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

net/publication/242145994

Implementation of the Contensou-Erismann Model of Friction in Frame of

the Hertz Contact Problem on Modelica

Article · October 2009

DOI: 10.3384/ecp09430006

CITATIONS READS

10 76

2 authors, including:

Ivan I. Kosenko
Moscow Aviation Institute
91 PUBLICATIONS   266 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Dynamics of the omni-wheeled vehicle View project

All content following this page was uploaded by Ivan I. Kosenko on 25 December 2014.

The user has requested enhancement of the downloaded file.

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

Implementation of the Contensou–Erismann Model of Friction in

Frame of the Hertz Contact Problem on Modelica
Ivan Kosenko Evgeniy Aleksandrov
Russian State University of Tourism and Service, Department of Engineering Mechanics
Glavnaya str. 99, Cherkizovo-1, Moscow reg., 141221, Russia

Abstract area arising in the Hertz model. It is known as the

Contensou–Erismann friction model [1, 2].
An approximate model to compute resulting wrench The model assumes the resulting wrench of the dry
of the dry friction tangent forces in frame of the Hertz friction tangent forces. The wrench consists of the to-
contact problem is built up. An approach under con- tal friction force and the drilling friction torque. An
sideration develops in a natural way the contact model approach under consideration continues in a natural
constructed earlier. Generally an analytic computa- way the contact model development started earlier [3].
tion of the integrals in the Contensou–Erismann model The normal contact force distribution is determined by
leads to the cumbersome calculation, decades of terms, the Hertz model while the tangent forces on an ele-
including rational functions depending in turn on com- mental level satisfy the Amontons–Coulomb law for
plete elliptic integrals. To implement the elastic bodies dry friction.
contact interaction computer model fast enough one The dry friction force and torque are integrated over
builds up an approximate model in the way initially the contact elliptic spot thus composing the resulting
proposed by Contensou. wrench. Generally an analytic computation of the in-
To verify the model built results obtained by several tegrals mentioned leads to the cumbersome calcula-
authors were applied. First the Tippe-Top dynamic tion including decades of terms depending on rational
model is used as an example under testing. It turned functions depending in turn on complete elliptic inte-
out the top revolution process is identical to one simu- grals.
lated with use of the set-valued functions approach. To keep an accuracy and to make the model fast
In addition, the ball bearing dynamic model was also enough an approach proposed initially by Contensou
used to verify different approaches to the tangent [1] is built up. The model under construction is one
forces computational implementation in details. A derived from the Contensou simplified model in the
model objects corresponding to contacts between balls following directions: (a) the model is anisotropic: total
and raceways were replaced by ones of a new class de- friction forces along the contact ellipse axes are differ-
veloped here. Then the friction model of the approxi- ent; (b) for the translatory and almost translatory rela-
mate Contensou type embedded into the whole bearing tive motions one uses the Amontons–Coulomb friction
dynamic model was thoroughly tested. law regularization [4]; (c) the approximate model for
Keywords: Hertz contact model; Contensou simplified the drilling torque also is under construction.
model; Contensou–Erismann model; Vil’ke model;
Tippe-Top; ball bearing model
2 Problem Formulation
1 Introduction The Hertz problem solution [5] to a normal pressure
distributed over the contact area of elliptic shape is de-
To make a contact model for the multibody dynamics fined [6] by the formula
more accurate and simultaneosly more efficient using s
the Hertz contact problem as a frame one has to de- 3N x2 y2
velop an approach taking into account nature of the σ(x, y) = 1− 2 − 2,
2πab a b
tangent forces acting along a contact spot area. The
simplest case one could encounter in this way is one where N is the total force of normal pressure, a and b
of the dry friction forces distributed over the elliptic are the contact spot ellipse semi-major and semi-minor

