Вы находитесь на странице: 1из 25

Journal of Sound and Vibration (1996) 189(3), 315–339

A MODAL SUPERPOSITION METHOD FOR


NON-LINEAR STRUCTURES
B. K
Tübitak-Sage, P.K. 16, 06261, Mamak, Ankara, Turkey

H. N. O̈
Department of Mechanical Engineering, Middle East Technical University, 06531, Ankara,
Turkey

(Received 30 September 1992, and in final form 16 February 1995)

The dynamic response of multi-degree of freedom (MDOF) non-linear structures is


usually determined by the numerical integration of equations of motion. This is
computationally very costly for steady state response analysis. In this study, a powerful and
economical method is developed for the harmonic response analysis of non-linear
structures. In this method, the equations of motion are first converted into a set of
non-linear algebraic equations, and then the number of equations to be solved is reduced
by employing linear modes of the system. The resulting coupled algebraic equations are
solved iteratively. The concept of the pseudo-receptance matrix is introduced in calculating
the steady state response. It is shown that the reduction in computational time can be
considerable as very accurate results can be obtained by using only a limited number of
modes. Several example problems are solved in order to illustrate the performance of the
method. Effects of higher modes and higher harmonic components of the response are
discussed in detail. In order to illustrate the accuracy and speed of the method suggested,
the results obtained by applying the so called Iterative Modal Method are compared with
those obtained by the Newmark method which is used as a classical time domain analysis
technique.
7 1996 Academic Press Limited

1. INTRODUCTION
In structural dynamics, numerical integration of equations of motion (time marching
analysis) is used frequently in order to find the steady state response of structures [1, 2].
The computational effort required for time marching analysis generally depends on the
damping and natural frequencies of the structure under consideration. Low damping levels
in the structure give rise to very long transients. Stiff structures with large degrees of
freedom require a very small time step to be used. Small time steps combined with long
transients result in huge amounts of computational time. In many branches of structural
dynamics, approximate frequency domain techniques are now being used as an alternative
to numerical integration procedures for steady state periodic response analysis of
non-linear multiple degrees of freedom structures.
Watanabe and Sato [3, 4] investigated the effects of clearance type non-linearities on the
harmonic response of structures. They used the describing function technique and a
bisection method to determine the response. Later, Murakami and Sato [5] experimentally
verified the method developed by Watanabe and Sato by testing a beam with clearance.
315
0022–460X/96/030315+25 $12.00/0 7 1996 Academic Press Limited
316 .   . . ̈̈
Choi and Noah [6] determined the steady state non-linear response of a rotor-support
system due to deadband and rubbing using harmonic balance and Newton–Raphson
techniques. They tried to predict the effects of sub-harmonic and super-harmonic terms
on the system response. Nataraj and Nelson [7] used trigonometric collocation to analyze
the periodic response of rotor dynamic systems. They used component mode synthesis to
model the linear part of the structure and developed a technique which reduces the total
number of co-ordinates affected by the non-linearities. Ostachowicz [8, 9] investigated
the periodic response of beams with dry friction dampers. He used the finite element
method and the harmonic balance technique in the formulations. Later, Ostachowicz [10]
developed a technique for reducing the size of the non-linear algebraic problem, which was
very similar to the one developed by Nataraj and Nelson. Shiau and Jean [11] used the
harmonic balance technique to analyze the periodic response of rotor dynamic systems.
The technique developed by Ostachowicz to deal with local non-linearities was also
developed independently by Shiau and Jean. Ferri and Dowell [12] used a special friction
model to investigate the effects of stick–slip behavior. They reduced the overall size of
the problem by using the component mode synthesis technique for the linear part of the
structure, and transformed the equations of motion to the frequency domain by using
Galerkin’s method and then obtained the solution by the Newton–Raphson method.
Budak and Ozgüven [13–16] developed a technique for the harmonic response analysis of
multiple degrees of freedom systems. They used a special complex algebra for non-linear
operations. They employed the method developed by Ozgüven [17] for the harmonic
response analysis of non-proportionally damped linear structures, to increase the speed of
computations especially for cases in which the non-linearities are localized. Very recently,
Tanrıkulu et al. [18] proposed a new formulation for fundamental harmonic response
analysis of structures having different kinds of non-linearities by using describing
functions, and the technique in reference [17] was employed to reduce significantly the
computational time for the analysis of very large structures with local non-linearities.
In recent years, reduction of computaton time for linear and non-linear dynamic
problems has become the focus of intense research efforts. The modal superposition
technique is a very effective reduction method for linear dynamic problems when only few
vibration modes are excited by the external loading, and several studies have been made
for improving its accuracy and efficiency [19]. The main target of the technique is to predict
the response with a substantial reduction in computational time by using only a few
vibration modes. Several studies have been made for the reduction of the order of
non-linear dynamic problems through the modal superposition technique, but most of
them are limited to classically damped systems where a damping proportional to the mass
or stiffness is assumed in order to uncouple the equations of motion [20, 21].
Stricklin and Haisler [22] gave a survey of the formulations and solution procedures for
non-linear structural dynamic analyses. They presented a new and very efficient method
based on modal analysis using the pseudo-force method. Bathe and Gracewski [23] studied
several mode superposition and substructuring procedures. They compared the results with
those found by direct integration of the complete system of equations for several sample
cases. Noor [19] summarized the status and some developments in the application of
reduction methods to non-linear static and dynamic problems. He discussed several aspects
of the reduction methods, and demonstrated the effectiveness of reduction methods with
numerical examples. Also, he identified a number of research areas which have high
potential for application of reduction methods. Subbiah, Bhat and Sankar [24] proposed
a modal reduction technique for solving rotor dynamic problems. They reduced the
resulting system matrices in size by considering only the contribution of the more
important lower modes. They performed unbalance response analyses of a simple Jeffcott
    -  317
rotor, and compared the results with those obtained by experiments. Chang and Mohraz
[25] presented two mode superposition procedures for the dynamic analysis of non-linear
structures with classical and non-classical damping. In their procedures, the non-linearities
were treated as pseudo-forces. They used undamped and damped eigensolutions to
uncouple the equations of motion for systems with classical and non-classical dampings,
respectively.
In this paper, a semi-analytical frequency domain technique based on the modal
superposition principle is presented. The theoretical basis of the method is presented in the
context of obtaining the fundamental harmonic response of structures with symmetrical
non-linearities although the method is applicable for virtually any kind of non-linearity
(symmetrical or asymmetrical). A complete formulation of multi-harmonic analysis is also
presented. The method is applied to several example cases in order to demonstrate the
performance of the method in respect to accuracy and computational effort.

