You are on page 1of 25

Modelling of frictional joints in dynamically

loaded structures – a review

HENRIK WENTZEL
Dept. of Solid Mechanics
Royal Institute of Technology (KTH)
(SE-100 44 Stockholm) Sweden
Email: henrik.wentzel@scania.com
Telephone : ++46+8 553 807 39

Introduction

The dynamic behaviour of joined structures is affected by the characteristics


of the joints, in particular the force-displacement relation for tangential
loading. Much research has been focused on this area triggered both by the
automotive industry’s need for shortened development cycles and by the
aeronautic/military industry’s need to reduce costly physical testing. It is a
challenging subject where non-linearities arise from different physical
phenomena such as friction, chatter and impact.

Often friction is a desired property as it will damp vibration amplitudes in


vehicle structures that otherwise would grow large and create fatigue
damage and/or noise. For example heavy vehicles with leaf-springs
suspension show desirable high damping characteristics compared to air-
spring suspended vehicles, which often makes the use of additional shock
absorbers unnecessary in the former (Johansson, 2005). On occasion friction
is unwanted in rotating machinery equipment as it increases energy losses
and destabilizes the system (Mottershead, 1997) or adds noise and chatter
(Ibrahim, 1994). Indeed, friction is an important factor for damping and may
account for up to 90% of the total damping in joined structures
(Padmanabhan, 1990).

Goodman presented a review on interfacial slip damping of joined structures


in 1959 (Goodman, 1959) and Ungar two more exhaustive studies on the
subject in 1964 and 1973 (Ungar, 1964) and (Ungar, 1973). The surveys
conclude that dry friction in joints only creates damping when the applied
forces are shearing the joint. Indeed, an experimental set-up consisting of
dozens of dry steel plates stacked on top of each other did not show any
measurable damping when loaded with a force normal to the contact area.
Shearing forces that are not sufficiently large to create global sliding will
still create some energy dissipation due to micro-slip in the joint. Shearing
forces create a difference in normal strains in the tangential direction in the
plates of the joint that results in a relative tangential motion. The frictional

1
forces oppose this motion and so energy dissipates. Figure 1 illustrates the
phenomenon. In (Earles, 1966) it was found that the maximum damping
capacity ratio for a riveted joint is obtained if all the shearing forces are
transferred via friction such that the rivet carries no shearing force. The
theories used in (Goodman, 1959), (Ungar, 1973), and (Earles, 1966) are
clearly comparable to Johnson’s theory on contact and sliding between
spherical objects (Johnson, 1990) and based on the assumption that
Coulomb friction governs the tangential interaction in the contact area.

Figure 1. Micro-slip in joints, a reproduction from (Segalman, 2001).

Except for in laboratory conditions, with constant clamping pressure over a


well defined area and uni-axial loading, the analytical theories fail to predict
the quantity of the energy loss. Part of the explanation is that a homogenous
contact pressure over a well-defined area is very rare in reality, particularly
so for bolted or riveted joints (Cullimore, 1964). However, the analytical
theories do yield valuable results as to how the energy dissipation ∆W varies
with Young’s modulus E, clamping force Fclamp, coefficient of friction μ and
amplitude
the amplitude of the applied force Fapplied . Both analytical and
experimental results show that the energy dissipation is influenced by these
parameters in the following way:

(
ΔW = ΔW E , Fclamp , μ , Fapplied
amplitude
)
∂ΔW ∂ΔW ∂ΔW ∂ΔW
< 0, < 0, < 0, >0
∂E ∂Fclamp ∂μ amplitude
∂Fapplied

Increasing the Young’s modulus makes all members of the joint stiffer,
thereby reducing the magnitude of all displacements including the relative
tangential slip responsible for energy loss. Increased clamping force and
increased coefficient of friction also makes the joint stiffer thereby
decreasing the amount of slip, but at the same time the slip that remains is
restrained by larger frictional forces, something that increases the energy

2
dissipation. However both analytical and experimental investigations show
that the overall result when increasing the clamping force or the coefficient
of friction is a decrease in energy loss. Finally the dissipation is strongly
influenced by the magnitude of the applied force in a positive way.

Specifically one of the analytical findings is that the energy dissipation is


proportional to the amplitude of the applied force raised to the power of 3.
In physical testing of actual joints this exponent is found to be closer to 2.5
(Ibrahim, 2005), (Gregory, 1999), (Smallwood, 2000) but still fairly
constant within some regions of the applied load amplitude. Extended
analytical studies by Song et al. (Song, 2005) show that the exponent 3 is
valid only for homogenous contact pressure but that other exponents apply
to other contact pressure distributions. A number of analytical and
experimental studies of bolted beam structures indicate that for a specific
load case there exists an optimal clamping force which maximises the
energy loss and hence the damping (Earles, 1966), (Beards, 1983), (Beards,
1985), (Ferri, 1995), (Ren, 1994), (Wentzel, 2005). Most presented studies
treat simple shear joints subjected to unidirectional shear and have not
investigated complicated load cases involving rotation of the joint, although
this type of loading may create much larger energy dissipation (Beards,
1977), (Wentzel, 2005).

Friction models

The Coulomb friction model is a phenomenological description of friction


and has experimentally been validated for global sliding between two rigid
bodies. For a sliding velocity v and a normal force FN the frictional force Ff
is:
F f = − sgn(v) ⋅ μ ⋅ FN

Coulomb friction is the most widely used model for sliding but also for
micro-slip (Gaul, 2001), and several modified friction models have been
derived from it. Stribeck investigated roll bearings and noted that the
coefficient of friction decreases at the onset of sliding until a certain
velocity before it increases again (Stribeck, 1902). This behaviour is called
the Stribeck effect.

