Computers and Structures 80 (2002) 957–965

www.elsevier.com/locate/compstruc

Evaluation of non-linear normal modes

for finite-element models
C.E.N. Mazzilli *, O.G.P. Baracho Neto
Departamento de Engenharia de Estruturas e Fundacßo~es, Escola Polit
ecnica da Universidade de S~
ao Paulo, Caixa Postal 61548,
S~
ao Paulo 05424-970, Brazil
Received 21 August 2001; accepted 18 January 2002

Abstract
In this paper, an alternative technique for the evaluation of non-linear normal modes is presented and applied to
finite-element models. It is based upon the method of multiple scales, so that non-linear normal modes are evaluated as
asymptotic expansions starting with the linear damped vibration modes. Non-linear normal modes are characterized
both by the time response of all generalised coordinates and by explicit non-linear relationships between them and the
modal variables. For the sake of an example, a non-conservative model of a cantilever beam, discretised with Bernoulli–
Euler finite elements, is analysed.
2002 Elsevier Science Ltd. All rights reserved.

Keywords: Non-linear normal modes; Non-linear dynamics; Finite-element models; Bernoulli–Euler beam theory; Method of multiple
scales; Non-proportional damping

1. Introduction defined. Yet, such a definition was not general enough to

Vibration modes play an important role in linear
dynamics. Hence, it comes as no surprise the desire to
extend to non-linear systems all the benefits of modal
analysis.
Vakakis and his co-workers [1] point out that
Lyapunov in 1907 had already tried to obtain modal
solutions for non-linear systems. But it was Rosenberg
[2] who first proposed a definition for non-linear normal
modes. According to Rosenberg, non-linear normal
modes are synchronous motions for which there is a
fixed relationship among the generalised coordinates,
that is, all generalised coordinates execute equiperiodic
motions, pass through the equilibrium position at the
same time and reach their maximum displacements si-
multaneously. Besides, if a value is attributed to a gen-
eralised coordinate, all the other ones will be uniquely
*
Corresponding author. Tel.: +55-11-3091-5232; fax: +55-
11-3091-5181.
E-mail address: cenmazzi@usp.br (C.E.N. Mazzilli).
include even the case of linear systems with non-pro-
portional damping, for which the normal modes behave
as non-standing waves, that is to say, they are non-
synchronous.
In the 1990s Shaw and Pierre [3] proposed a redefi-
niton of non-linear normal modes in which not only the
generalised displacements but also the velocities should
be taken into account. According to them, a non-linear
normal mode is a free-vibration motion which takes
place in a two-dimensional invariant manifold embed-
ded in the phase space of the system. This manifold must
contain the equilibrium point and be tangent to the
corresponding eigenspace of the linearised system. The
advantage of such a definition is that it incorporates
the previous one as a particular case and is suitable to
both conservative and non-conservative systems. They
also proposed a procedure to evaluate non-linear nor-
mal modes, which is here referred to as the invariant
manifold technique.
Nayfeh and co-workers [4,5] also contributed to the
subject. They studied the non-linear normal mode of
undamped single degree of freedom systems using the

0045-7949/02/$ - see front matter
2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 4 9 ( 0 2 ) 0 0 0 6 1 - 5
958 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965

method of multiple scales and showed the equivalence spectively for the mass, equivalent damping and stiffness
between their results and those of the invariant manifold coefficients introducing cubic non-linearities into (1).
technique for this simple case of a conservative system. Systems ruled by a linear-elastic constitutive law and
It has been Soares [6] who first succeeded in evalu- subjected to finite displacements (non-linear kinematics),
ating non-linear normal modes of full finite-element though small strains, fit into the class of problems
models. In his work, Soares applied the invariant man- modelled by (1) and (2)–(4). That is the case of planar
ifold technique to plane frame structures, discretised reticulated structures, as those studied in [7].
with Bernoulli–Euler elements [7]. In order to obtain the temporal description of all
In the present paper, a new approach to the problem generalised coordinates when the system is vibrating in a
of evaluating non-linear normal modes of discrete sys- particular non-linear normal mode, one can use the
tems, including finite-element models, is proposed. It method of multiple scales. Following the usual proce-
adopts Shaw and Pierre’s definition of non-linear nor- dure [8], the generalised coordinates and the time de-
mal modes, but relies upon the method of multiple rivatives are written as asymptotic expansions of a small
scales, rather than the invariant manifold technique, to dimensionless perturbation parameter e, that is,
calculate both the time response and the non-linear re-
pi ðtÞ ¼ epi1 ðT0 ; T1 ; T2 ; . . . Þ þ e2 pi2 ðT0 ; T1 ; T2 ; . . . Þ
lationships among the generalised coordinates of multi-
degree of freedom non-conservative systems. þ e3 pi3 ðT0 ; T1 ; T2 ; . . . Þ þ    ; ð5Þ
By the way, the very application of the method of
multiple scales to full finite-element models seems to be a Ti ¼ ei t; ð6Þ
pioneering accomplishment.
d
¼ D0 þ eD1 þ e2 D2 þ    ; ð7Þ
dt
2. Non-linear normal modes: temporal description of the
generalised coordinates
d2