© The Modelica Association, 2009 288 DOI: 10.3384/ecp09430006

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

axes respectively, see Figure 1, Pxyz is the contact lo- where vM/P (x, y) = [ω ω, r(x, y)], and r(x, y) = xi + yj is
cal coordinate frame oriented such that the x-axis is the current point M(x, y) radius vector within the con-
directed along the ellipse semi-major axis. All three tact spot, see Figure 1. The ellipse central point rela-
values: N, a, and b supposed already computed by the tive slip velocity vP is represented as follows
Hertz algorithm [3].  
α
vP = v  β  = vw, α = cos ψ, β = sin ψ,
0

where v is a relative slip velocity value at the point P,

and ψ is the angle between the axis Px and vector vP .
According to the Contensou–Erismann model [1, 2] to
compute the dry friction total force and torque vectors
one has to evaluate integrals over the contact elliptic
area in the following way
ZZ ZZ
Figure 1: The contact spot area F= τ (x, y)dxdy, T = [r(x, y), τ (x, y)] dxdy.
The body A supposed “below/behind” the picture of (3)
Figure 1 plane while the body B supposed “above/in For the further use it is suitable to introduce the dimen-
front of” it. In addition, all the forces under computa- sionless velocity u = v/aωz of relative slipping at the
tion here supposed to act to the body B from the body point P instead of a dimensioned one.
A. Consider a method to compute a wrench consist-
ing of the tangent force F = Fx i + Fy j and the resulting
torque T = Tz k arising due to couple of dry friction
3 Theoretical Background
forces distributed over the contact area. This latter one
It turned out the friction total force and drilling friction
usually called a drilling friction torque.
torque components are regular functions of the relative
According to the dry friction definition the tangent
sliding (dimensionless) velocity u, relative angular ve-
stress τ (x, y) at the contact spot any point M(x, y) is
locity supposed fixed parameter here, at a center of the
computed by the formula
ellipse such that for the exact force F and torque T we
vM (x, y) have
τ(x, y) = − f σ(x, y) , (1)
|vM (x, y)|
F(u) = F∞ , T(u) = T∞
¡ −2 ¢ −1 + O u−2
¡ ¢
where f is the dry friction coefficient, vM (x, y) is the 0 +O u 1u
relative slip velocity of the body B with respect to (4)
(w. r. t.) body A at the geometric point M. The right as u −→ ∞, and
hand side of Eqn. (1) isn’t regular. Because of that the
F(u) = F01 u + O u2 , T(u) = T00 + O u2
¡ ¢ ¡ ¢
(5)
set-valued functions calculus is applied frequently to
the problems including dry friction [7]. Let us try to 0 0
as u −→ 0. Here F∞ ∞
0 , T1 , F1 , T0 are constant vectors
build up a computational procedures for the dry fric-
defining the approximate model. The vectors F01 , T00
tion problems staying in frame of classical calculus
depend on complete elliptic integrals of the first and
and using the known results [8, 9] on asymptotic close-
the second kind depending in turn on the contact el-
ness for an exact and an approximate problems.
lipse eccentricity in the following way
Assuming the bodies A and B to be rigid from the kine-
matic viewpoint the body B relative motion, along the F10 x = −αAu, F10 y = −βBu, T00 z = −C
contact spot plane, is an instant planar (the relative ve-
locity normal component supposed to participate only with the constants
in the normal force computation), and subsequently
field of velocities over the spot is defined by the known 3 K(e) − E(e)
A = ,
Euler formula 2 µ e2
3
¶
E(e) − K(e)
B = K(e) + ,
 
vPx − ωz y 2
2 e
vM (x, y) = vP + vM/P (x, y) =  vPy + ωz x  , (2) 3
0 C = E(e),
8