A commonly used phenomenological model is the stiction model that


behaves similar to the Coulomb model but has a higher value of friction at
zero velocity (sticking) than in the sliding regime. In rubber tire applications
experimentally validated models with a coefficient of friction that strongly
depends on the velocity in a complex way are commonly used (Thorvald,
1998), (Gipser, 1990). These models are a mixture of friction and tire
models because some of the velocity dependence is caused by the relative
tangential deformations in the tire to ground contact and not by the local
tangential interaction in the contact surface (Deur, 2004). For low frequency

3
excitation of joined structures the velocities are low so the Stribeck effect is
rarely noticeable.

Physical models of friction have been proposed relating to the field of


tribology see for example (Hagman, 1993) or (Olofsson, 1995). The basic
assumption is that the (spherical) asperities on the contact surfaces interact
and deform elastically according to Hertz theory. The distribution and the
radii of asperities determine the coefficient of friction. There exist several
variants on this theory, notably the bristle-model (Haessing, 2001) where
the tangential interaction is seen as an interaction between bristles
populating the contact surfaces.

It has been noted that the energy dissipation in joints with machined contact
surfaces is influenced by the machined lay orientation (Rogers, 1975),
(Murty, 1982). Asperities are subject to plastic deformation and wear and
the influence of asperities is likely to change over time, this was measured
in (Padmanabhan, 1991) where the energy dissipation per cycle stabilized
after a couple of hundred load cycles.

Numerical implementation of friction is commercially available in several


Finite Element (FE) codes. ABAQUS (HKS, 1998) offers the possibility to
enforce the Coulomb friction model by means of Lagrange multipliers,
which sometimes is theoretically satisfactory but tends to create numerical
problems so that the solution does not easily converge. The stiction model
and other velocity dependent models are also widely available. Many FE
codes (HKS, 1998), (LS-Dyna, 1997) also include an elastic-slip friction
model where the surface interaction is elastic for small deformations. This
avoids the numerical difficulties of Lagrange multipliers and speeds up the
analysis considerably.

Experimental methods and results

Experimental studies of the dynamics of joined structures are usually


performed with hydraulic or electric shakers, or with the aid of an impact
hammer. The dynamic properties of structures are often visualised and
evaluated with Frequency Response Functions (FRF), which is the
equivalent of the systems transfer function. Commonly used FRFs are
admittance, mobility, and acellerance (Ewins, 1994). They are all based on
the Fourier transform of the response signal (u or u& or u&& ) divided by the
Fourier transform of the exciting force.
ℑ(u )
admittance:
ℑ(F )
ℑ(u& )
mobility:
ℑ(F )
ℑ(u&&)
accelerance:
ℑ(F )

4
A commonly used technique is to estimate the damping by using the half-
power points of the FRF. The damping as fraction of critical damping, ζ is
defined as the ratio between the dissipated energy per cycle and the peak
kinetic energy during vibration (times 4π). For a single Degree Of Freedom
(DOF) system with viscous damping it may be extracted from the mobility
plot as:

Δf
ζ = ,
2 fn

where Δf is the distance between the half-power points around the peak at
the natural frequency fn. This method also provides a good estimation of the
equivalent viscous damping for other types of damping and for multiple
DOF systems if the damping is small (ζ < 10 %) and the natural frequencies
are well isolated (Ewins, 1994).

Modern methods to determine the parameters of visco-elastic system models


usually involve the minimization of the error between a model FRF and a
measured FRF, (Balmes, 2005):

J = FRFmeasured (ω ) − FRFmodel (ω ) ,
2

where the following three parameter model FRF may be used for SDOF
systems,
A
FRFmodel (ω ) = 2 .
ωn − ω + 2iζωω n
2

For multi-DOF visco-elastic models this technique may be generalised. The


FRF of a multi DOF model may be written in the polynomial Laplace form
as

A(s )
FRFmodel (s ) = with
B (s )

J = FRFmeasured (s )B(s ) − A(s )


2

being a reasonable choice of the error function J. The frequency response


function may also be used to estimate joint model parameters instead of
global system parameters. Ren and Beards assumed a linear visco-elastic
joint model described by mass, stiffness and damping matrices and
developed a general methodology to extract these matrices from
experimental FRF data (Ren, 1995). Ratcliffe and Lieven assumed the same
linear joint model and enhanced the parameter identifying methodology
(Ratcliffe, 2000). Inamura has proposed a method for joint model parameter
identification that not only uses the FRF but also the mode shapes (Inamura,

5
1979). There are numerous other published studies on this approach, see for
example the review paper by Ibrahim (Ibrahim, 2003).

For linear systems the FRF is a system property and the same regardless of
which time history force is applied to the system. However, because of the
non-linear behaviour of joints and the increase in energy dissipation as
function of applied load amplitude, joined systems have signal dependent
FRF’s, i.e. for different input signals different FRF’s are obtained see for
example (Ewins, 2000).

Experimentally this is well illustrated in (Gaul, 1993) where measurements


were performed on a system consisting of two blocks of steel connected
with a single bolt lap joint. The entire system was suspended with weak
springs and excited with a sine-sweep signal by means of an electric shaker.
In that particular study different amplitudes of the excitation signal resulted
in different Frequency Response Functions, Figure 2.

Figure 2. Frequency Response Functions for a system with a bolted joint for different levels
of harmonic excitation (Gaul, 1993).

It may be seen from Figure 2 that as the amplitude of excitation and


response increases so does the equivalent viscous damping. The figure also
reveals that the stiffness decreases as the amplitude increases, because the
deformation in the joint increases with the amplitude. This is probably an
inherent property of systems with frictional joints. For a treatise on non-
linear systems in general and their FRF’s refer to (Thompson, 1993).