2
¼ D20 þ 2eD0 D1 þ e2 2D0 D2 þ D21 þ    ; ð8Þ
Non-linear free-vibration equations of a fairly gen- dt
eral class of elastic finite-element models can be written
as: dn
Dni ¼ : ð9Þ
dTin
Mrs €ps þ Drs p_ s þ U;r ¼ 0; ð1Þ
Taking (2)–(9) into (1) and collecting terms of the same
where Mrs , Drs and U;r stand for the generic elements of order of e, it is possible to arrive at differential equations
the mass matrix, equivalent damping matrix and elastic whose solutions and solvability conditions allow for the
force vector, respectively; r and s vary from 1 to n characterization of the non-linear normal modes.
(number of degrees of freedom of the model); summation
over the repeated index s from 1 to n is implied. The
2.1. Equations of order e
notation U;r is used to denote the partial derivative of the
strain energy U with respect to the generalised coordinate
The equations of order e correspond to the linear
pr , evaluated at the equilibrium position. It is further
damped system. They read
assumed that Mrs , Drs and U;r depend on the generalised
0
coordinates pi , i ¼ 1; 2; . . . ; n, and Drs may also depend Mrs D20 ps1 þ 0 Drs D0 ps1 þ 0 Krs ps1 ¼ 0: ð10Þ
on the velocities p_ i , i ¼ 1; 2; . . . ; n, as indicated:
The solution of (10) is sought in the form
Mrs ¼ 0 Mrs þ 1 Mrsi pi þ 2 Mrsij pi pj ; ð2Þ
ps1 ¼ As1 eWT0 ; ð11Þ
0 1 2
Drs ¼ Drs þ Dirs p_ i þ Dijrs p_ i pj ; ð3Þ where As1 ¼ As1 ðT1 ; T2 ; . . .Þ and W may be complex
numbers. It follows from (11) and (10):
U;r ¼ 0 Krs ps þ 1 Krsi pi ps þ 2 Krsij pi pj ps : ð4Þ
20 
W Mrs þ W 0 Drs þ 0 Krs As1 ¼ 0 8r; s ¼ 1–n; ð12Þ
In (2)–(4) summation over the repeated indices i and j, which characterizes a damped eigenvalue problem. The
from 1 to n, is also implied; 0 Mrs , 0 Drs and 0 Krs are the solvability condition, in this case, coincides with the
linear-theory generic elements of the mass, equivalent characteristic equation of the problem:
damping and stiffness matrices, respectively; 1 Mrs , 1 Drs

and 1 Krs stand respectively for the mass, equivalent det 0 S ¼ 0; ð13Þ
damping and stiffness coefficients introducing quadratic
non-linearities into (1); 2 Mrs , 2 Drs and 2 Krs stand re- where 0 S ¼ W2 0 M þ W 0 D þ 0 K.
 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965 959

The roots of (13) are written in the form Taking (19) and (21) in (20) and collecting terms in
eWT0 :
Wu ¼ au þ ibu ; u ¼ 1–n; ð14Þ

