Trindade Wolter Sampaio 2005 JSV

JOURNAL OF
SOUND AND
VIBRATION
www.elsevier.com/locate/jsvi
Journal of Sound and Vibration 279 (2005) 10151036
KarhunenLo" eve decomposition of coupled axial/bending
vibrations of beams subject to impacts
M.A. Trindade, C. Wolter, R. Sampaio*
Department of Mechanical Engineering, Pontif!cia Universidade Cat ! olica do Rio de Janeiro,
rua Marqu# es de S* ao Vicente, 225, 22453-900 Rio de Janeiro, Brazil
Received 15 March 2002; accepted 25 November 2003
Abstract
This work presents a study of the oscillations of a vertical slender beam, clamped in its upper extreme,
pinned in its lower one and constrained inside an outer cylinder in its lower portion. The beam is subject to
distributed axial loads, due to its own weight, leading to geometric softening of its lower portion and thus
yielding a large number of vibroimpacts with the outer cylinder. This is due to the axialbending coupling,
often called geometric stiffening and largely discussed in the last two decades. Here, it is accounted for by
using a non-linear nite element model proposed in a previous work, in which non-linear strain
displacement relations are considered. To help understand this non-linear coupled vibro-impact problem,
the KarhunenLo" eve decomposition, also known as the proper orthogonal decomposition, is applied to its
simulated dynamics. The results show that the micro-impacts, accompanying the beamhole impacts and
mainly due to the beam compressive softening, and the reaction forces at the top and bottom positions, are
well represented only when using a non-linear axialbending coupling. It is also shown that 15 proper
orthogonal modes are sufcient to reconstruct the dynamics of the impacting beam under a 3% error
margin.
r 2004 Elsevier Ltd. All rights reserved.
1. Introduction
It is well-known that exible beams subject to axial loads present strong stiffness variations.
This is due to what is often called geometric stiffening effect in the literature [1]. It may also be seen
as a consequence of the coupling between axial and bending strains. In the last two decades,
several methodologies have been proposed to account for the geometric stiffening effect. In
particular, Simo and Vu-Quoc [2] showed that modelling beams under large rotations using linear
ARTICLE IN PRESS
*Corresponding author.
E-mail address: [email protected] (R. Sampaio).
0022-460X/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jsv.2003.11.057
beam theories results in a spurious loss of stiffness and hence they proposed a consistent
linearization using steady state values for the axial internal force. Kane et al. [3] proposed a
methodology, that uses higher order strain measures, and applied it to the dynamics of a
cantilever beam attached to a moving base under prescribed large translation and rotation.
This idea was later generalized to arbitrary exible bodies by Banerjee and Dickens [4],
Wallrapp and Schwertassek [5], and more recently by Urruzola et al. [6], to treat structural
elements other than beams. Some of these models were summarized and compared by Trindade
and Sampaio [7] using a general non-linear model, resulting from non-linear straindisplacement
relations. Their conclusion is that a non-linear model accounting for the coupling between
axial and bending vibrations is required for an accurate representation of the dynamics of
slender beams.
Due to the non-linearity induced by the geometric stiffening, augmented by the intrinsic non-
linear behavior of vibroimpacting structures, one is obliged to consider non-linear analysis
techniques. The KarhunenLo" eve (KL) decomposition, also known as the proper orthogonal
decomposition, is a powerful tool for obtaining spatial information and providing a basis for
model reduction of non-linear structural systems [810]. It consists in obtaining a set of
orthogonal eigenfunctions (or proper orthogonal modes) where the dynamics is to be projected.
This set of KL modes are optimal in the sense that it minimizes the error of the approximation for
any number of modes considered, meaning that no other linear expansion may lead to a better
representation of the dynamics response with the same number of modes. Indeed, Steindl and
Troger [10] concluded, in their comparative work, that KL modes are by far the best choice for a
standard Galerkin approximation. Practically, the KL decomposition is obtained by constructing
a spatial autocorrelation tensor from the simulated or measured dynamics of the system.
Thereafter, performing its spectral decomposition, one nds that the spatial autocorrelation
tensor eigenfunctions provide the required proper orthogonal modes and its eigenvalues represent
the mean energy contained in that projection. This technique was previously used for the analysis
of vibroimpact problems by Azeez and Vakakis [9] and Wolter and Sampaio [11]. Other
model reduction techniques applied for a similar vibroimpact problem were also presented by
Friswell et al. [12].
In the present work, the non-linear oscillations of a typical conguration of drillstring used for
oilwell drilling are studied. Drillstring dynamics can present complex vibrational states and there
is a strong need to understand it in order to better control the drilling process. Several works on
drillstring dynamics have been published in the literature. Yigit and Christoforou [13] developed a
model to study the transverse vibrations of drillstrings caused by axial loading and impact with
the wellbore wall. In a later work, the same authors extended their model to account also for
torsional vibrations [14]. Tucker and Wang [15] have presented more recently an integrated model
for drillstring dynamics accounting for axial, bending and torsional vibrations.
Here, the drillstring is represented by a vertical slender beam, clamped in its upper extreme,
pinned in its lower one and constrained inside an outer cylinder in its lower portion. The beam is
subject to distributed axial loads, due to its own weight, leading to geometric softening of its lower
portion and thus yielding a large number of vibroimpacts with the outer cylinder. To help
understand this non-linear coupled vibroimpact problem, the KL decomposition is applied to its
simulated dynamics, which is evaluated using an extension of the non-linear nite element (FE)
model proposed in a previous work [7].
ARTICLE IN PRESS
M.A. Trindade et al. / Journal of Sound and Vibration 279 (2005) 10151036 1016
2. Non-linear model formulation
Let us consider an initially straight and slender cylinder, of undeformed length L and outer and
inner radii R
o
and R
i
; undergoing large displacements and small deformations as shown in Fig. 1.
2.1. Displacements and strain measures
Small deformations are assumed so that the beam cross-section rotation angle b is small. Also,
the assumption of negligible shear strains, leading to b v
0
; is considered. Notice that the prime
denotes the derivative with respect to the axial co-ordinate x: Consequently, the displacement
vector p of a given point with position X in the xz plane is
p
u zv
0
v
_ _
for X
x
z
_ _
; 1
where x and z directions are such that 0pxpL and R
o
pzpR
o
: The deformation gradient F
relative to the reference position is dened as
F
dp
dX
I
_ _
2
with I dening the identity operator. The Lagrangian strain tensor E reads
E
1
2
F
T
F I
1
2
dp
dX
_ _
dp
dX
_ _
T
dp
dX
_ _
T
dp
dX
_ _
_ _
: 3
Here, only the axial component of the strain tensor e
xx
E
11
is considered. Therefore, dening
the axial displacement as u
0
u zv
0
; the non-linear axial strain e
xx
may be written in the form
e
xx
u
0
0