Hartwigsen used a 3-piece beam structure forming a bolted double lap joint,
which was suspended in weak elastic cords in the experimental study

6
(Hartwigsen, 2004). The structure was excited with an impact hammer and
the FRF was measured for different levels of excitation. The equivalent
viscous damping was extracted from the FRF for several modes of vibration
and it was presented in an amplitude-of-vibration to damping diagram,
Figure 3.

Figure 3. Amplitude-of-vibration to damping relation for mode no. 1 in Hartwigsen’s


experimental setup (Hartwigsen, 2004).

Locally in the joint the rate of energy dissipation increases rapidly with
increased amplitude of motion and globally this is observed as an increase
of the damping.

As the length of the slip region in the joint increases the stiffness of the joint
decreases (refer to Figure 1). For the global system this is seen as a
reduction of the resonance frequency for higher amplitudes.

The response surface methodology used in (Padmanabhan, 1992) provides a


method to estimate the reliability of the damping measurements and
predictions. Padmanabhan performed quasi-static measurements on a
machine joint and varied the normal pre-load and the level of excitation of
the joint. The energy dissipation is described as a polynomial function of the
level of excitation and the clamping force, Figure 4. From the surface
gradient it is apparent how the energy dissipation varies with the clamping
force and the applied load amplitude.

7
Figure 4. Energy dissipation per cycle as a surface function of clamping force, P, and
amplitude of excitation, Tm (Padmanabhan, 1992).

The non-linear properties clearly visualized in (Padmanabhan, 1992), (Gaul,


1993) and (Hartwigsen, 2004) are imperative to include in a joint model.

Joint modeling

A good joint model should be as simple as possible and still capture all the
important physical properties of the actual joint. The important properties
considered here are the force-displacement behaviour, the resulting
hysteresis loop i.e. the damping, and the influence of velocity. The wear
aspect will not be considered at this stage nor will the possibility to create
failure criteria for the joint be investigated.

Six different joint modelling techniques from the scientific literature are
presented here together with a brief description:

Linear Visco-Elastic model, a damping matrix C is chosen for the global


system such that the viscous energy dissipation equals the expected dynamic
frictional dissipation in the joints. This is by far the most widely used
technique.

Linear Complex stiffness model, a complex stiffness matrix K is chosen for


the global system such that the energy dissipation equals the expected
dynamic frictional dissipation in the joints. This technique has with some

8
success been used to model high-frequency vibration and acoustics in the
frequency plane.

Iwan networks model, a network of springs and sliders replaces the actual
joint. The network is calibrated to produce the desired quasi-static or
dynamic force-displacement characteristics (including the dissipation) of the
actual joint.

Valanis model, a first order differential equation from Valanis plasticity


model replaces the actual joint. The parameters in the equation are
calibrated to produce the desired quasi-static or dynamic force-displacement
characteristics of the actual joint.

Bouc-Wen model, an equation containing time derivates and sign functions


of the state variables replaces the actual joint. The parameters are chosen to
produce the desired quasi-static or dynamic force displacement
characteristics of the actual joint.

Detailed FE model, a detailed FE model is used to simulate the micro-slip in


the joint and calibrated to produce the quasi-static force displacement
characteristics of the actual joint.

Each of the above mentioned techniques are examined closer in the


following.

Linear visco-elastic joint model

The most common technique for modelling dynamics of systems with


frictional joints is to use a linear elastic joint model for the stiffness in
combination with viscous Rayleigh-damping. The value of Rayleigh
damping is chosen such that approximately the same amount of energy is
dissipated viscously as the frictional losses in the joints would amount to.
Thus the dissipation is not localized in space but evenly distributed in the
structure; see (Pavic, 2005) for a theoretical discussion on the validity of
this approach. The joint parameters are the linearized stiffness for each joint
and the global systems modal damping. The linearized stiffness is often
computed with FEM while the global systems modal damping is estimated
with rules of thumb (Chang, 1964) or measured in physical testing see for
example (Ellison, 1972) or (Fischer, 2000). This modelling technique is
unable to capture neither the variations in damping nor the changes of
stiffness in the joint region discussed in the previous section. It is uses a
damping force that is proportional to the velocity in order to model the
frictional forces which are in general only marginally dependent of velocity.
Despite the method’s theoretical limitations it is very popular in the
automotive industry and has also been used successfully in the aerospace
industry see for example (Chang, 1969), (Ellison, 1972).

9
In order to estimate the energy loss during vibration and hence the damping
Ellison and Jones used substructure testing of satellites transported by the
Saturn rocket (Ellison, 1972). Modal damping for mode r is defined as

ΔWr
ζr = ,
4π ⋅ Trpeak

where Trpeak is the peak kinetic energy contained in mode r and ΔWr is the
energy loss per cycle of mode r.

Each joint in the structure is characterised by a force to energy dissipation


relation. The energy dissipations in the individual joints are measured
experimentally for typical load cases of typical amplitudes. In this way a
relation between the energy dissipation and the applied load amplitude F is
obtained for each joint i

ΔW i = ΔW i ( F i ) .

Each mode shape r contribute with a force on each joint i proportional to the
amplitude of that mode qr as

F i = ∑ a ri q r ,
r

where the coefficient ari depends on the geometry and the mode shape.

Thus when the structure vibrates at a single mode r the energy loss is given
by the expression

ΔWr = ∑ ΔW i = ∑ ΔW i (a ri q r ) .
i i

Recognising that for vibration at the resonance frequency the peak kinetic
energy is well determined if the amplitude of motion is known,

1 1
Trpeak = ∫ u&&2 dm = qr2ωr2 ∫ φr2 dm ,
2 mass 2 mass

where the last integral is unity if the mode shapes are normalized with
respect to the modal mass.