2. THEORY

2.1.    - 


The equation of motion of a non-linear structure modeled as an n degrees of freedom
discrete system can be expressed in matrix form as
[M ]{ẍ}+[C]{ẋ}+i[H ]{x}+[K ]{x}+{g}={f }, (1)
where [M ], [C], [H ] and [K ] are linear mass, viscous damping, structural damping and
stiffness matrices, respectively. {x}, {ẋ} and {ẍ} are the generalized displacement, velocity
and acceleration vectors, respectively, and i represents the unit imaginary number. {f } and
{g} represent the external forcing and the internal non-linear forces, respectively (a list of
nomenclature is given in Appendix B). For a harmonic excitation that can be expressed
in complex notation as
{f }={F} eivt , (2)
where {F} is the amplitude vector of the forcing and v is the excitation frequency, the
response can be represented in terms of Fourier series as

a a
{x}= s {x}r= s {X}r eirvt . (3)
r=1 r=0

The response of the structure consists of the bias term {x}0 , the fundamental harmonic
term {x1 } and the superharmonic terms {x}r (r=2, 3, 4, . . .). The terms with even
coefficients are due to non-linearities with asymmetrical characteristics. If interest is
confined to symmetrical non-linearities only, equation (3) simplifies to
{x}={X} eivt , (4)
provided that the following conditions are satisfied: (i) the system to be analyzed should
act as a narrow-band filter so that the sub-harmonic and the super-harmonic components
will be heavily attenuated, and an integer multiple of the excitation frequency v should
not coincide with any of the natural frequencies of the linear part of the system; (ii)
{g(x, ẋ)} should have finite partial derivatives with respect to {x} and {ẋ}.
In equation (4), the subscript (1) is dropped for convenience, and {X} is the amplitude
vector of harmonic oscillations, which is complex in general to accommodate phase
318 .   . . ̈̈
information. Since the response is assumed to be harmonic, the internal non-linear forces
can also be assumed to be harmonic and written as
{g}={G} eivt , (5)
where {G} is the complex amplitude vector of the internal non-linear forces. Substitution
of equations (2), (4) and (5) into equation (1) yields
[a]−1{X}+{G}={F}, (6)
where [a] is the receptance matrix of the linear part of the structure, which is given by
[a]=(−v 2[M ]+iv[C]+i[H ]+[K ])−1 , (7)
and can be obtained by modal superposition. In a recent study, Budak and Ozgüven [14]
developed a new and efficient method called the ‘‘Iterative Receptance Method’’ (IRM)
for the harmonic vibration analysis of non-linear structures in which the amplitude vector
of the internal non-linear forces is expressed as
{G}=[D]{X}. (8)
Here [D] is response dependent, the so-called ‘‘non-linearity matrix’’. Budak and Ozgüven
gave a formulation for obtaining the elements of [D] for the polynomial type non-linearities
(symmetrical and asymmetrical) by using complex variables, and employed this method
in further studies [15]. In the formulation, displacement and velocity dependent
non-linearities are considered such that the internal non-linear force vector can be written
in two parts as
{g}={S(x)}+{D(ẋ)}, (9)
where {S(x)} and {D(ẋ)} are vectors that represent the displacement dependent stiffness
and the velocity dependent damping forces: i.e.,

m n m n
Sr (x)= s s prjk (xr−xj )k+prrk xrk , Dr (ẋ)= s s qrjk (ẋr−ẋj )k+qrrk ẋrk , (10, 11)
k=3,5 j=1 k=3,5 j=1
j$r j$r

where prjk and qrjk are the coefficients of polynomials, and m is the highest power considered
in the polynomials. It was shown [14] that the elements of [D] can be expressed as

$ %
m n
Drr= s m(k) s grjk =Xr−Xj =k−1+grrk =Xr =k−1 , r=1, 2, . . . , n, (12)
k=3,5 j=1
j$r

m
Drj=− s m(k)[grjk =Xr−Xj =k−1 ], (r$j), r, j=1, 2, . . . , n, (13)
k=3,5

where
m(k)=(2/zp )G[(k+2)/2]/G[(k+3)/2], grjk=prjk+iv kqrjk , (14, 15)
in which G is the Gamma function. Further details about obtaining [D] and the method
can be found in reference [13].
Very recently, Tanrıkulu et al. [18] proposed a semi-analytical quasi-linear approach to
determine the elements of [D] for different kinds of non-linearities (including polynomial
    -  319
type non-linearities) by using the well developed describing function theory [26, 27]. The
elements of [D] can be found from describing functions nrj as [28]

n
Drr=nrr+ s nrj , r=1, 2, . . . , n, Drj=−nrj , r$j, r=1, 2, . . . , n,
j=1
j$r

(16, 17)
where

g
2p
i
nrj= nrj e−ic dc, c=vt, (18, 19)
pYrj 0

and Yrj is the complex amplitude of yrj which is defined as


yrj=xr−xj if r$j, yrj=xr if r=j. (20a, b)
Since xr and xj are harmonic functions, yrj can also be represented by a harmonic function
so that
r sin (vt+gr ),
xr=X j sin (vt+gj ),
xj=X rj sin (vt+brj ).
yrj=Y (21a–c)
nrj represents the non-linear force element between co-ordinates r and j for r$j, and
between ground and co-ordinate r for r=j. Note that nrj can be expressed as a function
of the inter-co-ordinate displacement yrj and its derivatives.

