See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/253929464

Fractional Calculus – A Different Approach to the

Analysis of Viscoelastically Damped Structures

Article in AIAA Journal · May 1983

DOI: 10.2514/3.8142

CITATIONS READS

707 1,030

2 authors, including:

Peter Torvik
Air Force Institute of Technology
80 PUBLICATIONS 4,324 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Damping of Hard Coatings View project

All content following this page was uploaded by Peter Torvik on 11 August 2016.

The user has requested enhancement of the downloaded file.

 VOL. 21, NO. 5, MAY 1983 AIAA JOURNAL 741

Fractional Calculus—A Different Approachttho e Analysis

of Viscoelastically Damped Structures
RonaldL . Bagley*
United StatesA ir ForceA cademy, Colorado Springs, Colorado
and
Peter J. Torvikf
Air Force Institute of Technology, Wright-Patterson Air Force Base, Ohio

Fractional calculus is used to construct stress-strain relationships for viscoelastic materials. These relation-shipsar e
usedithn e finite element analysiso f viscoelastically damped structuresan d closed- form solutions to the equations of
motion are found. The attractive feature of this approach is that very few empirical parameters
are required to model the viscoelastic material and calculate the response of the structure
2016 | http://arc.aiaa.org | DOI: 10.2514/3.8142

for general loading conditions.

Nomenclature = {<t>n } nt h mode shape

=
{ M }T = column vector,ro w vector w frequency
< >= argument of operator (~) = associated Withth e expanded equations f
= [ ] square matrix motion
A ba r area, example
Introduction
As
b
= area of shear pad, example
= model parameter THE successful determination ofdamped
the responses of a viscoelastically
structure to prescribed loading
c =time-scaling factor, example
8,July
[C] = damping matrix time histories hingesoth n e successful solution oftw o
= problems. The first problem is that of describing the
Da fractional derivative of ordera viscoelastic material's mechanical properties in a
E ba= r modulus, example mathemati-cally rational manner that enhancesth e
Downloaded by AFRL D'Azzo Wright-Patterson on

= column vector of applied forces prospects f suc-cessfully attacking the second problem—
= Laplace transformo f force vector solving the resulting equations f motionfo thr e structure.
[K] = stiffness matrix Fractional derivative stress-strain constitutive relationships
L =bar length, example for viscoelastic solidsno t only describeth e mechanical
[M] = mass matrix properties of some materials, but lead to closed-form
nij = modal constant solutionsth f e finite element equations of motionfo r
s = Laplace parameter viscoelastically damped structures.1 We will present the basic
t =time form af fractional derivative modelfo r a viscoelastic
tf = thickness of shear pad, example materialan d then outline constructionan d solution thf e
{x ( t ) } = column vector of nodal displacement resulting finite element equationso f motion.
= transform of displacement vector Previous attempts to describe the mechanical properties
= column vector of displacements for im-
pulsive loading of viscoelastic solids have suffered becauseth e
=r transform of displacement vector for mathematical models describing the behavior of these
impulsive loading materials have not been clearly linkedttho e physical
a,P = model parameters principles involvedTh. e engineerah s been forcedt o adopt
e(t) = strain history phenomenological (em-pirical) approaches to describe
e*(s),e*(/co)= Laplacean d Fourier transformsth f e mechanical properties of these materials. Somethof e
strain history more popular approachesar ae complex constant as
\n •= nth eigenvalue material modulus, numerical methods in the transform
fjL*(s)tn*(iw) = Laplace and Fourier transforms of the domain, and the standard linear viscoelastic model
shear relaxation modulus presented in textbooks.
p ba r density, example The simplest of these approaches is using a complex
a(t) = stress history constant as the material modulus. This approach is motivated
= Laplace and Fourier transforms of the by observing the relationship between sinusoidal stress and
stress history ' sinusoidal straini n viscoelastic materialsTh. e strain lagsth e
= dimensionless time, example; also, stressan thd e imaginary parto thf e complex constant
dummy variable adequately describes this phenomenon. The limitation of this
approachi s thati it s restricted o sinusoidal motionthf e
Presented as Paper 81-0484 at the AIAA/ASME/ASCE/AHS 22nd material. Crandall has shown2 that application of this method
Structures, Structural Dynamicsan d Materials Conference, Atlanta, tho e general motion thf e material leadst o serious
Ga., April 6-8, 1981; submitted April 15, 1981; revision received July mathematical problems.
19, 1982.This paper is declared a work of the U. S. Government and Th use eo f numerical methodsithn e transform domaint o
thereforei si nth e public domain. describe the frequency-dependent mechanical properties of
*Associate Professor, Department of Engineering Mechanics. viscoelastic materials is cumbersome at best. The major
tProfessor and Head, Department of Aeronautics and Astro- drawbackiths e substantial effort requiredt o calculate
nautics. Associate Fellow AIAA. numerically the transform inversion integral for every
point in timea t whicth e responseothf e structureis
n
e
e
d
e
d
.
 742 RL. : BAGLEYAN DP J. . TORVIK AIAA JOURNAL