1
2
u
0
0
2
v
0
2
: 4
ARTICLE IN PRESS
u
i
o
R
R
x
z
g
v
Fig. 1. Slender cylinder undergoing large displacements and small deformations.
2.2. Strain and kinetic energies
From the assumption of negligible shear strains and also neglecting the contribution of
transversal normal stress s
zz
; the strain potential energy of the beam is
H
1
2
_
Ee
2
xx
dV; 5
where E is the Youngs modulus of the beam. Considering a symmetric beam cross-section with
respect to z-axis and using the axial strain denition (4), the potential energy (5) of the beam may
be written in terms of the mean axial u and transversal v displacements only. Thus,
H
1
2
_
L
0
EA u
02
u
0
v
02
1=4v
04
u
03
1=4u
04
1=2u
02
v
02
_ _ _
EI v
002
3u
0
v
002
3=2u
02
v
002
3h
2
=20v
004
1=2v
02
v
002
_ __
dx; 6
where A and I are the area and moment of inertia of the beam cross-section and h is its thickness.
Single underlined terms in Eq. (6) are due to the presence of term v
0
2
; quadratic in the cross-
section rotation angle b v
0
; in the axial strain e
xx
: Notice that they appear only in the
membrane strain component, unlike double and triple underlined terms that are present in both
membrane and bending components of the strain energy. The term u
0
0
2
; quadratic in the axial
displacement derivative, in the axial strain (4) leads to the double underlined terms in the strain
energy function, while triple underlined terms are due to the coupling between the two quadratic
terms of the axial strain. It is worthwhile to notice also that the assumption of a linear
straindisplacement relation eliminates all underlined terms of Eq. (6).
In the present work, only the contributions of cubic and lower order terms in u
0
0
and v
0
are
retained in the potential energy (6). From the denition of u
0
; this leads to the simplied potential
energy
H
s

1
2
_
L
0
EA u
02
u
0
v
02
u
03
_ _
EI v
002
3u
0
v
002
_ _ _ _
dx: 7
The kinetic energy of the beam may also be written in terms of the main variables u and v:
Hence, starting from the general form of the kinetic energy in terms of the total displacement of
the beam, one gets
T
1
2
_
r p
T
p dV; 8
where r is the beam mass density and p is the velocity vector of a given point X of the beam. From
Eq. (1), dening p; and assuming a symmetric beam cross-section with respect to z-axis, the kinetic
energy of the beam may be written in terms of the main variables u and v as
T
1
2
_
L
0
rA u
2
v
2
rI v
02
dx; 9
where the terms in Eq. (9) correspond to inertia contributions due to translation, in x and z
directions, and cross-section rotation.
ARTICLE IN PRESS
The beam is subject to its own weight, which may be expressed as the following vertical forces,
that is in the x direction, due to the gravity eld:
f
g

rg
0
_ _
10
and, assuming symmetric cross-section, their work may be written as
W
_
p
T
f
g
dV
_
L
0
rgAu dx: 11
2.3. Variational formulation
Using the expressions for strain (Eq. (7)) and kinetic (Eq. (9)) energies and work due to gravity
forces (Eq. (11)) presented above, a variational formulation is used in this section to derive the FE
model.
The virtual variation of the simplied strain energy H
s
is decomposed in linear and non-linear
contributions arising from the non-linear straindisplacement relations (4) and, hence, is written
as
dH
s
dH
sl
dH
sn
; 12
where the linear dH
sl
and non-linear dH
sn
contributions are expressed in terms of the variations
du
0
; dv
0
and dv
00
as
dH
sl

_
L
0
du
0
EAu
0
dv
00
EIv
00
dx; 13
dH
sn

_
L
0
1
2
du
0
EA3u
02
v
02

3
2
du
0
EIv
002
3dv
00
EIu
0
v
00
dv
0
EAu
0
v
0
_
dx: 14
Notice that the non-linear contributions dH
sn
come from the underlined terms in Eq. (7). On
the other hand, the linear contributions dH
sl
are the standard ones for EulerBernoulli beams.
The virtual variation of the kinetic energy T may be derived from Eq. (9), leading to
dT
_
L
0
rAd u u d v v rId v
0
v
0
dx; 15
which through integration in time is equivalent to
_
t
2
t
1
dT
_
t
2
t
1
_
L
0
rAdu. u dv. v rIdv
0
. v
0
dx: 16
This expression may also be interpreted as the virtual work done by the inertial forces,
composed of translation in x and z directions and cross-section rotation in the xz plane.
ARTICLE IN PRESS
The virtual work done by the gravity forces is obtained from Eq. (11) and is written in terms of
du only since these forces keep their vertical x direction under deformation:
dW
_
L
0
rgAdu dx: 17
2.4. Non-linear FE model
The FE model is constructed through discretization of the virtual variations of strain (Eq. (12))
and kinetic (Eq. (16)) energies. It is clear that neglecting the other terms in the strain energy
(Eq. (6)) leads to the absence of several non-linear terms in the model. However, it is assumed for
now that the main contributions to the axialbending coupling should be accounted for by the
terms considered in Eq. (7).
The discretization is carried out using Lagrange linear shape functions for the axial
displacement u and Hermite cubic ones for the transversal deection v: This leads to an element
with six degrees of freedom q
T
e
fu
1
v
1
b
1
u
2
v
2
b
2
g; where b
1
; b
2
v
0
1
; v
0
2
: Moreover, the
axial and transversal displacements are discretized as
u N
u
q
e
; v N
v
q
e
; 18
where, dening the element length c and the adimensional axial position x x=c;
N
u
f 1 x 0 0 x 0 0 g;
N
v
0 1 3x
2
2x
3
xc1 x
2
0 x
2
3 2x x
2
cx 1
_ _
: 19
Replacing the discrete expressions for the displacements and their derivatives in the elementary
versions of linear and non-linear contributions to the virtual variation of strain energy, Eqs. (13)
and (14), leads to
dH
e
sl
dq
T
e
K
e
e
q
e
; dH
e
sn
dq
T
e
K
e
g
q
e
; 20
where the expressions for the elementary linear elastic stiffness K
e
e
and non-linear geometric
stiffness K
e
g
matrices are
K
e
e