© The Modelica Association, 2009 289

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

where K(e), E(e) are complete elliptic integrals of the

first and second kind respectively. Remark that really
the values A, B, C are a variable functions of time be-
cause the contact ellipse eccentricity e can vary while
the simulation process.
Note here the dry friction total wrench simplified
model doesn’t require any noticeable computational
resources because complete elliptic integrals men-
tioned are already calculated while computing the total
normal force according to the Hertz model.
As Contensou [1] remarked the main effect in the
Contensou–Erismann dry friction model if the con-
Figure 2: x-components of the force vector for ex-
tact area is non-zero besides the drilling friction torque
act and approximate models (similar picture for y-
arises is that the total friction force decreases mono-
components)
tonically to zero as a function of u.
On the other hand one can easily note from (5) that a
steepness of the total friction force change, as a func- infinity and being defined by the function Tz∞ (u). An
tion of v, grows as aωz −→ 0. In this case either con- equation specifying the value u∗z has the form
tact spot area decreases to zero or the drilling angu- T1z∞
lar velocity vanishes. Finally, for the value aωz small C=− .
u∗z
enough and A, B ≥ aωz /δ, where δ is a regulariza-
tion parameter for the case of dry friction, we have
Comparison of graphs for the Contensou–Erismann
the almost point contact case already implemented ear-
model functions is represented in Figures 2 and 3.
lier [4] as a dry friction model regularization. Thus in
Functions of the exact model correspond to solid lines,
the current simplified Contensou model resulting com-
and ones of the approximate model correspond to the
puter model always uses the “regular” case of the fric-
dotted lines.
tion force decreasing, though sometimes steep, to zero.
Taking into account that according to (4) for u large
enough the simplified friction force differs from its
Coulomb’s value by the magnitude of the second order
of smallness and following the Contensou proposal [1]
let us simplify our model such that the friction force
supposed
£ ∗ to¢ be of the Amontons–Coulomb £ ∗ ¢type for
u ∈ ux,y , ∞ and linear one for u ∈ 0, ux,y . Note,
the friction force has an anisotropy here: constants u∗x
and u∗y along axes Px and Py respectively are in gen-
eral different. Evidently, we can find these values from
equations
Au∗x = 1, Bu∗y = 1.
Figure 3: The drilling friction torque for exact and ap-
Now we can represent the Contensou approximate
proximate models
model for the (dimensionless) functions Fx (u), Fy (u),
Tz (u) as follows
If while the simulation process all the values u∗x , u∗y ,
½
Au for u ∈ [0, u∗x ) , u∗z found become less than δ/aωz then we arrive at
XC (u) = −α the regularized Coulomb model implemented earlier.
½ 1 for u ∈£[u∗x , ∞)
¢,
Bu for u ∈ £0, u∗y ¢, Thus the approximate Contensou model implemented
YC (u) = −β here really is a simplest generalization of the regu-
1 for u ∈ £u∗y , ∞¢ ,
½
−C for u ∈ £0, u∗z ¢, larized Coulomb one mentioned regularizing it even
TC (u) = ∞ −1 more by introducing the parameters u∗x , u∗y , u∗z enhanc-
T1z u for u ∈ u∗z , ∞ ,
ing initial use of the parameter δ. Such an improve-
where u∗z is a sewing point for the horizontal “shelf” of ment simply is a consequence of the contact spot ex-
height C and a branch of the hyperbola decreasing at istence in the exact model. Thus as a result, with the

© The Modelica Association, 2009 290

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

exception of the cases of a = 0 and ωz = 0, we can After the value c = c∗ needed has been computed then
avoid use the set-valued functions being able to apply to find the values A, B, C mentioned above we should
the procedures of classical calculus. calculate complete elliptic integrals of the first and sec-
The approximate model under construction here has ond kind using theta-functions [14]. First of all for any
several differences from a piece-wise linear approxi- c∗ ∈ [0, 1) one can use the expansion
mation built up in the paper [1]:
∞
2

