aluminium alloy sheets carried out previ-
ously by the authors [2,14,16]. Finally, following from the present
study suggestions for future work are offered.

2. Riveted specimen fabrication

The geometry of the riveted lap joints used in the fatigue tests Fig. 1. Geometry of: (a) riveted specimen; (b) rivet.
is shown in Fig. 1a, whilst the dimensions are given in Table 1.
Table 1
Dimensions of the riveted specimens.
Sheet thickness t Rivet diameter d Rivet hole diameter Protruding rivet length
(mm) (mm) do (mm) h (mm)
1.0 3.18 3.3 4.35
1.6 3.98 4.1 6.33
2.0 4.77 4.9 7.11
sheets were connected using three rows of universal head
MS20470-AD rivets of which the geometry is presented in Fig. 1b
and the dimensions are given in Table 1. The measured mechanical
properties of the 2024-T3 Alclad sheet material were slightly above
those required by MIL-HDBK-5H. The rivets were installed under
load control, as detailed in Refs. [16,17], so that for each sheet
thickness two series of joints characterized by the rivet driven
head diameter (D) to the rivet shank diameter (d) ratio of 1.3 and
1.5 were assembled (cf. Fig. 1b). The above D/d values are near
the lower bound and upper bound respectively of the range
repre- sentative of the aircraft industry practice. The so called
edge effect, i.e. a premature fatigue crack initiation at rivets in
the edge col- umns, was avoided by installing these rivets with
the squeeze force
10% higher compared to the other rivets. For each combination of
the sheet thickness and the D/d ratio, in some joints the mating
surfaces of the sheets were coated with a 0.05 mm thick Teflon foil
in order to eliminate friction on the contact surface.
3. Quantities representative of the riveting process
A squeeze force (Fsq) necessary to obtain the desired D/d ratio
was selected from the D/d vs. Fsq data produced for each sheet
thickness within a D/d range between 1.2 and 1.8 [17]. Because a
larger sheet thickness (t) requires a larger rivet diameter (cf.
Table 1), the Fsq value corresponding to a given D/d ratio increases
with t, which makes the Fsq vs. D/d data for different thicknesses
separate from each other. However, the presentation of the D/d
ratio in terms of the squeeze stress defined as
2
Ssq d Fsq =pd =4
shown in Fig. 2a makes the results for t = 1.6 and 2 mm collapse
into a common scatter band, though the data for the 1 mm thick
sheet remain still slightly apart.
If the squeeze stress is determined alternatively as
2
Ssq D F sq =pD
3 addition, two other rivets 3b were symmetrically installed at a
a somewhat worse consolidation of the results is obtained, as
shown in Fig. 2b. The reason is a difficulty in the measurements
of D caused by the barrel shape of the rivet driven head.
According to Schijve [18], the D/H ratio, where H is the rivet dri-
ven head height (cf. Fig. 1b), is the best indicator of the amount of
squeezing the rivet. H is obtained from the assumption of the
con- stant volume of the protruding part of the rivet as
2 2
H d h=D
where h denotes the protruding length of the rivet (cf. Fig. 1b).
With the D/H vs. Ssq d plot produced in Fig. 2c, the data for t = 1
and 2 mm merge, whilst the t = 1.6 mm data are slightly separated.
As said earlier and substantiated in Ref. [2], radial expansion of
the rivet hole can be considered an important quantity capable of
representing in the fatigue life prediction model the effect of the
riveting process. At sufficiently large hole expansion compressive
residual tangential stresses, whose area increases with hole
expan- sion, are generated around the rivet hole [1(pp.7399)].
Other con- sequences of increased expansion, beneficial for
the fatigue
Specimen width Specimen length L Rivet pitch p Number of rivets
W (mm) (mm) (mm) in row
144 148 16 9
160 226 20 8
144 280 24 6
properties of riveted joints, are the diminished sensitivity for the
surface finish of the hole and reduced rivet tilting [19].
Hole expansion is defined as
d e do
he 100% 5
do
where do and de are the rivet hole diameter prior to riveting and the
expanded rivet hole diameter respectively.
Comparisons between the he values measured at the mid-
thickness and at a distance of 0.5 mm from the faying surface for
all sheet thickness and D/d configurations considered for the riv-
eted joints (cf. Section 2) are provided in Fig. 3. The he results were
obtained for the do and de diameters measured with an accuracy of
0.01 mm using a method described elsewhere [17]. No systematic
dependency of he on the sheet thickness can be observed in Fig. 3,
but a trend evident for each sheet thickness is an increase in he
with the squeeze force (represented by D/d). It is further seen in
Fig. 3 that he is always larger in the sheet next to the rivet driven
head than in the sheet adjacent to the manufactured head. Both
trends were also observed by others, including the present authors
[2,4,20,21].