a
Th e standard linear viscoelastic model,a serieso f a(0 +bD*(a(t)> =E0e(t) +E1D -<e(t)> (3)
derivatives actingothn e stress fields relatedtoa serieso f time
derivatives on the strain fields Th e fractional derivativesar e definedb y

(1)
is also cumbersomet o useFo. r viscoelastic materials having This fractional derivative operatorahht s e propertyiht n e
mechanical properties thatar e strongly frequency dependent over
Laplace transform domain
many decades of frequency, the number of time
derivatives, M and TV, in the series is large. Consequently the
(5)
number of empirical parameters in the model is large. As a
result,th e modelis time consumingto manipulate and, when that the transform of the fractional derivative of x(f) is
pu t intoth e equations f motion,
produces high-order dif-
equal to 5" times the transform of x(t). This property can be
ferential equationsTh. e resultsar e that considerable effort demonstrated by taking the Laplace transform of Eq. (4),
mustb e expendedt o obtainth e eigenvalues (resonances) d
eigenvectors (mode shapes) of the equations of motion. = x(t)e~ dt
st
(6)
We will show that the formulation of the equations of JO
motion using derivatives of fractional order also produces
http://arc.aiaa.org | DOI: 10.2514/3.8142

higher order matrix equations to solve. However, our frac-

A similar relationship exists in the Fourier transform domain.
tional derivative model of the frequency-dependent
By takingth e Fourier transform,
mechanical properties typically requires only five empirical
parameters. This is fewer parameters than are usually required
witth e corresponding standard linear viscoelastic modelW. e
(7)
believe that relativelyfe w empirical parameters
are requiredithn e fractional derivative models because they
ar e consistent thwith e physical principles involved. Con- Eqof . (4),a relationship
sequently, we feel that fractional derivative models are at- similartEqo (5. ) results.
tractive as a tool of engineering analysis.
(8)
The Fractional Derivative Viscoelastic Model
Early observations of the mechanical properties of The Fourier transform of the fractional derivative of order a. of
Downloaded by AFRL D'Azzo Wright-Patterson on July 8, 2016 |

