Académique Documents
Professionnel Documents
Culture Documents
General objectives
General objectives
Single impacts
Multiple impacts
Single impacts
Single impacts
We are going to review some collision mappings:
q + = F(q , q, parameters)
Single impacts
Classes of impact dynamics modeling:
Models (2) may feed models (1) with analytical expressions for
restitution coefficients.
10
Single impacts
Assuming that the impact is instantaneous, then it is an easy
matter to deduce that the contact force is impulsive (a Dirac
measure) and that the impact dynamics is an algebraic relation
between velocties and impulses (the impulse being the Dirac
measure magnitude). For two bodies colling at a single point this
gives a relation of the type :
+
i
i
= Pi , i = 1, 2
(1)
MAi
VA+i VAi
11
Single impacts
+
It remains 13 unknowns +
i , VAi for i = 1, 2, and pn . We have 12
equations. The system may be completed by a restitution law:
12
Single impacts
Notice that if MAi is not diagonal (inertial couplings between
normal and tangential directions) then even without friction or
tangential deformation one may have jumps in i and Vt,Ai .
The system is solvable with a unique post-impact velocity and a
unique impulse with Newtons rule of impact.
The kinetic energy loss is given by:
TL =
1 m1 m2
(e 2 1)(nT (VA1 VA2 ))2
2 m1 + m2 n
X 1
[0 PiT ](A+i + Ai )
2
i =1,2
X 1
+
+ vn,i
pn,i (vn,i
2
i =1,2
+
) = pn,i , pn,1 = pn,2 and the Netwons impact law
vn,i
mi (vn,i
the result follows.
14
Single impacts
Let us turn our attention to the Darboux-Keller dynamics of
impact.
Three main definitions of the restitution coefficient:
Kinematic (Newton)
Kinetic (Poisson)
Single impacts
The Darboux-Keller model is based on some assumptions:
These assumptions may not be verified, as well as the fact that the
impact is instantaneous, or that it should not create kinetic energy
(vibrating bodies that collide may create energy at the impact
macroscopic level...).
This is an extension of Rouths graphical method that applies to
2D impacts only.
16
Single impacts
(2)
The collision
occurs on [0, tf ], and F (t) > 0 on (0, tf ) so that
R
p(t) = [0,t] F (s)ds is strictly increasing: one can safely perform a
time-scale, replacing t by p.
(there exists a strictly increasing f () such that p = f (t), f (0) = 0,
so that t = f 1 (p), v (t) = v f 1 (p) = v (p)).
17
Single impacts
dv
dp (p)
1
m
(3)
18
Single impacts
We obtain:
v (p) v (0) =
1
p
m
so that
1
1
(ptc p(0)) = ptc = v (0)
m
m
and thus ptc = mv (0) > 0 since v (0) < 0 (there is an impact).
Poissons restitution model states that
v (ptc ) v (0) =
ep =
p(tf ) p(tc )
( 0)
p(tc )
19
Single impacts
20
Single impacts
The energetic model of restitution (Stronge) states that:
elastic energy released during the expansion phase
elastic energy released during the compression phase
and is found to be equal to
e2 =
e2 =
Wn,e
( 0)
Wn,c
where
Wn,e =
v (p)dp, Wn,c =
[p(tc ),p(tf )]
v (p)dp
[0,p(tc )]
are the works performed by the normal force during the expansion
phase (resp. compression phase).
(it was used that F (t)v (t)dt = v (p)dp, and due to infinite tangential
stiffnesses the elastic energy is entirely due to the normal deformation).
21
Single impacts
One computes (recall that v (p(tc )) = 0):
Wn,e =
so that
e2 =
1
1 2
v (p(tf )) and Wn,c = (v 2 (0))
2
2
v 2 (p(tf ))
and v (p(tf )) = e v (0)
v 2 (0)
Single impacts
It easily follows that the loss of kinetic energy is given by
1
TL = T + T = m(e 2 1)v 2 (0)
2
so that TL 0 e [1, 1].
(4)
23
Single impacts
In the previous one degree-of-freedom case all three coefficients are
equal. Lets consider a 2-dimensional problem of a lamina colliding
a plane (without friction).
f0
h0
AAAAAAAAAAAAAA
AAAAAAAAAAAAAA
AAAAAAAAAAAAAA
Figure: Lamina colliding an anvil.
24
Single impacts
The kinematics at the contact point yields:
vn = f0 and vt = h0
so that in particular if f0 and h0 6= 0 one has
vn (p(tc )) = 0 vt (p(tc )) = 0 :
sliding vanishes when compression ends. So from the basic
assumptions the collision is:
(compression vn < 0 + sliding vt > 0)
followed by
(expansion vn > 0 + sliding vt < 0 )
25
Single impacts
and (t
Poissons law: Let I be the moment of inertia of the lamina
w.r.t. the rotating point O. Then after integration over the
compresion and expansion phases:
n (tf )) (0))
I ((p
= f0 pn (tf )
p(tf )p(tc )
p(tc )
f ))
= (p(t
= en .