The capability of the Ssq d; Ssq D, D/H and he parameters to
take into account the effect of the riveting process in the fatigue
life prediction model will be examined later on in this paper.
4. Friction on the faying surface of riveted joints
Friction on the faying surface of a riveted joint cannot be easily
identified by direct testing. The present authors attempted to
2 investigate the friction force evolution for a riveted joint under
cyclic loading through direct measurements. The specimens shown
in Fig. 4 assembled from 1.6 mm thick and 25 mm wide 2024-T3
Alclad strips using MS20470-AD, 4.1 mm dia rivets were intended
to replicate friction conditions on the faying surface below the
rivet head. Strip 1 was incised and then broken in the hole section.
Sub- sequently, its both parts were pushed close to each other,
=4 fixed, and mounted between two straps 2 using rivet 3a. In
distance of
20 mm from rivet 3a in the straps utilizing distance pieces 4 in
order to reproduce for rivet 3a residual clamping conditions possi-
bly similar to those for an actual riveted joint. Two different rivet
squeeze forces resulting in D/d ratios of 1.3 and 1.5 were applied
to assemble the specimens.
The experiment involved cyclic pulling out and slipping back
4 both parts of the broken strip by clamping their ends in the grips
of a fatigue machine operating under stroke control with a stroke
value of 0.25 mm. During the tests, the relative displacement of
the strip parts was measured using a 30 mm gauge length exten-
someter. The friction force (T) versus displacement diagrams cap-
tured at various stages of the tests on two specimens mounted
with various rivet squeeze forces are shown in Fig. 5.
In cycle 1, only one strip part was observed to slide, Fig. 5a. Prior
to continuing the test this part was drilled through and blocked
with a certain backlash relative to the straps utilizing pin 5, which
allowed identification of friction forces necessary to move either
part of the strip in the subsequent cycles, as illustrated in
Fig. 2. Correlation between squeeze stress (note two different definitions) and two indicators of the rivet driven head dimensions: (a) D/d vs. Ssq(d); (b) D/d vs. Ssq(D); (c) D/H
vs. Ssq(d). For the nomenclature see text.
Fig. 5b and c. During a few initial cycles the friction force for either
strip part was observed to slightly decrease, Fig. 5b. Later on, how-
ever, a reverse trend occurred. Fig. 5c reveals that already after
100 cycles T considerably increased, whilst the displacement
was reduced. A rise in the coefficient of friction due to the
development of fretting damage and the gradual accumulation of
wear debris on the sliding surfaces are responsible for the above
behaviour. With the further cycling, the relative sliding of the
strip parts was observed to vanish and for the larger rivet
squeeze force the T vs. displacement diagrams adopted a shape
similar to that observed in tests on integral specimens, as
illustrated in Fig. 5d for the
D/d = 1.5 specimen. Such a behaviour is consistent with a holo-
graphic study of the deformation in riveted lap joints of 2024-T3
Alclad sheets which indicated that at low applied load the rivet
and the surrounding material appeared to move as a single body
because the frictional forces were the primary mechanism of load
transfer [22]. This effect may be enhanced by local welding
between the mating surfaces possible on small areas within the
fretted regions due to the frictional heat development [23,(p.
439)24]. Applying in cycle 2001 a larger stroke, and therefore a
higher load on the specimen, disrupted the connections on the fret-
ting surface and allowed again the sliding movements on the con-
 Fig. 3. Effect of sheet thickness on hole expansion: (a) D/d = 1.3; (b) D/d = 1.5.
Fig. 4. Specimen for friction force tests.
tacting surfaces. The subsequent loading with the above stroke
value promoted a behaviour similar to that observed earlier for
the smaller stroke value, namely an increase in the squeeze force
with the elapsed cycles until the local welding connections
recurred and prevented the relative slip.
With the present experiments, the effect of the rivet squeeze
force is manifested through higher friction forces and a more sig-
nificant displacement reduction with the number of cycles in the
case of the D/d = 1.5 specimen compared to the D/d = 1.3
specimen, as especially evident in Fig. 5d.