Since the energy loss for each mode is known as a function of the modal
coordinate the energy equivalent modal damping is obtained with the
relation

10
∑ ΔW (a q )
i i
r r
ζr = i
.
2πq r2ω r2

The method of using substructure testing and then computing the modal
forces may work on systems with low damping. However, Bowden showed
that for highly damped systems this approach leads to erroneous results,
(Bowden, 1988). For lightly damped systems the method does produce a
damping that is energy equivalent to the dissipation in the joints when the
structure vibrates at a single mode at specific amplitude. It is not calibrated
for and is not likely to produce correct damping for vibrations at several
simultaneous modes or for dynamic processes of varying amplitude.

An alternative to Rayleigh damping is to model only the actual joint as a


visco-elastic element. In this way a simple frequency analysis of the
structure gives not only the natural frequencies and the mode shapes but
also a corresponding modal damping. The damping is largest for the modes
that create the largest relative velocities in the joint region. For highly
damped systems this model may be used in large-displacement analyses in
the actual coordinates (without transformation to the modal space).

The joint parameters are the stiffness and the viscosity of the joint material
model or the spring dashpot. In (Baraco, 1981) joints of sheet metal were
experimentally characterised by the number of bolts, clamping force and the
way the loading was applied. Different local energy equivalent viscous
damping coefficients were computed for different joints and load cases.
Coefficients that later mat be used for linear dynamic FE simulations. A
similar approach is used in (Dubigeon, 1982) for bending of bolted joints,
resulting in very simple viscous models that are treated analytically. In
(Bowden, 1988) the joints are modelled with a parallel spring and dashpot.
For any modal vibration a joint participation factor is computed which is a
measure on how much the joints are exercised during vibration.

Although the usage of viscous damping models seem to be somewhat out of


date in the scientific literature it remains the most common tool in the
industry. One of the reasons for its popularity is that dynamic analyses of
large structures are generally performed in modal coordinates for which
linear viscous damping is very easily incorporated.

Linear complex stiffness joint model

A variant of the linear visco-elastic joint model is the use of complex


Young’s modulus in the material model of the joints. Complex eigenvalue
solvers are commercially available and permit the extraction of complex
eigenvalues (Neumark, 1969). The approach permits the computation of a
deformation dependent modal damping, where the modes that create the

11
largest strain in the joint region are damped the most. In the frequency
domain the FRF of a complex stiffness problem may be formulated as:

1
FRF (ω ) =
− ω M + (1 + iγ sgn (ω ))K
2

It should be remarked that when transformed to the time domain this


equation does automatically fulfil causality, see (Inaudi, 1995), and complex
stiffness models are indeed generally used exclusively in the frequency
domain for acoustic applications. The method is not restricted to usage of an
imaginary stiffness that is proportional to the real stiffness although this
guarantees that the modal vectors are unchanged. Also for this approach it
remains to define the magnitude of the complex stiffness such that the
energy dissipation corresponds to the actual dissipation in the joints, a task
equally challenging as that of defining a viscous damping. The energy
equivalent viscous damping for a specific mode of vibration excited at the
natural frequency is related to that mode's corresponding complex
eigenvalue with the relation:

Re(λ )
ζr =
Im(λ )

This modelling technique is unable to describe the variations in damping as


the amplitude of motion increases, but it has with some success been used
for high frequency analysis of engine components (Fischer, 2000), and for
analysis of break squeal (Lou, 2004).

Joint models with Iwan networks

Many joint models are derived from friction or material models that include
plasticity. Iwan’s material model (Iwan, 1967) consisting of a network of
spring and slider elements is currently the basis for many joint models.
These models are simplifications of the joint but permit some liberty in the
design of the force-displacement characteristics.

A typical Iwan network is depicted in Figure 5. The sliders are non-linear


elements that implement the Coulomb friction model with a predetermined
normal force and coefficient of friction resulting in a break-free force fi.

12
Figure 5. Schematics of parallel Iwan networks.

For a finite number of spring-slider units, n, the force displacement


relationship is given by:
n
F (u ) = ∑ k i (u − xi ) ,
i =1
where xi is the current displacement of slider i. The above relation results in
the hysteresis curve depicted in Figure 6, where n = 4.

13
Figure 6. Force displacement hysteresis loop of a discrete parallel Iwan network of four
spring-slider units with stiffness 1 and break-free forces 1,2,3, and 4 respectively.

Of particular interest are parallel Iwan networks with an infinite number of


spring-slider units where all the springs have the same stiffness k. The
computation is performed using a population distribution ρ(f) of the break-
free force of the slider elements. For such a network the force displacement
relationship is written (Iwan, 1967):


F (u ) = ∫ ρ ( f )k [u − x( f )]df ,
0

where x(f) is the current displacement of all sliders with break-free force f.
Segalman proposes the use of the moments of distribution (Segalman, 2001)
of ρ(f) to facilitate the computation. The moments of distribution are defined
as:

θ
Λ n (θ ) = ∫ ρ ( f ) f n df
0

In (Segalman, 2001) it is demonstrated that for distributions of the type

ρ ( f ) = Rf χ ,

the energy dissipation during harmonic loading with a force of constant


amplitude F depends as:

14
ΔW ∝ F 3+ χ

Refer to Gregory and Smallwood et al. (Gregory, 1999), (Smallwood, 2000)


where it was experimentally shown that this exponent, ( 3 + χ ) may be
somewhere between 2.5 and 3 indicating a negative value of χ.