(a) the model is anisotropic and is suitable for the θ3 (q) = 1 + 2 ∑ qn

n=1
elliptic contact area of any eccentricity;
with fast conversion such that complete elliptic inte-
(b) for the cases of instant translatory and almost in- gral of the first kind can be computed by the formula
stant translatory bodies in the contact relative mo-
tion with the conditions π
K(c) = θ23 (q),
2
∗ ∗ ∗
ux aωz < δ, uy aωz < δ, uz aωz < δ
where nome q is calculated with a very high accuracy
fulfilled simultaneously we apply the dry friction using equations [14, 15]
regularization proposed in [8] and [11];
1 1 − (1 − c)1/4
ε= · ,
(c) the approximate model used also for the drilling 2 1 + (1 − c)1/4
friction torque.
q = ε + 2ε5 + 15ε9 + 150ε13 + 1707ε17 + . . . ,
Computations show the force/torque expressions rep-
resented here give an approximation of the Contensou- and the terms enumerated above are sufficient for the
–Erismann model more accurate in compare with the accuracy level of order not less than one for the value
linear-fractional approximation satisfying boundary 1/221 .
conditions at zero and infinity. If we use the Pade ap- It is convenient for complete elliptic integral of the
proximations with the polynomials of the second and second kind to use the formula [16]
third degrees [12] then the resulting accuracy is im- " #
proved but computations become more significant. 2−c π2 1 ∞
q2n
E(c) = K(c) + −2 ∑ .
3 K(c) 12 2n 2
It is known [13] the V. G. Vil’ke formula gives an ap- n=1 (1 − q )
proximation for the contact interaction normal elastic
force decent enough in a wide range of eccentricities. If the value of c is small then to regularize the expres-
Computer implementation of such a model runs no- sion
E(c) − K(c) dE
ticeably faster than the implementation of the exact =
2c dc
Hertz model. The main reason for that is a necessity in
the latter case to resolve the transcendental equation one can use hypergeometric expansions converging
well enough in this case [17].
1 K(c)
− (1 − c) = g (0 ≤ c < 1, 0 < g ≤ 1) (6)
2 K0 (c)
4 Implementation Specifics
w. r. t. c = e2 which is the contact spot eccentricity
According to experience while developing the models
squared. Here we use the elliptic integral modulus
for elastic contacting of rigid bodies interactions in the
squared c as an argument of complete elliptic integral
multibody dynamics a flexibility provided by Model-
of the first kind, as it has been done in [14].
ica can be used to utilize a wide variety of different
The V. G. Vil’ke algorithm to compute the normal con-
properties concerning a contact of solids. The proper-
tact force doesn’t require to know the current value
ties are mainly of the following categories:
of c but the Contensou–Erismann anisotropic friction
model does. To keep the gain has been gotten while the (a) geometric properties for surfaces in vicinity of
normal force calculation and don’t waste the computer the contact spot (gradients of the functions defin-
time to resolve the equation (6) this time to compute ing surfaces, their Hesse matrices);
the tangent friction force it turned out to be possible
that the solution mentioned can be reduced to the ex- (b) a model to compute the contact area dimensions
plicit linear formula once applied. and normal elastic force;

© The Modelica Association, 2009 291

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

Figure 4: The model of mechanical contact by stages of inheritance.

(c) model for the normal viscous force of resistance; ...

replaceable
(d) model for the tangent forces along the plane of NormalElasticForce
the contact area. normalElasticForce;
replaceable
A submodel of the geometry properties is to describe NormalViscousForce
analytically algebraic surfaces of the structure com- normalViscousForce;
plex enough. To implement the normal force computa- replaceable
tion one can choose from at least two approaches: the TangentForce
Hertz model and its volumetric modification. Force tangentForce;
of viscous resistance also can be modeled in several replaceable
SurfacesOfConstraintDifferential
different ways: linear, non-linear, etc. In the mod-
geometry;
els for tangent forces one can adopt either “simplest” ...
approaches based on the Amontons–Coulomb friction end ContactConstraintTemplate;
or more complex ones represented by the Contensou–
Erismann, and other models. To develop complete model one can move along differ-
While developing a mechanical contact model archi- ent ways. Class parametrization implemented in Mod-
tecture we used the base class Constraint described elica is the facility in line to apply to the problem under
earlier [18] as a starting point to construct its inher- description. In our case we have four class parameters
itor ContactConstraintTemplate being simulta- corresponding to the submodel categories enumerated
neously a base class of new family of models to sim- above. An example to construct specific contact inter-
ulate mechanical contacts. Really this class is a base action model see in Figure 4. The example includes
template represented as a container having four “sock-
two stages of inheritance:
ets” to instantiate there the specific parameter classes
of four types enumerated above, see its visual model 1. to derive a template with the forces models,
in Figure 4 at a top left corner, and its Modelica code namely: the Hertz model for normal force, non-
can be outlined as follows linear viscous force, the Contensou–Erismann
partial model ContactConstraintTemplate model for the dry friction forces (to “fill in” three
extends Constraint; sockets in the middle of the base template visual

© The Modelica Association, 2009 292

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

model, see the derived template visual model at a

central position of the Figure 4);

2. to complete the whole construct one should de-

fine a specific geometry submodel for the sur-
faces in contact (to “seal” the socket for geom-
etry properties, see the complete visual model at
a bottom right corner of the Figure 4).
The Modelica code for the intermediate derived tem-
plate can be represented in the following way
partial model
ContactConstraintTemplate...
extends ContactConstraintTemplate(
redeclare
NormalElacticForceHertzDiff Figure 5: The Tippe-Top geometric properties
normalElasticForce,
redeclare
NormalViscousForceNonLinear
normalViscousForce,
redeclare
TangentForceContensou
tangentForce);
...
end
ContactConstraintTemplate...;

