You are on page 1of 11

Available online at www.sciencedirect.

com

Computers and Structures 85 (2007) 1858–1868


www.elsevier.com/locate/compstruc

Hysteretic damping of structures vibrating at resonance:


An iterative complex eigensolution method based
on damping-stress relation
George D. Gounaris, Eleftherios Antonakakis, Chris A. Papadopoulos *

Machine Design Laboratory, Mechanical Engineering Department and Aeronautics, University of Patras, 26500-Patras, Greece

Received 4 September 2006; accepted 19 February 2007


Available online 16 April 2007

Abstract

In this paper the damping is examined as an engineering property used in analysis and design of structures and machines. The design
engineer needs to know not only the stresses of his structure or machine, under steady state conditions but also the stresses under res-
onance conditions. Then the material damping, as a function of the stress of the structure, has an important role to play and ignoring the
damping the calculated stresses are far from reality. The nonlinearity here is due to the dependence of the hysteretic damping on the stress
of the structure. Specifically here two problems are investigated in the following way:
Firstly the direct problem is solved. The direct problem is to find the maximum bending stress at the resonance when the relation of
the dissipating energy (or of the hysteretic damping) vs. the bending stress is known in advance. To perform this calculation, a useful tool
for the design engineer, the structure is modelled using the continuum mechanics analytical approach or the finite elements (FE) method.
Then the eigenvalues are calculated and using an iterative procedure the real stress. The procedure presented here is called iterative com-
plex eigensolution method (ICEM). Secondly the inverse problem is solved. The inverse problem is to find the relation between the hys-
teretic damping and the bending stress. For this purpose the logarithmic decrement is experimentally measured, the eigenvalues and the
maximum bending stress of the structure, excited at the eigenvalue, when the damping is the same as the measured one, are computed
using the finite elements method. Once the bending stresses are found in each discrete element of the structure, then the mathematical
expression of the relation of the dissipating energy and the stresses can be specified by minimizing a suitably formed objective function.
 2007 Elsevier Ltd. All rights reserved.

Keywords: Hysteretic damping; Damping-stress relation; Damping at resonance; Iterative complex eigensolution method (ICEM); Damping auto-
determination; Hysteretic; Damping; Stress; Resonance; Eigensolution

1. Introduction extremely useful studying i.e. the seismic response of struc-


tures in the time domain. For completeness purposes these
1.1. Hysteretic damping models models are presented here.
Non causal constant hysteretic model: The constant
Several models for the hysteretic damping are intro- hysteretic model is a pathological model because it is
duced last decade in order to find the response in the time non-causal. This means that the model responds prior to
domain viscoelastic linear analysis, besides the well known the application of an impulsive excitation.
Maxwell, Kelvin–Voigt and their different combinations x h xi
such as the standard linear solid model. These models are GðxÞ ¼ G1 þ iG2 sgn ¼ G1 1 þ icsgn ; ð1Þ
e e
where G1 is the frequency-independent storage modulus,
*
Corresponding author. Tel.: +30 261 096 9426; fax: +30 261 099 6258. G2 sgnðx=eÞ is the loss modulus, x is the frequency variable
E-mail address: chris.papadopoulos@upatras.gr (C.A. Papadopoulos). and e is an arbitrary positive real number with units of rad/s

0045-7949/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2007.02.026
G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868 1859

that is present to make the sign function dimensionless, sensitivity-based methods [8] solve the nonlinear eigenvalue
c ¼ G2 =G1 the material loss factor [1]. problem using iterative techniques based on the eigenfre-
Biot hysteretic model: Biot [2] since 1958 introduced the quency convergence.
hysteretic model represented in the frequency domain by Present model: iterative complex eigensolution method
the following form: (ICEM) based on damping-stress relation: The method pre-
" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi #
2c x2ffi 2c   sented here could be defined as an iterative complex eigen-
1 x solution method based on the nonlinear relation between
GðxÞ ¼ G1 1 þ ln 1 þ þ i tan : ð2Þ
p e p e damping and stress. This method is used in the frequency
domain, since from the point of view of the designer the
The Biot hysteretic damping model is also used for the
knowledge of the steady state at the resonance is the
deterministic and stochastic analysis of nonlinear systems.
subject.
More specifically Spanos et al. [3] use the Biot hysteretic
model involving an infinite collection of elements whose
1.2. Hysteretic damping-stress relation at the resonance
dynamic characteristics are specified through a probability
density-like function. The numerical results are further
It is well known that hysteretic damping affects the
used to assess the accuracy of a statistical linearization pro-
dynamic behavior of the structures in the resonance zone
cedure adopted in determining the response of the hysteret-
where maximum stress is expected. It is also known that
ically damped system to white noise.
the hysteretic damping as well as the dissipating energy
Causal hysteretic damping: The dynamic stiffness of the
are both functions of the stress in the nonlinear zone
causal hysteretic model introduced by Makris [4] can be
[10,11]. This dependency becomes clear in Fig. 1.
described by
 On the other hand, the stress (only bending stress is con-
2c x x
sidered for most of engineering problems) depends on
GðxÞ ¼ G1 1 þ ln   þ icsgn ; ð3Þ
p e e damping. As it is obvious, when designing, a dynamically
loaded structure, it is necessary to know in advance the
where c ¼ G2 =G1 a positive real number named hysteretic
maximum stress developed in the structure under various
damping coefficient, and x is the frequency variable [4].
loads. Consequently it is necessary to know the relation-
Laguerre polynomial approximation (LPA): The
ship between damping and bending stress as it was given
Laguerre polynomials are used for the state space formula-
tion of linear viscoelastic systems by Palmeri et al. [5,6]. In
these papers, a state space formulation of viscoelastic
systems with memory is proposed, which overcomes the
concept of memory by enlarging the state vector with a
number of internal variables that bear the information
about the previous history of the system. First a state space
representation of the generalized Maxwell model is shown,
then a new state space model is presented in which the
relaxation function is approximated with Laguerre polyno-
mials. This approach is extended [6] to single-degree-of-
freedom systems subjected to random, nonstationary
excitation.
Modal strain energy method (MSE): The modal strain
energy method predicts damped vibrations behaviour by
employing eigenvectors fUg of the associated undamped
structure [7,8]. If [D] and [K] are the damping and the stiff-
ness matrices of the structure then the damping loss factor
ci that corresponds to the ith vibration mode can be calcu-
lated from the ratio of the dissipated ED over the strain
energy ES:
T
ED fUgi ½DfUgi
ci ¼ ¼ : ð4Þ
ES fUgTi ½KfUgi
MSE method is also used to predict the damping in struc-
tures with constrained viscoelastic layers [9] from un-
damped normal mode results.
Complex eigensolution methods (CEM): These are meth-
ods based on the complex formulation of the eigenvalue Fig. 1. (a) Damping energy and (b) loss coefficients curves for mild steel
problem. The modified CEM as well as the complex eigen- [10] (in this figure of Lazan g instead of c is used).
1860 G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868