Song et al. (Song, 2004) developed a finite element based on Iwan networks
labelled the Adjusted Iwan Beam Element (AIBE) that was used for dynamic
simulation of a joined structure. The Iwan network consisted of an infinite
number of spring-slider units all with the spring stiffness k. The break-free
forces of the spring-slider units are defined by the population distribution:
β
ρ( f ) =
2 fy
[
H ( f − (1 − β ) ⋅ f y ) − H ( f − (1 + β ) ⋅ f y ) ]
In addition, the network is adjusted by adding an extra spring with stiffness
kextra = k α without a slider in series (or, equivalently, a slider with break-
free force f = ∞). Thus, the initial stiffness of the network is k(1+α). The
AIBE element consists of two such networks, Figure 7, and can transfer
both shear forces and bending moment. It is an eight parameters model with
the parameters:

{k , α , f
1 1 y1 , β1 , k 2 , α 2 , f y 2 , β 2 }

Figure 7. Schematic of the AIBE element, (Song, 2004).

If the initial stiffness is known in both directions the model is reduced to six
parameters, and if in addition β = 1, which implies that micro-slip occurs for

15
infinitesimal small loading the model is further reduced to four parameters.
Song et al. trained a neural network to identify the four parameters from
hammer excitation loading. Pettit used the AIBE element to model
variability in joints (Pettit, 2004) and developed a method to identify the
parameters from harmonic loading (Pettit, 2005). Comparison of simulated
data to experimental results showed good agreement, particularly for the
envelopes of the signals.

Parallel Iwan networks show capability to model complex energy


dissipation behaviour and can interpolate any energy dissipation that
increases convexly with the applied force (Schindler, 2005). It is a static
model where the energy dissipation and stiffness are independent of
frequency and velocity. Furthermore Iwan networks are relatively easy to
handle numerically when the number of spring-slider units is small or
described by a few parameter population distribution.

Joint models based on Valanis’ plasticity model

Valanis model of plasticity (Valanis, 1980) derives from a theory originally


intended to unify the kinematic and the isotropic hardening models. Lenz
and Gaul used this model to reconstruct measured force-displacement
characteristics of a dynamically excited lap-joint (Lenz, 1995), (Gaul,
1997). The governing equation of the model they used is the first order
differential equation:

⎡ λ ⎤
E 0 x& ⎢1 + sgn ( x& ) (Et x − F )⎥
⎣ E0 ⎦,
F& =
λ
1 + κ sgn ( x& ) (Et x − F )
E0
where E0, Et, λ, and κ are constants and model parameters.

Just like the Iwan model the Valanis model fulfils what is sometimes
referred to as the Masing hypothesis (Segalman, 2006): the force-
displacement characteristics during cyclic loading may be obtained from
reflection, translation and scaling of the force-displacement characteristics
during monotonic loading.

The Valanis model seems able to capture some of the non-linear phenomena
in the joint transfer behaviour and is clearly easily implemented in existing
FE-codes.

16
The Bouc-Wen model

Wen has developed a model to describe the restoring force in a system with
hysteresis (Wen, 1976), (Wen, 1980) based on a model first introduced in
(Bouc, 1967). According to the model the following relation describes the
total restoring force in a hysteretic system:

F ( x, x& ) = g ( x, x& ) + z ( x(τ ),τ = ] − ∞,τ ] )

where g is the non-hysteretic component and z is the hysteretic component.


The value of z is governed by a first-order differential equation:

n −1 n
z& = Ax& − α ⋅ x& ⋅ z − β ⋅ x& ⋅ z , with z(0)=0

where A, α, β, and n are parameters to be determined. The parameter A


corresponds to the initial stiffness of the joint and the other parameters
govern the shape of the hysteresis loop. Wen concluded that this
mathematical model may be used to describe a very wide range of hysteretic
behaviours. Oldsfield studied a particular case, rotation of an isolated joint,
and used this model to describe hysteretic behaviour (Oldsfield, 2003).

Detailed finite element modeling of joints

As the modelling capabilities of available FE software have increased,


several attempts have been made to model the physics in joints with detailed
FE-models. Notably the contact pressure distribution and the stick-slip state
in the contact surface during loading may be studied with detailed FE
models. In theory this permits computation of the frictional dissipation.

In the 1990’s several two dimensional FE-models appeared in various


publications. Generally these models correlate well with the analytical
theories of two dimensional frictional problems and some of the
experimentally investigated non-linearities may be predicted. See for
example Lobitz numerical study (Lobitx, 2001) of the experimental
investigations conducted by (Gregory, 1999) and (Smallwood, 2000) or
(Gaul, 1993), where slip regions and power-law dissipation for a simple
jointed resonator are correctly predicted. The two dimensional models are
however of limited practical interest since so few real joints exhibit the
plane stress or strain state and furthermore it is required that the load is
applied in particular directions. The two dimensional models generally fail
to quantify the dissipation even for very simple joints.

17
Three dimensional models including friction have been used in order to
investigate the ultimate failure of open bolted joints (Bursi, 1997) and
regular lap joints (Chung, 2000). These authors found acceptable agreement
in force-displacement characteristics under monotonic loading up to the
point of maximum force. Pratt studied a conical-head bolted lap joint with
three dimensional finite element models and compared force-displacement
characteristics with experiments during cyclic loading (Pratt, 2002) using a
rather coarse mesh but still obtaining seemingly good results. The three
authors use three different solvers with slightly different numerical
implementation of the frictional interaction and they all stress the
importance of correctly choosing the analysis parameters. This is clearly an
area of ongoing research.

Figure 8. a) Details of the finite element model used by Bursi, (Bursi, 1997). b) Details of
the finite element model used by Chung, (Chung, 2000). c) Details of the finite element
model used by Pratt, (Pratt, 2002).