(0)
26
Single impacts
e2
v (p )dpn
[0,pn (tc )] n n
one obtains:
e2 = ep en
27
Single impacts
Let us now pass to the case where friction is present during the
collision between two bodies.
28
Single impacts
According to [Whittaker, 1904] one writes the Coulombs model in
terms of the contact impulses, not the contact forces:
pt = f pn sgn(vt )
or more precisely (we deal with instantaneous impacts that imply
discontinuous velocities):
pt f pn sgn(vt+ )
Obviously this may introduce some errors when the tangential
velocity changes its sign (velocity reversal) during the impact,
because then the ratio tangential/normal impulses is no longer
equal to the ratio tangential/normal forces.
29
Indeed:
TO BE DONE
30
31
Single impacts
Lets first define an impulse ratio
=
pt
pn
1 m1 m2
(vr ,n (t ))2 (1+en )[(en 1)+2r +(1+en )2 ] (5)
2 m1 + m2
Single impacts
The case of a particle against a wall
The impact dynamics is
Lets try
m(v + v ) = m
vt+ vt
vn+
vn
pt
pn
pt = f pn sgn(vt+ )
with f > 0, and
vn+ = en vn , vn < 0
33
Single impacts
Simple calculations yield:
(6)
Single impacts
35
Single impacts
36
Single impacts
Lets try an explicit way: pt = f pn sgn(vt ), then we obtain:
vt+ = vt + f (1 + e)vn sgn(vt )
Let vt > 0, then vt+ = f (1 + e)vn + vt and the sign of vt+
depends on f , e and the pre-impact normal velocity magnitude: is
there some sound mechanical behaviour behind this?
Moreover vt = 0 implies vt+ f (1 + e)|vn |[1, 1], so the
mapping is multivalued (no single value of the post-impact
velocity) and the energetical behaviour is not clear.
It is always better to work with implicit formulations of the
unilateral inclusions.
37
Single impacts
The effect of inertial couplings
Lets consider a rod falling on the ground. The dynamics is given
by, with q = (x, y , )T :
0
0
1
n +
t
m
q = mg +
1
0
0
l sin()
l cos()
(7)
0 n h(q) = y l cos() 0
t = f n sgn(x + l sin())
38
Single impacts
The impact dynamics is deduced as:
m(x + x ) = pt
m(y + y ) = pn
p 0, y + l sin() 0, y + + l + sin() 0
n
(8)
39
Single impacts
Let us denote =
pt
pn ,
1
M(, )
TL = m(1 + en )N(, ) (1 en ) + r 2
2
M0
(9)
with:
N(, ) =
M(, ) =
(y +l sin())2 mI
I
2
+l sin2 ()l 2 sin() cos()
m
mI +l 2 cos2 () l 2 cos() sin()
I
+l 4 cos2 () sin2 ()
m
40
Single impacts
When sgn(vt+ ) = sgn(vt ) then one may have TL > 0 for
e [0, 1], = f sgn(vt+ ), f > 0, see [Kane and Levinson, 1985]
for the double-pendulum.
The fundamental reason for this loss of coherency is that the
orthogonality of tangent and normal velocity in the local contact
frame (euclidean metric), does not transport to orthogonality in
the configuration space of generalized coordinates (in which the
natural metric is the kinetic metric: x T M(q)y ).
Similar issues exist for sliding motions and Painleve paradoxes.
41
Single impacts
This has motivated researchers to propose various extensions of
Whittakers law to avoid such drawback.
e.g. extension of Coulombs law at the impulse level [Smith, JAM
1991]:
|vr,t |vr,t + |vr+,t |vr+,t
pt
=
pn
|vr,t |2 + |vr+,t |2
Single impacts
About the use of an impulse ratio =
pt
pn :
Single impacts
These experiments show the limitations of neglecting the possibly
finite tangential stiffness.
Single impacts
45
Single impacts
[Wu et al, Proc. R. Soc. A, 2009] conducted FEM simulations of a
disk against a half-space, and computed the ratio f as a function
of:
2(11 )
21
= c1 + c2 tanh(c3 + c4 )
f
where ci s are material property dependent and fitted, and
for c .