_
c
0
EAN
0T
u
N
0
u
EIN
00T
v
N
00
v
dx; 21
K
e
g

_
c
0
EA
3
2
N
0T
u
N
0
u
q
e
N
0
u

1
2
N
0T
u
N
0
v
q
e
N
0
v
N
0T
v
N
0
u
q
e
N
0
v
_ _ _
EI
3
2
N
0T
u
N
00
v
q
e
N
00
v
3N
00T
v
N
0
u
q
e
N
00
v
_ _
dx: 22
The mass matrix is obtained from the discretization of the inertial forces virtual work, Eq. (16).
Hence, replacing the discrete expressions of the displacements in Eq. (16) leads to
_
t
2
t
1
dT
e
dt
_
t
2
t
1
dq
T
e
M
e
. q
e
dt; 23
ARTICLE IN PRESS
where the elementary mass matrix M
e
is
M
e
_
c
0
rAN
T
u
N
u
rAN
T
v
N
v
rIN
0T
v
N
0
v
dx: 24
The elementary virtual work done by the gravity forces is also discretized using the discrete
expression of the axial displacement yielding
dW
e
dq
T
e
F
e
g
; 25
where the elementary load vector resulting from the gravity eld is
F
e
g

_
c
0
rgAN
T
u
dx: 26
Therefore, the discretized virtual variations may be introduced in Hamiltons principle,
d
_
t
2
t
1
T H
s
W dt 0; 27
which, from Eqs. (20), (23) and (25) and after assembling all elements, yields the following
discretized equations of motion:
M. q K
e
K
g
qq F
g
; 28
where . q denes the acceleration vector. Point forces and damping matrices can be imposed a
posteriori on the system. The element mass matrix (24) may be decomposed in translation M
e
tr
and
rotation M
e
rot
contributions which are
M
e
tr

rAc
420
140 0 0 70 0 0
156 22c 0 54 13c
4c
2
0 13c 3c
2
140 0 0
sym 156 22c
4c
2
_
_
_
_
;
M
e
rot

rI
c
0 0 0 0 0 0
6=5 1=10c 0 6=5 1=10c
2=15c
2
0 1=10c 1=30c
2
0 0 0
sym 6=5 1=10c
2=15c
2
_
_
_
_
: 29
The symmetric linear elastic stiffness matrix K
e
e
corresponds to that of standard EulerBernoulli
beams with axial and bending stiffnesses. K
e
g
states for the geometric stiffness which, as presented
previously, depends on the conguration and thus corresponds to the non-linear terms in the
ARTICLE IN PRESS
equations of motion. These stiffness matrices are
K
e
e

E
c
A 0 0 A 0 0
12I=c
2
6I=c 0 12I=c
2
6I=c
4I 0 6I=c 2I
A 0 0
sym 12I=c
2
6I=c
4I
_
_
_
_
; 30
K
e
g