2.2. -    - 


Representation of non-linear forces in a matrix form opens a new avenue in the
harmonic vibration analysis of multi-degree-of-freedom non-linear structures. This
representation was the key point in the previously developed two methods [14, 18], and
the method proposed in this study is based on the matrix representaton of the
non-linearity. In this section, the pseudo-receptance matrix of a non-linear structure for
a given forcing will be introduced by making use of the non-linearity matrix [D].
By using equation (8), equation (6) can be simplified to
{X}=(−v 2[M ]+iv[C]+i[H ]+[K ]+[D])−1{F}, (22)
where the coefficient matrix of {F} can be identified as the response level dependent
‘‘pseudo-receptance matrix’’ of the structure, derived by using first order approximations:
[U]=(−v 2[M ]+iv[C]+i[H ]+[K ]+[D])−1 . (23)
It should be noted that the term ‘‘pseudo’’ is used, since [U] cannot be called the
‘‘receptance matrix’’ of the structure. It is valid only for the given forcing configuration
and is used in order to find the response from the equation
{X}=[U]{F}. (24)
A detailed discussion of the physical significance of [U] can be found in reference [18].

2.3.               

The linear structural damping matrix [H ] can be expressed as
[H ]=[H ]p+[H ]n , (25)
320 .   . . ̈̈
where [H ]p and [H ]n are the proportional and the non-proportional parts of the structural
damping matrix, respectively. Just for convenience, [H ]p can be assumed to be proportional
to [K ] such that
[H ]p=p[K ]. (26)
Upon making use of equation (26), equation (22) takes the form
(−v 2[M ]+iv[C]+i[H ]n+(1+ip)[K ]+[D]){X}={F}. (27)
The undamped modal data can be obtained by solving the real eigenvalue problem
[K ]{f}=v 2[M ]{f}. (28)
Solution of the eigenvalue problem gives the mass normalized undamped modal vectors
{f r} as well as the undamped natural frequencies vr . The co-ordinate transformation
{X}=[F]{h0 } (29)
yields the modal equations
(−v 2[I ]+iv[C*]+i[H*]+(1+ip)[vr2 ]+[D*]){h0 }={F *}, (30)
where {h0 } is the complex amplitude vector of the principal (or modal) co-ordinates, [I]
is the identity matrix, and
[C*]=[F]T[C][F], [H*]=[F]T[H ]n [F], (31, 32)
T
[D*]=[F] [D][F], {F *}=[F] {F}.
T
(33, 34)
Equation (30) can also be written as
[[a*]−1+(iv[C*]+i[H*]+[D*])]{h0 }={F *}, (35)
where [a*] is the receptance matrix of the proportionally damped linear part of the
structure in terms of modal co-ordinates and it is given by the equation
[a*]=[[vr2 ](1+ip)−v 2[I]]−1 . (36)
Premultiplying all terms of equation (35) by [a*] gives
[[I]+[a*](iv[C*]+i[H*]+[D*])]{h0 }=[a*]{F *}, (37)
{h0 }=[[I]+[a*](iv[C*]+i[H*]+[D*])]−1{a*}{F *}. (38)
[a*] is a diagonal matrix whose elements are given by
a* 2 2
jj =1/(vj (1+ip)−v ), j=1, 2, . . . , n. (39)
When the excitation frequency v is very close to an undamped natural frequency vj , a* jj
increases without bound if p=0. Thus, in the neighborhood of vj , the role of p becomes
more pronounced for a* jj . Even if the structural damping matrix is not included in the
model of the structure, a very small artificial value for the loss factor should be introduced
for numerical stability. This will not only improve the computatonal accuracy
considerably, but also it will represent a more realistic case, since every structure will
have some uniformly distributed structural damping no matter how small this value is.
However, if the proportional part of the damping is introduced for numerical stability, the
loss factor should be kept very small so as not to increase the overall damping of the
structure artificially.
    -  321
From equation (38), the pseudo-receptance matrix of the non-linear structure defined
in terms of modal co-ordinates, [U*], can be expressed as
{h0 }=[U*]{F *}, (40)
where
[U*]=[[I ]+[a*](iv[C*]+i[H*]+[D*])]−1 [a*]. (41)
The relation between [U*] and the pseudo-receptance matrix [U] of the non-linear
structure can be obtained from equations (29), (34) and (40) as
[U]=[F][U*][F]T. (42)
The method described in this study is an application of mode superposition for
non-linear systems. In principle, the use of the mode superposition method simply involves
a co-ordinate transformation from the generalized co-ordinates to modal co-ordinates.
This is particularly effective if only a few modal co-ordinates are employed. Very often,
the response of linear systems can be predicted very accurately by using a greatly truncated
modal series rather than the full n modes, as, for harmonic forcing, only a few vibration
modes of the structure are excited. The accuracy of the mode superposition method
suggested here for non-linear structures will be demonstrated by some numerical examples.
When a reduced number of modal co-ordinates are taken into account for the response
prediction by truncating the modal series, the computational effort required in the analysis
is reduced drastically. For example, the modal matrix [F] used in equation (29) for
co-ordinate transformation is then to be constructed by using only a few eigenvectors and
therefore it is a rectangular matrix of size n×m, where mWn. Computational time
comparisons will also be made for the numerical applications to be presented. Effects of
higher harmonic terms on the response also will be discussed in detail. Note that, since
the modes of a non-linear structure are difficult to define (if they exist), the term ‘‘modes’’
refers to the modes of the linear undamped part of the structure.

2.4.   