=1
46
Single impacts
Tangential restitution versus Coulombs friction:
Mimicking the normal restitution model:
vt+ = et vt
Remark: In an instantaneous impact framework ( algebraic
impact dynamics) all four coefficients , en , et , f satisfy
relationships. For instance for two particles colliding one has
=
1 + et vr,t
1 + en vr,n
et = 1 + (1 + en )
vr,n
vr,t
47
Single impacts
Physical meaning of et :
48
Single impacts
Some authors [Lun and Bent, Powder and Grains 1993] choose to
the following model for disk against wall:
+
vt = et,0 vt if sticking (|pt | < fpn )
+
vt = et vt
if sliding (|pt | = fpn )
with:
et = 1 + f (1 + en ) 1 +
(10)
mR 2
2
|vn |
|vt |
49
The works of Maw, Barber and Fawcett 1976, 1977, 1981, that
evidence the role of Coulomb friction, stick, slip, and the incidence
angle.
TO BE DONE
50
Single impacts
Some works [Brach, 1991; Walton 1992; Jenkins JAM 1992]
propose the use of a bilinear law of the form:
pn = (1 + en )mvn
pt = min{pn , (1 +
et )m|vt |}
(11)
sgn(vt )
Compared with the basic (en , f ) law there is one more parameter
et and this law is shown to better fit with the above experimental
data (figure 2).
This is further extended in [Chatterjee and Ruina, JAM 1998].
[Moreau 198] also adds a tangential restitution.
51
Single impacts
Some experimental results for en and et [Antonyuk et al, Granular
Matter 2010]:
TO BE DONE
53
Single impacts
Those analytical and experimental results indicate that:
The normal deformation process is independent of the
tangential one (impact angle varied from 0 to 80 degrees)
confirming other experiments [Calsamiglia et al, JAM 1998].
The tangential restitution coefficient varies with the impact
angle (transitions from rolling without slipping, to sliding at
large angles), which demonstrates that Coulombs like
phenomena are behind it (so et is a super-macroscopic
coefficient!).
For sphere/sphere or sphere/plane oblique impacts, f may
vary with |vt .
The problem raised by inertial couplings and Kane-Levinsons
example with TL > 0 is a fundamental issue: one can estimate
separately en and f from suitable experiments, but inserting
them into the Whittakers law of impact with friction no
longer works: the physical validity of en and f seems to be
54
Single impacts
The underlying issue is that using Newtons coefficient and
Coulombs friction model at the impulse level does not yield a
generalized equation for the post-impact velocities, with good
properties like maximal monotonicity (that would assure existence
and uniqueness of the solutions).
This has motivated researchers to use other approaches:
Single impacts
Lets consider a 2-dimensional problem of a lamina colliding a
plane (with friction).
f0
h0
AAAAAAAAAAAAAA
AAAAAAAAAAAAAA
AAAAAAAAAAAAAA
Figure: Lamina colliding an anvil.
56
Single impacts
f0 +fh0
f0 fh0 en
e2 = ep en
where f > 0 is the friction coefficient.
57
Single impacts
f0 +fh0
f0 fh0
1
2
e and
en < e < ep
(12)
58
Single impacts
(13)
59
Single impacts
By construction of the energetical coefficient one has necessarily
e 1 (which is the advantage of using it). The above upper
bounds may then be refined:
ep
2fh0
1+
f0 fh0
2
2fh0
1
f0 fh0
2
Single impacts
Lower bounds:
e 0 by definition.
61
Single impacts
Let us consider the impulse ratio
R
pt (tc )
[0,t ] Ft (s)ds
= R c
=
pn (tc )
[0,tc ] Fn (s)ds
From the Coulomb model Ft = f Fn sgn(vt ) and the fact that the
sliding reverses at p(tc ) we get that Ft = f Fn sgn(p(t) p(tc )).
It follows that
if pn pn (tc )
fpn
pt (pn ) =
(14)
Single impacts
It follows that
=
f (1 ep )
f ((2pn (tc ) pn (tf ))
=
pn (tf )
1 + ep
Single impacts
The parameter
fh0
f0
fh0
f0
64
Single impacts
Single impacts
Logically, impacts with friction should be prone to the same
difficulties as sliding motion with friction (frictional paroxisms for
some configurations).
However some of the Painleve paradox effects may not exist in
shock dynamics since one works with impulses and not forces: as
shown in [Genot and Brogliato, EJM A/Solids 1998] at some
plane, Fn while its impulse pn < .
points of the (, )
We retrieve here a big advantage of working with impulses,
similarly to what happens in time-stepping schemes `a la
Moreau-Jean.