EA
60c
2
90* u 31 3rc
%
b 12* v c1 3r4c
%
b 5b
2
c 3* v
0 72* u1 3r 6c* u1 3r
0 6c* u1 3r 8c
2
* u1 3r
90* u 31 3rc
%
b 12* v c1 3r4c
%
b 5b
2
c 3* v
0 72* u1 3r 6c* u1 3r
0 6c* u1 3r 2c
2
* u1 3r
_
_
90* u 31 3rc
%
b 12* v c1 3rc
%
b 5b
2
c 3* v
0 72* u1 3r 6c* u1 3r
0 6c* u1 3r 2c
2
* u1 3r
90* u 31 3rc
%
b 12* v c1 3rc
%
b 5b
2
c 3* v
0 72* u1 3r 6c* u1 3r
0 6c* u1 3r 8c
2
* u1 3r
_
_
;
31
where r I=A: Notice that the geometric stiffness matrix K
e
g
depends on the variables * u u
2
u
1
;
* v v
2
v
1
and
%
b b
1
b
2
: Moreover, the bending stiffness 2=151 3rEA* u varies linearly with
the relative axial displacement * u: That is, this stiffness increases when * u is positive and decreases in
the opposite case. This is in agreement with the notion that the beam is stiffer in bending when
under extension and, on the contrary, it is less stiff when under axial compression.
2.5. Accounting for initial deformed conguration
In this section, the non-linear FE model is applied to a typical conguration of drillstring used
for oilwell drilling, which can be represented by a vertical slender cylinder clamped at its top
position, axially sliding at its bottom position and subject to its own weight. Hence, the boundary
conditions considered here are: all degrees of freedom locked at the top position and transversal
displacement locked at the bottom position. Fig. 2 presents the idealized undeformed and
deformed congurations for the drillstring. Three a posteriori forces are now introduced in the FE
ARTICLE IN PRESS
model. In addition to the constant gravity forces, an also constant negative axial point force,
represented by the force vector F
f
; is applied at the beam free end, the drill bit, to simulate the
static reaction force when the bit touches the formation. Notice that the gravity force F
g
and the
reaction force F
f
are time-invariant. Two external forces are also applied to the drillstring, namely
a contact force F
c
due to the beamhole impacts (Fig. 2(c)) and a perturbation force F
p
due to
induced vibrations. Therefore, the equations of motion (28) may be rewritten as
M. q K
e
K
g
qq F
g
F
f
F
c
F
p
: 32
In the practical case, the drillstring is lowered until the drill bit touches the formation. In the event
of continued lowering, the reaction force of the formation F
f
; applied to the drill bit, grows and
the lower part of the drillstring is compressed. In the present work, it is supposed that after this
quasi-static lowering and when the reaction force reaches a given value, the axial displacement of
the drill bit is locked (Fig. 2(b)). Therefore, further motions occur around this initial deformed
conguration, which is the solution of the equation
K
e
K
g
q
s
q
s
F
g
F
f
: 33
This equation holds because F
c
and F
p
are initially zero, that is, there is neither beamhole contact
nor external perturbation.
Since the drillstring is supposed to be initially straight and only axial components of both
gravity F
g
and reaction F
f
forces are non-null, one may expect that only the elements
corresponding to axial displacements in K
g
q
s
will not vanish (see Eq. (31)). Moreover, in this
case, the elements of K
g
q
s
; e.g., 3EA* u=2c
2
; are negligible compared to those of K
e
; e.g., EA=c;
since generally * u{c: Hence, the initial static displacement vector q
s
may be found through the
ARTICLE IN PRESS
s
x, u
z, v
1
2
3
L
L
L
d
(a) (b) (c)
Fig. 2. (a) Undeformed, (b) initial deformed and (c) deformed congurations for a typical drillstring.
solution of the previous algebraic equation with K
g
q
s
0; leading to
q
s
K
1
e
F
g
F
f
: 34
This is the initial deformed conguration of the drillstring, which is represented in Fig. 2(b). Let
us dene then a new displacement vector % q relative to the static one q
s
as
% q q q
s
: 35
Substituting q by % q q
s
in Eq. (32) and accounting for Eq. (34), one can write a new set of
equations of motion in terms of the relative displacement vector % q;
M
.
% q K
e
K
g
% q q
s
% q F
c
F
p
: 36
The axial displacement of the drill bit is then locked into its static value, such that % u
L
0; or
u
L
u
L
s
: The last equation then becomes
M
r
.
% q
r
K
er
K
gr
% q q
s
% q
r
F
cr
F
pr
; 37
where % q
r
is the reduced relative displacement vector. Notice that the reduced mass and stiffness
matrices are those for axial and transversal displacements locked at the bottom position (that is, a
clampedhinged beam, as shown in Fig. 2(c)). The global response is then obtained through
summation of the relative displacements, augmented by the nil relative axial displacement at the
bottom position f% q
T
r
0g
T
; with the static displacements q
s
:
The contact force vector F
c
in Eq. (36) is composed of nodal impact forces F
j
c
that depend on
whether the corresponding FE node j is in contact with the borehole, and is subject to the
following law:
F
j
c
t
0 if jv
j
tjpe;
kfjv
j
tj e signv
j
tg if jv
j
tj > e;
_
38
where e is the distance between the outer surface of the drillstring and the borehole wall and v
j
is
the transversal displacement of the node j: One may notice from Eq. (38) that the impact force is
supposed to be either linear elastic, with spring constant k when there exists beamhole contact, or
zero otherwise. One could also consider a non-linear spring and/or a damping mechanism to
account for inelastic impacts, however in this work simple assumptions for the impact are made to
focus attention on the geometric stiffening effects. The impact force is also supposed to have only
transversal components, that is, the beamhole friction effect is neglected. However, a modal
damping of 5%, relative to the linear elastic model, was considered a posteriori to approximate all
forms of damping of the system.
3. Numerical results
In the three following sections, the dynamics of a typical conguration of drillstring, studied
experimentally by Sotomayor et al. [16], is simulated using the non-linear FE model presented
previously and analyzed with the help of the KL decomposition. The strategy shown in the previous
section to account for the initial deformation state is also used. The geometrical and material
properties of the drillstring were adapted from Ref. [16] and are shown in Table 1. The drillstring is
divided in three different cross-sections, as shown in Fig. 2. The upper portion, composed of drill
ARTICLE IN PRESS
pipes, is normally subject to large traction forces and hence is much less exible in bending than the
rest of the drillstring; it is consequently designed with small outer diameter and thin wall. On the
other hand, the lower portion, composed of drill collars, is highly compressed by the weight of the
upper portion and thus is subject to higher bending effects. That is why it is designed to have a
larger outer diameter and a thicker wall. As for the transition portion, denoted as the heavy weight
drill pipes section, located between the drill pipes and drill collars sections, it is designed to have the
outer diameter of the drill pipes and the inner diameter of the drill collars.
The lower portion of the drillstring is conned inside a borehole of radius R
h
0:156 m and has
two stabilizers located 25 and 50 m away from the drill bit x 1975 and x 1950 m;
respectively). In this work, the stabilizers are accounted for by locking the beam transversal
displacements at their positions. The spring constant considered for the elastic impact forces is
k 10
8
N=m and the clearance e is obtained from the difference between the radii of the borehole
and drillstring sections, e R
h
R
o
: As explained in the previous section, the axial displacement
of the drill bit is locked into its static deformed position. This is done by considering an axial static
reaction force at the bottom position of 200 KN. In addition, a sinusoidal perturbation moment
of 50 sin 2pt KN m is applied through F
p
to the hinged bottom position to simulate bit-formation-
induced lateral vibration. Since, in the present study, the axial displacements are supposed to be
initially at their static values, they can be excited only through coupling with bending vibrations.
3.1. Drillstring dynamics simulation
The evolution of the axial displacement, relative to the static one, at a position 6:25 m from the
drill bit x
m
1993:75 m is shown in Fig. 3, for both linear and non-linear models. One can
observe that the axial displacement is very small for the linear model (Fig. 3(a)). This is due to the
fact that, in this model, axial displacement is not coupled to the transversal displacement, which is
the only one excited by the perturbation force considered. Indeed, these values for the axial
displacement in the linear model are believed to be due to numerical integration errors. On the
other hand, it is clear from the results for the non-linear model (Fig. 3(b)) that the axial
displacement is indeed excited. Although much smaller than the static axial displacement, which is
the reason why it is generally neglected, the effect of the variation of axial displacement relative to
the static one will be evidenced later in this work.
Fig. 4 shows the transversal displacement on the bottom portion of the drill collar
x
m
1993:75 m; evaluated through integration in time of the linear (Fig. 4(a)) and non-linear
(Fig. 4(b)) equations of motion. The bottom portionhole clearance is also shown in the gure, by
ARTICLE IN PRESS
Table 1
Geometrical and material properties of the drillstring
Section 1 Section 2 Section 3
Youngs modulus E (GPa) 210 210 210
Mass density r kg m
3
7850 7850 7850
External radius R
o
(m) 0.064 0.064 0.102
Internal radius R
i
(m) 0.054 0.038 0.038
Length L (m) 1700 100 200
ARTICLE IN PRESS
0 2 4 6 8 10
0.06
0.04
0.02
0
0.02
0.04
0.06
Time (s)
0 2 4 6 8 10
Time (s)
D
e
f
l
e
c
t
i
o
n

(
m
)
0.06
0.04
0.02
0
0.02
0.04
0.06
D
e
f
l
e
c
t
i
o
n

(
m
)
(a) (b)
Fig. 4. Transversal displacement at a position 6:25 m from the drill bit x
m
1993:75 m using (a) linear and (b) non-
linear models.
(a) (b) (c) (d)
Fig. 5. Deformed congurations at integration times: (a) 0, (b) 3.5, (c) 6.5 and (d) 10 s (not in scale, drawing window
for each instant 0:936 m 395 m; ; undeformed FE nodal points; 3; deformed FE nodal points.
0 2 4 6 8 10
0
2
4
6
2
4
6
x 10
9
x 10
4
Time (s) Time (s)
V
a
r
i
a
t
i
o
n

o
f

a
x
i
a
l

d
i
s
p
.