The expression used to calculate [U*] in equation (41) contains the non-linearity matrix
which depends on the vibration amplitudes. Therefore, the non-linear complex algebraic
equations defined by equation (40) can only be solved iteratively. The following iteration
strategy can be followed at a particular frequency v:
{h0 }i+1=[U*]i {F *}. (43)
{h0 }i+1 is the complex amplitude vector for the principal co-ordinates at the (i+1)th
iteration step and [U*]i is calculated by using {h0 }i . The iterations can be continued until
the Euclidian norm of the displacement error vector {e}, defined as

b b
(h0 ) ji+1−(h0 ) ji
e j= , j=1, 2, . . . , m, (44)
(h0 ) ji

drops below a pre-defined tolerance value. The superscript j denotes the jth element. The
iteration scheme can be improved if instead of {h0 }i+1 one uses {h'0 }i+1 to determine [U*]i+1
such that
{h'0 }i+1=12 ({h0 }i+1+{h0 }i ). (45)
In practical cases, response information is required for a certain frequency range of
lower limit vl and upper limit vu in order to generate frequency response curves. Moreover,
322 .   . . ̈̈
structures with certain types of non-linearities may show multiple response behavior in
the frequency range of interest. In predicting multiple responses, the proximity of the
initial guess {h0 }'1 to the actual solution plays an important role. The initial guess at each
frequency can be selected as follows.
Low to high frequency sweep: at v=vl , the response of the proportionally damped
linear part of the structure is used as {h0 }1 . At other frequencies, vlQvEvu , the response
obtained at v is used as {h0 }1 for iterations at frequency v+Dv.
High to low frequency sweep: at v=vu , the response of the proportionally damped
linear part of the structure is used as {h0 }1 . At other frequencies, vuqvevl , the response
obtained at v is used as {h0 }1 for iterations at frequency v−Dv.
It can be easily seen from equation (41) that the solution procedure requires an inversion
operation to find [U*] at each iteration step. However, inversion is known to be sensitive
to ill-conditioning in the matrix to be inverted. This problem can be overcome by using
the method developed by Ozgüven [17] for the dynamic analysis of non-proportionally
damped structures. In this method, Ozgüven obtained the receptance matrix of a
non-proportionally damped structure by utilizing the undamped receptance and the
damping matrix of the structure. The damping forces are treated as if they are acting
externally on the undamped part of the structure. He has shown that the computational
effort to calculate the receptance matrix is reduced significantly if damping affects only a
few co-ordinates. The same approach is used here by taking the internal non-linear and
damping forces as external (pseudo-) forces acting on the linear proportionally damped
part of the structure. Since the same steps are followed here as those applied in Ozgüven’s
method, the method will be described very briefly. The original method is well explained,
with several examples, in reference [17].
One can define a new matrix [D] as

[D]=iv[C*]+i[H*]+[D*]. (46)

Then {h0 } can be found from equation (35) as

{h0 }=[a*][{F *}−[D]{h0 }]. (47)

The complex amplitude of the pth modal co-ordinate can be expressed as

m m m
h0p= s a* s − s a*
ps F* ps s Dsk h0 .
k
(48)
s=1 s=1 k=1

U*pj can be obtained from equation (48) by dividing all terms by F j* and setting all other
forcing terms to zero:

m m
U* pj − s a*
pj =a* ps s Dsk U*
kj . (49)
s=1 k=1

If only the kth column of [D] is considered, equation (49) becomes

m
U* pj − s a*
pj =a* ps Dsk U*
kj . (50)
s=1
    -  323
If one takes p=k, U*
kj can be obtained as

>0 1
m
U*
kj =a*
pj 1+ s a*
ks Dsk , j=1, 2, . . . , m. (51)
s=1

After the U*kj , (k=k; j=1, 2, . . . , m) have been calculated the remaining elements of [U*]
can be found from equation (50). The above procedure gives [U*] under the effect of only
the kth column of [D]. If the calculated [U*] is treated as [a*] in equations (50) and (51),
a new [U*] can be obtained by considering another column of [D]. After repeating this
procedure for all columns of [D] (i.e., k=1, 2, . . . , m), the final receptance matrix which
gives the required [U*] can be calculated.

3. MULTI-HARMONIC RESPONSE

3.1. 
For non-linear structures, the first order approximation given for the response prediction
is usually sufficient, unless the non-linearities are very strong. However, sometimes, it
may be required to consider higher harmonic terms as well [28–30]. In this section, the
formulation presented for fundamental harmonic response will be extended such that the
higher harmonics of the response are included in the analysis. If the structure is excited
by external forcing as given by equation (2), the displacement vector with higher order
terms can be expressed as given in equation (3). The kth element of {X}m which is the
complex amplitude vector for the mth harmonic can be written as
k )m ei(gk )m ,
(Xk )m=(X (52)
where (Xk )m and (gk )m are the displacement amplitude (real) and the phase, respectively,
of the mth harmonic component of the response for the kth co-ordinate. Also, one should
note that
}0 .
{X}0={X (53)
Similarly, inter-co-ordinate displacements can be represented as

a a
ykj= s ( ykj )m= s (Ykj )m ei(mv)t , (Ykj )m=(Xk )m−(Xj )m , k$j. (54, 55)
m=0 m=0

(Ykj )m=(Xk )m , k=j, kj )m ei(bkj )m ,


(Ykj )m=(Y kj )0 .
(Ykj )0=(Y (56–58)
If the internal non-linear forces are assumed to be periodic, then they can be expressed
as

a a
nkj= s (nkj )m= s (Akj )m ei(mv)t , kj )m ei(zkj )m .
(Akj )m=(A (59, 60)
m=0 m=0

kj )0 and (Akj )m can be calculated from the Fourier integrals


(A

g g
2p 2p
1 i
kj )0=
(A nkj dc, (Akj )m= nkj e−imc dc, me1, (61, 62)
2p 0
p 0
324 .   . . ̈̈
where c=vt. In equations (61) and (62), the following form of ykj should be used in
evaluating integrals:
a
kj )m sin (mc+(bkj )m ).
ykj= s (Y (63)
m=0

The describing function of order m, (nkj )m , corresponding to nkj can be expressed as


(nkj )m=(Akj )m /(Ykj )m . (64)
By using equations (61), (62) and (64), (nkj )m can be determined as

g g
2p 2p
1 i
(n̄kj )0= nkj dc, (nkj )m= nkj e−imc dc, me1. (65, 66)
kj )0
2p(Y 0
p(Ykj )m 0