20 
where au is minus the product of the damping ratio by W Mrs þ W 0 Drs þ 0 Krs As2


the undamped linear frequency of mode u, and jbu j is the ¼ 2W 0 Mrs þ 0 Drs D1 As1 : ð22Þ
damped linear frequency of mode u.
Choosing a non-zero Auv1 (for a certain degree of One must bear in mind that in (13) it was imposed that
freedom v and mode u)—see Eq. (11)—as modal coor- the matrix 0 S should be singular. So, for the system (22)
dinate, it is possible to normalise the amplitudes of the to have solution, it is required that the determinant of
other degrees of freedom with respect to this one: the matrix 1 S—which results from the substitution of
one of the columns of 0 S by the vector of independent
Aus1 terms of (22)—be null. This is the solvability condition
¼ /us ) Aus1 ¼ /us Auv1
Auv1 to be imposed on terms of order e2 :

 s ¼ 1–n;

¼ cus þ idus Auv1 ; ð15Þ det 1 S ¼ 0: ð23Þ
no sum in u:

Of course, /uv ¼ 1. Note that /us represents a term of the One arrives at a differential equation of the type:
complex eigenvector.
ð f þ igÞD1 A1 ¼ 0; ð24Þ
It is proposed from this moment on a simplification
in the notation with the elimination of the indexes u and
v, which stand for the selected mode and the degree of where f and g are real numbers. Eq. (24) is satisfied if
freedom related to the modal variable, respectively. This D1 A1 ¼ 0, or A1 ¼ A1 ðT2 Þ.
simplification implies no restriction to the procedure, Note that the term in eWT0 in (21) does not add any
because it can be applied to any mode, provided any new feature to the solution, since there is already a term
non-zero generalised coordinate is chosen as the modal of this type in ps1 (see (19)). It will therefore be disre-
variable. Let’s write then garded from this point on.
Collecting now terms in e2aT0 , one obtains:
Wu ¼ W ¼ a þ ib; ð16Þ
20 
4a Mrs þ 2a 0 Drs þ 0 Krs Es2
Auv1 ¼ A1 ; ð17Þ

¼ W2 1 Mrsi þ WW 1 Dirs þ 1 Krsi A i1 As1

21 i 
Aus1 ¼ As1 ¼ /s A1 ¼ ðcs þ ids ÞA1 : ð18Þ ¼ W Mrs þ WW 1 Dirs þ 1 Krsi /i /s A  1 A1 : ð25Þ

Since the solution of (10) must be real:

So,
ps1 ¼ As1 e WT0  s1 eWT0 ;
þA ð19Þ  1 A1 ;
 1 A1 ¼ ðrs þ iss ÞA
Es2 ¼ gs A ð26Þ

 s1 and W represent the complex conjugates of As1

where A where rs and ss are real numbers, provided the matrix of
and W, respectively. coefficients is non-singular. Collecting finally terms in
e2WT0 , it is seen that
2.2. Equations of order e2
20 
4W Mrs þ 2W 0 Drs þ 0 Krs Bs2
After collection of terms of order e2 in (1), one gets:

¼ W2 1 Mrsi þ W2 1 Dirs þ 1 Krsi Ai1 As1
0
Mrs D20 ps2 þ 0 Drs D0 ps2 þ 0 Krs ps2

¼ W2 1 Mrsi þ W2 1 Dirs þ 1 Krsi /i /s ðA1 Þ2 : ð27Þ
¼ 2 0 Mrs D0 D1 ps1 0 Drs D1 ps1 1 Mrsi pi1 D20 ps1
1 Dirs ðD0 pi1 D0 ps1 Þ 1 Krsi pi1 ps1 : ð20Þ So,

Terms on the right-hand side of (20) depend upon the
solution of order e. The solution of (20) is sought in the
form
ps2 ¼ As2 eWT0 þ Bs2 e2WT0 þ Es2 e2aT0 þ c:c:;
where c.c. stands for complex conjugate.
Bs2 ¼ qs ðA1 Þ2 ¼ ðs þ ins ÞðA1 Þ2 ; ð28Þ
where s and ns are real numbers, provided the matrix of
coefficients is non-singular. In the case of a system with
ð21Þ one eigenvalue exactly equal to twice another one, such
as the case of internal resonance 2:1 in undamped
models, the solution of (27) cannot be written as in (28).
960 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965

2.3. Equations of order e3 Eq. (32) may be written as:

After collection of terms of order e3 in (1), one gets:


W2 0 Mrs þ W 0 Drs þ 0 Krs As3
0
Mrs D20 ps3 0
þ Drs D0 ps3 þ Krs ps3 0

   1 ðA1 Þ2 e2aT0 ;
¼ ðxr þ iyr ÞD2 A1 þ ðwr þ izr ÞA ð33Þ
¼ Mrs 2D0 D1 ps2 þ 2D0 D2 þ D21 ps1
0

0 Drs ½D1 ps2 þ D2 ps1  1 Mrsi pi1 ½D0 ps2
 where xr , yr , wr and zr are real numbers.
þ 2D0 D1 ps1  þ pi2 D20 ps1 1 Dirs fD0 pi1 ½D0 ps2
Applying the solvability condition, which is analo-
þ D1 ps1  þ D1 pi1 D0 ps1 þ D0 pi2 D0 ps1 g gous to that discussed for the terms of order e2 , one
 
1 Krsi ½pi1 ps2 þ pi2 ps1  2 Mrsij pi1 pj1 D20 ps1 arrives at the following differential equation:
   
2 Dijrs D0 pi1 pj1 D0 ps1 2 Krsij pi1 pj1 ps1 : ð29Þ

 1 ðA1 Þ2 exp 2a
Having in mind that D1 A1 ¼ 0 and, therefore, that qÞD2 A1 þ ð^
ðp^ þ i^ r þ i^sÞA T2 ¼ 0; ð34Þ
e2
D1 ps1 ¼ 0 and D1 ps2 ¼ 0, for i ¼ 1–n, Eq. (29) reads
0
Mrs D20 ps3 þ 0 Drs D0 ps3 þ 0 Krs ps3
 where p^, q^, r^ and ^s are real numbers.
¼ 0 Mrs ½2D0 D2 ps1  0 Drs ½D2 ps1  1 Mrsi pi1 D0 ps2 Expressing the amplitude A1 in polar coordinates,

þ pi2 D20 ps1 1 Dirs ½D0 pi1 D0 ps2 þ D0 pi2 D0 ps1  one obtains:
 
1 Krsi ½pi1 ps2 þ pi2 ps1  2 Mrsij pi1 pj1 D20 ps1
   
2 Dijrs D0 pi1 pj1 D0 ps1 2 Krsij pi1 pj1 ps1 : ð30Þ 1
A1 ¼ aeih ; ð35Þ
2
Therefore, the solution of (30) should be sought in the
form
with a ¼ aðT2 Þ and h ¼ hðT2 Þ real functions.
ps3 ¼ As3 eWT0 þ Cs3 e3WT0 þ c:c: ð31Þ Substituting (35) in (34) and solving the corre-
sponding differential equations, one arrives at
After substitution of (19), (26), (28) and (31) in (29) and
collection of terms in eWT0 , it comes out

20  a0
W Mrs þ W 0 Drs þ 0 Krs As3 aðtÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
1 þ a ðea0 Þ2 #a ðea0 Þ2 e2at
#
¼ 2W/s 0 Mrs D2 A1 /s 0 Drs D2 A1 ð36Þ
h j # #
1 Mrsi 2W/i qs A  1 ðA1 Þ2 þ 2a/i gs A  1 ðA1 Þ2 hðtÞ ¼ h0 ln 1 þ ðea0 Þ2 ðea0 Þ2 e2at ;
2# a a
i h 2
þ 2a/i gs A  1 ðA1 Þ2 e2aT0 1 M i W /s qi A  1 ðA1 Þ2
rs
i
þ W2 /s gi A  1 ðA1 Þ2 þ W2 /s gi A  1 ðA1 Þ2 e2aT0 where
h
1 Dirs 2WW/i qs A  1 ðA1 Þ2 þ 2aW/i gs A  1 ðA 1 Þ2 8 9
>
> >
>
i h < =
 1 ðA1 Þ2 e2aT0 1 Di 2WW/s qi A  1 ðA1 Þ2 r^p^ þ ^sq^
þ 2aW/i gs A rs #¼      ;
i >
> 2 2 >
>
: 4 p^ þ q^ ;
þ 2aW/s gi A  1 ðA1 Þ2 þ 2aW/s gi A  1 ðA1 Þ2 e2aT0
h a0 ¼ að0Þ;
1 Krsi /i qs A  1 ðA1 Þ2 þ /i gs A  1 ðA1 Þ2 8 9 ð37Þ
i h >
> >
>
 1 ðA1 Þ2 e2aT0 1 K i /s qi A  1 ðA1 Þ2 < r^q^ ^sp^ =
þ /i gs A rs
i j¼      ;
>
> 2 2 >
>
þ /s gi A 1 ðA1 Þ2 þ /s gi A  1 ðA1 Þ2 e2aT0 : 4 p^ þ q^ ;
h
2 Mrsij W2 /i /j /s A  1 ðA1 Þ2 þ W2 /i /j /s A  1 ðA1 Þ2 h0 ¼ hð0Þ:
2
i h
þ W /i /j /s A  1 ðA1 Þ2 e2aT0 2 Dij WW/i /j /s A
rs
 1 ðA1 Þ2
i Again, as it happened to the solution of order e2 , the
þ W2 /i /j /s A  1 ðA1 Þ2 þ WW/i /j /s A  1 ðA1 Þ2 e2aT0 term in eWT0 in (31) does not add any new feature to the
h solution, since there is already a term of this type in ps1
2 Krsij /i /j /s A  1 ðA1 Þ2 þ /i /j /s A  1 ðA1 Þ2
(see (19)). It will therefore be disregarded from this point
i
þ /i /j /s A  1 ðA1 Þ2 e2aT0 : ð32Þ on.
Collecting, finally, terms in e3WT0 , one obtains:
 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965 961