r
e
l
a
t
i
v
e

t
o

s
t
a
t
i
c

o
n
e
V
a
r
i
a
t
i
o
n

o
f

a
x
i
a
l

d
i
s
p
.

r
e
l
a
t
i
v
e

t
o

s
t
a
t
i
c

o
n
e
0 2 4 6 8 10
0
0.5
1
1.5
2
2.5
3
3.5
(a) (b)
Fig. 3. Variation of axial displacement relative to the static one u
m
t=u
m
s
at a position 6:25 m from the drill bit
x
m
1993:75 m using (a) linear and (b) non-linear models.
which one can notice that the drill collar is continuously under impact with the borehole wall. This
is in part due to the excitation at the drill bit and in part due to the extra bending exibility caused
by the compression of the lower portion of the drillstring. To improve clarity, the rst impact
instant is also shown in detail for linear and non-linear models. A comparison between Figs. 4(a)
and (b) shows that the non-linear model leads to a larger number of micro-impacts than the linear
one. This is probably due to the higher exibility in bending of this portion of the drillstring,
caused by the axialbending coupling, only accounted for in the non-linear model.
The deformed congurations of the drillstring bottom portion at four different integration
times, 0, 3.5, 6.5 and 10 s; are represented in Fig. 5. Notice that they are not in scale to facilitate
the observation of the drillstring bottom portion bending vibrations and the impacts with the
borehole at multiple locations of the drill collars section. As expected, most of the bending
vibration is concentrated in the bottom portion of the drillstring, that is, in the heavy weight drill
pipes and drill collars. That is why the upper portion of the drillstring is not shown in Fig. 5. It is
also possible to observe in this gure, the static axial displacement (at t 0 s) compared to the
undeformed conguration, represented by the symbols.
The effect of non-linear coupling between axial and transversal vibrations is specially evidenced
in the following analysis of the reaction forces. Fig. 6 shows the reaction forces at the top and
bottom positions using linear (Fig. 6(a,b)) and non-linear (Fig. 6(c,d)) models. Comparison
between Figs. 6(a) and (c) shows that the linear model, for which axial and transversal vibrations
are uncoupled, is unable to capture the vibrations transmitted to the rotary table (at the top
position, where the drillstring is clamped). Only the static force, due to the drillstring weight, is
well represented by the linear model. On the contrary, the non-linear model is able to capture the
variation of the force at the top position compared to its static value, though it is small, as shown
ARTICLE IN PRESS
0 2 4 6 8 10
7.335
7.34
7.345
7.35
7.355
7.36
7.365
7.37
x 10
5
x 10
5
x 10
5
x 10
5
Time (s)
0 2 4 6 8 10
Time (s)
Time (s)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

t
o
p

p
o
s
i
t
i
o
n

(
N
)
7.335
7.34
7.345
7.35
7.355
7.36
7.365
7.37
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

t
o
p

p
o
s
i
t
i
o
n

(
N
)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

b
o
t
t
o
m

p
o
s
i
t
i
o
n

(
N
)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

b
o
t
t
o
m

p
o
s
i
t
i
o
n

(
N
)
0 2 4 6 8 10
Time (s)
0 2 4 6 8 10
2
1.5
1
0.5
0
0.5
1
3.2
3
2.8
2.6
2.4
2.2
2
1.8
(a) (b)
(c) (d)
Fig. 6. Reaction forces at the top and bottom positions using (a,b) linear and (c,d) non-linear models.
in Fig. 6(c). However, the results for the reaction forces at the bottom position using the linear
and non-linear models present larger and more important differences. From Fig. 6(b), one
concludes that the linear model allows the reaction force at the bottom position to be positive.
This means that the drill bit is pulled from the bottom position, making it possible for the drill bit
to loose contact with the formation for the larger values of deection. This is clearly prevented
here, since the axial displacement of the drill bit is locked at its static value. Nevertheless, the
decrease in the force magnitude leads to a worse drilling performance. On the other hand, analysis
of Fig. 6(d) shows that this behavior is not observed, for the present case, when using the non-
linear model. It is thus only a consequence of neglecting the coupling between axial and bending
strains. In fact, the use of the non-linear model leads to an increase in the force magnitude due to
lateral vibration. This may be explained by the fact that the transversal displacement induces an
increase in the axial displacement, leading to an augmentation of the compression in the bottom
portion of the drill collars section, instead of inducing its traction. Notice that this effect of
counterbalance between compression and traction may vary according to the parameters of the
drillstring and to the static value of reaction force.
Another interesting analysis stems from the evaluation of the axial position in the drillstring that
separates its compressed (lower) and stretched (upper) portions. This position is quite important in
practice, since it should always lie inside the stiffer region of the drillstring, generally in the drill collars
section, to prevent buckling. Therefore, Fig. 7 shows the evolution of this position in time due to the
perturbation-induced vibrations, using linear and non-linear models. Due to the parameters chosen
here, the static value of this position is somewhere near the center of the drill collars section. For the
linear model, this position oscillates between its static value and 10 m below it (Fig. 7(a)). This means
that larger portions of the drill collars section become stretched. For the non-linear model, however,
this position is oscillating about its static value in a range of 15 m (Fig. 7(b)).
3.2. KarhunenLo" eve decomposition of the response
In this section, the direct method of KarhunenLo" eve decomposition [11,17] is applied to the
drillstring dynamics. For this purpose, the time response qt is subtracted from its mean Eqt to
obtain the deviation q
d
t qt Eqt of the time response. Consequently, the new vector q
d
t
has elements with zero mean. As the time response qt; and thus also q
d
t; results from time
integration of the discretized equations of motion (37), it is in fact sampled in M instants of time
t
1
; t
2
; y; t
M
; chosen in the time integration algorithm. Hence, the time response qt may be
ARTICLE IN PRESS
0 2 4 6 8 10
1906
1908
1910
1912
1914
1916
1918
Time (s)
0 2 4 6 8 10
Time (s)
A
x
i
a
l