66
Single impacts
Proceeding in the same way, the case of two bodies colliding with
Coulomb friction gives the dynamics:
dvr,n
dpn
dvr,t1
dpn
dvr,t2
dpn
f cos()
1
= M f sin()
1
(15)
i =1,2
1
1
1
r3i2 Ii1
,12 + r3i r1i Ii ,32 + r3i r2i Ii ,13 r2i r1i Ii ,33
Single impacts
The 3D case is much more involved. Darboux in his 1880 paper
states some results:
Proposition
If during a soft shock process a sliding phase ends, and if sliding
resumes before the end of the collision, then the direction and
orientation of the relative tangential velocity on this subsequent
period is constant.
This of course relies on the above stringent assumptions on the
impact behaviour...
To the best of the speakers knowledge, no experimental results
exist that corroborate any of the studies on Darboux-Kellers approach...which is somewhat worrying...
68
Single impacts
69
70
Single impacts
Single impacts
+en q
, = {q Rn | h(q) 0}, T (q) is the tangent
with w = q 1+e
n
cone to at q, NT (q) (w ) is the normal cone to the tangent cone,
evaluated at w .
72
Single impacts
73
Single impacts
An equivalent formulation is [Payr and Glocker 2005]:
q = v + v , q + = v en v
T (q) v (q)
v (q)
(16)
N (q)
75
Single impacts
M(q)(q + q ) = h(q)p
U + = hT (q)q + , U = hT (q)q
0 U + + en U p 0
(17)
76
Single impacts
77
Single impacts
Graphical interpretation:
Single impacts
The analysis is also led with Poissons CoR in [Payr and Glocker
2005].
[Payr and Glocker 2005] analyze various impact rules and extend
Moreaus restitution law to comply with some of the basic
requirements (like spanning of the whole admissible post-impact
velocity space).
79
Single impacts
The algebraic impact law using en and some tangential restitution
et of f or , has been the object of many analysis and extensions
in order improve the basic Whittakers law.
All of these models are gross approximations of the impact phenomenon.
It seems difficult to meet all the requirements (even for 2D
impacts!):
Single impacts
81
Single impacts
82
Single impacts
83
Single impacts
The CoR obtained from a linear spring/daspot model
m = c k (sometimes called Kelvin-Voigt):
en = exp
1 2
c .
2 km
84
Single impacts
Single impacts
2 1 2
arctan
en = exp
22 1
1 2
86
Single impacts
These two models are rather poor since en does not depend
on vn (0), contradicting experiments and more sophisticated
analysis.
Single impacts
The compliance relations obtained from Hertz theory
(statical theory of elastic contact)
Basic assumptions:
88
Single impacts
In case of two identical spheres with radii R and mass m colliding
one may then write for the collision dynamics (rate independent
materials):
m
3
d 2 n
2
=0
+
k
n
dt 2
m2
2RE
with m = mm11+m
, k = 3(1
2 )m and n is the local normal
2
deflection. For rate dependent materials:
3
p dn
d 2 n
2
+
k(
=0
n
+
c
n
dt 2
dt
where c is a constant that is a function of viscosity parameters
(not obvious to determine analytically).
the viscous dissipation is nonlinear as well...
89
Single impacts
Remark: the widely used (in some fields like robotics) Hunt and
Crossley model [Hunt and Crossley, JAM 1975]:
Fn = c||m km
3
viscous term c|| 2 and not c n d n ). It is rather used because
of its integrability property.
For low velocities it gives en 1 vn (0) for some constant and
thus reproduces a general tendency (for some materials) that en
decreases with increasing vn (0) and en = 1 for very small impact
velocity.
90
Single impacts
3Ad
2
2
3E
2
5
vn (0)
m2
, Ad =
1
1 (32 1 )2 (1 2 )(12)
.
3 32 +21
E 2
91
Single impacts
According to [Johnson, Contact Mechanics, 1985] the relation
Fn (n ) for elasto-plastic contact is not precisely defined, so one has
to resort to approximate analysis.
Assumptions:
92
Single impacts
Then
r
en 3.8
Yd
E
1 mvn2 (0)
2 Yd R 3
81
93
Single impacts
The obtained value of en in [Johnson, 1985] when compared to
experimental data (steel, aluminium alloy, brass) provides
overestimation of the real CoR.
several subsequent studies to enrich Johnsons works to better
match with experimental measurements.
[Mangwandi et al, Chem. Eng. Sci. 2007] compared various
expressions of en with experimental measurements with granules
(calcium carbonate, polyethylene glycol).
They conclude that the existing results (elastoplastic with full
plasticity during the loading phase [Johnson, 1985]; same but with elastic
contribution during loading [Thorton, 1997]; finite plastic deformation
[Wu et al 2003) yield over- or under-estimation of the CoR.