viscoelastic materialsb y Nutting indicated that e stress x(t)i s (/w) - timesth e Fourier transform ofx(t). Takingth e
relaxation phenomenon appeared to be proportional to time Fourier transform of the five parameter viscoelastic model,
3
raised to fractional powers. Later observations by Gemant Eq . (3), yields
showed that e frequency dependenceothf e mechanical
properties varies as frequency raised to fractional powers, and (9)
he suggested that differentials of fractional order be used to
45 where a*(/co) and e*(/co) are the transform of the stress and
model materials. ' More recently, Scoft-Blair proposed that
fractional derivatives could be used to relate time-dependent strain histories, respectively. Factoringan d dividing termsi n
6
stress and strain in viscoelastic materials. Scott-Blair's use of this operation produces
fractional calculus simultaneously modeledth e observations of
Nuttingan d Gemant. Fifteen years ago, Caputoin -
dependently echoed Scott-Blair's observations on the use of (10)
fractional calculus. Caputo used fractional derivativest o
78
model the behavior of viscoelastic geological strata. ' In
addition, Caputoan d Minardi concluded from experimental The model suggests that the frequency dependent modulus is
observations that fractional derivative relationships were valid a function f fractional powers f frequency. Thisi s consistent
9
for some metals and glasses. witth e observationo f Gemant alludedto earlier.
The fractional derivative model put forward here is a The five parameters are determined by a least squares fit of
generalization f both Caputo'san d Scott-Blair's models. this model to the frequency-dependent mechanical properties
This modified modeli s basedoth n e observed mechanical ofth e material. Such fits have been obtainedfo r several
10
behavior 3f 0 materials (elastomersan d glassy enamels). I t materials. Figure 1 demonstrates the excellent agreement
is important o note that fractional derivative relationships, between the model and the mechanical properties at 550°C of
11 a Corning glass doped with oxidesof aluminum, sodium,an d
derived from first principles, for uncrosslinked polymer
solidsar e almost identicalin formt o Scott-Blair's modelan d cobalt (7.5% A12O3, 3% Na2O, 1% Co2O3). The parameters
to the model presented here. of the model are
Th e general formouf r fractional derivative viscoelastic modelis
9 2
JE0 = 4.15x'10 N/m (11)
11 0 641 2
(2) '7 = 1.09xio N-(s) - /m (12)
631
6= 3.50(s)°- (13)
whereth e time-dependent stress fieldsar e relatedt o time- a = 0.641 (14)
dependent strain fields through serieso f derivativeso f
fractional order. Experimental observations indicate that many =0 0.631 (15)
viscoelastic materialsca nb e modeledb y retaining only the first
fractional derivative termi n each seriesiEqn . (2). These results are typical. The five parameters consistently
The result is a viscoelastic model with five parameters, b, E0, describe two high-order curves simultaneously. In many cases,
EI} a, and/3. taking a= /3 producesa very satisfactory fit. Data given.
MA Y 1983 ANALYSIS OF VISCOELASTICALLY DAMPED STRUCTURES 743

10' formulation of the stiffness matrices for the finite element in

— FIVE PARAMETER MODEL the viscoelastic regions of the structure, however, must be,
° DATA, STORAGE MODULUS limitedto those methods thatdno t constrainth e stressesin
€10" ° DATA, DISSIPATION MODULUS each finite element to be in equilibrium with the forces at the
nodes of the element. This restriction is generated by,the
CO
ID
10'°
requirement that viscoelastic stressesb e dependento n strain-
_J
=> time histories. Finite element methods constraining stresses
Q
O within an element to be in equilibrium with the nodal forces,
e.g., assumed stress methods,arnoe consistent withth e
assumption that stressesb e dependentoth n e local strain
t

history.14 Consequently, assumed displacement methods for

5
I08 IO" !0'4 IO"3 I0~2 10"' 10° I01 IO2 IO3 constructing viscoelastic finite element stiffness matrices are
F R E Q U E N C Y - Hz usedin formulating the equationsof motion.
Fig. 1 Mechanical properties of Corning 10. The stiffness matrix of a viscoelastic finite element is
constructed using the elastic-viscoelastic correspondence
15
principleTh. e stiffness matrixi s first formulateda s though
12
in Fig1. were obtainedfo ar range of temperatures and the material were elastic. The element stiffness matrix is then
frequenciesan d reducedt oa single temperature throughth e separated twinto o matrices, one containing those terms
reduced frequency procedure.13 These data and curves proportionaltho e Lame* constantXan d the other containing
describeth e same material t 610°Ci thf e frequencyi s those terms proportional to the Lame* constant /*,
10.2514/3.8142