p
o
s
i
t
i
o
n

w
h
e
r
e

a
x
i
a
l

s
t
r
a
i
n

c
h
a
n
g
e
s

s
i
g
n

(
m
)
A
x
i
a
l

p
o
s
i
t
i
o
n

w
h
e
r
e

a
x
i
a
l

s
t
r
a
i
n

c
h
a
n
g
e
s

s
i
g
n

(
m
)
1890
1895
1900
1905
1910
1915
(a) (b)
Fig. 7. Position where the axial strain changes sign using (a) linear and (b) non-linear models.
written as a sampling matrix of dimension M N;
q
q
1
t
1
q
2
t
1
y q
N
t
1
q
1
t
2
q
2
t
2
y q
N
t
2
^ ^ & ^
q
1
t
M
q
2
t
M
y q
N
t
M
_
_
_
; 39
where each column represents the time response of a given degree of freedom from the FE mesh,
with N being the total number of degrees of freedom or the dimension of qt: Alternatively, each
row represents the spatial distribution of the response at a given time instant, that is, a point in the
N-dimensional phase space.
Using the ergodicity assumption, the mean value Eqt may be obtained by the time
average of q; that is
M
i
qt
i
=M: Hence, the deviation q
d
t with respect to the mean may also be
written as a sampling matrix, obtained by subtracting from each line of q the time average of all
rows,
q
d
q q
E
; where q
E

1
M
M
i1
q
1
t
i

M
i1
q
2
t
i
y

M
i1
q
N
t
i
^ ^ & ^
M
i1
q
1
t
i

M
i1
q
2
t
i
y

M
i1
q
N
t
i
_
_
_: 40
The spatial autocorrelation matrix is then written in terms of the zero-mean time response
sampling matrix as
R
1
M
q
T
d
q
d
; 41
where R is, by denition, symmetric and positive semi-denite. Hence, its eigenvectors form an
orthogonal basis and its eigenvalues are non-negative. In this case, the eigenvectors C
j
are the
coherent structures or proper orthogonal modes (POMs) and the corresponding eigenvalues l
j
; or
proper orthogonal values (POVs), give a measure of the mean energy contained in each mode.
These are dened as
RC LC; 42
with
C C
1
C
2
? C
N
and L diagl
1
; l
2
; y; l
N
:
Notice that the autocorrelation matrix R has dimension N N: Hence, its dimension, and the
number of proper orthogonal modes and values, depend only on the number of degrees of
freedom and not on the time instants used for the sampling of the time response. The procedure is
summarized in Fig. 8, which shows the direct method implementation algorithm of the
KarhunenLo" eve decomposition used here.
The time response sampling matrix q may then be reconstructed, in a truncated basis, by
expressing q in terms of its time average matrix q
E
and a reduced number K of evaluated POMs as
q
K
j1
A
j
C
T
j
q
E
; 43
ARTICLE IN PRESS
where the vectors of time coefcients A
j
are easily found by projecting the time response onto each
POM C
j
; that is,
A
j
qC
j
: 44
Notice that, by denition, the POVs also respect the following relation with the time coefcients
l
j

1
M
M
i1
A
T
j
A
j
: 45
Next, this procedure is applied to the drillstring dynamics response resulting from the
integration of the equations of motion (37). Moreover, one expects to obtain additional
information through analysis of the POMs and POVs. These are also used to construct an optimal
basis, with minimum dimension, that allows us to project the equations of motion and obtain a
reduced order model, well representing the main features of the response.
ARTICLE IN PRESS
1200
1300
1400
1500
1600
1700
1800
1900
2000
(a) (b) (c) (d) (e)
Fig. 9. Deection in the rst ve proper orthogonal modes evaluated from the response. (a) l
1
0:55; (b) l
2
0:34; (c)
l
3
0:03; (d) l
4
0:02; (e) l
5
0:05:
Time
average
Spatial
correlation
Eigenvectors
Eigenvalues
POMs
Energy
R
11
R
12
... R
1N
R
22
R
2N
R
NN
...
.
.
.
.
.
.
q
Data
q
d
1
2
M
+
_
1
2
N
E[q]
sym
1
2
M
Fig. 8. Direct method implementation algorithm of KL decomposition.
Fig. 9 shows the transversal displacement in the rst ve POMs evaluated from the drillstring
dynamics response q; zoomed in the drillstring bottom portion, and the corresponding POVs l
j
: It
is worthwhile noting that the POVs where normalized so their sum is unitary. One may observe
then that most of the energy (89%) is contained in the rst two POMs and that only the bottom
portion of the drillstring presents transversal deections. This is clearly due to the fact that the
upper portion of the drillstring is under traction and thus is much stiffer in bending than the
bottom portion that is under compression.
Although the rst ve POMs are responsible for 97% of the response, one should not be able to
obtain a perfect reconstruction by considering only these ve modes. Indeed, it is generally
recommended to consider a number of modes sufcient to sum 99.9% of the response energy. In
order to analyze the quality of the reconstruction using the obtained POMs, Fig. 10 shows the
reconstruction of the transversal displacement response using the rst ve, nine, 15 and 30 proper
orthogonal modes. The percentage of the energy contained in these rst POMs are respectively,
96.6567%, 99.1286%, 99.8613% and 99.9918%. Thus, one should need a little more than 15
modes to reconstruct the response, although even in this case there would be a great reduction of
the model dimension, since the full FE model contains 87 degrees of freedom. Nevertheless, one
could also use other measures to quantify the quality of the reconstruction. Indeed, from Fig. 10,
one may observe that the overall behavior of the transversal displacement is captured even with
only ve POMs, the effect of bottom-hole impact being the main source of error. Another
measure of the reconstruction quality was then considered, consisting of the time average of the
relative error modulus between the reconstructed response v
R
and the response of the full model
v
F
; e

M
j1
j1 v
j
F
=v
j
R
j=M: Using this measure, one may see that using only the rst ve POMs
leads to a reconstruction within 14% response error. This error drops to 8%, 3% and less than
1% when using the rst nine, 15 and 30 POMs respectively. In the zoomed window present in Fig.
10, one may observe that there is a clear improvement of the reconstruction of the response, near
the impact region, when the number of POMs is increased. In fact, when using the rst 30 POMs
one may capture precisely all the micro-impacts observed in the response. Nevertheless, although
using only the rst 15 POMs leads to small differences in the response inside the impact region,
ARTICLE IN PRESS
0 2 4 6 8 10
0.06
0.04
0.02
0
0.02
0.04
0.06
Time (s)
D
e
f
l
e
c
t
i
o
n