Then, the internal non-linear forces can be written as


a a a
nkj= s (nkj )m (Ykj )m e imc , {g}= s {g}m= s {G}m e imc , (67, 68)
m=0 m=0 m=0

where
n
(Gk )m= s (nkj )m (Ykj )m . (69)
j=1

If equations (2), (3) and (68) are substituted into equation (1), the following can be
obtained by grouping the terms with the same frequencies:
[K ]{X}0+{G}0={0}, ([K ]−v 2[M ]+i[H ]+iv[C]){X}1+{G}1={F}, (70, 71)
([K ]−(mv)2[M ]+i[H ]+i(mv)[C]){X}m+{G}m={0}, m=2, 3, . . . (72)
Furthermore, {G}m , m=0, 1, 2, . . . , can be expressed as
{G}m=[D]m {X}m , (73)
n
(Dkk )m=(nkk )m+ s (nkj )m , k=1, 2, . . . , n, (Dkj )m=−(nkj )m , k, j=1, 2, . . . , n.
j=1
j$k

(74, 75)
If equation (73) is substituted into equations (70)–(72), and the structural damping matrix
[H ] is expressed by equations (25) and (26), one can obtain
([K ]+[D]0 ){X}0={0}, ((1+ip)[K ]−v 2[M ]+i[H ]n+iv[C]+[D]1 ){X}1={F},
(76, 77)
((1+ip)[K ]−(mv)2[M ]+i[H ]n+i(mv)[C]+[D]m ){X}m={0}, m=2, 3, . . . (78)
By employing linear undamped modal data, the complex amplitude vector for the mth
harmonic can be expressed as
{X}m=[F]{h}m , (79)
where {h}m is the complex amplitude vector of the principal (or modal) co-ordinates
corresponding to the mth harmonic. If equation (79) is substituted into equations
    -  325
T
(76)–(78), and the resulting equations are pre-multiplied by [F] , the following equations
can be obtained:
([vr2 ]+[D*]0 ){h}0={0}, (80)
2 2
((1+ip)[v ]−v [I]+i[H*]+iv[C*]+[D*]1 ){h}1={F *},
r (81)
((1+ip)[vr2 ]−(mv)2[I]+i[H*]+i(mv)[C*]+[D*]m ){h}m={0},
m=2, 3, . . . , (82)
where
[D*]m=[F]T[D]m [F], m=0, 1, 2, . . . (83)

3.2.  


Equations (80)–(82) can be expressed in a more compact form as
{T}m={0}, m=0, 1, . . . , (84)
where
{T}0=([Z]0+[D]0 ){h}0 , {T1 }=([Z]1+[D]1 ){h}1−{F *}, (85, 86)
{T}m=([Z]m+[D]m ){h}m , m=2, 3, . . . , [Z]0=[vr2 ], (87, 88)
2 2
[Z]1=(1+ip)[v ]−v 2[I],
r [Z]m=(1+ip)[v ]−(mv)2[I ],
r m=2, 3, . . . , (89, 90)
[D]0=[D*]0 , [D]1=i[H*]+iv[C*]+[D*]1 , (91, 92)
[D]m=i[H*]+i(mv)[C*]+[D*]m , m=2, 3, . . . (93)
[Z]m can be called the dynamic stiffness matrix of the linear proportionally damped part
of the structure at frequency (mv) defined in terms of modal co-ordinates. Equation (84)
constitutes a set of non-linear complex algebraic equations which should be solved
iteratively as explained in section 2.4.

4. CASE STUDIES
Several case studies are presented to demonstrate the application of the method.
Comprehensive discussion of the accuracy obtained in the case discarding higher modes
is given. Effects of higher harmonics on the response predictions are also investigated in
detail. The responses computed by using the formulation presented in this study are

Figure 1. Twenty d.o.f. cantilever beam (local non-linearity).


326 .   . . ̈̈
T 1
Properties of the cantilever beam
Length, L 0·787 m
Width, b 0·0635 m
Height, h 0·0127 m
Modulus of elasticity, E 2×1011 N/m2
Density, r 7800 kg/m3
Proportional structural damping, p 0·01

compared with those obtained by the classical time domain analysis technique for accuracy
validation.

4.1.          -



In this section, a cantilever beam discretized to 20 degrees of freedom is analyzed
by using only a single mode (see Figure 1). The properties of the cantilever beam are shown
in Table 1. The first two undamped natural frequencies of the linear part of the structure
are calculated as v1=105·24 rad/s and v2=659·54 rad/s. Cubic stiffness, piecewise
linear stiffness and Coulomb damping type non-linearities are considered (see Figure 2).
The mathematical expressions and the corresponding description functions of these
non-linearities are given in Appendix A. In all cases, the non-linearity is taken to be local
and between the 19th co-ordinate and the ground.
Figures 3 and 4 show the non-linear response (X 19 ) around the first resonance for low
to high and high to low frequency sweeps, respectively, for a cubic stiffness type of
non-linearity (cs=1×108 N/m3, F19=1 N). The response is calculated by considering only
one mode and compared with that obtained by including 20 modes. It is observed that
one mode is sufficient to predict the response quite accurately since the first mode is well
separated, and multiple solutions can fully be simulated by using one mode. The effect of
mode truncation on computational time is also investigated and it is concluded that
truncation results in an extremely important reduction. Comparison of computational time
in one iteration with or without mode truncation is given in
Table 2.
Second, the same structure is analyzed with a piecewise linear stiffness type
of non-linearity, (k1=10 N/m, k2=2500 N/m, d=5×10−4 m, F19=1 N). As can be
seen from Figures 5 and 6, the non-linear response (X 19 ) can be obtained very

Figure 2. Schematic diagrams of the non-linearities analyzed in this study. (a) Coulomb damping; (b) piecewise
non-linear stiffness; (c) cubic stiffness.
    -  327

19 ) of the system in Figure 1 around first resonance as obtained by using one
Figure 3. Non-linear response (X
q and 20 (—×—) modes (low to high frequency sweep, local cubic stiffness).
(−−)

accurately for low to high and high to low frequency sweeps at all frequencies around the
first resonance by using even a single mode. Figure 7 compares the linear and non-linear
responses.
Lastly, the response of the structure is obtained for a Coulomb damping type
non-linearity, (a=0·45 N, F19=1 N), and is depicted in Figure 8. The results obtained by
considering one mode are in very good agreement with those obtained by including all 20
modes of the system. The reduction in computational time when using only one mode can
be observed in Table 2 for the above cases.