Although the detailed three dimensional FE models seem to be able capture


many of the important physical phenomena in the joints the extensive
demand of computational resources is a restraint, particularly so for analysis
of dynamic processes.

To overcome this problem studies have been made to use the detailed FE-
models in quasi-static simulation to compute simplified joint model
parameters. Oldsfield studied a detailed FE-model of an isolated joint and
used the results from static FE-simulations to design a parallel Iwan-
network and a Bouc-Wen model with similar properties, (Oldsfield, 2003).
Wentzel used a detailed FE-model to compute the energy dissipation in
joints during loading and computed an equivalent modal viscous damping
for the global structure (Wentzel, 2005).

Dynamics of joined structures and future development

The primary aim of simplified joint models is to permit simulation of


dynamic processes. Distinction is made between cases where only the
steady-state solution to a particular harmonic loading is sought for and cases
where the actual time history of a long dynamic process is to be determined.

18
Steady state solutions can be computed analytically for systems with a very
limited number of DOF's and Coulomb friction (Nosonovsky, 2004). This is
of limited interest in vehicle systems where steady state but rarely is reached
and the models considered often are complex with many degrees of
freedom.

In transient response analysis a common technique is to use some kind of


reduction method. One reduction technique is Gyuan reduction (Gyuan,
1965) where the mass of the internal points is neglected and only the
external points of the structure are retained. A more common variant is
Component Mode Synthesis (CMS), (Bampton, 1968), (Hurty, 1971). The
CMS method retains a few internal DOFs such that the lowest internal mode
shapes are retained, thus not completely neglecting the mass of the internal
points. The method makes use of a transformation matrix composed by the
local eigenvectors. The constituting parts of the global system that are
approximately linear are reduced so that only the external (and a few
internal if CMS is used) DOFs are retained. A linear model (constant
stiffness and mass matrices) denoted linear substructure is replacing the
reduced parts. The different linear substructures may be interconnected to
each other via non-linear joint models such as the Iwan-, Valanis-, or Bouc-
Wen-model. It has also been envisaged to connect linear substructures to a
detailed (not simplified) joint model (Gaul, 1993). However, not much is
gained with this approach since it is the computation of the dynamic
frictional interaction that is computationally expensive. In the foreseeable
future it remains interesting to model the complex joint mechanics with
simplified models. This is particularly important for applications where the
actual joint is not the primary aim of the study, but where the force-
displacement characteristics are needed in order to compute the structural
response. This may for example be time-domain dynamic analyses of large
joined structures.

Of particular interest in the future are applications where multiple modes are
excited simultaneously. Multiple mode vibration complicates joint
mechanics even more and has scarcely been treated at all. Another area
where the industry demands improvement is in the parameter estimation of
the simplified models. Most of the authors in the scientific literature on joint
models use experimental data to find suitable model parameters. Some
attempts have been made to extract parameters for simplified models from
detailed FE-models, (Oldsfield, 2003), (Wentzel, 2005). This approach will,
if proven successful, have substantial impact in the industry.

19
Concluding remarks

The force displacement characteristics of frictional joints play an important


role in the dynamic behaviour of joined structures. Depending on the
available test resources and knowledge of the system different joint models
are available. In general, for simulations where the primary interest is to
model the dynamics far from the joint, a phenomenological description of
friction or damping is generally apt to provide satisfactory results. However
it is necessary to further investigate the behaviour of these simplified
models in systems with multiple simultaneous modes of vibrations.

Detailed FE-models may in many cases capture the important characteristics


during static or quasi-static loading. If these detailed models may be used to
estimate parameters for simplified joint models much is won. The ongoing
research on simplified joint models has so far provided several interesting
alternatives of which Iwan networks is perhaps the most promising. The
need for predictive joint models in the industrial product development phase
and simplified joint models for dynamic simulations will continue to drive
the development for years to come.

References

Balmes, E., Methods for vibration design and validation, Ecole Centrale
Paris, 2005.
Bampton, M.C.C., Craig, R.R., Coupling of substructures for dynamic
analyses. AIAA Journal 6-7 pp.1313-1319, 1968.
Baraco, A., Blanze, C. Amortissement dans les assemblages boulonnes,
DGRST no 80-7.0654, 1981.
Beards, C.F., Williams, J.L., The damping of structural vibration by rotating
slip in joints, Journal of Sound and Vibration 53 pp. 333-340, 1977.
Beards, C.F., The damping of structural vibration by controlled interface
slip in joints, American Society of Mechanical Engineers, Journal of
Vibration, Acoustics, Stress analysis, Reliability and Design 105, pp. 369-
373, 1983.
Beards, C.F., Woodwat, The control of frame vibration by friction damping
in joints, American Society of Mechanical Engineers, Journal of Vibration,
Acoustics, Stress Analysis, Reliability, and Design 107 pp. 27–32, 1985.
Bouc, R., Force vibration of mechanical systems with hysteresis. Proc. 4th
Conf. on Nonlinear Oscill., 1967.
Bowden, M., Dugundji, J., Effects of joint damping and joint nonlinearity
on the dynamics of space structures, 29th Structures, Structural Dynamics
and Materials conference, 1988.