by Lazan [10] for the mild steel. But since damping and viscous damping theoretically and under the assumption
bending stress are interconnected, a new method, also that the eigenvalues and eigenvectors are known.
called auto-determination method, was proposed by Goun- Gounaris et al. [13] have solved the inverse problem for
aris et al. [12,13] and is based on successive iterations using the hysteretic damping, in a floating Timoshenko beam,
proposed and calculated damping until the two values of moving forward and excited by waves (simulating a ship).
damping converge. In [13] the steady state equilibrium is In this work it is supposed that the bending stresses are mea-
achieved by taking all possible loops of damping and sured near to the resonance under the excitation of the
adopting the objective function to be minimized. In the waves and it is observed that (a) measuring of stress distri-
same work a method called predictor–corrector, based on bution is not an easy task, especially regarding to the irreg-
neural networks is also adopted for solving the inverse ularity of wave loading and (b) Since the exciting frequency
problem. Besides this method was proved to be an intelli- is not at the resonance, but near to the resonance point, the
gent and fast method that minimizes the time of damping coefficients are different. This difference is smaller
calculations. than this obtained directly from the damping-stress dia-
Kume et al. [14] introduced a damping-stress diagram gram. This is due to bad sensitivity, since small differences
for a cantilever beam. When the maximum bending stress in the damping of elements correspond to the same stress
is known, then the damping may be estimated using this distribution. The above remarks lead to the conclusions
diagram. The same could be applied in each vibration that experiments must be done at exactly the resonance
mode, under steady state equilibrium, i.e. equal input and point, to measure the total loss hysteretic damping coeffi-
calculated damping. Kume et al. transformed the diagram cient, by measuring the logarithmic decrement in free decay
taken from Lazan [10] of dissipating energy vs. maximum or the half power bandwidth method. In these experiments
bending stress, to a diagram of total loss coefficient vs. the total loss coefficient vs. the maximum stress at the
maximum bending stress. Thus the steady state equilibrium clamped end of a cantilever beam, have to be determined,
(input damping = calculated damping) and thus maximum as it was given by Kume et al. [14]. A general method must
stress in each element is obtained as presented in [12,13,15] be independent of the boundary conditions. In this direc-
with various methods, at the resonance zone. tion the applicability of the method in cases with different
In these previous works the structure was excited near to boundary conditions has also to be investigated.
the resonance (but not exactly at the resonance). Obviously In this work a cantilever beam is investigated experimen-
in these previous works, the more the structure is excited tally as will be described later. The measured loss factor
near to the resonance frequency, the better of accuracy coefficients are combined with FEM codes to give the stres-
results are obtained. Thus the necessity to compute the pre- ses based on the damping measurements.
cise complex eigenvalues and correlate the corresponding These codes are extended here in order to solve the
response with that of the experiment at exactly the reso- inverse problem. The inverse problem is defined as the
nance frequency arises. determination of the relation between dissipating energy
In the present paper the complex eigenvalue is com- and bending stresses at various points of the elements,
puted, the excitation frequency causes the vibration of overcoming thus the boundary condition restrictions.
the structure exactly at the resonance and the correspond- The effect of the hysteretic damping is taken into
ing response influenced by the converged hysteretic damp- account here by assuming a complex valued modulus of
ing is also calculated. The experimental measurements, that elasticity as proposed by Lazan [10] and Nashif [11]. The
will be presented later, give good correlation with the pre- eigenvalues and responses are computed with the new
sented method. method proposed here as will be explained and validated
The method here presupposes that the relation between later.
dissipating energy or damping vs. bending stress is known Bishop and Price [19] described the eigenvalues compu-
in advance, as taken from Lazan [10]. In this work the dia- tation that was based on theoretical models and for viscous
gram taken from the experimental work performed by damping. Using suitable software, as for example EIS-
Lazan and Demer [10,16] in mild steel is given (see Fig. 1). PACK package, all eigenvalues are to be computed
Since the damping depends not only on the material whereas in this work only the eigenvalue or response of
quality but also on the bending stress distribution [10], interest is computed.
and the bending stress distribution is affected from the
stress concentration due to discontinuities like welds, rivets
etc., as referred by [13], the necessity of a new approach to 2. Mathematical model
this topic, with new experiments in connection with new
theory applications, becomes now necessary. 2.1. Euler–Bernoulli formulation
In this direction Dovstam and Dalenbring [17] deter-
mined experimentally, by solving the inverse problem, the The well known beam theory of Euler–Bernoulli equa-
material viscous damping, frequency dependent, in a plate tion (5) is used here. The hysteretic damping c is also con-
numerically using neural networks. Lancaster and Prells sidered and its effect is introduced by taking the modulus of
[18] solved this inverse problem for the determination of elasticity in the generalized standard complex form
G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868 1861