94
Single impacts
Their new model incorporates strain hardening [Mangwandi et al,
Chem. Eng. Sci. 2007]. It is given by:
en = 3
with
x=
9Ry
4El
2n+4
3
10
15El
16R 2
y2
Eul (x
(x + y )
and y =
+ y ) 2n+4
5
2n+4
i 12
k
(2+n)R n+1
(18)
i 1
(2n + 4)R n+1 mvn2 (0) 885735R 3
k
2
16394El4
95
Single impacts
Here are the results from [Mangwandi et al, Chem. Eng. Sci.
2007]:
96
Single impacts
Single impacts
Single impacts
This raises two fundamental questions:
99
Single impacts
Further results:
[Steven et al, Powder Techn. 2005] compared 8 definitions of the
CoR with experiments of stainless/stainless and
chrome-steel/chrome-steel collisions of two spheres:
Single impacts
Excerpts of the results for the CoR:
Single impacts
Excerpts of the results for CoR continued:
Single impacts
Excerpts of the results for collision duration:
Single impacts
104
Single impacts
CoR dependence on bodies temperature
The CoR may also depend on the temperature of the materials
that collide:
105
Single impacts
force/identation F ()
collision duration
106
Single impact
Also [Antonyuk et al Granular Matter 2010] show that the energy process
during pure compression (very low velocity vn (0) = 0.02 m/s) is not at
all the same as that during an impact (0.5 vn (0) 4.5 m/s): energy
absorption during pure compression than during impacts.
107
Single impact
However Hertz theory predicts well the impact process during the
elastic phases.
108
Single impacts
The effect of repeated impacts on plastic deformation
A recursive formula for the successive CoRs from one impact to
the next is given by [Weir and Tallon, Chem. Eng. Science, 2005] :
!
3
8
8
8
5
v
n,1
2
3
3
3
en,k+1
en,k
1 2.7
= en,k
+ en,1
c0
and en,k 1 as k +: plastic deformation becomes less and
less important.
[Weir and Tallon, Chem. Eng. Science, 2005] conducted
experiments of a sphere colliding an identical sphere and proved
that their analytical prediction fits well (errors 5% compared to
20% with Johnsons expression).
109
Single impacts
Single impacts
Similar tendency for rods impacting spheres (sphere/plane contact)
in [Seifried et al, Int. J. Imp. Eng. 2005] (aluminium alloy, steel):
111
Single impacts
Wave effects
In [Weir and Tallon, Chem. Eng. Science, 2005] the following CoR
expression is proposed for low velocities:
3 !
c0 vn (0) 5
en = exp 0.6
c2
c0
that takes into account wave losses, c2 and c0 are the shear and
the compressional waves velocities (resp.). When plastic
deformation holds (intermediate velocity):
en = 3.1
Yd
E
5
8
R1
R
3
8
c0
vn (0)
1
4
112
Single impacts
It seems from the above that even in the case of two spheres
colliding, some wave effects may be important for an accurate
prediction of the impact outcome.
113
Single impacts
Some general tendencies:
The normal CoR tends to 1 for zero normal incidence velocity vn (0),
and decreases exponentially with vn (0) for metals (steel, aluminium
alloy).
114
Single impacts
115
Single impacts
116
Single impacts
Many theoretical and experimental studies that concern:
Single impacts
The impact with friction phenomenon is extremely complex and
involves (too) many physical phenomena. Two general directions
may be drawn:
Single impacts
119
Single impacts
Initially no vibrations inside the bodies. Scenario:
P
1
2
Initial velocities and initial kinetic energy ( 4
i =1 2 mi vGi (0))
1 and 4 leave
The
kinetic energy is equal to the initial one and is
P4 final
1
2
i =1 2 mi vGi (0)
120
Multiple impacts
121
Multiple impacts
Let q Rn denote the vector of independent generalized
coordinates of the system in a free-motion mode (i.e. the contact
points of interest are supposed to be inactive). The inertia matrix
is denoted as M(q) assumed to be symmetric positive definite.
The gap functions hi (q) 0, 1 i m, are used to state the
non-penetrability of the contacting bodies. They are signed
distances.
We define the m gap functions hi : Rn R as differentiable
functions.
In general they are hard to compute analytically, so a numerical
estimation is necessary (collision detection algorithms).
122
Multiple impacts
M(q)
q + F (q, q,
t) = h(q)
0 h(q) 0
123
Multiple impacts
Definition
Let = {q Rn | h(q) 0} be the admissible domain of the
mechanical system. A multiple impact of order p (or a pimpact)
is an impact that occurs at a codimension p singularity of the
boundary bd().