20
Bursi, O.S., Jaspart, J.P., Calibration of a finite element model for isolated
bolted end-plate steel connections, J. Construct. Steel Res. 44(3), pp. 225-
262, 1997.
Chang, C.S., Damping in multi-beam vibration analysis, part II, Lockheed
Missiles & Space Co., LMSC-HREC A710166, 1964.
Chang, C.S., Ellison A.M., Robinson G.D., Prediction of structural damping
for future Saturn flights final report. 1969.
Chung, K.F., Ip, K.H., Finite element modelling of bolted connections
between cold-formed steel strips and hot rolled steel plates under static
shear loading, Engineering Structures 22 pp.1271-1284, 2000.
Cullimore, M.S.G., Upton, K.A., The distribution of pressure between two
flat plates bolted together. Int. J. Mechanical Sciences, 6, pp. 13-25, 1964.
Deur, J., Asgari, J., Hrovat, D., A 3d brush-type dynamic tire friction model,
Vehicle System Dynamics, 42(3) pp. 133-173, 2004.
Dubigeon, S., Kim, C.B., Raideur et amortissement structurels d’un
assemblage par boulonnage, DGRST no 79.7.1189, 1982.
Earles, S.W.E, Theoretical Estimation of the frictional energy dissipation in
a simple lap joint, Mechanical Engineering Science, 8(2), 1966.
Ellison, A.M., Jones, W. E., Modal Damping Predictions Using
Substructure Testing, in National Aerospace Engineering and
Manufacturing Meeting, 1972.
Ewins, D.J., Modal Testing, Theory and Practice, ISBN: 0863800173,
Research Studies Press, 1994.
Ewins, D.J., Basic and state-of-the-art modal testing, Sãdhãna, 25(3), pp.
207-220, 2000.
Ferri, A. A. Friction damping and isolation systems, Journal of Mechanical
Design, 117, pp.196-206, 1995.
Fischer, P., Engelbrechtsmüller, M., Local damping effects in acoustic
analysis of large FE engine structures, International Conference on Noise
and Vibration Engineering, 2000.
Gaul, L., Nackenhorst U., Willner v, Lenz J., Nonlinear vibration damping
of structures with bolted joints. Proc. Of 12th IMAC, pp. 875-881, 1993.
Gaul, L., Lenz J., Nonlinear dynamics of structures assembled by bolted
joints, Acta Mechanica, 125(1-4) pp. 169-181, 1997.
Gaul, L., Nitsche R., Dynamics of structures with Joint Connections, in
Structural Dynamics @2000: current status and future directions, edited by
Inman & Ewins, 2001.
Gipser, M. Zur modellierung des reifens in CASCaDE. Mitteilungen des
Curt-Risch-Institutes der Universität Hannover. Hannover, 1990.

21
Goodman, L., A review of progress in analysis of interfacial slip damping,
Structural Damping, papers presented at a Colloquim on structural damping
held at the ASME annual meeting in Atlantic City, NJ, edited by Jerome E.
Ruzicka, pp. 35-48, 1959.
Gregory, D.L., Smallwood, D.O., Coleman, R.G., Nusser M.A.,
Experimental Studies to Investigate Damping In Frictional Shear Joints,
Proc. 70st Shock and vibration symposium, 1999.
Gyuan, R.J., Reduction of stiffness and mass matrices, AIAA Journal, 3,
280, 1965.

Haessing, D.A., Friedland, B., On the modeling and simulation of friction,


journal of dynamic systems, measurements, and control, 1991.
Hagman, L. Micro-slip under surface deformation. Licenciate Thesis, Royal
Institute of Technology, Stockholm, 1993.
Hartwigsen, C.J., Song, Y., McFarland, D.M., Bergman, L.A., Vakakis,
A.F., Experimental study of non-linear effects in a typical shear lap joint
configuration, Journal of Sound and Vibration 277, pp. 327-351, 2004.
Hibbit, Karlsson & Sorensen, Inc, ABAQUS Theory Manual Version 5.8,
Pawtucket, 1998.
Hurty, W.C, Collins, J.D., Hart, G.C., Dynamic analysis of large structures
by modal synthesis techniques, Comp. Struct, 1, 535-563, 1971.
Ibrahim, R.A., Friction-induced vibration, chatter, squeal, and chaos – Part
I: Mechanics of contact and friction. Applied Mechanics Reviews,
47(7):209-226, 1994.
Ibrahim, R.A., Friction-induced vibration, chatter, squeal, and chaos – Part
II: Dynamics and modelling. Applied Mechanics Reviews, 47(7):227-253,
1994.
Ibrahim, R.A., Pettit, C.L., Uncertainties and dynamic problems of bolted
joints and other fasterners, Journal of Sound and Vibration 279, pp. 857-
936, 2005.
Inamura, T., Sata, T., Stifness and damping properties of the elements of a
machine tool structure, Annals of the CIRP 28, pp. 235-239, 1979.
Inaudi J.A., Kelly J.M., Linear hysteretic damping and the Hilbert
transform, Journal of Engineering Mechanics, 121(5) pp. 626-632, 1995.

Iwan, W.D., On a class of models for the yielding behavior of continuous


composite systems, Journal of Applied Mechanics, 89, pp. 612-617, 1967.
Johansson, A. personal communications, 2005.
Johnson, K.L. Contact Mechanics. Cambridge University Press, Cambridge,
1990.