Eð1 þ icÞ, where E is the static modulus of elasticity and is By substitution of (7) and (8) into (6) the following
assume to be equal to the dynamic one. equation is obtained in terms of A, B, C and D:
As it is referred, this model could be also extended to
ðA þ iBÞðC þ iDÞ þ 1 ¼ 0
structures with nonlinear behavior. The hysteretic damping
is not frequency dependent and the complex modulus of or ðAC  BD þ 1Þ þ iðAD þ BCÞ ¼ 0; ð9Þ
elasticity comprises an acceptable model for analytical pur- since now above frequency equation is a complex one, for the
poses even for nonlinear materials [10]. To explain more same reasons as explained below in Section 2.3 in the case of
this let us consider the Fig. 1 [10] where the dissipating FEM, at the case of resonance the value obtained by (9)
energy and the loss coefficient are presented for mild steel. should be the minimum and those will be the eigenvalue
Lazan approximates in three segments the damping and the ðkn ¼ An þ iBn Þ [19]. But to do so, the real and the imaginary
damping energy as functions of stress: the region of low, of part separately should be minimised at the same time. This
intermediate and of high stress. In this paper our interest is means that above Eq. (9) should now be transformed to give
on the second region, that of intermediate stress, where the the minimum at the resonance i.e. at the requested pairs of
relation between damping and stress is nonlinear. eigenvalues Ae and Be by iterations as follows:
Besides, the model used is to be experimentally con-
firmed later in this paper. jAC  BD þ 1j þ jAD þ BCj ! min : ð10Þ
The governing equation is The results of this approach are included in Table 2, under
4 2 the title ‘‘1st method: analytical eigenvalues calculation
o yðx; tÞ o yðx; tÞ
E I þ qA ¼ F ðx; tÞ; ð5Þ based on Eq. (10)’’. In this table the two first complex
ox4 ot2
eigenvalues are calculated in the form kn ¼ An þ iBn . The
where E ¼ Eð1 þ icÞ the complex modulus, I is the second- corresponding loss factor at the steady state is also pre-
ary moment of inertia, yðx; tÞ is the vertical displacement, q sented in the second column in connection with different
the mass density of the material, A the area of the cross sec- applied forces at the free end of a cantilever beam (first col-
tion and F is the applied force. umn of Table 2).

2.2. Free vibration problem – analytical computation of the 2.3. Finite elements complex formulation of the
eigenvalues non-homogeneous equation

By solving the Bernoulli equation (5) for a cantilever In the previous paragraph the complex eigenvalues of
beam when F ðx; tÞ ¼ 0 the following frequency equation the homogeneous form of Eq. (5) are computed for the case
is obtained [7,14]: of a cantilever beam. Using the well known finite element
rffiffiffiffiffi ! rffiffiffiffiffi ! formulation described analytically in [12] the following
ke ke
cosh L cos L ¼ 1; ð6Þ equation is obtained:
c c
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðK  k2n MÞfYg ¼ fFg; ð11Þ
where c ¼ Eð1 þ icÞI=l, ke ¼ Ae þ iBe , with Ae the real

part and Be is the imaginary part of the complex eigenvalue where K ¼ ð1 þ icÞK is the global complex stiffness matrix,
corresponding to the e mode and L = length of the beam, M the global mass matrix, and kn ¼ An þ iBn the modal
l = linear density, and I = moment of inertia. eigenvalues where An is the nth modal damping and Bn is
Since in the above Eq. (6) the circular and hyperbolic the nth eigenvalue. The matrices K and M are given, in de-
functions have complex arguments, after algebraic substi- tails in Refs. [12,13,20,21] for Timoshenko beams. After
tutions and manipulations, solutions with real arguments substitutions and operations in Eq. (11), and by defining:
at the above circular and hyperbolic functions are obtained
in the form K1 ¼ K þ ðA2n þ B2n ÞM; and K2 ¼ 2An Bn M þ cK ð12Þ
A B
rffiffiffiffiffi ! zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ and since {Y} is a complex, in the form Y ¼ YRe þ iYIm Eq.
ke cosðLdÞ Y 1 sinðLdÞ Y 1
cosh L ¼ ðe þ eY 1 Þ þi ðe  eY 1 Þ ¼ A þ iB (13) is obtained:
c 2 2  

 K1 K2  YRe
ð7Þ   ¼ fFg: ð13Þ
K K1  YIm
C D 2
rffiffiffiffiffi ! zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
ke cosðLgÞ Y 2 sinðLgÞ Y 2 From Eq. (13) it seems that, resonance occurs when the
cos L ¼ ðe þ eY 2 Þ þi ðe  eY 2 Þ ¼ C þ iD;
c 2 2 respective characteristic matrices take minimum value
[19,22]. To this respect a code was written for searching
ð8Þ
p4
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi out for the minimum of the determinant in Eq. (13) for var-
where Y 1 ¼ q4 L sin u3 , Y 2 ¼ q4 L cos u3 , q4 ¼ q1 q2 l=ðEIÞ, ious eigenvalues pairs (Ai, Bi). In addition it must be
pffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffi pffiffiffi
q1 ¼ q0 , q ¼ 1 þ c2 , q2 ¼ 4 q, q0 ¼ A2e þ B2e , u3 ¼ pointed out that the hysteretic damping does not influence
u1  u2 , u1 ¼ ðw þ 2epÞ=2, u2 ¼ ðu þ 2epÞ=4, tan gw ¼ the imaginary part of the eigenvalue [11] i.e. the eigenfre-
Be =Ae , tan gu ¼ 1c ffi c, g ¼ q4 sin u3 , d ¼ q4 cos u3 . quency and this is also confirmed in this work.
1862 G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868