124
Multiple impacts
Multiple impacts
cos(ij ) = q
126
Multiple impacts
case >
case =
(q0 , q 0 )
(q0 , q 0 )
0
(restitution coefficient e = 0)
(restitution coefficient e = 1)
127
Multiple impacts
A result of [Paoli, 2005] states that continuity with respect to
initial data holds under certain conditions, roughly:
hhi (q), M 1 (q)hj (q)i 0 if en = 0
(19)
(20)
(0 ij 2 )
(ij = 2 )
Multiple impacts
129
Multiple impacts
The rocking block
L
y
b'
h()
-g
AAAAAAAAAAAA
AAAAAAAAAAAA
AAAAAAAAAAAA
b
f()
a'
Multiple impacts
h1 (q) = + M (y ) 0, h2 (q) = M (y ) 0
where
M (y ) = arcsin
or equivalently
h1 (q) = y
2y
2
l + L2
l
arctan
L
L
l
L
l
cos() sin(), h2 (q) = y cos() + sin()
2
2
2
2
The two constraints are ortogonal in the kinetic metric if and only
if
l=
2L
131
Multiple impacts
The rod with two contact points
h2 (q)
h1 (q)
132
Multiple impacts
The rod at the impact:
L
d0
d1
d2
133
Multiple impacts
h1 (q) = y cos()+(d1 x) sin(), h2 (q) = y cos()+(d2 x) sin()
The (unnormalized) kinetic angle at the impact with the 2 points
(a 2-impact) is given by:
12
L
L
1
+
(d1 d0 )(d2 d0 )
12 =
m mL2
2
2
so 12 depends on the rod length and on its position w.r.t. the 2
points at the impact time:
1+
12
L
L
(d1 d0 )(d2 d0 ) > 0 12 [0, )
L2
2
2
2
1+
12
L
L
(d1 d0 )(d2 d0 ) < 0 12 ( , )
2
L
2
2
2
134
Multiple impacts
135
Multiple impacts
136
Multiple impacts
Let us now transform the Lagrange impact dynamics using some
specific state vetcor change.
The unitary normal vector to each hypersurface of constraint
hi (q) = 0, 1 i m, in the kinetic metric is
nq,i = q
M 1 (q)hi (q)
hiT (q)M 1 (q)hi (q)
137
Multiple impacts
Lets perform a specific state vector change as follows: Let
T
nq
=
tT
q
and let
M(q) = M(q)
The new vector of velocities is:
q norm
= M(q)q
q tan
that splits the generalized velocity into a normal and a
tangential components (in the kinetic metric).
138
Multiple impacts
Then the Lagrange dynamics is transformed into:
qnorm + F1 (q, q,
t) = nq Fq
qtan + F2 (q, q,
t) = 0
(21)
140
Multiple impacts
141
Multiple impacts
One may write a generalized frictionless impact law as follows for
each constraint hi (q). 1 i m:
+
= en,i q norm,i
q norm,i
(22)
1X 2
)2
(en,i 1)(q norm,i
TL =
2
(23)
i =1
142
Multiple impacts
Remark
143
Multiple impacts
Rocking block example
Considering only the left contact point one has:
q T q f1 (q)
q norm,1 = p
q f1 (q)T M 1 (q)q f1 (q)
2
y
l +L2 4y 2
with q T q f1 (q) = 2
q tang,1 =
and
+ ,
mx
)
2I
m
(t
y(t
k
k
4I +mL2
L 4I +mL2
(24)
144
Multiple impacts
Applying the above generalized Newtons rule to the first contact
point (i.e. q norm,1 )
2
2
+
+
y (t ) + (t ) = en
y(t
) + (t )
L
L
because at the impact times = 0, y =
l
2
2
l +L2 4y 2
and 2
= L2 .
2
2
2
2
)
+ ) = (2en + 1)(L + l ) + 4(l + L ) (t
(t
4l 2 + 16L2
145
Multiple impacts
This is to be compared with the widely used Housners model that
treats the rocking block as a one degree-of-freedom system and
states that
+ ) = e (t
)
(t
This tends to indicate that such a e depends on the block
dimensions and on en .
two normal generalised restitution coefficients certainly not
enough to describe rocking motion (already noticed by Moreau).
Well introduce later a matrix of generalised restitution coefficients.
146
Multiple impacts
147
Multiple impacts
(6) the ability of the impact rule to span the whole admissible
post-impact velocities domain,
Multiple impacts
149
Some items are peculiar to multiple shocks, like item (4) about
wave effects: waves through the bodies are responsible for the
dispersion of the energy.
Energy dispersion
This characterizes the fact that the kinetic energy is distributed
among the bodies of the system during the shock, as a result of
waves effects that travel throughout the mechanical system.
150
Multiple impacts
151
Multiple impacts
The textbooks solution concerns solely the case of one ball that
impacts a chain of balls at rest and in contact. Then q n+ = q 1 ,
while q i+ = 0 for all 1 i n 1, i.e. all the energy is transferred
from the first to the last ball.