If a 0.1 mm thick Teflon interlayer was inserted between the
mating surfaces, the T vs. displacement behaviour did not vary dur-
ing the cyclic loading and the friction forces showed a negligible
dependency on the D/d ratio. The T vs. displacement plots
observed after 2000 cycles for the Alclad-Alclad and Alclad-
Teflon contact surfaces of the D/d = 1.5 specimens are compared in
Fig. 6. A sim- ilar effect of the Teflon coating was observed for
the D/d = 1.3 specimens.
5. Fatigue test results for riveted lap joint specimens
The variables considered in the fatigue tests on the riveted lap
joint specimens described in Section 2 were the rivet squeeze force
represented by the D/d ratio, the sheet thickness combined with
the appropriate rivet diameter, and the condition of the sheet
interface. The applied constant amplitude loading at a stress ratio
of 0.1 simulated hoop stress variations in the aircraft fuselage skin
typically experienced in service [1(pp. 1126)]. The loading fre-
quency was 20 Hz. For all specimen series the same three maxi-
mum gross stress levels, namely Smax = 120, 100 and 80 MPa,
were considered. In addition, two D/d = 1.3, t = 1.6 mm joints were
fatigued at Smax of 140 MPa. The fatigue test results for all speci-
men series are shown in Fig. 7 in terms of the Smax stress versus
the number of cycles to failure (Nf). The lines plotted in Fig. 7 rep-
resent the Basquin equation fit to the S-N data. Note that the
power
Fig. 5. Evolution of friction force versus displacement records during cyclic loading on D/d = 1.3 and D/d = 1.5 specimens: (a) cycle 1; (b) cycles 210: (c) cycles 100110; (d)
cycle 2000.
Fig. 6. Comparison between friction force versus displacement loops after 2000
cycles for D/d = 1.5 specimens with Alclad-Alclad and Alclad-Teflon contact
surfaces.
function does not always offer an adequate approximation of the
test data, especially for the D/d = 1.3, t = 1.6 mm Alclad series
of the joints, Fig. 8a and the D/d = 1.5, t = 1.6 mm Teflon series
of the joints, Fig. 9b.
A major trend indicated by the fatigue test results is a
beneficial effect of applying a higher rivet squeeze force. If the
sheet thick- ness, faying surface condition and stress level remain
the same, always a longer fatigue life is observed for the
specimen riveted to obtain D/d of 1.5 than for the D/d = 1.3
specimen. The increase in life by a factor ranging from 1.4 (t
= 1 mm, Alclad, Smax =
100 MPa) to 4.7 (t = 1.6 mm, Alclad, Smax = 100 MPa) occurs, but
no dependencies of this behaviour on the test variables and stress
level are obvious.
The influence of sheet thickness on the fatigue life of riveted
joints with eccentricity is typically ascribed to secondary bending
which becomes enhanced for thicker sheets. If the magnitude of
secondary bending computed using a model by Schijve [25], as
detailed in Ref. [14], were the main determinant of the thickness
effect, then the fatigue life should rise with thinning the sheets,
with the results for t of 1.6 and 2 mm almost coinciding due to
sim- ilar bending factor values (kb of 1.121.26 and 1.141.28
respec- tively) and the results for t = 1 mm (kb of 0.951.07)
being more apart. As revealed in Fig. 8a and b for the joints with
the Alclad mating surfaces and in Fig. 9a and b for the joints with
the Teflon interlayer, such a behaviour is never exhibited in a
systematic way, i.e. at all stress levels for a given combination of
the D/d ratio and the type of interface. It is seen by comparing
Figs. 8 and 9 that the thickness effect is more pronounced in the
case of the Alclad contact surface and at the highest applied
stress. Evidently, the effect of secondary bending on the fatigue
life is partly covered by other mechanisms related to interactions
between the riveting process, interface type and applied stress
level.
It is clear from Fig. 7 that the friction condition on the faying
surface can influence the fatigue response of the joints. Except
for the D/d = 1.3, t = 1 mm specimen series, the fatigue lives of
the joints with the Teflon foil are longer (up to a factor of 2) than
in the case of the joints with the Alclad interface. The Teflon inter-
layer almost eliminates friction and totally prohibits fretting on
the contact surfaces of the overlapping sheets. Thus, longer
fatigue lives of the Teflon interface joints imply that the
detrimental effect of friction, which is the fretting corrosion on
the faying surface, outweighs its favourable effect, namely
releasing the bearing load on the rivet hole. It can be seen in Fig. 7
that if the favourable influ- ence of the Teflon interlayer occurs, it
is most pronounced at the