This Eq. (11) could be used in the following ways: Y ¼ Y 0 eT ðAþiBÞ ;
where e = the mode number, Y0 = the displacement at time
a. Free vibration (F = 0): The homogeneous form of Eq.
t at the free end, A þ iB is the first complex eigenvalue,
(11) is solved giving thus the complex eigenvalues of
Y = the displacement at the free end and at time t þ T ; T
the structure in the form kn ¼ An þ iBn . These eigen-
is the period of vibration. Then:
values are presented in Table 2 under the title ‘‘2nd
method: FEM eigenvalues calculation based on Eq. 1 iBT
Y ¼ Y0 e
(13)’’. eAT
b. Forced vibration (F 6¼ 0): The cantilever beam is and ignoring the complex term the following relation is
vibrating under the excitation of an applied force at obtained:
its free end. The excitation frequency is exactly the
imaginary part of the calculated eigenvalue Bn. The Y0 Y 1 Y
Y ¼ AT () ¼ AT () AT ¼ ln :
modal response (response of the structure excited at e Y0 e Y0
the eigenfrequency) and the maximum bending stress, Since by definition it is known that the logarithmic decre-
at the steady state vibration, is calculated. At this max- ment in the period T is
imum bending stress the hysteretic damping is calcu-
lated [12,13,15] based on the knowledge of the Y0
d ¼ ln or AT ¼ d:
damping-stress relation [10]. The solution of Eq. (11) Y
is then repeated using the new damping values until And since we know that T ¼ 2p=B then d ¼ 2pA=B
introduced and computed damping are equal, within which is the same equation as that given by Sekhar [23]
given accuracy. The method described here is called or A ¼ Bd=ð2pÞ and since in all above computation when
iterative complex eigensolution method (ICEM). finding eigenvalues, was demonstrated in all cases that:
Bc Bd d
The modal response, calculated for the first time, using A¼ ¼ () c ¼ ð14Þ
the ICEM when introduced and calculated damping are 2 2p p
(almost) equal, will serve in order to: that is also referred by other investigators. Therefore we
have found an explicit relation (14) which also gives the
• Compare the difference between, maximum bending real part of the eigenvalues when eigenfrequency and total
stress when computed at the complex eigenvalues loss damping coefficient are known in advance. Thus Table
(including damping) and at the eigenvalues (i.e. neglect- 2 is completed by the ‘‘3rd method: eigenvalues calculation
ing damping). based on Eq. (14)’’.
• Compare the diagrams of total loss coefficients vs. the Since the eigenvalues are confirmed to have nearly the
maximum bending stress diagrams computed with those same values for the numerical model using FEM and for
obtained by experiments described in Section 4.1. Since the analytical model presented in Section 2.2, having intro-
these experiments are done at the eigenmodes, the com- duced the damping in these computations using the com-
putation is necessary to be done at the eigenvalues. To plex modulus of elasticity (generally suitable for linear
this respect, such above diagrams taken by computa- materials), then it should be concluded that the restriction
tions or experiments are valid and this is irrespective on the linear material damping for analytic purposes,
of specimen’s dimensions [12,14]. should not be a prerequisite for this case [10].
• Solve the inverse problem when extended as will be Since the analytical eigenvalue computation, presented
explained later. The solution of the inverse problem here in Section 2.2, the finite elements complex formulation of
is referring to the fact on finding the new dissipating the non-homogeneous equation of Section 2.3 give the
energy-bending stress function for specimens of other same results as shown in Table 2 with the closed form cor-
materials or specimens affected by stress concentrations relation of the present Section 2.4, also called as empirical
(welding, riveting, cracks etc.). The experimental way one, the new proposed methods are supposed to be
for the construction of the diagram, of total loss coeffi- validated.
cient vs. maximum bending stress, shall be described in
Section 4.1. This new proposal, for the reasons 2.5. Steady state response at the eigenvalues and loss
explained in the introduction, is considered to be more coefficient computation
effective as easier to be done.
For the cantilever beam excited by a vertical force F at
the free end, Eq. (13) is solved, and the displacements,
2.4. Closed form correlation between eigenvalues and total the bending stresses and the eigenvalues are calculated, at
loss coefficients the last iteration when the introduced and the calculated
damping are almost the same. For the computation of
From the theory of Bishop and Price [19] it is known the bending stress using FEM, details are given in
that for the e mode, Y ¼ Y 0 eT ðAe þiBe Þ and for the 1st mode [12,13,20,21].
G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868 1863

For this purposes the damping (total loss coefficient) is evaluation of the total loss factor c and the loss factor
taken by definition [11]: for each element ce.
This was done here for a beam of orthogonal section of
1 W
c¼ ; ð15Þ length L and height 2h. Employing above Eqs. (15) and
2p U
(18) the total loss factor takes the form
where W is the energy loss per cycle for the entire volume V  B22  D1
RL Rh
of the beam A1 ri y
þ C 1 hr ri y
dy dx
0 0 hrf
Z c¼ R L R h ri y 2
f
; ð20Þ
p
W ¼ f ðrs ÞdV ð16Þ E 0 0 h
dy dx
V