(
m
)
Fig. 10. Reconstruction of the transversal displacement response using the rst ve, nine, 15 and 30 proper orthogonal
modes: full; , ve;--, nine; - -, 15;--, 30.
one may observe that the main micro-impacts effects are accounted for since this reconstructed
response follows almost exactly the full model response just after the impact. This means that this
reconstruction, within a 3% error, may be reasonably accurate to consider only the rst 15 POMs
in the reduced order model.
3.3. KL reduced order model results
In this section, the equations of motion (36) are projected onto a reduced basis formed by the
rst 15 POMs evaluated previously. This is done through the substitution of % qt in Eq. (36) by its
approximation in terms of the known truncated POMs matrix
*
C and mean value Eqt; given as
% qt qt q
s
E
*
Cat Eqt q
s
; 46
followed by the pre-multiplication of the resulting equation by C
T
; leading to the reduced order
equations of motion
M
C
. a K
eC
K
gC
aa F
cC
F
pC
F
kC
; 47
where
M
C

*
C
T
M
*
C; K
eC

*
C
T
K
e
*
C; K
gC
a
*
C
T
K
g
q
*
C;
F
cC

*
C
T
F
c
; F
pC

*
C
T
F
p
; F
kC

*
C
T
K
e
K
g
qEqt q
s
:
Notice that the reduced matrices and vectors of this new system have a reduced dimension of
15, which is the number of POMs considered. This system can then be solved numerically for the
vector of POMs coefcients at; subject to the initial conditions a0 and a0; which are meant to
approximate the static initial deformed conguration % q0

% q0 0; or q0 q
s
and q0 0:
The reduced order system was then used to simulate parametric disturbances in the impact force
spring constant and sinusoidal perturbation moment frequency values. These were varied in the
ranges k 10
7
; 5 10
8
N=m and o 4; 8 rad=s; respectively. Notice that the POMs were
evaluated from the time response of the full order system, with given values for these two
parameters k 10
8
N=m and o 2p rad=s: Hence, the response of the KL reduced order system,
with different values for these parameters, is expected to be only an approximation of the
correct full order system response, since the KL modes change when the system parameters or
the forcing conditions are changed. Nevertheless, the KL modes will be signicantly different only
when the parameters variation is large. Hence, hopefully, the original KL modes will be able to
approximate the new system response.
Variation of the excitation frequency has a large effect on the time response, mainly because it
induces a change in the duration of impact. This effect can be observed in Fig. 11(a), which shows
the deection at 6:25 m from the drill bit x 1993:75 m: One can see that there is a good
agreement between the deections evaluated with full order and KL reduced models. However,
one can also see in Fig. 11(b) that the reduced model approximation for the axial displacement is
not as good. This may be explained by the fact axial displacements, relative to the static ones, are
small compared to transversal displacements and thus the rst POMs are meant to represent
mainly the latter.
ARTICLE IN PRESS
Unfortunately, axial displacements are quite important to the evaluation of the reaction forces
at the top and bottom positions. That is why these reaction forces are not well represented by the
KL reduced order model, with only 15 POMs, as shown in Figs. 12(a) and (b).
The following gures show the effect of varying the impact force spring constant on the time
response. The decrease of this parameter leads mainly to a higher penetration of the drillstring in
ARTICLE IN PRESS
0 0.5 1 1.5 2
7.34
7.345
7.35
7.355
7.36
7.365
x 10
5
x 10
5
Time (s) Time (s)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

t
o
p

p
o
s
i
t
i
o
n

(
N
)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

b
o
t
t
o
m

p
o
s
i
t
i
o
n

(
N
)
0 0.5 1 1.5 2
4
3.5
3
2.5
2
1.5
1
0.5
(a) (b)
Fig. 12. Reaction forces at (a) the top and (b) bottom positions evaluated with full order model (light lines) and KL
reduced model (dark lines) for three values of perturbation moment frequency: , o
1
4 rad=s; - -, o
2
6 rad=s; --,
o
3
8 rad=s:
0 0.5 1 1.5 2
0.06
0.04
0.02
0
0.02
0.04
0.06
Time (s) Time (s)
D
e
f
l
e
c
t
i
o
n

(
m
)
0 0.5 1 1.5 2
1.0872
1.0873
1.0873
1.0874
1.0874
1.0875
1.0875
A
x
i
a
l

D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
(a) (b)
Fig. 13. (a) Transversal and (b) axial displacements evaluated with full order model (light lines) and KL reduced model
(dark lines) for three values of impact force spring constant: , k
1
10
7
N=m; - -, k
2
10
8
N=m; --, k
3

5 10
8
N=m:
0 0.5 1 1.5 2
0.06
0.04
0.02
0
0.02
0.04
0.06
Time (s) Time (s)
D
e
f
l
e
c
t
i
o
n

(
m
)
0 0.5 1 1.5 2
1.0872
1.0873
1.0873
1.0874
1.0874
1.0875
1.0875
A
x
i
a
l

D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
(a) (b)
Fig. 11. (a) Transversal and (b) axial displacements evaluated with full order model (light lines) and KL reduced model
(dark lines) for three values of perturbation moment frequency: , o
1
4 rad=s; - -, o
2
6 rad=s; --, o
3
8 rad=s:
the borehole wall. This effect may be observed in Fig. 13(a), which shows also a good agreement
between the deections evaluated by the full order and reduced order models. However, as also
observed for the excitation frequency variation, here the reduced order model is unable to capture
the behavior of axial displacements.
As noted also in the previous case, the KL reduced order model would require a larger number
of POMs to yield a good approximation to the axial displacements and, thus, to the reaction
forces at the top and bottom positions. This conclusion also holds for the variation of the impact
force spring constant, as shown in Figs. 14(a) and (b).
4. Conclusions
The non-linear oscillations of a non-rotating drillstring, represented here by a vertical slender
cylinder, clamped in its upper extreme, pinned in its lower one and constrained inside an outer
cylinder in its lower portion, were studied in the present work. The drillstring was supposed as
being subject to distributed axial loads, due to its own weight, leading to geometric softening of its
lower portion and, thus, to a large number of vibroimpacts with the borehole wall. It was shown
that one should account for the axial displacement dynamics, using non-linear strain
displacement relations, since the coupling of axialbending dynamics may be very important in
the dynamical behavior of general slender beams such as a drillstring. In particular, the micro-
impacts, accompanying the bottom portionhole impacts and mainly due to the beam
compressive softening, are well represented only when using a non-linear axialbending coupling.
Moreover, the use of standard linear beam models may yield false predictions of the reaction
forces at the top and bottom positions of the drillstring, that is the forces at the rotary table and at
the formation.
Furthermore, the KarhunenLo" eve decomposition was applied to the simulated dynamics of
this particular system in order to obtain additional information on the system through analysis of
the POMs and POVs and also to construct an optimal reduced order model. The results for the
case treated here have shown that at least 15 POMs are required to reconstruct the dynamics of
the impacting drillstring under a 3% error margin. These results are encouraging if one compares
the dimension of the reduction basis (15 modes) with that of the original FE model (87 degrees of
ARTICLE IN PRESS
0 0.5 1 1.5 2
7.34
7.345
7.35
7.355
7.36
7.365
x 10
5
Time (s) Time (s)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