9W2 0 Mrs þ 3W 0 Drs þ 0 Krs Cs3 p_ s ¼ Gs1 U þ Gs2 V þ Gs3 U 2 þ Gs4 UV þ Gs5 V 2


 þ Gs6 U 3 þ Gs7 U 2 V þ Gs8 UV 2 þ Gs9 V 3 ; ð43Þ
¼ 1 Mrsi 2W/i qs ðA1 Þ3 1 Mrsi W2 /s qi ðA1 Þ3



1 Dirs 2W2 /i qs ðA1 Þ3 1 Dirs 2W2 /s qi ðA1 Þ3 with s ¼ 1–n and U and V are, respectively, a chosen

 non-zero generalised coordinate pv and its correspond-
1 Krsi ð/i qs ÞðA1 Þ3 1 Krsi /s qi ðA1 Þ3 ing velocity p_ v , that is:



2 Mrsij W2 /i /j /s ðA1 Þ3 2 Dijrs W2 /i /j /s ðA1 Þ3
U ¼ pv ; ð44Þ


2 Krsij /i /j /s ðA1 Þ3 ; ð38Þ
V ¼ p_ v : ð45Þ
and the solution is of the type
From (41) one obtains p_ s by derivation with respect to
Cs3 ¼ ks ðA1 Þ3 ¼ ðls þ ims ÞðA1 Þ3 ; ð39Þ time. If, now, ps and p_ s , as determined from (41), are
introduced on the left-hand side of (42) and (43), and
where ls and ms are real constants, provided the matrix U ¼ pv and V ¼ p_ v , as determined from (41), are intro-
of coefficients is non-singular. In the case of a system duced on the right-hand side of (42) and (43), one ends
with one eigenvalue exactly equal to three times another up with expressions in terms of the several coefficients
one, such as the case of internal resonance 3:1 in un- Fsi , Gsi . After equating terms of the same power in e, one
damped models, the solution of (38) cannot be written as obtains equations in which only linearly independent
in (39). functions are present. The evaluation of Fsi and Gsi
At this point, it is possible to write the time response comes from the imposition that the coefficients of such
for the generalised coordinates when the system is vi- functions on both sides of the equations should be the
brating according to a certain non-linear normal mode. same. By doing so, it is possible to arrive at linear sys-
In complex notation tems which, once solved, allow for the determination of
all coefficients in (42) and (43),
ps ¼ eAs1 eWt þ e2 Bs2 e2Wt þ e2 Es2 e2at þ e3 Cs3 e3Wt The evaluation of the coefficients Fsi is considered
    first.
¼ e As1 eWt þ A s1 eWt þ e2 Bs2 e2Wt þ Bs2 e2Wt