and U is the strain energy for cycle and the entire volume V where ri is the modulus of the maximum bending stress
of the beam at the middle of the element i and y is the distance from
Z 2 the neutral axis (A1 ¼ 6895; B22 ¼ 2:3; C 1 ¼ 41350, and
1 rs dV D1 ¼ 8Þ.
U¼ : ð17Þ
2 V E Performing appropriate integrations and accounting for
the discrete form of the structure, the loss factor gets the
The dissipating energy function f ðrs Þ is given in form:
[10,12,14] by experiment for the mild steel as
B1 D1 P
n
rs rs 3E
ðA3 S Bem3 þ C 3 S Dem3 Þ
f ðrs Þ ¼ A1 þ C1 ; ð18Þ p
em¼1
rf rf c¼ ; ð21Þ
P
n
r2em
where rf is the fatigue limit stress of the material, whereas em¼1
A1 ¼ 6895; B1 ¼ 2; C 1 ¼ 41350 and D1 ¼ 8 (constants to
change the Unit System from English to International). where S em ¼ rem =rf , rem is the modulus of the maximum
Using Eq. (18) the following two different methods are bending stress at the middle of the element e, n is the total
applied to compute the first modal damping for a mild steel number of discretized elements, (A3 ¼ 2089:5; B3 ¼
cantilever beam. The total loss coefficient, calculated using 2:3; C 3 ¼ 4595:5, and D3 ¼ 8).
both methods for comparison purposes, at the steady state Eq. (21) relates the total loss factor to the bending stres-
equilibrium of the damping, with the same force F at the ses developed in the structure. This equation follows from
free end. the definition of the total loss factor Eq. (15) assuming that
In the first method proposed by Kume et al. [14] the fol- strain energy and unit dissipating energy may be approxi-
lowing relationship for the total loss factor c is obtained mated by summing up corresponding energies for each ele-
[12,14] ment of the structure. Since it is assumed that each element
! of the structure has its local loss factor, it may be evaluated
rBm21 rDm21 for each element accounting on individual energy levels of
c ¼ E A2 B22 þ C 2 D22 ; ð19Þ
rf rf each one element. Thus

where A2 ¼ 2144:8, B21 ¼ 0:3, B22 ¼ 2:3, C 2 ¼ 8231, 3E A3 S Bem3 þ C 3 S Demn


3
ce ¼ ; ð22Þ
D21 ¼ 6, D22 ¼ 8 p r2em
In this equation, E is the dynamic modulus of elasticity
that is assumed to be constant as happens in most of engi- ce being here the loss factor for each element alone of the
neering applications. By rm the maximum bending stresses discretized structure.
attained at clamped end is denoted. Eq. (19) results from The developed stresses in the structure depend on the
analytical point of view where mode shapes and unit dissi- loss factor. The loss factor controls maximum amplitudes
pating energy function (18) are known in closed form and of vibration and of stresses. These above Eqs. (21) and
appropriate integrations appearing in Eqs. (15)–(17) are (22) may be used for the evaluation of damping if the stres-
carried out analytically in order to obtain Eq. (19). ses are known from the FEM. The methods used are exten-
In this Eq. (19) the stress rm at the clamped end is sively explained by Gounaris et al. [12,13,15]. Here it
required at the steady state equilibrium and at the eigen- should be repeated that the method referred to the total
values (first) and same is obtained by applying FEM as loss coefficient determination at the steady state equilib-
explained above, or analytically. rium is applied at the resonance i.e. at the exact complex
The second method, by Gounaris–Anifantis [12], to eigenvalues.
solve problems by FEM, with structures with complicated In previous work [12] the steady state equilibrium was
geometry where was not possible to perform above integra- searched at the eigenfrequency ðkn ¼ Bn Þ and not at the
tions explicitly. complex eigenfrequency ðkn ¼ An þ iBn Þ. The results
The finite element piecewise approach for the analysis of obtained from the experiments at the resonance (Section
vibration problems as stated above is utilized here for the 4.1) have to be correlated to those obtained by numerical
1864 G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868

any material or (b) for any structure under stress


concentrations.
B D
rs rs
f ðrs Þ ¼ A þC : ð23Þ
rf rf

The fatigue limit rf is supposed to be known.


The procedure to be followed is:

• A cantilever beam is used, as it is described in Section


4.1, for the experimental determination of the loss fac-
tor. The maximum displacement at the free end and
the corresponding maximum bending stress at the
clamped end, have been recorded. Then the total loss
coefficient is taken through the measured logarithmic
decrement. Following the first eigenvalue is computed.
• Then the force F at the free end that gives the same dis-
placement, as this of the experiment, is found. Then the
maximum bending stress at the middle of each element is
computed.
• Knowing these stresses the proposed method by
Gounaris et al. [13] is applied to solve the inverse prob-
Fig. 2. Flow chart of numerical procedure. lem and to find the corresponding damping ce for each
element.

methods at the eigenvalues. This is for solving the inverse The inverse problem is a minimization problem. Here
problem in a different way from that proposed by [12,13] the direct method is used where all possible combination
due to reasons explained above. of the elements damping ce are used and the predictor–cor-
The technique for the evaluation of the steady state rector method which minimize these loops based on some
equilibrium between inserted and calculated damping, criteria, but lacks sometimes in sensitivity. The objective
and thus total loss factor and stress distribution, is the function to be minimized is the sum of the absolute differ-
same for both above methods, the difference being in the ence of bending stresses computed in each cycle of iteration
computation of this damping coefficient. Kume et al. [14] and known (measured) bending stresses. Thus at the mini-
use Eq. (19) while the other uses Eq. (21). mum of the objective function, the hysteretic damping of
In this technique an initial guess is given for the c, and iter- each element ce, is now computed. During this procedure
atively this value is improved, approaching its exact value. the eigenvalues, are remaining constants and equal to those
At each iteration step, the stresses and loss factor c are eval- computed.
uated, and compared to those found in previous step, respec- Substituting Eq. (23) in Eq. (20), the following equation
tively. Convergence is attained when the differences between is obtained:
the two last computed values are lower than some small  
constant. Fig. 2 shows a flow chart of computation in accor- 3E A C
ce r2em ¼ S Bem þ S Dem : ð24Þ
dance with this iterative scheme. This simple computational p Bþ1 Dþ1
method converges to the solution of the steady state equilib-
rium problem at the resonance frequency. To solve the new inverse problem and to find the
Eq. (22) will be used later, for the solution of the inverse unknown parameters A, B, C, D, the minimization referred
problem, when finding the parameters of the dissipating procedure is also applied here. The objective function to be
energy function (see Eq. (18)) for new materials or struc- minimized here is the sum of the absolute differences,
tures with strain concentration problems, etc. between the first and the second part of the above Eq.
(24) for every element. This happens because the first part
of this equation is now known. Since the unknown here are
3. Inverse problem four, at least four discrete elements are needed.
It must be emphasized that, once the dissipating energy
The inverse problem is defined as the determination of function of Eq. (23) is known, then the same procedure
the dissipating energy function. The dissipating energy could be applied in order to find the stress distribution at
function (Eq. (18)) as given by Lazan [10] for specimens the designing stage and in this case for every boundary con-
of mild steel is, now given here with the unknown A, B, dition. But in order to solve the inverse problem i.e. Eq.
C, D to be determined (a) for any uniform structure of (24), experimental and numerical (FEM) results have to
G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868 1865