t
o
p

p
o
s
i
t
i
o
n

(
N
)
R
e
a
c
t
i
o
n

f
o
r
c
e

a
t

b
o
t
t
o
m

p
o
s
i
t
i
o
n

(
N
)
0 0.5 1 1.5 2
4
3.5
3
2.5
2
1.5
1
0.5
x 10
5
(a) (b)
Fig. 14. Reaction forces at (a) the top and (b) bottom positions evaluated with full order model (light lines) and KL
reduced model (dark lines) for three values of impact force spring constant: , k
1
10
7
N=m; - -, k
2
10
8
N=m; --,
k
3
5 10
8
N=m:
freedom). A reduced order model was then developed, through projection of the FE equations of
motion onto the rst 15 POMs, and applied to the simulation of the dynamical response of the
drillstring under variation of the impact force spring constant and excitation frequency. The
results show that the KL reduced order model is capable of well representing the drillstring
deections in all cases considered. However, in order to represent the axial displacements as well,
one should consider using more POMs in the projection or magnifying the axial displacements
before evaluating the POMs. Future works are being directed to the improvement of these results
through alternative techniques for evaluating and selecting the POMs. In addition, a three-
dimensional analysis of the drillstring is also being considered through the use of a geometrically
exact model proposed by Rochinha and Sampaio [18] together with a singularity-free rotation
representation [19].
Acknowledgements
The authors gratefully acknowledge the nancial support of Funda@* ao Carlos Chagas Filho
de Amparo " a Pesquisa do Estado do Rio de Janeiro (FAPERJ), through grants nos. 172.038/00,
151.188/00 and 150.687/00.
References
[1] I. Sharf, Geometric stiffening in multibody dynamics formulations, Journal of Guidance, Control, and Dynamics 18
(4) (1995) 882890.
[2] J. Simo, L. Vu-Quoc, The role of non-linear theories in transient dynamic analysis of exible structures, Journal of
Sound and Vibration 119 (3) (1987) 487508.
[3] T.R. Kane, R.R. Ryan, A.K. Banerjee, Dynamics of a cantilever beam attached to a moving base, Journal
of Guidance, Control, and Dynamics 10 (2) (1987) 139151.
[4] A.K. Banerjee, J.M. Dickens, Dynamics of an arbitrary exible body in large rotation and translation, Journal of
Guidance, Control, and Dynamics 13 (2) (1990) 221227.
[5] O. Wallrapp, R. Schwertassek, Representation of geometric stiffening in multibody system simulation,
International Journal for Numerical Methods in Engineering 32 (8) (1991) 18331850.
[6] J. Urruzola, J.T. Celigueta, J.G. De Jalon, Generalization of foreshortening through new reduced geometrically
nonlinear structural formulation, Journal of Guidance, Control, and Dynamics 23 (4) (2000) 673682.
[7] M.A. Trindade, R. Sampaio, Dynamics of beams undergoing large rotations accounting for arbitrary axial
deformation, Journal of Guidance, Control, and Dynamics 25 (4) (2002) 634643.
[8] L. Sirovich, Turbulence and the dynamics of coherent structures, Part I: coherent structures, Quarterly of Applied
Mathematics 45 (3) (1987) 561571.
[9] M.F.A. Azeez, A.F. Vakakis, Proper orthogonal decomposition (POD) of a class of vibroimpact oscillations,
Journal of Sound and Vibration 240 (5) (2001) 859889.
[10] A. Steindl, H. Troger, Methods for dimension reduction and their application in nonlinear dynamics, International
Journal of Solids and Structures 38 (1013) (2001) 21312147.
[11] C. Wolter, R. Sampaio, KarhunenLo" eve bases: applications to solid mechanics, in: Proceedings of the
First Brazilian Workshop in Applications of Dynamics and Control, ABCM/SBMAC; S* ao Carlos, Brazil, 2001,
pp. 129172.
[12] M.I. Friswell, J.E.T. Penny, S.D. Garvey, The application of the IRS and balanced realization methods to obtain
reduced models of structures with local non-linearities, Journal of Sound and Vibration 196 (4) (1996) 453468.
ARTICLE IN PRESS
[13] A.S. Yigit, A.P. Christoforou, Coupled axial and transverse vibrations of oilwell drillstrings, Journal of Sound and
Vibration 195 (4) (1996) 617627.
[14] A.S. Yigit, A.P. Christoforou, Coupled torsional and bending vibrations of drillstrings subject to impact with
friction, Journal of Sound and Vibration 215 (1) (1998) 167181.
[15] R.W. Tucker, C. Wang, An integrated model for drill-string dynamics, Journal of Sound and Vibration 224 (1)
(1999) 123165.
[16] G.P.G. Sotomayor, J.C. Pl " acido, J.C. Cunha, Drill string vibration: how to identify and suppress, in: Proceedings
of Fifth Latin American and Caribbean Petroleum Engineering Conference and Exhibition, Rio de Janeiro, Brazil,
1997, p. SPE#39002.
[17] C. Wolter, M.A. Trindade, R. Sampaio, Obtaining mode shapes through the KarhunenLo" eve expansion for
distributed-parameter linear systems, Shock and Vibration 9 (45) (2002) 177192.
[18] F.A. Rochinha, R. Sampaio, A consistent approach to treat the dynamics of exible systems, Journal of the
Brazilian Society of Mechanical Sciences 19 (1997) 228241.
[19] M.A. Trindade, R. Sampaio, On the numerical integration of rigid body nonlinear dynamics in presence of
parameters singularities, Journal of the Brazilian Society of Mechanical Sciences 23 (1) (2001) 4962.
ARTICLE IN PRESS

Trindade Wolter Sampaio 2005 JSV

Uploaded by

Copyright:

Available Formats

Trindade Wolter Sampaio 2005 JSV

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Trindade Wolter Sampaio 2005 JSV

Uploaded by

Copyright:

Available Formats

JOURNAL OF

You might also like