multiplied by IO2, at 690°C if the frequency is multiplied by

4
anIO , da t 840°C i thf e frequencyi s multipliedb y IO6. (19)
The following section is an outline of the construction and
solutionthf e finite element equations f motionfo r A t this pointth e transformsthf e viscoelastic moduli, JJL*(s)
| DOI:

viscoelastically damped structures when Eq. (3) is used to an d X*(s),ar e substituted n placeothf e Lame* constants
modelth e damping material/The Laplace transformi s usedin Since this developmenti s limitedtho e consideration f
|http://arc.aiaa.org

this development. Before proceedingi it s usefult o presentth e uniform, uniaxial shear strain, the viscoelastic stiffness
relationship between the transform of stress a*(s) and strain matrix takesth e form
e*(s)fo thr e five parameter model,

(16) Substituting the five parameter model for IJL*(S) [Eq. (18)]
July 8, 2016

into Eq. (20) produces

or
-Patterson on

(21)
(17)
D'Azzo Wright

where At this point the stiffness matrices for both the elastic and
viscoelastic finite elements in the structure are used to con-
(18) structth e stiffness matrixothf e total structureithn e
traditional manner. The resulting stiffness matrix differs
from conventional stiffness matrices in that some matrix
This modulusi s usedto constructth e stiffness matricesothf elements are functionsth f e Laplace parametersTh . e
by AFRL

e
viscoelastic elements in the damped structure. resulting equations f motionfo thr e total structureithn e
To this point, only a one-dimensional or uniaxial stress- transform domain are
Downloaded

strain relationship has been described. However, the frac-

tional calculus doesno t suffer from sucha limitationTh. e s2 [M] {X(s)} + (K(s) ] (X(s) } = (F(s) (22)
presentation of a complete three-dimensional relationship is
given n 1Ref . . Also presentedi n 1Ref . i as complete
Since the stiffness matrix contains functions of the Laplace

derivation of the finite element equations of motion for parameter,i i t s clear that conventional techniquesfo r
structures damped with multiple viscoelastic materials. What decoupling the equations of motion do not apply. Con-
follows here is an outline of this derivation. It is assumed, for sequently,a different approachfo r decouplingth e equations
the sake of simplicity in notation, that a single viscoelastic of motionis adopted.
material undergoing uniform, uniaxial stress (strain) is used The method used is similar to the method developed by
to dampth e structure. Foss16 for decoupling the equations of motion having non-
proportional viscous damping, i.ea. , damping matrixno at
The Constructioan d Solution linear combinationothf e massan d stiffness matrices. Those
of the Equations of Motion equations are
Having made several observations about fractional
derivative material models,i it s appropriate o presentth e [M] [ x ( t ) } + [C] ( x ( t ) } + [K] (x(t) } = i f ( t ) } (23)
method usedithn e analysiso f viscoelastically damped
structures. Because of its popularity and the flexibility it These equations of motion are posed in the form
affordsi thn e analyseso f structureso f engineering interest, M
th e following discussion s presentedi n termso f a finite ^llfAlr l ~ ! ° if * i_
elementUsformulation . eo thf e fractional calculuismodel ,
of course,no t limited to finite element applications. c JUlTL o !^Ji7J~
Th e cornerstoneanf y finite element approachiths e (24)
construction of the stiffness matrices for each of the finite
elements in the structure. Normally, the stiffness matrix for where the top set of partitioned matrix equations is satisfied
each elastic finite elementi s constructedb y using assumed identicallyanthd e lowerse ot f matrix equationsisthe original
displacement methods or assumed stress methods. The equations of motion [Eq. (23)]. The efficacy of this approach
 744 R. L. BAGLEY AND P. J. TORVIK AIAAJOURNAL