be correlated, for the same boundary conditions, here the These results are presented in Table 3 and the corre-
cantilever beam. sponding diagrams are shown in Fig. 3 and are compared
This correlation could be extended also for other bound- with the available experimental data. The loss factors pre-
ary conditions. Having the diagram of the total loss coeffi- sented in these diagrams are in good agreement with the
cient vs. the maximum bending stress, at the eigenvalue, it experiment. Therefore, this is also considered here as a val-
is easy to solve the direct problem at the designing stage. idation in the response involved on the above numerical
Thus the solution of the inverse problem Eq. (24) will not methods in which complexity of the modulus of elasticity
be necessary. But this has to be proven as valid for other is also attained.
cases also, as it was done here for the cantilever beam. Above results are obtained within the numerically accu-
To this respect, for the cantilever beam, both above racy of FEM computations, for mild steel material and
methods, of Kume et al. [14] and Gounaris–Anifantis under the assumption that static and dynamic modulus
[12], as explained above are applied here and demonstrated of elasticity are the same. In future work numerical accu-
below (in the chapter of application) to give comparable racy has to be improved, experiments with other materials
results. have to be done and the change in the dynamic modulus of
elasticity has to be taken into consideration.
4. Analytical, numerical and experimental validation The maximum bending stresses are on the nonlinear
zone since they are above the 2 MPa or 250 psi (Fig. 1).
In this chapter the analytical, numerical and experimen- In Table 4, substantial difference is observed at the steady
tal validation of the above exposed methods is presented. state equilibrium and on resulting pairs of total loss coeffi-
The geometrical and material parameters for the cantilever cient and maximum bending stress near the clamped end
beam which is used for the computations as well as for the (at the middle of the first element) and at the same applied
experiment are given in Table 1. force F. One pair corresponds to the eigenvalues and the
In Table 2 results are given of eigenvalue computations other at the eigenfrequency only i.e. with A = 0.
by the above three methods. In Section 2.2 some comments This substantial difference on these pairs (total loss coef-
about the validation of these methods are presented. Here ficient–bending stress) confirms the necessity of eigenvalues
the assumption that the complex modulus of elasticity is computation and finding the steady state equilibrium at the
not only restricted to linear materials is made and is also resonance i.e. at the eigenvalues and this because:
validated from the results.
In the Table 3 the computed total loss factor vs. maxi- • At the design stage the maximum bending stress is
mum bending stress diagrams, are given for both above needed to be known in advance.
methods, presented by Gounaris–Anifantis [12] and by • At the solution of the inverse problem since results taken
Kume et al. [14]. from the experiments at the resonance and so the steady
state equilibrium, to which results have to be correlated,
should have to be at the resonance i.e. at the
Table 1
Geometry and material properties of the beam
eigenvalues.
Property Symbol Value Units
For the solution of the inverse problem, an algorithm is
Beam height H 0.008 m made in accordance with above Section 2.2. But in this case
Beam width W 0.008 m
Beam length L 0.5 m
the dissipating energy function representing by Eq. (18) at
Modulus of elasticity E 210 GPa the first stage, is taken to be known for the mild steel same
Shear modulus G 79.3 GPa as the specimen of the experiment thus ce of each element,
Fatigue limit stress rf 186 MPa in Eq. (22) are computed at the steady state equilibrium

Table 2
Application of the three different methods for the calculation of the complex eigenvalues
Eigenvalues calculated by three different methods
Applied force Loss factor c at the 1st method: analytical eigenvalues 2nd method: FEM eigenvalues 3rd method: eigenvalues
at free end steady state calculation based on Eq. (10) calculation based on Eq. (13) calculation based on Eq. (14)
1st eigenvalue 2nd eigenvalue 1st eigenvalue 2nd eigenvalue 1st eigenvalue 2nd eigenvalue
A B A B A B A B A B A B
0.0002 0.0040 0.34 169.8 2.13 1064 0.34 169.8 2.12 1063 0.34 169.8 2.12 1063
0.0004 0.0060 0.51 169.8 3.19 1064 0.51 169.8 3.20 1063 0.51 169.8 3.18 1063
0.0006 0.0063 0.53 169.8 3.35 1064 0.54 169.8 3.34 1063 0.53 169.8 3.34 1063
0.0007 0.0065 0.55 169.8 3.46 1064 0.54 169.8 3.45 1063 0.55 169.8 3.45 1063
0.0008 0.0070 0.59 169.8 3.72 1064 0.60 169.8 3.73 1063 0.59 169.8 3.72 1063
0.0009 0.0072 0.61 169.8 3.83 1064 0.62 169.8 3.84 1063 0.61 169.8 3.82
1866 G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868