   • order e
þ e2 Es2 þ Es2 e2at þ e3 Cs3 e3Wt þ C s3 e3Wt ð40Þ     
1 a Fs1 cs
¼ ; ð46Þ
0 x Fs2 ds
or, in real notation

ps ¼ a^eat ½cs cos ðbt þ hÞ ds senðbt þ hÞ with a, cs and ds defined in (18) and x ¼ ðd=
1 2 2at dtÞðbt þ hÞ ¼ b þ h_ ffi b:
þ ð^ aÞ e ½s cos 2ðbt þ hÞ ns sen 2ðbt þ hÞ þ rs  • order e2
2 2 38 9
1 3 3at 1 a a2 < Fs3 =
þ ð^ aÞ e ½ls cos 3ðbt þ hÞ ms sen 3ðbt þ hÞ; 6
4 2 2 7
4 1 a ðx a Þ 5 Fs4
ð41Þ : ;
0 x 2xa Fs5
8 rs þs
ru þu  9
where a^ ¼ ea and the other symbols are defined in (15), < 2
> 2
Fs1 ðru a þ u a nu xÞFs2 >
=
(26), (28), (36) and (39). ¼ s þ u Fs1 þ 2ðu a nu xÞFs2
>
: >
;
ns þ nu Fs1 þ 2ðu x þ nu aÞFs2 ;
ð47Þ
3. Regeneration of the generalised coordinates as func-
tions of the modal variable with s , ns and rs defined in (26) and (28).
• order e3
The aim here is to characterize the non-linear normal
mode by writing the generalised coordinates in the form: 2 38 9
1 a ða2 x2 Þ ða3 3ax2 Þ > F
> s6 > >
60 0 x2 3ax2 <F >
7> =
6 7 s7
ps ¼ Fs1 U þ Fs2 V þ Fs3 U 2 þ Fs4 UV þ Fs5 V 2 6 7
40 x 2xa 3xa2 5> >
> Fs8 >>
: > ;
þ Fs6 U 3 þ Fs7 U 2 V þ Fs8 UV 2 þ Fs9 V 3 ; ð42Þ 0 x 2xa 2 3
ð3xa x Þ Fs9
962 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965
8 9 8 9
> ls lu Fs1 3ðlu a mu x ÞF
s2 2u Fs3  > > 3ðls a ms xÞ lu Gs1 3ðlu a mu xÞGs2 >
>
> > >
> >
>
> 3ðu a nu xÞFs4 þ 4 u x a þ 2nu ax Fs5 >
>
2 2
>
> > 2u Gs3 3ðu a nu xÞGs4
> >
>
>
> >
> >
> 
  >
>
>
> > > þ 4  x 2
a 2
þ 2n ax G >
>
> 4 ls þ 4 lu Fs1 þ 4 ðlu a
mu xÞFs2 ðru u ÞFs3 >
3 3 9 >
>
>
>
>
u u s5 >
>
>
>
> >
>