is that n orthogonal transformationca b n e found that

diagonalizes the two real, square, symmetric matrices in Eq. [0]
(24), while obtaining an orthogonal transformation for the {0}
original equations [Eq. (23)]i s usuallyno t possible.
To put the equations of motion of our viscoelastically (31)
damped structure into a form similar to Eq. (24), we multiply
the equations of motion by the denominator of /**(?), which is {0}
(1 + bs ). Next we segregate matrix elements of the equation
of motion13 into matrices having common powers of the
Laplace parameter. The resulting form of the equations of Notice that the two large matrices are real, square, and
motion is symmetric because the submatrices [Aj ] are themselves real,
+s2 [M] +s"E][K]]+seb[K+2.] [K3]] square,an d symmetric. Also note that e lowestse ot f
partitioned matrix equations are the equations of motion for
the structure [Eq. (26)]an ald ol thf e upper setso f matrix
(25) equations are satisfied identically.
where the terms involving [Kj ] through [K3 ] are the stiff- This expanded form of the equations of motion differs
from Eq. (24) in that the equations are posed in the transform
ness matrix in Eq. (22) multiplied by (I + bs$).
domain,bu tht e fractional powers sf i thn e submatrices of
The next step is to identify the smallest common the column matrices correspond to derivatives of fractional
denominator thf e fractionsan d 0, referredt oa s m.
order. Equation (27) may be returned to the time domain, or
Having obtainedm th, e left sideoEqf . (25)i s expresseda as Eq . (24) broughtttho e transform domainTh. e major dif-
Patterson on July 8, 2016 | http://arc.aiaa.org | DOI: 10.2514/3.8142

matrixsu man thd e equationso f motion taketh e form ference, then, between Eq. (27), the expanded equations of
motion,an Eqd . (24)i s that e latterar e posed in termso f
[Aj] (si"»X(s)} = (1+ bs?) (F(s)} (26) derivatives of integer order.
Iti s clear that e orderthf e expanded equations for
j=o structureso f engineering interest couldb e very largean thd e
Clearly, some ofth e matrices [Aj] willbe zero,an thd e sizeoth f e matrices prohibitivet o manipulate even onth e
nonzero [Aj] come from [M] and [Kj] through [K3] inEq.
electronic computer. However, this doesno t diminishth e
(25). Note that /= m(2 + 0). value of the expanded equations of motion for the orthogonal
The problem now is how to pose the equations of motion in transformation decouplingth e expanded equationsof motion
termsotwf o real, square, symmetric matricesfo r whichw e b can e constructedan th d e decoupling procedureac -
can obtain an orthogonal transformation and decouple the complished without directly usingth e expanded equations f
motion.1
equations of motion. The answer is to cast the equations of
The orthogonal transformation for the expanded equation
motioninth e following format: is constructed using homogeneous solutionsfo thr e original
(27) equations f motion [Eq. (25)]. These solutions satisfyth e
equation

2 m
ri [M]
Downloaded by AFRL D'Azzo Wright-

[0] [0] [0 ] [Aj]

(32)
[0] [0] [Aj] [Aj.,]
(28) which is the homogeneous form of Eq. (25). The eigenvalues an\n
thd e eigenvectors (0^)fo r this equationca nb e found using
matrix iteration techniques similar to those commonly
[0] [Aj]• [A3] [A2] used for finding mode shapes and resonant frequencies for
1
undamped structures.
[Aj] [ A j _ , ] • [A2] [A,]
Given that e expanded equations f motionca bn e
decoupled using constructions basedo n homogeneous
[0] [0] [0] [-Aj] [0] solutions to Eq. (25), it is appropriate to present the general
[0] form of the solution to the expanded equations of motion,1
[0] [0] [-AJ] [-AJ-,]
[0] [0] [-A,,] [-A,2] (0} (33)
[*] =

This solution is the transform of the structural displacement

[-AJ] l-Aj.,} [-A2] (0} time histories [X(s)} in terms of the transforms of the applied
loads (F(s)}. {0Jan d \ar the e solutionstEqo .
[0] [0] [0] [0] [A0] (32), (1+ bs i)ths e dominatoro f n*(s),an d TVthis e
(29) orderthf e
matrices[M an ] d
13
[K]i thn e expanded equationso f motion.