152
Multiple impacts
Let us illustrate here the issue of continuity of the trajectories with
respect to the initial data. We consider chains impacting a wall:
(a)
(b)
(c)
Multiple impacts
The dynamics with two balls impacting a wall is given by:
m1 q1 (t) = m1 g + 1
m2 q2 (t) = m2 g 1 + 2
0 1 h1 (q) = q1 q2 R1 R2 0
0 2 h2 (q) = q2 R2 0
(25)
m1
cos(12 ) =
<0
(26)
m1 + m2
so that 0 < 12 < 2 and from [Paoli 2005] one may expect
discontinuity w.r.t. initial data when en 6= 0.
154
Multiple impacts
men,1
1+en,1
+
q ++ = m(1+en,1 ) q e 1en,1 m q
n,2 1+m
1
2
2
1+m
with m =
m1
m2 .
155
Multiple impacts
The sequence of impacts B1 /B2 (1 ) and B2 /wall (2 ) and then
B1 /B2 (1 ) again, produces the outcomes
q 1++ =
men,1
1+en,1
men,1
q + 1+m
q 2
1+m
1+m
n 1
o
1en,1 m
1+en,1 m(1+en,1 )
q
+
q
en,2 1+m
1
2
1+m
1+m
(28)
n
o
m(1+en,1 ) men,1
1+en,1
+++
=
q
+
q
1+m
1+m
1+m 2
o
n 1
1en,1 m m(1+en,1 )
1en,1 m
en,2 1+m
q
+
q
1
2
1+m
1+m
156
Multiple impacts
Clearly the final values in (27) and (28) are not the same.
157
Multiple impacts
Similar calculations and conclusions may be drawn for the
three-ball chain by considering two different sequences of
impacts.Let us now consider the three-ball chain with initial gaps
equal to zero, and Newtons law at each contact. Calculations give:
+ 1
q 1 = 3 [(1 2en,1 )q 1 + (1 en,2 + 2en,1 )q 2 + (1 + en,2 )q 3 ]
+ 1
q 3 = 3 [(1 + en,1 )q 1 + (1 + 2en,2 en,1 )q 2 + (1 2en,2 )q 3 ]
(29)
158
Multiple impacts
Suppose that q 2 = q 3 = 0:
So if en,i
2
2
TL 0 en,1
+ en,2
+ en,1 en,2 3
2
= 0 one has en,j
3 > 1, i 6= j.
159
160
Multiple impacts
The upper bounds of the coefficients vary with the initial data!
There is not a unique choice for en,1 and en,2 even for a given
2 + e2 + e e
energetical behaviour: TL = 0 means en,1
n,1 n,2 = 3
n,2
in the second case.
Multiple impacts
162
Multiple impacts
Three-ball chain with linear springs
The governing equations at the instant of impact are (before at
least one contact opens) :
dP1
(t) = k(x1 (t) x2 (t))
dt
dP2
(t) = k(x1 (t) x2 (t)) k(x2 (t) x3 (t))
dt
dP3
(t) = k(x2 (t) x3 (t))
dt
(30)
Multiple impacts
It is possible to show (analytically) that:
If + then q 1+ = 13 m/s
lim0 q 1+
If 0 then
lim0 q 3+ = 1 m/s.
= 0 m/s,
q 2+ = q 3+ =
lim0 q 2+
2
3
m/s
= 0 m/s,
164
Multiple impacts
It is not clear whether or not linear springs may well model wave
effects through the chain, because Hertz contact brings
nonlinearity which makes waves behave differently (even in the
case of pre-compression in the chain).
165
Multiple impacts
Complexity of the force/displacements in the three-ball chains:
from [Acary and Brogliato MIT Conf. Comp. Fl. Solid Mech., 2003]
166
Multiple impacts
Multiple impacts
168
Multiple impacts
[Reinsch 1994, Am. J. Phys.] computes masses and stiffnesses ratios so
that a linear chain is dispersion-free, which means that if n left-balls
impact m right-balls at rest, then n right-balls leave the chain while m
left-balls stay at rest after the shock.
Some contacts may last very long under some configurations and
initial data.
Multiple impacts
Multiple impacts
One may start from a quite general point of view and derive a
general restitution mapping for generalized velocities using for
instance the above (q norm , q tan ):
+
q norm
q norm
(31)
=E
+
q tan
q tan
171
Multiple impacts
172
Multiple impacts
173
Multiple impacts
174
Multiple impacts
Darboux-Keller approach is extended to the case of multiple
impacts:
175
Multiple impacts
Starting from the Lagrange equations and proceeding for instance
as Pfeiffer and Glocker (1995,1996) we may obtain the shock
dynamics as:
Md q WdP = 0
The mass matrix M and the jacobian matrix W remain unchanged
during the impact by assumption. The relative velocity of the
contact points is expressed as
i (q, t)
i (q, t) = wiT (q, t)q + w
The matrix W collects the terms wi , and dP is the vector of
normal impulses.
