Secondary Bending Joint PDF
Secondary Bending Joint PDF
Secondary Bending Joint PDF
A R C H I V E
VOL. LV
O F
M E C H A N I C A L
E N G I N E E R I N G
2008
Number 4
Key words: riveted joints, secondary bending, theoretical models, fatigue life
For riveted joints with eccentricities of the load path, bending moments referred
to as secondary bending are induced under nominally tensile loading conditions.
Two simple theoretical models proposed in the literature to estimate the
associated bending stresses are evaluated in the paper. Both approaches have been
implemented in computer programs and applied to estimate the effect of several
variables on the calculated bending stresses in the lap joint. Possibilities of the
experimental and numerical verification of the models are also considered. Finally,
a correlation between the secondary bending computed by one of the simple models
and the observed fatigue properties of riveted specimens, as reported in the literature,
is investigated. It is shown that deviations of the experimental results from the
theoretical expectations stem from additional to secondary bending factors, like the
inhomogeneous load transmission through the joint and the residual stresses induced
by riveting process. These phenomena are known to be relevant to the fatigue
behaviour of riveted joints, but they are not accounted for by the simple models.
A conclusion from the present study is that despite the limitations and approximations
inherent in the simple models, they provide reliable estimates of nominal bending
stresses at the critical rivet rows and can be utilized in currently used semi-empirical
concepts for predictions on the fatigue life of riveted joints.
1. Introduction
Riveting is one of the major methods for holding together sheet panels,
stringers and stiffeners of the fuselage of an aircraft and its use will continue
in the foreseeable future despite alternatives like welding and bonding.
Among primary advantages of the riveted joints are their low production
cost, utilization of conventional metal-working tools and techniques,
possibility of the riveting process automation, ease of inspection, possibility
of their repeated assembling and disassembling for the fabrication
AGH University of Science and Technology, Al. Mickiewicza 30, 30-059 Krakw,
Poland; E-mail: korbel@agh.edu.pl
370
replacement or repair, good hole filling properties of the rivets and, last
but not least, a long-standing experience of the industry with riveted joints.
At the same time, however, the riveted joints represent a fatigue critical
element in metallic airframe construction. For example, the present problem
of aging aircraft is associated with fatigue of riveted lap joints in pressurized
fuselage structure [1]. Understanding the fatigue process within riveted joint
requires a detailed knowledge of the local stress state. The local stresses
are affected by factors associated with high stress concentration at the rivet
hole, the load transfer through the rivet and the rivet installation. The rivet
installation process imparts residual stresses in the vicinity of the holes.
Another result of rivet installation are frictional forces between the mating
sheets induced due to the clamp-up which contribute to fretting damage at
the faying surface. For riveted joints with eccentricities of the load path,
bending stresses which occur under a nominally tensile loading on the joint
must also be considered. Bending caused by the tensile load on the joint is
referred to as secondary bending. Crack path eccentricities are inherent in
riveted joints typically present in aircraft fuselages, namely longitudinal lap
joints and circumferential single strap joints.
The contribution of secondary bending is often quantified by the bending
factor defined as
kb =
Sb
S
(1)
6Mb,imax
Bti2
Nominal stress is computed for the cross section neglecting the stress concentration.
(2)
371
where subscript i indicates the critical location (in sheet 1 or 2), Mb,imax is
the maximum bending moment and B and ti denote the specimen gross width
and thickness respectively.
Because outside the overlap region either sheet carries the full load P
coming from the pressurization of the fuselage, the maximum applied stresses
equal
Si =
P
Bti
(3)
372
(4)
(5)
with
Fig. 2. Nomenclature for the calculation of the bending moment for segment i of a riveted joint
373
In the case of hinged clampings of the sheet ends, the fixing moments
M1A and Mn,B and the reaction V equal zero.
The general solution of Eq. (4) is in the form
wi (xi ) = Ai cosh (i xi ) + Bi sinh (, xi ) + Ci xi + Di
(6)
i = (P/Gi )1/2 ,
(7)
where:
Ci = V /P and Di = Mi,A/P
(8)
dwi (xi )
.
dxi
Though very much alike, the concepts of Schijve and Das et al differ,
however, in some details. The effect of the joint eccentricities is covered
in either model in a distinct way, as schematically shown in Fig. 3 for the
case of a hinged (a) and rigid (b) clamping of the sheet ends. According to
Schijves approach, often referred to as the neutral line model, the neutral
axis is stepped by the eccentricities between the segments both prior to and
after the joint deformation. In case of eccentricity ei between segment i and
i + 1 he assumes
0
where wi (xi ) =
(9)
Fig. 3. Modelling the eccentricities for the lap joint from Fig. 1 according to [4] and [5] in case
of hinged (a) and rigid (b) clamping of the sheet ends
374
(10)
(11)
(12)
(13)
(14)
(15)
375
conditions on the bending stress value can be avoided if the distance between
the clamping edge and the outer rivet row (the so called free length lf , Fig.