mn is a modal constant, similar in form to a modal mass,

r (J-l)lm and is defined as
(J-2/m)
(34)
(30) where {4>n } is the nth eigenvector of the expanded equations
of motion. [4>in] sconstructed from an\n d { < / >th„) , e
solutions to Eq. (32).
Having presented the transform of the structural response,
the next issue to be confronted is the existence of the inverse
 MAY 1983 ANALYSIS OF VISCOELASTICALLY DAMPED STRUCTURES 745

transformothf e structural response, namelyth e existenceo f

the time history of the motion of the modes of the finite
element modelW. e are particularly interestedithn e structural
responsefo r impulsive loading, i.e., whenth e applied forces
ar e equalttho e Dirac delta function. Observing that e
transform of the delta function is one, the transform of the
impulse responseis

Fig. 2 Elastic rod supported by elastomeric pads.

(35)
mtt(s'"»-\n)
describedb thy e five parameter model, i.e.,
The inverse transform of { X(s)}, [x^t)}, exists and is real,
continuous, and causal when 1) '(X(s)} is analytic for
Re(s)>0,2 ) (X(s)]i s realfo sr realan d positive,an 3d )
(X(s)} is order s~ , where y> 1, for Is I large in the right half
s planeTh. e proof7 that these three conditionsarmee i t s Asl/t, 0
1
straightforward. (38)
Consequently,th e time-dependent responseothf e structure 0 As2/t2
to impulse loading, existsan di s real, continuous,an causal.
d

Using contour integration and the residue theorem to Th e resulting numerical worki s simplifiedb thy e introduction
http://arc.aiaa.org | DOI: 10.2514/3.8142

determine the form of the impulse response produces of a dimensionless scaled time r defined through

i rt ' • t = cr (39)
{*(0}=- W{ (X(re- *)]e- dr\
/ \J 7T0 For this example, 5 will be taken as the dimensionless Laplace
transform parameter corresponding to the dimensionless
(36) Thtime . e scalemafactor yb e incorporatedthinto e damp
nii and mass matrices to yield
The essential details of the process used in passing from Eq.
(35) to Eq. (36) are given in the Appendix. M r_PAL \4 1 ~ (40)
Note that e responsethf e structureha tws o parts,on e part 6c 2
[i 4
Downloaded by AFRL D'Azzo Wright-Patterson on July 8, 2016 |

being a sum of decaying sinusoids and the other an in-tegral that ~

decreases with increasing Thtime . e integral does not decrease AE 2 -1
K (41)
exponentially. In fact, the integral is asymptotic to t~ifo r large /, ~~L -1 2
where i rj s greater than one. Therefore,th e integral dominates
the response for a time long after the loading. This componentthf
e response describesth e nonoscillatory return of the structure to \.* 0A*i Hi.
its unloaded equilibrium positionI. it s alsoa manifestationthf e
3 I
power law stress relaxation phenomena observed by Nutting. D= l +b ( s / c ) 0 As2/t2 (42)

Example Damper properties are described by /*0, #7 , and b. For

Since the method introduces unfamiliar notations and convenience, both fractional derivatives have been taken to be !/z.
procedures,a n exampleoitf s applicationi s helpfulW. e will Each shearpa di os f contact area buAs, th e thicknesses tj and t2
demonstrate the process followedi n analyzing systems are different to insure that the damping matrix is
described by fractional derivatives through the example nonproportionalAl. l rodsar oe f lengthL , density anp, d
depicted in Fig. 2. The tension member is fixed at both ends, cross-sectional area A.
but supportedatw t o interior pointsb y Introducing the following numerical values:
shear dampers,
consisting f padso af material described thy e five-parameter
fractional derivative model. Only longitudinal motion resulting
from forces appliedtht e nodal points s m
considered.
The finite element representation of this system, after P = 1.25xl04 kg/m3 9
= 200xl0 N/m
2

transformingttho e Laplace domain, s 7 2 5 /2 2

> 0 =4xl0 N/m N-s' /m
2
s [M] +[X] [D(s)] +[X] [K] =[X] IF] (37)

The mass and stiffness matrices are developed from rod 4

elements,an thd e damping matrix s developed fromth e nodal anda convenient scale factor, =5xlO~ s leads to the
force relationship that results when the shear pads are following system, see Eq. (22)

72 1
0.33 0.0825 s'/'Xl.O + O.ls' )- -2
s2 0.0825 0.33 -2 , 4.