Table 3
Comparison of the experimentally measured loss factor coefficient and the results of Refs. [12,14]
Force F at free end Max stress at clamped end, Total loss coefficient Max stress at clamped end, Total loss coefficient
in N MPa MPa
FEM, Ref. Experiment % FEM, Ref. Experiment %
[12] [14]
0.0004 25.2 0.0060 0.0065 8 19.5 0.0065 0.0050 30
0.0006 29.7 0.0063 0.0075 16 28.2 0.0075 0.0073 3
0.0007 34.1 0.0065 0.0095 32 33.0 0.0077 0.0092 16
0.0008 38.5 0.0070 0.0110 36 34.4 0.0078 0.0095 18
0.0009 43.2 0.0072 0.0120 40 42.6 0.0085 0.0108 21

4.1.1. Half power bandwidth method


A clamped-free beam was used in this experiment. The
beam was clamped on a harmonically vibrating table in
the vertical direction. A mechanical device, bolted on the
vibration table, was constructed to carry the vibrating
beam. A kind of bolted vice was used to support the beam.
The displacement at the free end was measured by an accel-
erometer. The experimental determination of the loss fac-
tor with this method was done in a steel axial specimen
of mild steel with same material characteristics as those
described in Table 1.
The test vibration table was so adjusted to create a sinu-
soidal excitation at the fixed end with regulated displace-
ment and exciting frequency. The exciting frequency was
at the resonant zone while the displacement was limited
so that the stresses at the fixed end to be below the fatigue
Fig. 3. Comparison of the experimental results and the Refs. [12,14], in
log–log diagram. limit.
These diagrams of displacement (at free end) to exiting
frequency was done by maintaining constant the exiting
displacement at the fixed end (one other accelerometer
and at the first eigenvalue. Secondly these ce now are taken as gives the indication) meanwhile the exciting frequency
known to the new algorithm which solves the inverse was changed between 5% and +5% of the eigenfrequency
problem on finding A, B, C, D parameters of Eq. (24). To (with a step of 0.02 Hz). Finally the total loss coefficient c
this respect, in all cases examined, above A, B, C, D para- was computed and plotted as a function of maximum dis-
meters are confirmed to be, as was expected, the same with placement at the free end (or maximum bending stress at
Eq. (18). the fixed end). This drawing was made each time at differ-
Final validation of this proposed method for the solution ent maximum displacement at the free end, but always at
of the inverse problem will be given in the future with spec- the same first eigenfrequency.
imens of different materials or with stress concentrations.
4.1.2. Logarithmic decrement method
4.1. Loss factor experimental measurement in a cantilever The experimental determination of the total loss factor c
beam was done in a cantilever beam axle of the same as above
mild steel material and by the use of the logarithmic decre-
Both methods, as explained below, give nearly the same ment method. A heavy vice was used to realize the clamped
results, i.e. nearly the same diagrams of total loss coefficient end of the beam and the exciting displacement was in the
vs. the maximum bending stress at the clamped end. direction that gives stiffer clamping, perpendicular to the

Table 4
Observed difference in bending stress and displacement when and A = 0
Applied force 1st complex eigenvalue, Incoming total loss Outcoming total loss Response
in N at the free end A + iB coefficient c coefficient c
A B, rad/s Bending stress Vertical displacement
in MPa at the free end, m
0.0009 0.61488 169.8 0.0072 0.0071 40.2 3.76 E-3
0.0009 0 169.8 0.0030 0.0028 1.8 1.69 E-4
G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868 1867

Fig. 4. Comparison of experimental results of damping based on 3 dB method ( ) and on the logarithmic decrement ( ) in log–log scale.

claws of the vice. This setup tends to be perfect giving a on both above methods, given in the Fig. 4, are in good
small difference between the analytical and the experimen- agreement.
tal natural frequency of the order of 3%. This was obtained
by giving an initial displacement at the free end, and then
leaving the specimen to vibrate freely. An accelerometer 5. Conclusions
measured the response at the free end, and the logarithmic
decrement d was computed. After that by using the known A new method called iterative complex eigensolution
relation c ¼ d=p, (see Section 2.4), the total loss coefficient method (ICEM) based on the damping – stress relation is
c was plotted vs. the measured logarithmic decrement d. A presented in this paper. The complex eigenvalues of a can-
diagram was made for the total loss factor c as a function tilever beam taking into account the hysteretic damping are
of the maximum displacement at the free end, or as a func- computed with three new (because of the use of the ICEM)
tion of the equivalent maximum bending stress at the fixed different ways (analytical, numerical and empirical) and are
end. The loss factor depends on the current displacement found to be in good agreement. Experiments, made with
or on the equivalent bending stress [10,24]. Therefore the two different ways, both at the resonance, give nearly the
experimental determination of the logarithmic decrement same results for the total loss coefficient and the maximum
d has to be applied for two consecutive cycles. By doing bending stress at the constrained end. The experimental
this the logarithmic decrement will correspond, each time, results have proven to be comparable with those obtained
at the proper displacement or bending stress of each cycle. by the two different numerical methods. The use of the
The equivalent relation between maximum displacement complex modulus of elasticity suitable for linear models,
at the free end and maximum bending stress at the fixed in this case of nonlinearity, due to the nonlinear damp-
end was found: (a) introducing the displacement measured ing-stress relation, is proven to be not far from the reality.
at the free end and the total loss coefficient c into the FEM The new proposed method, for solving the inverse prob-
algorithms the eigenvalues are computed and then, by iter- lem i.e. finding the parameters of the dissipating energy
ation of the displacement at the free end, the requested function, was validated for the mild steel. For structures
bending stress at the clamped end is found and (b) by the of other material or structures with stress concentrations,
use of the following Eq. (25) taken from Kume et al., similar dissipating energy functions have to be found in
Rao [25] and Fertis [26]. the future and many experiments have to be repeated. In
" # this work a big difference, as it was expected, was found
ðsinkLþsinh kLÞ
2
ðsin kx þ sinh kxÞ  ðcoskLþcosh kLÞ
 ðcos kx þ cosh kxÞ at the response at the steady state equilibrium when this
rmax ¼ Erk y 0 ;
ðsinkLþsinh kLÞ
ðsinkL  sinhkLÞ  ðcoskLþcosh kLÞ
 ðcos kL  cosh kLÞ (equilibrium) was obtained at the complex eigenvalue
ðkn ¼ An þ iBn Þ and at the eigenfrequency ðkn ¼ Bn Þ.
ð25Þ
Having the diagram of total loss coefficient vs. bending
stress at the clamped end for the cantilever beam for a dif-
where k ¼ 1:875 for the first eigenmode, E is the modulus ferent material, as explained above, could be used to obtain
of elasticity, r is the distance from the neutral axis, L is dissipating energy parameters Eq. (24) which then could be
the length of the beam, y0 displacement at the free end used to solve all cases with any boundary conditions. The
given by experiment, x is the distance from the clamped Eq. (23) based on experimental measurements correlates
end and rmax is the maximum stress at the clamped end the dissipating energy to the bending stress, for any bound-
(i.e. at x = 0). Both above procedures have proven to give ary conditions but for the same material, if stress concen-
comparable results and the results of the experiments based tration is not present. Otherwise if stress concentration
1868 G.D. Gounaris et al. / Computers and Structures 85 (2007) 1858–1868