> 9 l a m x þ 3 l G þ 9 ðl a m xÞG >
> >
>
> þ 32 u a 2nu x ru a þ 12 x Fs4 >
> >
> >
>
>
>   >
> >
> 4 s s 4 3u s1 4 u u s2 >
>
< 2u ð2x2 a2 Þ Fs5 ½6n ax þ 2ru a2 Fs5 = >
> ðr
u u ÞGs3 þ 2 u a 2nu x >
>
¼
u
: >
> >
>
> > >
> r a þ 1
x G 2 ð 2x 2
a2
Þ >
>
>
>
3 3 9
4 ms þ 4 mu Fs1 þ 4 ðlu x þ mu aÞFs2 þ 2nu Fs3 >
> < u 2 s4 u =
>
> >
> ¼ þ 6nu ax þ 2ru a Gs5 2
;
> þ 5 u x þ 3nu a þ 1 ru x Fs4
> >
>
>
> 2 2 >
> >
> >
>
> 2 2
> þ ½6u ax þ 2nu ð2a x Þ þ 2ru axFs5 >
> > 9 3 9
> 4 ðls x þ ms aÞ þ 4 mu Gs1 þ 4 ðlu x þ mu aÞGs2 > >
>
>
> >
> >
> 5  >
>
> >
> >
> 1 >
>
> m
> s m F 3 ð l x þ m aÞF 2n F >
> >
> þ 2n u G s3 þ 2
 u x þ 3n u a þ 2
ru x G s4 >
>
: u s1 u u s2 u s3 ; >
> 2 2 >
>
3ðu x þ nu aÞFs4 4½2u ax þ nu ða x ÞFs5 2 2 >
> þ ½ 6 u ax þ 2n u ð 2a x Þ þ 2r u ax G s5 >
>
>
> >
>
ð48Þ >
> 3ð l x þ m aÞ m G 3 ð l x þ m a ÞG >
>
>
> s s u s1 u u s2 >
>
>
> 2n G 3 ð  x þ n a ÞG >
>
>
: u s3 u u s4 >
;
with ls and ms defined in (39). 4½2u ax þ nu ða x ÞGs5 2 2
The coefficients Gsi are evaluated from systems of
ð51Þ
linear algebraic equations which are very similar to the
with ls and ms defined in (39).
former ones (46)–(48), the differences being limited to
the vectors of independent terms.
• order e
     4. Numerical example
1 a Gs1 cs a ds x
¼ : ð49Þ
0 x Gs2 cs x þ ds a For the sake of an example, the elastic cantilever
beam model of Fig. 1, with non-proportional damping,
with a, cs and ds defined in (18) and x ¼ was analysed and its second non-linear normal mode
ðd=dtÞðbt þ hÞ ¼ b þ h_ ffi b: evaluated. Large-amplitude free-vibration motion was
• order e2 considered, so that geometric non-linearity should be
2 38 9
1 a a2 < Gs3 >
> =
taken into account. The beam was discretised with 10
6
 7 Bernoulli–Euler finite elements (30 active degrees of
4 1 a x2 a2 5 Gs4
>
: >
; freedom). The equations of motion can be written in the
0 x 2xa Gs5 form (1). Details on the beam-element formulation are
8
 9
>
> aðrs þ s Þ xns ru þ u
Gs1 >
> presented in [7].
< 2 =
ðru a þ u a nu xÞGs2 Fig. 2 shows a plot of the time evolution of the ver-
¼ ;
>
> 2ðs a ns xÞ þ u Gs1 þ 2ðu a nu xÞGs2 >
> tical displacement of the beam tip (the 29th degree of
: ;
2ðs x þ ns aÞ þ nu Gs1 þ 2ðu x þ nu aÞGs2
ð50Þ

with s , ns and rs defined in (26) and (28).

• order e3
2 38 9
1 a ða2 x2 Þ ða3 3ax2 Þ > Gs6 >
6 7<>
> >
>
60 0 x2 3ax2 7 Gs7 =
6 7
6 0 x 3xa2 7
4 2xa 5>>
> Gs8 >
>
>
2 3
: ; Fig. 1. Cantilever beam and finite-element model. Mechanical
0 x 2xa ð3xa x Þ Gs9 constants: q ¼ 7800 kg/m3 , E ¼ 2:1  1011 N/m2 , c ¼ 0:7 N s/m.

Fig. 2. Time evolution of the tip of the beam vertical displacement.

 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965 963

Table 1
Coefficients of the manifold associated with the second non-linear normal mode of the beam for p30 and p_ 30
Coefficient Multiple scales Invariant manifold
p30 p_ 30 p30 p_ 30
U 2.3896 4.0551 2.3896 4.0551
V 0.0005959 2.3907 0.0005959 2.3907
U2 0.0 0.0 0.0 0.0
UV 0.0 0.0 0.0 0.0
V2 0.0 0.0 0.0 0.0
U3 1.0372 46.0378 0.2045 42.533
U 2V 0.0001147 9.2005 0.00287 7.9776
UV 2 0.0004572 0.01787 0.0006399 0.01703
V3 0.0 0.0004522 4.155  10 7 0.0006394