8
xlO - m/N (43)
X2 (s)
 746 R. L. BAGLEY AND P. J. TORVIK AIAA JOURNAL

Clearingth e Laplace parameter fromth e denominatorothf e Table 1 Eigenvalues and eigenvectors for the example problem
damping matrix produces equationsthn e formEqf . (25),
0.33 0.0825 0.33 0.0825 X, = 1.071768 +/1.099794 -f L
1. .040 + /
1 I 0.0165
0.0825 0.33 0.0825 0.33
, = 1.071768-/1.099794 2
I' 1.040-/0.0165J
2.0 0.0 4.0 -2.0 ( 0 3 i =[ ;;
5 (0.1) , = -1.073498 +/1.028611 .001 + /Q.0238

.0 .01.4 -2 .0 0
,= -1.073498 -/1.028611
1.001-/0.0238
4.2 -2.0 1.000+/0.0000 i
(44) \5 = 1.581998 +/1.601989 -0.9713+/0.01203
-2.0 4.01 X2(s)
load 1.000+/0.0000 ^
Th e load vectorhaF s been manipulated intoane w f = 1.581998-/1.601989
vector Fthrough this process. (07i= [_!:! -0.9713 -/0.0120 J

Th e homogeneous equations f motion producea charac-

teristic value problem X7= -1.584631+/1.547150
.002 + /0.0240>>
0.03300 0.00825 0.3300 0.0825
,= -1.584631 -/1.547150
Downloaded by AFRL D'Azzo Wright-Patterson on July 8, 2016 | http://arc.aiaa.org | DOI: 10.2514/3.8142

-1.002-/0.0240
0.00825 0.03300 0.0825 0.3300
,= -9.997418+/0.0
0.60 -0.20 4.2 -2.0
1.000 + /0.0
,= - 9.99385+/0.0
-0.20 0.41 -2.0 4.01
(45)
where Xcorresponds to square roots of the Laplace parameter
s. This fifth-order eigenvalue problem for a two degree-of-
freedom system will lead to 10 (complex) eigenvalues Xand 10 Since the solutions are not in a familiar form, however, the
(complex) eigenvectors {</>} . expected response characteristics may not be recognized in the
We now solve the homogeneous system of the expanded numerical resultsI. it s usefult o lookithn e results for what
equationso f motion (26)an d (27),Fo r this example, one would expect to see in the response of a two degree-of-
freedom system. First,w e expectseo twe o eigenvaluesan d
-2.0 two eigenvectors corresponding totw o normal modesTh. e
(46) two eigenvalues should correspond to two natural frequen-
4.01 cies. If the dampers were not present, the second frequency
should be twice the value of the first. The mode shape
correspondingttho e first frequency should indicate in-phase
0.60 -0.20 motionoth f e nodesanth d e mode shapefoth r second
e