On all the stages of inheritance the templates con-

sidered have an internal information interconnections
between the submodels to be instantiated. These in-
terconnections are implemented via the set of equa-
tions hidden behind the visual models and can vary
Figure 6: The top axis of symmetry evolution
for different models requiring different variables for
the algorithms to compute normal and tangent forces
of the complete model. So the whole picture remind investigated as an example of the first case. All the
us known construct of a card with the sockets and the parameters and initial conditions are exactly the same
interconnection wiring in its internal layers as a base as in the paper [19] whose authors got these data in
template, and a chips to be instantiated in the sockets turn from the work [20]. The only difference is that in
as a models of four types from above. With one exclu- our case we considered an unrestricted problem with
sion: we have the derived template playing a role of the contact ellipse, including depth of penetration and
additional card with its own additional wiring servic- normal force, being computed dynamically.
ing already instantiated models “covering the card” of
The top body, supposed geometrically rigid, com-
the base template.
posed by two balls, Figure 5, one of larger radius
One can remark finally an approach under presentation
R = 1.5·10−2 m, and another, smaller, one of the radius
allows us to create and to change fast enough different
r = 0.5 · 10−2 m. The top mass center location sup-
types of an elastic contact models while developing the
posed “under” the larger ball geometric center on its
multibody dynamics systems simulators.
axis of symmetry at a distance of a0 = 3 · 10−3 m and at
the distance of a1 = 16 · 10−3 m to the smaller ball cen-
5 Numeric Experiments ter. The top mass is equal to m = 6 · 10−3 kg. The top
body supposed dynamically symmetric, and the cen-
The tangent forces model under presentation here has tral principal moments of inertia are the following: an
been verified by two stages: (a) for the case of cir- equatorial moment equals to 8 · 10−7 kg · m2 , and a po-
cular contact; (b) for the case of elliptic non-circular lar one has the value of 7 · 10−7 kg · m2 . A material the
contact. The known Tippe-Top dynamical model was top and the horizontal floor the top rolls on are made

© The Modelica Association, 2009 293

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

Figure 7: The contact indicators evolution. Some fragments zoomed in and rescaled.

of the wood with Young’s modulus E = 9.1·109 N/m2 . elastic forces, changes. Then the contact spot changes
If we suppose the Poisson ratios as 0.3 then an effec- also, and so for its radius. Thus the top really under-
tive Young’s modulus has to be E∗ = 5 · 109 N/m2 just goes the vertical microvibrations. And, as one can see
as in [19]. The dry friction coefficient supposed to be from the above simplified model, the drilling friction
equal to the value f = 0.3. torque also changes causing in general a consequences
The top center of mass supposed resting at initial in- for the top motion.
stant of motion. Besides the top itself, more accurately In the model under development here we consider an
its larger ball, assumed without any initial penetration unrestricted contact problem that is the normal force is
with the horizontal surface. The smaller ball is located computed from the Hertz (or V. G. Vil’ke) model with
on the upper hemisphere of the larger ball, and ini- addition of some nonlinear viscous term. Simultane-
tially the top axis of symmetry bends w. r. t. verti- ously the contact area is computed too. Then all the
cal by the angle θ0 = 0.1rad. Initial angular velocity data have been gotten are used to calculate the tangent
ω0 = 180s−1 is the same as in [19] and directed along force and the drilling friction torque in frames of the
axis of the top symmetry. simplified Contensou model.
Note that in [19] contact problem is interpreted as Remarkably, a computational experiment showed the
usual in so to speak “restricted” sense: the contact top revolution from “feet”, the larger ball in contact, to
area supposed constant and predefined corresponding “head”, the smaller ball in contact, scenario obtained
to the normal force of the static equilibrium. This as- in [19] using another approach to the problem, based
sumption concerns the contacts for both the balls with on the set-valued functions calculus, repeated in our
the same contact area radius. Actually, while motion model with a high degree of accuracy. One can get
the normal reaction force, being implemented here by an access to the paper [19] visiting, for instance, the