Figure 4. As Figure 3 but high to low frequency sweep.


328 .   . . ̈̈
T 2
Computational time comparison for different non-linear elements
Type of non-linear element Number of modes considered Time for one iteration (s)
Local cubic stiffness 1 0·58
Local cubic stiffness 20 72·75
Local piecewise linear stiffness 1 0·58
Local piecewise linear stiffness 20 73·00
Local Coulomb damping 1 0·59
Local Coulomb damping 20 70·00
Distributed cubic stiffness 1 0·60
Distributed cubic stiffness 20 74·00

4.2.         


-
The cantilever beam whose properties are tabulated in Table 1 is considered in this
section with cubic stiffness type non-linearities (cs=2·5×106 N/m3 ) between all
translational co-ordinates and ground (see Figure 9). The non-linear response (X 19 ) for
low to high and high to low frequency sweeps when the beam is excited from all
translational co-ordinates, with F=1 N, is illustrated in Figures 10 and 11, respectively.
Linear and non-linear responses are compared in Figure 12. From Figures 10 and 11, it
can be observed that the harmonic response of the structure can be accurately simulated
by using only a single mode. As can be noticed from Table 2, there is not a notable
computational time difference between those required for distributed and local
non-linearities, respectively, and using a single mode reduces the computational time
considerably.

19 ) of the system in Figure 1 around first resonance as obtained by using one
Figure 5. Non-linear response (X
q and 20 (—×—) modes (low to high frequency sweep, local piecewise linear stiffness).
(−−)
    -  329

Figure 6. As Figure 5 but high to low frequency sweep.

4.3.        


The system considered in section 4.1 is used in this case study, (cs=−2×107 N/m3 ).
Figure 13 shows the linear and non-linear response (X 19 ) around the first resonance by
using 20 modes. It is observed that the third harmonic component has negligible influence
on the total response, and therefore it is not shown in the figure. Although some
discrepancies between multi-harmonic and fundamental harmonic solutions are observed,
it can be concluded that, in general, the response is well approximated by the fundamental

Figure 7. Linear and non-linear response (X19 ) of the system in Figure 1 around first resonance (local piecewise
linear stiffness). ——, Linear; , non-linear, low to high sweep; · · · ·, non-linear, high to low sweep.
330 .   . . ̈̈

q 20 modes, non-linear;
Figure 8. As Figure 7 but local Coulomb damping. —×—, One mode, non-linear; −−,
——, linear.

harmonic solution. The maximum difference between the multi-harmonic and fundamental
harmonic solutions in this case study is found to be 4·2%. The multi-harmonic responses
obtained by using only a single mode and all 20 modes are compared in Figure 14. Since
the first mode is well separated, the response can be obtained quite accurately by using
only a single mode. Comparison of the computational times spent in one iteration when
using one mode and all 20 modes is given in Table 3.
Consider a two d.o.f. system as shown in Figure 15. The undamped linear natural
frequencies of the system are v1=11·9 rad/s and v2=32 · 53 rad/s. Figure 16 shows a
comparison of the amplitude of the first harmonic component of the multi- and
fundamental harmonic solutions. The amplitude of the third harmonic component of the
response is also given in the figure. The third harmonic component possesses two peaks,
one at 4 rad/s and the other at around 12 rad/s. The frequency corresponding to the peak
at 4 rad/s is 12 (=3×4) rad/s which is the first natural frequency of the undamped linear
part of the system. However, this peak does not contribute to the total response
significantly. The second peak around 12 rad/s corresponds to 36 (=3×12) rad/s which
is close to the second natural frequencies of the undamped linear part of system. The peak

Figure 9. Twenty d.o.f. cantilever beam (distributed non-linearity).


    -  331

19 ) of the system in Figure 9 around first resonance as obtained by using


Figure 10. Non-linear response (X
q and 20 (—×—) modes (low to high frequency sweep, distributed cubic stiffness).
one (−−)

is 20% of the fundamental harmonic component. Hence, it can be concluded that


distribution of the modes of the system plays a very important role in the analysis.

4.4.        


The fundamental harmonic response of a five d.o.f. system (Figure 17) is now compared
with the time domain solution in respect to accuracy and computatonal effort. Cubic
stiffness type non-linearity (cs=2·5×106 N/m3, F1=1 N) is considered between the first
co-ordinate and the ground. Figure 18 shows the frequency response (X 1 ) of the system

Figure 11. As Figure 10 but high to low frequency sweep.


332 .   . . ̈̈

19 ) of the system in Figure 1 around first resonance (distributed


Figure 12. Linear and non-linear response (X
cubic stiffness). Key as Figure 7.

around the first two resonances as obtained by using two modes of the linear part of
system. The response of the system obtained by numerical integration is shown in the same
figure as well. The steady state response of the system was evaluated by the Newmark-b
method which is the conventionally used method for solving non-linear dynamic problems
by a time marching procedure. The time increment taken for this particular problem is
about 1/40th of the period corresponding to the highest undamped natural frequency of
the system. Figure 18 shows that there is a very good agreement in the fundamental
harmonic response obtained by the method suggested and the time integration results.

Figure 13. Results of multi- (——) and fundamental harmonic (· · · ·) analyses of the system in Figure 1
(obtained by using 20 modes). - - - -, Linear solution.
    -  333

19 ) of the system in Figure 1 obtained by using one (—×—) and


Figure 14. Multi-harmonic response (X
q modes.
20 (−−)

Furthermore, it is observed that the computational time for the Newmark-b method in
obtaining the steady state response is extremely high. The comparison of computation
times for the fundamental harmonic response by using the method suggested with two
modes, all five modes and the Newmark-b method, respectively, as given in Table 4.
As the last case study, a two d.o.f. system is considered (see Figure 15). Linear system
parameters are chosen such that undamped linear natural frequencies are calculated as
v1=10·78 rad/s, v2=32·34 rad/s (=3×v1 ). A cubic stiffness type non-linearity
(cs=1×106 N/m3 ) is taken between the first co-ordinate and the ground, and the system
is excited at 10·78 rad/s (F1=1 N). Figure 19 shows the steady state linear, single harmonic
(non-linear), multi-harmonic (non-linear) and time domain (non-linear) solutions. At this
particular frequency, the accuracy of the response prediction by employing single harmonic
analysis drops due to violation of the first condition given in section 2.1. The excitation
frequency coincides with integer multiples of resonances (v1=1×v, v2=3×v). Hence,
a maximum error between the single harmonic and time domain solutions is found to be
25%. If multi-harmonic analysis is performed, the maximum error drops to 7% when the
third harmonic component is included in the calculations.