1) is sufficiently long. Muller [2] demonstrated numerically that for sheet
thicknesses below 2 mm the minimum free length lf of 50t suffices to make
the change in the Sb value due to the change of the clamping conditions
less than 1%. This is substantiated by the plots in Fig. 4 which present the
sensitivity of the kb -value to the type of clamping according to the model by
Schijve [4]. It is seen in Fig. 4 that for a given sheet thickness the bending
factor reaches a steady level at a certain limiting lf /t-value which increases
with t. A practical conclusion for laboratory fatigue tests is that in order to
avoid the influence of the fixture type, much shorter specimens suffice in the
case of thinner sheets compared to thicker sheets. The behaviour of model
[5] is similar to that shown in Fig. 4.
Fig. 4. Influence of the type of clamping the ends of the lap joint from Fig. 1 on the bending
factor (kb ) depending on the sheet thickness (t) according to model [4].
The applied stress S=100 MPa
376
Fig. 5. Effect of the applied stress level (S), sheet thickness (t) and the rivet row pitch (p) on the
bending factor (kb ) and the bending stresses (Sb ) at the critical rivet rows for the lap joint from
Fig. 1 according to model [4]
which shows variations of the bending stresses along the lap joint computed
for two sheet thicknesses at the applied stress S=120 MPa. It is seen that
for t=2 mm both solutions give very close results on the peak Sb -levels at
the outer rivets (136 MPa [4] and 130.2 MPa [5]), whilst for t=0.8 mm the
Sb -values differ quite significantly (Sb =99.1 MPa [4] and 66.5 MPa [5]). For
t=1.2 mm the corresponding Sb values (not shown in Fig. 5 for clarity) are
109.9 MPa [4] and 91.3 MPa [5].
Fig. 6. Variations of the bending stresses along the lap joint according to model [4] and [5]. The
applied stress S=120 MPa
Differences between the results from both models are further quantified
in Figs. 7 and 8. In Fig. 7, Sb [4] and Sb [5] denote the peak bending stresses
computed using the model of Schijve [4] and Das et al [5] respectively.
Except at very low applied stresses, the Sb [4]/Sb [5] ratio is above unity which
377
Fig. 7. Comparisons between the estimates of peak bending stresses for the lap joint from Fig. 1
according to model [4] and [5] for several sheet thicknesses. The rivet row spacing p=25 mm
Fig. 8. Comparisons between the reduction in secondary bending due to increasing the rivet row
spacing in the lap joint from 25 mm to 40 mm predicted according to [4] and [5] for several
sheet thicknesses
378
379
puted and measured data compared less favourably which can be attributed
to neglecting in the model differences in the loads transmitted by the sheets
within the overlap. Also, Brenner and Hu bsch [9] reported a satisfactory
agreement between the bending stresses computed for a single strap joint
using a method similar to Schijves model and those measured with strain
gauges bonded at a distance of 2 mm from the rivet rows and half way
between the rivet columns.
The regions of the most severe stress concentration which are of primary
concern for fatigue are located beneath the rivet heads and at the faying
joint surface, in either case hidden from the capabilities of conventional
experimental stress analysis techniques. These critical areas are, however,
accessible for numerical analyses. For the lap joint configuration considered
in the FE analyses by Rans et al. [7], the neutral line model by Schijve was
found to provide accurate predictions of secondary bending only up to a
distance of three rivet diameters from the rivet row centre line. More near
the hole the simple model proved inaccurate due to significant variations in
secondary bending along the joint width.
Altogether, the available literature evidence cited in this section suggests
that the model of Schijve can produce reliable estimates of nominal bending
stresses for riveted joints with eccentricities.
Das et al [5] provided favourable comparisons between the strain gauge
readings, 3-D FE results and local stresses computed based on the nominal
bending stresses derived from their simple model for a padded riveted lap
joint.