© The Modelica Association, 2009 294

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

Figure 8: Comparison of three models

one of the authors Home Page, see [21]. Graph in into the modulated pulse decreasing afterwards. One
Figure 6 illustrates well the Tippe-Top revolution pro- can match the problem parameters such that the pulse
cess: similar to [19] it shows the θ(t) angle evolution. amplitude will grow and the top can start to bounce
If we compare this plot with one from [19] then soon over the floor thus distorting all the following dynami-
the complete identity can be observed. The only dif- cal predictions of its revolution. In the upper subfigure
ference is that in Figure 6 one can find additional vi- we can observe also the change of the balls at contact,
brations of small amplitude evidently existing due to before the instant of time = 2 seconds. Then for the
elastic compliance in direction normal to the contact case of the smaller ball contacting the floor we observe
area. Similar identity show the curves of the contact the larger depth of penetration. Indeed, in this case we
indicators for the balls the top composed of and the have a smaller area of the contact spot.
horizontal surface, see Figure 7. The indicator for the The bottom subfigure reflects the revolution process
pair (larger ball, floor) marked by the solid line while inself. Here the whole graph zoomed in vicinity of the
the (smaller ball, floor) contact indicator pictured as a time instant of 1.825 seconds, and we see that the rev-
dashed curve. olution process is implemented by two attempts: two
Really an indicators are the distances between an op- times the solid humps alternate the dashed ones. Thus
posite points for the surfaces being tracked for contact. first time the Tippe-Top “head” touches the floor then
The indicator is strictly positive if contact is absent. it once more is forced to loose a contact temporarily,
Otherwise it is less than (if the bodies are in a state and only then the head–floor contact becomes perma-
of mutual penetration) or equal to (if the bodies touch nent. The right subfigure illustrates the depth of pen-
one another exactly at one point) zero. Let us describe etration for the larger and smaller balls by the vertical
the Figure 7 in more details. Initially the top smaller rescaling over the same interval as for the bottom sub-
ball is out of contact, and corresponding indicator is figure.
positive, dashed curve. But other indicator is not equal In addition, yet another verification procedure has
to zero. Instead it is negative, see the vertically scaled been performed, this time using the results of the pa-
subfigure at the upper edge of Figure 7. Here at the per [10] (one can access the paper [10] using the Sci-
very left side we see that initially indicator set to be of enceDirect on-line library [22]). Namely, exact for-
the zero value. Then the penetration develops and the mulae for the friction force and for the drilling fric-
whole top sinks into the floor by very small depth un- tion torque, case (a), were applied to the top dynamics
til the vertical quasi-equilibrium is reached. After that computer model implemented on Modelica language
we can observe the vertical micro-oscillations develop in frame of the unrestricted, in sense mentioned above,

© The Modelica Association, 2009 295

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

the case (a) while the dashed one represents the case
(b). A values of the contact spot eccentricity squared
appeared to be constant equal to 0.687 in the case (a)
and 0.643 in the case (b).

Figure 9: The Tippe-Top 3D-animation

contact model. In the same dynamical frame the sim-

plified Contensou model, case (b), as well as a linear-
fractional Pade approximation for the friction force
and torque, case (c), were also implemented. The re-
sults of the inclination angle evolution are shown in
Figure 8, where the cases (a), (b), and (c) correspond
to the solid, dashed, and dotted curves respectively.
One can see easily the revolution scenarios are mu-
tually closest in cases (a) and (b). The 3D-animation
shot is shown in Figure 9.
Note in addition, one can easily obtain a behavior typi-
cal to the Tippe-Top, revolution to “head”, in frame of
the “regularized” Amontons–Coulomb friction. One
has to understand regularization in a sense proposed
in the works [8, 11] and used in [4] in case of the
point contact. We only have to “bend” graph for the
friction force dependence on the relative slip velocity
in vicinity of zero replacing its discontinuity by the
linear function. The more flat slope of the graph the
sooner one can find out the Tippe-Top revolution ef-
fect. As the simplified Contensou model shows that
just this slope appears in the corresponding graph for Figure 10: One of the friction force component evolu-
the friction force dependence on the velocity, this time tion. A final stage of the simulation zoomed in.
in frame of the exact Contensou–Erismann model.
The dynamical model of the ball bearing was consid-
ered in a way similar to the paper [3] while the verifica- 6 Conclusions
tion second stage. This time the contact area is essen-
tially elliptic one. The main goal for the numeric sim- Summarizing the results described above we can re-
ulations was to compare two approaches: (a) the stan- mark the following.
dard Hertz model for the normal force plus the Con-
tensou simplified model for the friction forces; (b) the • The Tippe-Top “on head” revolution effect is
simplified model of V. G. Vil’ke for the normal elastic caused completely by the dry friction force “reg-
force plus the Contensou simplified model for the fric- ularization” in vicinity of zero value for the ve-
tion forces. As it was observed in [3] for the case of locity of relative slip. Such a regularization takes
the regularized Coulomb friction force here dynamical place exactly in the Contensou–Erismann model.
models of the cases (a) and (b) differ one from another Numeric experiments showed if the slope of fric-
in a slightest degree too. Simultaneously, the model tion force graph in vicinity of the zero velocity in
(b) is faster than (a) by 20% meaning the CPU time the regularized Coulomb model is steep enough
needed. To illustrate this in Figure 10 we compare the then the Tippe-Top effect either isn’t observed at
cases (a) and (b) for one component of the tangent fric- all or arising during short time after a long evolu-
tion force at a contact between one of the bearing ball tion then vanishes quickly. And only noticeable
and its inner raceway. The solid curve corresponds to decreasing of the slope mentioned immediately