T 3
Computational time comparison for multi-harmonic and fundamental harmonic analysis
Analysis Number of modes considered Time for one iteration (s)
Multi-(H) 1 12
Multi-(H) 20 160
Fundamental H 1 1
Fundamental H 20 70
334 .   . . ̈̈

Figure 15. Two d.o.f. non-linear system.

5. CONCLUSIONS
In this paper, a semi-analytical frequency domain method has been presented for the
forced harmonic response analysis of non-linear structures. The method can be used as
an alternative to numerical integration procedures, which are very expensive, and to other
non-linear analysis methods which are restricted to a few degrees of freedom. Application
of the method for virtually any kind of non-linearity is possible and it reduces the
computational effort considerably.
The basic theory is based on the describing function approach which has been frequently
used to solve many non-linear control engineering problems. The non-linear ordinary
differential equations of motion are converted into a set of non-linear algebraic equations
which should be solved simultaneously. The number of equations to be solved is reduced
by making use of the modes of the linear part of the structure of interest. The resulting
coupled algebraic equations are solved iteratively. In order to predict multiple solutions
and to increase the speed of computations, a special solution technique is employed.

Figure 16. Multi-harmonic (M.H.) and single harmonic (S.H.) solutions of the system in Figure 15. ,
M.H., first harmonic; · · · ·, S.H. ——, M.H., third harmonic.
    -  335

Figure 17. Five d.o.f. non-linear system.

Jump behavior as a typical non-linear frequency response phenomenon can be fully


simulated by using the technique. Reduction in the number of equations by using principal
co-ordinates provides drastic drops in the computational time. Significant savings can be
obtained even for structures with a relatively small number of degrees of freedom.
Furthermore, reducing the number of modal co-ordinates and discarding the insignificant
higher modes of the linear part of the structure does not diminish the accuracy of the
response predictions. While some of the non-linear vibration analysis methods given in
the literature are developed only for structures having localized non-linearities [11], the
method proposed in this study can analyze structures having distributed and localized
non-linearities with similar computational effort.
The internal non-linear forces are expressed in matrix form by using describing
functions. This makes it possible to define the pseudo-receptance matrix in harmonic
motion analysis. It is important to use the receptance matrix concept in non-linear
vibration analysis since this provides compatibility with the well developed standard linear
modal analysis procedures. In modal testing which employs frequency response functions,
the effects of non-linearities on the linear identification procedures can be investigated [31].
The concept of receptances can be used in non-linear structural coupling, modification
analysis, detection and identification of non-linearities.

Figure 18. Fundamental harmonic and time domain solutions of system in Figure 17. , I.M.M.; *, time
domain; - - - -, linear.
336 .   . . ̈̈
T 4
Time comparisons with time domain analysis (based on one
frequency)
Analysis Time (s)
Fundamental H (two modes) 2
Fundamental H (five modes) 14
Newmark-b 1100

The case studies presented show that higher harmonic components of the response are
usually much smaller in amplitude compared to the fundamental harmonic component,
but neglecting their presence can decrease the accuracy in certain frequency ranges which
are determined by the higher modes of the structure. It is observed that the computation
time required for obtaining a solution with a defined accuracy, in general, is increased as
more harmonics are included in the analysis. Thus, a multi-harmonic analysis can be very
costly for structures with a very large number of degrees of freedom and should be
attempted only if multiples of the excitation frequency coincide with a resonance of the
structure.
The accuracy of the method was investigated by comparing the results with a classical
time domain analysis technique considered as an exact analysis. It is concluded that very
accurate results can be obtained by the method suggested. In addition, drastic reductions
in the computational time (which is a critical factor in the analysis of structures with a
large number of degrees of freedom) can be obtained.
It is believed that the method proposed will form the basis of the harmonic vibration
analysis of non-linear structures in further studies.

Figure 19. Linear, single harmonic, multi-harmonic and time domain solutions of the system in Figure 15.
    -  337
REFERENCES