176
Multiple impacts
We suppose that the force/indentation mapping at the contact
point i is:
i = Ki (i )
(32)
= i
(33)
dt
dPi dt
dPi
177
Multiple impacts
In terms of the compliant model expressed by (32), we have
di
= Ki (i )1 i = Ki (i )1 wiT q
dt
(34)
i
Ki
1
(35)
i
i di = Ki wiT qdP
(36)
178
Multiple impacts
Z
i (Pi (t)) = ( + 1)
Pi (t)
0
i
Ki wiT qdP
+1
(37)
179
Multiple impacts
Ki
Kj
1
+1
Pi (t) T
i
wi qdP
R0
Pj (t) T
j
wj qdP
0
+1
(38)
180
Multiple impacts
Pi (t)
i,
wiT qdP
Ej =
Pj (t)
0
j
wjT qdP
(39)
181
Multiple impacts
Let us introduce the ratios of contact stiffnesses ji = Kj /Ki , and
define
R Pj (t) T
j
wj qdP
Ej
0
0, (j = 1, 2, . . . , s, j 6= i ) (40)
Eji =
= R P (t)
i
Ei
i
wT qdP
0
+1
dPi , j = 1, 2, . . . , s, j 6= i
(41)
182
Multiple impacts
Multiple impacts
Energetical constraints for local energy loss
According to the definition given by Stronge and his predecessors
like Routh and Boulanger, the energetic constraint es,j is given by:
2
e,j
R Pj (tc ) T
R Pj (tc )
j dPj
j
wj qdP
Wr ,j
0
= R P0 (t )
= R P (t )
=
j
j
f
f
Wc,j
j
wT qdP
j dPNj
Pj (tc )
Pj (tc )
(42)
184
Multipe impacts
Multiple impacts
186
d q
dP
=W
if Eji (Pj , Pi ) 1 for all j 6= i
dPi
dPi
(43)
with
1
+1
dPj
= ji+1 (Eji (Pj , Pi ))
dPi
Eji =
Ej (Pj ) =
Pj (t)
(44)
Ej (Pj )
, 1 i m, 1 j m
Ei (Pi )
j , Wr ,j =
wjT qdP
Pj (tc )
0
j , Wc,j =
wjT qdP
(45)
Z
Pj (tf )
Pj (tc )
j
wjT qdP
(46)
and the time tc at the contact j is calculated from j (tc ) = 0, while tf is
2
calculated from the energy constraint Wr ,j = es,j
Wc,j .
187
Multiple impacts
This is an extension of the mono-stiffness model, in which the
stiffness varies fom the compression to the expansion phases.
The relationship for the compression phase at the contact point j
is expressed as:
c,j = Kj (c,j )
and the one for expansion phase is
e,j r ,j
e,j = m,j
m,j r ,j
(47)
(48)
Multiple impacts
O
OM
Gr
GM
189
Muliple impacts
Remarkably enough, the distributing rule adapts to the bi-stiffness
model with several compression/expansion phases, and to impacts
with pre-compression:
O
OM1
M1
OM2
M2
OR
B GR
GM2 GM1
Multiple impacts
The bilinear stiffness model for energy loss: may be mechanically
justified, see for instance [Antonuyk et al Gran. Matter 2010]
Multiple impacts
Multiple impacts
Numerical results have been obtained for:
Three-ball chains
Five-ball chains
Bernoullis system
193
Multiple impacts
Let us consider the column of beads studied in [Falcon et al,
Europ. J. Phys. B, 1998]:
1
h
Wall
Multiple impacts
Initial height h
Hertz elasticity
195
N influences very little the maximum force during the impact (waves
effects).
196
197
198
200
Multiple impacts
201
An important point for multiple impact modeling is: how may the
studies on waves in chains of balls be used in a multiple impact law
?
For instance: [Nesterenko] solitary wave in a window of 5 balls
when no pre-compression
most probably good parameters should be estimated from
groups of 5 balls in the chain (if the chain is long enough)
Problem: high sensitivity w.r.t. initial data (initial velocities and
initial pre-compression, type of elasticity linear or nonlinear)
However it seems that much more has to be done on waves
analysis to be able to really enrich a multiple impact law with some
macroscopic model that reflects the waves transmissions
throughout the chain, for any initial data.
202