© The Modelica Association, 2009 296

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

causes the top revolution on the “head” with the References

subsequent long precession in this position.
[1] Contensou, P., Couplage entre frottement de gli-
The Contensou–Erismann model creates a prop-
ssement et frottement de pivotement dans la
erty just as one described above. Note the drilling
théorie de la toupie. In: Kreiselprobleme Gyro-
friction torque role is reduced to a dissipative ef-
dynamics: IUTAM Symposium Celerina, 1962,
fect with subsequent gradual “fall” of the top ap-
Berlin: Springer, 1963, pp. 201–216.
proaching it to the static stable configuration.
[2] Erismann, Th., Theorie und Anwendungen des
• Since complete elliptic integrals used in the Con-
echten Kugelgetriebes. Z. angew. Math. Phys.,
tensou simplified model are already found in
1954, Vol. 5, No. 5, pp. 355–388.
frame of the Hertz algorithm while computing the
normal force then from the computational view- [3] Kosenko I. I., Alexandrov E. B., Implementa-
point application of this model is practically “cost tion of the Hertz Contact Model and Its Volumet-
free”. If, in addition, we will take into account ric Modification on Modelica. In: Bachmann, B.
an effect of the regularization provided by the (Ed.) Proceedings of the 6th International Model-
Contensou–Erismann model then we arrive at a ica Conference, Bielefeld, Germany, March 3–4,
unexpected from the first sight result: a numeric 2008, Bielefeld: The Modelica Association, and
simulation of the Hertz model for the normal University of Applied Sciences Bielefeld, 2008,
force and the Contensou–Erismann for the tan- pp. 203–212.
gent force and the drilling friction torque turned
out to be faster than the combination of the Hertz [4] Kossenko, I. I., Implementation of Unilat-
model and the “simple” Amontons–Coulomb dry eral Multibody Dynamics on Modelica. In:
friction. It is evident such a deceleration in lat- Schmitz, G. (Ed.) Proceedings of the 4th Interna-
ter case surely concerns the large stiffness of the tional Modelica Conference, Hamburg–Harburg,
problem while the almost rolling mode. Germany, March 7–8, 2005, Hamburg–Harburg:
The Modelica Association, and The Depart-
• Though for isotropic case, one of the circu- ment of Thermodynamics, Hamburg University
lar contact area, the tangent forces average val- of Technology, 2005, pp. 13–23.
ues for the Amontons–Coulomb and Contensou–
Erismann models differ not so much, however in [5] Hertz, H., Über die Berührung fester elastischer
anisotropic case the first model becomes inad- Körper. J. reine und angewandte Mathematik,
equate while the second one continues to serve 1882, B. 92, S. 156–171.
correctly the contacting process simulation. Such
a property has an importance for instance in case [6] Landau, L. D. and Lifshitz, E. M., Theory
of the ball bearing simulation with the contact ar- of Elasticity. 3rd Edition. Landau and Lifshitz
eas of essentially elliptic form. Course of Theoretical Physics. Volume 7. Oxford
– Boston – Johannesburg – Melbourne – New
Regarding the directions of a future work one can Delhi – Singapore: Reed Educational and Pro-
enumerate possible development and testing for dif- fessional Publishing Ltd., 1999.
ferent kinds of the contact properties combinations:
normal-elastic-force / normal-viscous-force / tangent- [7] Leine, R. I. and Nijmeijer, H., Dynamics and Bi-
force+drilling-friction-torque to match various engi- furcations of Non-Smooth Mechanical Systems.
neering applications. It would be for instance different Berlin – Heidelberg – New York: Springer Ver-
types of lubrication, or any new types of the normal lag, 2004.
elastic volumetric models etc.
[8] Novozhilov, I. V., Conditions of Stagnation in
Systems with the Coulomb Friction. Mechanics
7 Acknowledgement of Solids, 1973, Vol. 8, No. 1, pp. 8–14.