1. M. A. D and K. S 1989 Computers and Structures 32, 1371–1386. A survey of
direct time-integration methods in computatonal structural dynamcs: I, explicit methods.
2. M. A. D and K. S 1989 Computers and Structures 32, 1387–1401. A survey of
direct time-integration methods in computatonal structural dynamcs: II, implicit methods.
3. K. W and H. S 1988 Journal of Vibration, Acoustics, Stress and Reliability in Design
110, 410–411. A modal analysis approach to non-linear multi-degrees-of-freedom system.
4. K. W and H. S 1988 Journal of Vibration, Acoustics, Stress and Reliability in Design
110, 36–41. Development of non-linear building block approach.
5. K. M and H. S 1990 Journal of Vibration and Acoustics 112, 508–514. Vibration
characteristics of a beam with support accompanying clearance.
6. Y. C and S. T. N 1987 Journal of Vibration, Acoustics, Stress and Reliability in Design
109, 255–261. Non-linear steady-state response of a rotor-support system.
7. C. N and H. D. N 1989 Journal of Vibration, Acoustics, Stress and Reliability in
Design 111, 187–193. Periodic solutions in rotor dynamic systems with non-linear supports: a
general approach.
8. W. O 1989 Computers and Structures 33, 851–858. Forced vibrations of a beam
including dry friction dampers.
9. W. O 1989 Journal of Sound and Vibration 131, 465–473. The harmonic balance
method for determining the vibration parameters in damped dynamic systems.
10. W. O 1990 Computers and Structures 30, 721–728. A discrete linear beam model to
investigate the non-linear effects of slip friction.
11. T. N. S and A. N. J 1990 Journal of Vibration and Acoustics 112, 501–507. Prediction
of periodic response of flexible mechanical systems with non-linear characteristics.
12. A. A. F and E. H. D 1988 Journal of Sound and Vibration 124, 207–224. Frequency
domain solutions to multi-degree-of-freedom dry friction damped systems.
13. E. B 1989 M.Sc. Thesis, Middle East Technical University, Ankara, Turkey. Dynamic
analysis of non-linear structures for harmonic excitation.
14. E. B and H. N. Ozgüven 1990 Proceedings of the 15th International Seminar on Modal
Analysis, Leuven 2, 901–915. A method for harmonic response of structures with symmetrical
non-linearities.
15. E. B and H. N. O̈ 1991 Proceedings of the 4th International Conference on Recent
Advances in Structural Dynamics, Southampton, 791–800. Harmonic vibration analysis of
structures with local non-linearities and validity assessment of some linearization techniques.
16. E. B and H. N. O̈ 1993 Mechanical Systems and Signal Processing 7, 75–87.
Iterative receptance method for determining harmonic response of structures with symmetrical
non-linearities.
17. H. N. O̈ 1987 Journal of Sound and Vibration 117, 313–328. A new method for harmonic
response of non-proportionally damped structures using undamped modal data.
18. O. T, B. K, H. N. O̈ and M. I ̈ 1993 American Institute of
Aeronautics and Astronautics Journal 31, 1313–1320. Forced harmonic response analysis of
non-linear structures using describing functions.
19. A. K. N 1981 Computers and Structures 13, 31–44. Recent advances in reduction methods
for non-linear problems.
20. A. J. M, K. M. V and C. W. G 1976 Journal of Pressure Vessel Technology 98,
151–156. Application of normal mode theory and pseudo force methods to solve problems with
non-linearities.
21. V. N. S, G. J. B and A. N. N 1979, Journal of Pressure Vessel Technology
101, 134–141. Modal superposition method for computationally economical non-linear
structural analysis.
22. J. A. S and W. E. H 1977 Computers and Structures 7, 125–136. Formulations
and solution procedures for non-linear structural analysis.
23. K. B and S. G 1981 Computers and Structures 13, 699–707. On non-linear dynamic
analysis using substructuring and mode superposition.
24. R. S, R. B. B and T. S. S 1989 Journal of Vibration, Acoustics, Stress and
Reliability in Design 111, 360–365. Dynamic response of rotors using modal reduction
techniques.
25. C. J. C and B. M 1990 Computers and Structures 36, 1067–1080. Modal analysis of
non-linear systems with classical and non-classical damping.
338 .   . . ̈̈
26. D. P. A 1975 Non-linear Control Engineering. New York: Van Nostrand Reinhold.
27. A. G and W. E. V V 1968 Multiple-input Describing Functions and Non-linear
System Design. New York: McGraw-Hill.
28. O. T 1991 M.Sc. Thesis, Imperial College of Science, Technology and Medicine,
London. Forced periodic response analysis of non-linear structures for harmonic excitation.
29. B. K 1992 M.Sc. Thesis, Middle East Technical University, Ankara, Turkey. Dynamic
analysis of harmonically excited non-linear structures by using iterative modal method.
30. R. E. M 1984 Journal of Sound and Vibration 94, 456–460. Comments on the method of
harmonic balance.
31. H. N. O̈ and I. I ̈ 1993 International Journal of Analytical and Experimental Modal
Analysis 8, 151–164. Complex modes arising from linear identification of non-linear systems.

APPENDIX A: HARMONIC INPUT DESCRIBING FUNCTIONS


Mathematical expressions and the corresponding harmonic input describing functions
for the non-linearities considered in this paper are listed below.
Cubic stiffness:
n=csy 3, n=34 csY 2. (A1, A2)
Piecewise linear stiffness:
n=k1 y, =y=Qd, n=(k1−k2 )d+k2 y, =y=ed. (A3a, b)

$ 0 1 0 1X 0 1 %
d d d
2
2(k1−k2 )
Qd,
n=k1 , Y n= arcsin + 1− +k2 , ed.
Y
p 
Y 
Y 
Y

(A4a, b)
Coulomb damping:
n=a sgn (ẏ), n=i4a/pY
. (A5, 6)

APPENDIX B: NOMENCLATURE

(Akj )m complex amplitude of internal non-


linear forces corresponding to nkj and terms of modal co-ordinates
mth mode [C] linear viscous damping matrix
brj phase angle of inter-co-ordinate dis- [D] non-linearity matrix
placement yrj [F] mass normalized modal matrix
grjk generalized non-linear coefficient [H] linear structural damping matrix
i unit imaginary number [H]p proportional part of the linear struc-
n number of degrees of freedom tural damping matrix
nrj non-linear restoring force element [H]n non-proportional part of the linear
nrj describing function corresponding to nrj structural damping matrix
p proportionality constant [I] identity matrix
prjk kth power stiffness coefficient between [K] linear stiffness matrix
rth and jth co-ordinates [M] linear mass matrix
c generic angle [U] pseudo-receptance matrix of the struc-
qrjk kth power damping coefficient between ture
rth and jth co-ordinates [U*] pseudo-receptance matrix of the struc-
t time variable ture defined in terms of modal co-ordi-
v excitation frequency nates.
vr rth natural frequency {D(ẋ)} non-linear damping vector
yrj inter-co-ordinate displacement {h0 } complex amplitude vector of modal
(Ykj )m complex amplitude of (ykj )m co-ordinates
[a] linear receptance matrix {f } external forcing vector
    -  339
{F} complex amplitude vector of external
forcing {ẍ} vector of accelerations
{f r} rth eigenvector {X} complex amplitude vector of displace-
{g} vector of internal non-linear forces ments
{G} complex amplitude vector of internal {x}r rth harmonic component of the re-
non-linear forces sponse
{S(x)} non-linear stiffness vector {X}r complex amplitude vector of the rth
{x} vector of displacements harmonic component

Вам также может понравиться