5. Effect of secondary bending on the fatigue performance of riveted
joints
Fatigue tests of Hartman and Schijve [3] were conceived to investigate
the dependence of the riveted joint fatigue performance on the amount of
secondary bending. The geometry and dimensions of their specimens are
shown in Fig. 9. Differences in the kb -factors for the two series of lap joints
(A and B, Fig. 9) were introduced by varying the rivet row spacing. The
differences for the single strap joints (C, D and E, Fig. 9) were obtained by
varying the number and the thicknesses of the straps. A symmetrical double
strap joint (F, Fig. 9) for which secondary bending does not occur served as
a reference case. Measures were taken by the authors to make the differences
in the fatigue behaviour of the specimens stem mainly from the differences
in the amount of secondary bending. With this end in view, all specimens
were cut from the same batch of the material (2024-T3 Alclad) and the rivet
380
type, the rivet diameter (Do ) and the rivet driven head diameter, D = 1.5Do ,
were the same for all specimens.
Fig. 9. The geometry and dimensions of the riveted specimens tested by Hartman and Schijve [3]
The Sb vs. S dependence for each specimen can be derived utilizing the
neutral line model of Schijve [4]. The model application to the lap joint
has already been considered earlier in this paper, see also Fig. 1. As said
previously, for the type A and B configuration from Fig. 9, the maximum
bending stresses and, at the same time, the maximum total tensile stress
(S + Sb ), where S is computed for the local sectional area neglecting the rivet
holes, always occur at the outer rivet rows (location A and B in Fig. 1). The
deflected neutral axis of a single strap joint and the division of the joint into
three segments is schematized in Fig. 10. Due to the joint symmetry it is
enough to only consider half of the configuration. The bending stresses were
computed at four sections (, , and ) shown in Fig. 10. For specimen C
and D, the critical locations where the highest total tensile stresses occur are
381
at the inner rivets (section , Fig. 10). This holds also true for specimen E
at S 25 MPa, but for larger applied stress levels the critical location shifts
to section . Because all constant amplitude fatigue tests of Hartman and
Schijve were carried out at the same applied mean stress of 70 MPa, the
above implies that at the maximum of a fatigue cycle the peak total tensile
stress (S + Sb ) in specimen E always occurred in section . At the same time,
the total stress amplitude (S + Sb )a was always higher for section than for
section .
Fig. 10. The deflected neutral line and the division into segments for a single strap joint
Fig. 11. The computed according to Schijves model peak total stress at the critical rivet row
against the applied stress amplitude for the specimens from Fig. 9
Plots in Fig. 11 show the peak total tensile stress at the critical location
calculated at the maximum (notation max) and at the minimum (notation
min) of a fatigue cycle presented against the applied stress amplitude Sa .
For specimens A, B, C and D the fatigue crack nucleation sites observed
by Hartman and Schijve on the specimen fracture surfaces agree with the
locations of the computed maximum tensile stresses. With specimens E,
382
however, only at the highest applied stress amplitudes (Sa 56 MPa) the
failure changed over from the longer strap (section , Fig. 10) to the sheet at
the outer rivet row (section ). Such a behaviour can hardly be linked with
the bending stress performance described above.
Fig. 12. Correlation between the fatigue lives observed in test by Hartman and Schijve [3] and:
(a) the applied stress amplitude; (b) the peak total stress amplitude
In Figs 12a and b, the observed fatigue lives (Nf ) for all specimens are
correlated in terms of the applied stress amplitude Sa and the total stress
amplitude (S + Sb )a . Also shown are the corresponding trend lines. Equally
large scatter in the data points in seen for both ways of the presentation.
If the joint endurance were only dependent on (S + Sb )a , then taking into
account that Sb =0 for specimen F the following equation should be satisfied
for a given fatigue life
(S + Sb )a,SB = Sa,F
(16)
383
where the subscript SB refers to any specimen for which secondary bending
occurs.
Shown in Fig. 13 by the dashed lines are the Sa vs. Nf plots for all
specimens with eccentricities ensuing from Eq. (16). Sb is computed from
Schijves model and Sa,F comes from the trend line representing the fatigue
test results for specimens F. The actual mean curves Sa vs. Nf for the other
specimens are also presented in Fig. 13 as the full lines. Evidently, the
predicted from Eq. (16) reduction in the fatigue strength due to the secondary
bending is larger than observed because for every specimen the Sa vs. Nf
curve falls significantly below the Sa vs. Nf curve.