The paper was prepared with partial support of Rus- [9] Novozhilov, I. V., Fractional Analysis: Methods
sian Foundation for Basic Research, projects 08-01- of Motion Decomposition, Boston: Birkhauser,
00600-a, 08-01-00718-a, 08-08-00553-a. 1997.

© The Modelica Association, 2009 297

Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009

[10] Zhuravlev, V. F., The Model of Dry Friction [19] Leine, R. I. and Glocker, Ch., A Set-Valued
in the Problem of the Rolling of Rigid Bodies. Force Law for Spatial Coulomb–Contensou Fric-
J. Appl. Math. Mech., 1998, Vol. 62, No. 5, tion. Europian Journal of Mechanics A/Solids,
pp. 705–710. 2003, Vol. 22, No. 2, pp. 193–216.
[11] Rooney, G. T. and Deravi, P., Coulomb Friction [20] Friedl, C., Der Stehaufkreisel. Master’s thesis.
in Mechanism Sliding Joints. Mechanism and Augsburg: Institut für Physik, Universität Augs-
Machine Theory, 1982, Vol. 17, Iss. 3, pp. 207– burg, 1997.
211.
[21] http://www.zfm.ethz.ch/~leine/
[12] Kireenkov, A. A., Three-Dimensional Model publications.htm
of Combined Dry Friction and Its Applica-
tion in Non-Holonomic Mechanics. In: van [22] http://www.sciencedirect.com/
Campen, D. H., Lazurko, M. D. van den science/journal/00218928
Oever, W. P. J. M., (Eds.) Proceedings of ENOC-
2005, Fifth EUROMECH Nonlinear Dynam-
ics Conference, Eindhoven, August 7–12, 2005,
Eindhoven, The Netherlands: Eindhoven Univer-
sity of Technology, 2005.
[13] Aleksandrov, E. B., Vil’ke, V. G. Kosenko,
I. I., Hertzian Contact Problem: Numerical Re-
duction and Volumetric Modification. Computa-
tional Mathematics and Mathematical Physics,
2008, Vol. 48, No. 12, pp. 2226–2240.
[14] Whittaker, E. T., Watson, G. N., A Course of
Modern Analysis, Cambridge – New York – Mel-
bourne – Madrid – Cape Town: Cambridge Uni-
versity Press, 2002.
[15] Janke, E., Emde, F., Lösch, F., Tafeln Höherer
Funktionen, Stuttgart: B. G. Teubner Verlagsge-
sellschaft, 1960.
[16] Milne-Thomson, L. M., Elliptic Integrals,
Abramowitz, M., Stegun, I. A., (Eds) Hand-
book of Mathematical Functions: with Formulas,
Graphs, and Mathematical Tables, New York:
Dover Publications Inc., 1972.
[17] Bateman, H., Erdélyi, A., Higher Transcenden-
tal Functions. Volume 3. New-York – Toronto
– London: Mc Graw-Hill Book Company, Inc.,
1955.
[18] Kosenko, I. I., Loginova, M. S., Obraztsov, Ya. P.
and Stavrovskaya, M. S., Multibody Sys-
tems Dynamics: Modelica Implementation and
Bond Graph Representation. In: Kral, Ch. and
Haumer, A., (Eds.) Proceedings of the 5th Inter-
national Modelica Conference, Vienna, Austria,
September 4–5, 2006, Vienna: The Modelica As-
sociation, and arsenal research, 2006, pp. 213–
223.

© The Modelica Association, 2009 298

View publication stats