freedom, taken together with its corresponding velocity
as modal variables) for certain initial conditions, just for
the sake of an example. Three curves are presented in the
plot, namely, the linear solution, the one obtained by the
technique here proposed and the one obtained by the
invariant manifold technique [6]. The last two almost
coincide, which is seen as a validation test for the tech-
nique proposed here.
Re-generation—as proposed in Section 3—of the
generalised coordinate associated with the rotation of
the tip of the beam (the 30th degree of freedom) in
terms of the modal variables, which are here chosen
to be the vertical displacement of the tip of the beam
and its velocity, leads to the coefficients presented in
Table 1.
In spite of apparent numerical discrepancies in the
coefficients of cubic terms according to the multiple-scale
approach and the invariant manifold technique, they have
an overall effect which is amazingly equivalent. In fact,
Figs. 3 and 4 show that the temporal responses are in ex-
cellent agreement. It should be noted that this agreement is
not because the non-linear coefficients are unimportant,
since it can also be observed in these figures that the non-
linear responses are quite different from the linear one.
Another interesting feature of this problem, explored
by Soares in his doctoral thesis [9], is that even the linear
solution does not characterize a standing wave for the
normal modes, that is, the nodal points are not fixed
during the structure vibration. As mentioned before, this
is typical of systems with non-proportional damping and

Fig. 3. Time evolution of the tip of the beam rotation.

Fig. 4. Time evolution of the tip of the beam rotation velocity.

964 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965

Fig. 5. Time evolution of the displacements of the cantilever beam with non-proportional damping. Inital conditions: U ð0Þ ¼ 0:201,
V ð0Þ ¼ 0:000 (i.e. a0 ¼ 2:15, h0 ¼ 0:052).

the same feature has also been correctly captured by
the technique here presented for the non-linear system
(see Fig. 5). Note that the nodal point is moving along the
beam axis as the motion takes place. It stays approxi-
mately at 0.5 m from the beam tip most of the time, but
also reaches the beam tip at t ¼ 0:022 and 0.0643 s.
5. Conclusions
In this paper it is presented an alternative method for
the evaluation of non-linear normal modes of finite-
element models.
The procedure relies upon the definition of non-linear
normal modes of vibration proposed by Shaw and Pierre
[3] and is based on the application of the method of
multiple scales to the non-linear free-vibration equations
of motion. Both the time evolution of the generalised
coordinates and the topological characterization of the
invariant manifold associated with the non-linear nor-
mal mode are pursued.
As an illustrative example, it is chosen a model of a
cantilever beam with non-proportional damping. The
results obtained are compared with those of the linear
theory and the invariant manifold technique, with ex-
cellent agreement with the latter.
 C.E.N. Mazzilli, O.G.P. Baracho Neto / Computers and Structures 80 (2002) 957–965
Acknowledgements
The authors acknowledge the support to their re-
search project provided by FAPESP, Fundacß~ao de
Amparo a Pesquisa do Estado de S~ao Paulo, under the
grant 98/13.776-5, and by CNPq, Conselho Nacional de
Desenvolvimento Cientifico e Tecnol ogico, under the
grants 300345/96-8 and 509185/99-01. The authors
would also like to express their gratitude to Dr. Mario
Soares who helped them with the numerical example
presented in this paper.
References
[1] Vakakis AF, Manevitch LI, Mikhlin YV, Pilipchuk VN,
Zevin AA. Normal modes and localization in nonlinear
systems. Wiley Series in Nonlinear Science; 1996.
965
[2] Rosenberg RM. On nonlinear vibrations of systems with
many degrees of freedom. Adv Appl Mech 1966;9:155–242.
[3] Shaw SW, Pierre C. Normal modes for nonlinear vibratory
systems. J Sound Vibr 1993;164(1):85–124.
[4] Nayfeh AH, Nayfeh SA. On nonlinear modes of continuous
systems. J Vibr Acoust 1994;116:129–36.
[5] Nayfeh AH, Nayfeh SA, Chin C. Nonlinear normal modes
of a cantilever beam. J Vibr Acoust 1995;117:477–81.
[6] Soares MES, Mazzilli CEN. Nonlinear normal modes of
planar frames discretised by the finite element method.
Comput Struct 2000;77:485–93.
[7] Brasil RMLRF, Mazzilli CEN. A general formulation of
nonlinear dynamics applied to accessing the statical loading
effect upon the dynamic response of planar frames. Appl
Mech Rev 1993;46(11):S110–7.
[8] Nayfeh AH, Mook DT. Nonlinear oscillations. Wiley; 1979.
[9] Soares MES. Non-linear modes of vibration in systems
discretised by the finite-element method. PhD thesis, Escola
Politecnica da Universidade de S~ao Paulo, 1998 (in Portu-
guese).