Fig. 13. Comparisons between the actual and computed using model [4] mean S N curves for
the specimens with eccentricities
Altogether, the results presented in Figs 12 and 13 imply that the (S +Sb )a
parameter is not capable of consolidating the data points for specimens of
various configurations along a single SN curve and leads to an overestimate
of secondary bending detrimental effects on the joint fatigue properties. Even
qualitatively some misjudgments are obtained since, according to the fatigue
tests, the order of joints with decreasing fatigue strength is F, B/E, A, C, D
whilst, according to the increasing kb value the order is F, E, B/D, A, C.
The lack of correlation between the calculated bending and the fatigue
test results is not surprising since it is well known that the fatigue cracking
of a notched component is controlled by the local stresses at the crack
nucleation site rather than by the nominal stresses. As already said in
the Introduction, the stress state near the rivet hole depends on a number
of factors. Consequently, depending on a specific combination of the joint
geometry, rivet material and type, sheet material and the riveting process the
same nominal stress amplitude (S + Sb )a may be associated with different
384
local stress amplitudes. Certainly with the tests by Hartman and Schijve,
the differences in the bending stresses are not the sole cause of the
observed differences in the fatigue performance of their specimens. First, the
load transmission is very different for the double strap joint (specimen F)
compared to the single strap joints C, D and E. Moreover, within the latter
joints the load transfer must have been also diversified due to the different
strap numbers and thicknesses. Fatigue tests on riveted specimens indicate
considerable fatigue life improvements due to increasing the rivet hole
expansion (e.g. by applying a larger riveting force or by plastically
expanding the hole prior to the rivet installation) and generating in this
way a more beneficial residual stress field [2, 10]. At the same time, the
experimental work by Muller demonstrates that the hole expansion becomes
smaller for a larger sheet thickness in spite of the same driven head diameter
[2]. With the experiments of Hartman and Schijve, the above implies that
installing the rivets to obtain the same D/Do ratio for all specimens could
lead to a diversification in the hole deformation depending on the joint total
thickness. It can be concluded that only for specimens A, B and C both
the load transmission and the residual stress field were very much alike.
Consequently, only for these joints the differences in the fatigue behaviour
can be fully attributed to secondary bending. The above reasoning is backed
up by comparing the scatter of the S N data for specimens A, B, and C.
Whilst for the Sa vs. Nf presentation of the results the correlation coefficient
is 0.87, its value jumps up to 0.96 when the fatigue life is correlated using
(S + Sb )a .
A question arises about the significance of the simple models [4,5]
for predictions on the fatigue life of riveted joints. Only a semi-empirical
prediction approach, like for example the concepts by Das et al [5] or by
Homan and Jongebreur [11], is possible because it would be not feasible to
account analytically for all the influences involved. Both methods referred
to above are based on a similarity principle, namely it is assumed that the
same local stress amplitude at the critical location for two different riveted
joints yields the same fatigue life. The predictions for the actual joint are
extrapolated from a known S N curve for a so called reference riveted
joint. The approach requires calculation of the peak local stress for the
actual riveted joint and for the reference joint. This local stress level can be
approximated by the superposition of three components, namely the stresses
induced by the bypass load, the transfer load and secondary bending. Each
stress component is expressed as the product of the nominal stress and
Bypass load is the part of the load passing the rivet hole, i.e. remaining in the sheet; transfer
load is the part of the load transmitted by the rivet to another sheet.
385
The authors acknowledge a financial support from the Eureka project No.
E!3496.
386
387
modeli oraz dostpne eksperymentalne i numeryczne dane literaturowe sugeruj, e obie koncepcje umoliwiaj poprawn ocen nominalnych napre zginajcych w krytycznym rzdzie
nitw. Przeprowadzone analizy wykazuj, e odnotowan w literaturze redukcj wytrzymaoci
zmczeniowej poczenia nitowego w zalenoci od jego konfiguracji geometrycznej mona powiza ze wzrostem wtrnego zginania. Pokazano, e niezadowalajca ilociowa korelacja pomidzy
wynikami przewidywanymi przy uyciu jednego z rozwaanych modeli a trwaoci obserwowan
w badaniach zmczeniowych prbek nitowanych wynika z nieuwzgldnienia w rozwaanych koncepcjach nierwnomiernego transferu obcie przez zcza, a take pominicia wpywu procesu
nitowania. Rwnoczenie stwierdzono, e omawiane tu proste modele s dogodnym narzdziem
do oceny wpywu wtrnego zginania i mog znale zastosowanie we wspczenie stosowanych
pempirycznych metodach przewidywania trwaoci zmczeniowej zczy nitowych.