exists, then Eq. (23) takes effect only for exactly the same [12] Gounaris GD, Anifantis NK. Structural damping determination by
geometry. finite element approach. Comput Struct 1999/10/12; 73(1–5):445–52.
[13] Gounaris GD, Papazoglou VJ, Anifantis NK. Inverse vibration
The future work of the authors includes the determina- problem solution of floating timoshenko beam. Comput Struct
tion of Eq. (23) for other materials, with or without the [submitted for publication].
presence of stress concentration. The effect of a fatigue [14] Kume Y, Hashimoto F, Maeda S. Material damping of cantilever
crack on the damping of a structure is also to be examined. beams. J Sound Vib 1982/1/8;80(1):1–10.
[15] Gounaris GD, Papazoglou VJ, Anifantis NK. Dynamics of fractured
timoshenko beams moving in wavy fluids. Comput Struct 2001/
References 2;79(4):431–39.
[16] Lazan BJ, Demer LS. Damping elasticity and fatigue properties of
[1] Makris N, Zhang J. Time-domain viscoelastic analysis of earth temperature resistant materials. Anonymous 1951:51.
structures. Earthquake Eng Struct Dyn 2000;29(6):745–68. [17] Dovstam K, Dalenbring M. ER – damping function estimation based
[2] Biot MA. Linear thermodynamics and the mechanics of solids. on modal receptance models and neural nets. Comput Mech
Proceedings of 3rd US National Congress of Applied Mechanics, 1997;V19(4):271–86.
ASME 1958:1–18. [18] Lancaster P, Prells U. Inverse problems for damped vibrating
[3] Spanos PD, Tsavachidis S. Deterministic and stochastic analyses of a systems. J Sound Vib 2005/5/20;283(3–5):891–914.
nonlinear system with a biot visco-elastic element. Earthquake Eng [19] Bishop RED, Price WG. Hydroelasticity of ships. Cambridge: Cam-
Struct Dyn 2001;30(4):595–612. bridge University Press; 1979.
[4] Makris N. Causal hysteretic element. J Eng Mech 1997;123(11): [20] Gounaris G, Papazoglou VJ. Three-dimensional effects on the natural
1209–14. vibrations of cracked timoshenko beams in water. Comput Struct
[5] Palmeri A, Ricciardelli F, De Luca A. State space formulation for 1992/3/3;42(5):769–79.
linear viscoelastic dynamic systems with memory. J Eng Mech [21] Gounaris GD, Papadopoulos CA. Crack identification in rotating
2003;129(7):715–24. shafts by coupled response measurements. Eng Fract Mech 2002/
[6] Palmeri A, Ricciardelli F, Muscolino G. Random vibration of 2;69(3):339–52.
systems with viscoelastic memory. J Eng Mech 2004;130(9):1052–61. [22] Klein W, Pommer C, Stoustrup J. Stability of rotor systems a
[7] Dimarogonas AD. Vibration engineering. West Group 1976. complex modelling approach. Zeitschrift Für Angewandte Mathe-
[8] Lin MR, Lim MK. Complex eigensensitivity-based characterization matik Und Physik. J Appl Math Phys 1998:644–55.
of structures with viscoelastic damping. J Acoust Soc Am [23] Sekhar AS, Kumar Dey J. effects of cracks on rotor system
1996;100(5):3182–91. instability. Mech Mach Theory 2000/12/1;35(12):1657–74.
[9] Johnson Conor D, Kienholz David A. Finite element prediction of [24] Bovsunovsky AP. The mechanisms of energy dissipation in the non-
damping in structures with constrained viscoelastic layers. AIAA J propagating fatigue cracks in metallic materials. Eng Fract Mech
1982;20(9):1284–90. 2004/11;71(16–17):2271–81.
[10] Lazan BJ. Damping of materials and members in structural [25] Rao SS. Mechanical vibrations. Prentice-Hall; 1990.
mechanics. Pergamon Press; 1968. [26] Fertis DG. Mechanical and structural vibration. Wiley-Interscience;
[11] Nashif AD, Jones DI, Henderson JP. Vibration damping. John 1995.
Wiley & Sons, Inc.; 1985. pp. 480.

You might also like