22
Lenz, J., Gaul, L., The influence of microslip on the dynamic behaviour of
bolted joints, Proceedings of the 13th International Modal Analysis
Conference, pp. 248-254, 1995.
Lobitz, D.W., Gregory, D.O., Smallwood, D.O., Comparison of finite
element predictions to measurements from the Sandia microslip experiment,
Proceedings of International Modal Analysis Conference, Orlando, FL,
2001.
Lou, G., Wu, T.W., Bai, Z, Disk brake squeal prediction using the ABEL
algorithm, Journal of Sound and Vibration, 272 (3-5), pp. 731-748, 2004.
LS-DYNA User's manual-version 940, Hallquist, J.O., Livermore Software
Technology Corporation, 1997.
Mottershead, J.E., Ouyang, H., Cartmell, M.P., Friswell, M.I.,
Parametric resonances in an annular disc, with a rotating system of
distributed mass and elasticity; And the effects of friction and damping,
Procedings of the Royal Society of London series A-Mathematical physical
and engineering sciences, 453 (1956), pp. 1-19, 1997.
Murty, A.S.R., Padmanahban, K.K., Effect of surface topology on damping
in machine joints, Precision Engineering 4(4) pp. 185-190, 1982.
Neumark, S., Concept of complex stiffness applied to problems of
oscillation with viscous and hysteretic damping, Aeronautical research
council reports and memoranda No. 3269, 1957.
Nosonovsky, M., Adams, G.G., Vibration and stability of frictional sliding
of two elastic bodies with a wavy contact interface, Journal of applied
mechanics-transactions of the ASME 71(2), pp. 154-161, 2004.
Oldsfield, M., Ouyang, H., Mottershead, J.E., Kyprianou, A., Modelling and
simulation of bolted joints under harmonic excitation, Materials Science
Forum 440-441, pp. 421-428, 2003.
Olofsson, U., Cyclic micro-slip under unlubricated conditions, Tribology
International, 28(4) pp. 207-217, 1995.
Padmanahban, K.K., Murty, A.S.R., Damping in structural joints subjected
to tangential loads, Proc. Inst. Mechanical Engineers 205, pp. 121-129,
1991.
Padmanabhan, K.K., Prediction of damping in machined joints,
International journal of machine tools manufacturing, 32(3), pp. 305-312,
1992.
Pavic, G., The role of damping on energy and power in vibrating systems,
Journal of Sound and Vibration, 281(1) pp 45-71, 2005.

Pettit, C., Measurements and Modeling of Variability in the Dynamics of a


Bolted Joint, AIAA-2004-1622, 45th Structural Dynamics and Materials
Conference, 2004.

23
Pettit, C., Parameter Identification and Investigation of a Bolted Joint
Model, AIAA-2005-2378, 46th Structural Dynamics and Materials
Conference, 2005.
Pratt, J.D., Pardoen, G., Numerical modelling of bolted lap joint behaviour,
Journal of aerospace engineering, pp. 20-31, January 2002.
Ratcliffe, M.J., Lieven, N.A.J., A generic element-based method for joint
identification, Mechanical systems and signal processing, 14(1), pp. 3-28,
2000.
Ren, Y., Beards, C.F., An experimental study on the dynamic response of a
beam structure containing a pseudo joint, Proc. Instn. Mechanical Engineers
208, pp. 321-328, 1994.
Ren, Y., Beards, C.F., Identification of joint properties of a structure using
FRF data, Journal of Sound and Vibration, 186(4), pp 567-587, 1995.
Rogers, P.F., Boothroyd, G., Damping at metallic interfaces subjected to
oscillating tangential loads., Journal of Engineering for Industry, august
1975.
Segalman, D.J., An initial overview of Iwan modelling for mechanical
joints, Sandia report SAND2001-0811, 2001.
Segalman, D.J., Modelling joint friction in structural dynamics, Structural
Control and Health Monitoring, 13 pp. 430-453, 2006.

Smallwood, D.O., Gregory, D.L., Coleman, R.G., Nusser M.A., Damping


Investigations of a Simplified Frictional Shear Joint, Proc. 71st Shock and
vibration symposium, 2000.
Song, Y., Hartwigsen, C. J., Bergman, L. A., Vakakis, A. F., A three-
dimensional nonlinear reduced-order predictive joint model, Earthquake
Engineering and Engineering Vibration, 2(1), 2003.
Song, Y., Hartwigsen, C.J., McFarland, D.M., Vakakis, A.F., Bergman
L.A., Simulation of dynamics of beam structures with bolted joints using
adjusted Iwan beam elements, Journal of Sound and Vibration 273, pp. 249-
276, 2004.
Song, Y., McFarland, D.M., Bergman, L. A., Vakakis, A. F., Effect of
pressure distribution on energy dissipation in a mechanical lap joint, AIAA
Journal 43(2), pp. 420-425, 2005.
Schindler, A., Scania Report RTCC05013, 2005.
Stribeck, R., Die wesentlichen Eigenschaften der Gleit- und Rollenlager.
Zeitschrift des Vereins deutscher Ingenieure, 46(38,39) pp. 1342-48, 1432-
37, 1902.
Thomson, W., Theory of vibration with applications, Fourth edition,
Chapman & Hall, 1993.

24
Thorvald, B., On truck tyre modelling, Doctoral thesis, Dept. of Vehicle
Engineering KTH, 1998.
Ungar, E.E., The status of engineering knowledge concerning the damping
of built-up structures, Journal of Sound and Vibration, 26, pp.141-154,
1973.
Ungar, E.E., Energy dissipation at structural joints; mechanisms and
magnitudes, Bolt Technical Documentary Report No. FDL-TDR-64-98, Air
Force Flight Dynamics Lab, 1964.
Valanis, K.C., Fundamental consequences of a new intrinsic time measure.
Plasticity as a limit of the endochronic theory. Archive of Mechanics, 32, pp
171-191, 1980.
Wen, Y.K. Method of random vibration of hysteretic systems, J. Eng. Mech.
– ASCE 102, pp. 249-263, 1976.
Wen, Y.K. Equivalent linearization for hysteretic systems under random
excitation, J. Appl. Mech.-T ASME 47, pp. 150-154, 1980.
Wentzel, H., Olsson, M. Numerical prediction of damping in structures with
frictional joints, Int. Journal of Vehicle Noise and Vibration, in press.
Wentzel, H., Olsson, M., Influence from contact pressure distribution on
energy dissipation in bolted joints, SAE world conference 2006, in press.

25

You might also like