(47) frequency should indicate out-of-phase motionI. n addition,

0.20 0.41 one would expect the response of the system to be described
by real functions of time. With these anticipated charac-
teristicsothf e physical responsein mind,le ut s returnttho e
0.0 0.0 numerical results.
IA2]= (48) The eigenvaluesX were introduced as square rootsoth f e
0.0 0.0 dimensionless Laplace parameter s. It follows that non-
dimensional eigenvalues correspond to square roots f
dimensionless complex natural frequencies. Squaring the
0.3300 0.0835 first, second, fifth, and sixth eigenvalues in Table 1 produces
[A4] = (49) two conjugate pairs, where each conjugate pair corresponds
0.0825 0.3300
to a natural frequency of the system. We find o>7 = Xj
= -0.0609+ /2.357, o>=2 Xf .= -0.0637+ £5.069. It shouldb e
noted that eacho f these frequenciesha s been computed with
0.03300 0.00825
[A5] = (50) the frequency-dependent material properties appropriate to
0.00825 0.03300 that frequencyTh. e dimensionless frequenciesar e readily
converted into radians/second by dividing by the scale factor
The expanded equations of motion are of order 10, leading to c. The frequencies are 750 and 1614 cps. For the hypothetical
10 complex eigenvalues andth e complex eigenvectors. The elastomer usedin this example,th e magnitudeothf e modulus
eigenvaluesan d eigenvectors were found usinga modified increases 52% over that range. The magnitude of one
formo f matrix iteration,l with resultsa s shownin Table A1. s frequency is approximately twice the magnitude of the other,
expected, the complex eigenvalues, and eigenvectors occur in as expected. Since the real parts are small, the system is lightly
complex conjugate pairsOn. cae no w returntEqo . (33)an d dampedTh. e remaining eigenvaluesdno t denote further
construct the transform of the response of the structure for a resonances, but describe the nonoscillatory behavior of the
general loadingTh . e computationthf e inversefo thr e case damped structure. These eigenvalues are recognized by the
ofa n impulse loadingis demonstratedinth e Appendix, fact that the phase angle of their squares do not lie in the
The solutions to the homogeneous problem contain much range -180<0<180. Thus,th e natural frequencies are
important information concerningth e responseothf e system. predicted and identified.
 MA Y 1983 ANALYSIS OF VISCOELASTICALLY DAMPED STRUCTURES 747

Th e eigenvectors associated withth e eigenvalues identifying The integrals on the segments of the closed contour are
resonant behavior, (the first, second, fifth,an d sixth) also evaluated for the case where the length of segment 1 is ex-
occur in conjugate pairs. The first and second eigenvectors tended indefinitely in the positive and negative imaginary
combineto produceth e first mode shapean d indicate real, in- directions,
phase motion of the nodes. Similarly, the fifth and sixth
st st
eigenvectors combineto produceth e second mode shapean d [X(s)}e ds= ' [X(s)}e ds (A4)
indicates real, out-of-phase motion. The remaining six
eigenvectors combine with the conjugate temporal terms to and, as a result, Eq. (A2) becomes
show that the nonoscillatory portion of the motion of the I 7p+ 100 ^
structure is also real. — (X(s)}eads
2iriJ 7-ico'
Conclusion 1 6
We conclude that the equations of motion for = -y-: D
-
(A5)
viscoelastically damped structuresca nb e constructedan d
solvedi an reasonably straightforward manner whenth e showing that one need evaluate the right side of Eq. (A5) to
mechanical behavior of the viscoelastic material is portrayed obtainth e inverse transform.
by the fractional calculus model. The attractiveness of these To maintain the continuity of the closed contour, the radii
models is that they are consistent with the physical principles of segments 2 and 6 are increased indefinitely and segments 3
and that they accurately describe the mechanical properties of and 5 are extended indefinitely in the negative real direction.
a variety of materials over wide ranges of frequency. The Whenth e radiio f contours2an d6 ar e increased indefinitely,
Downloaded by AFRL D'Azzo Wright-Patterson on July 8, 2016 | http://arc.aiaa.org | DOI: 10.2514/3.8142

small number of empirical parameters needed with these it can be demonstrated that the resulting value of the integrals
models facilitatesth e process f obtaining least squaresfi t along these two segments is zero. Similarly, it can be shown
to the mechanical properties of the materials and reduces the that e integral along contour4 goesto zeroaths e radiuso f
effort required to solve the resulting equations of motion. the contour is decreased indefinitely. The integrals along
The fractional calculus approach is an encompassing contours 3 and 5 are not zero, nor do they add to zero.
methodfo thr e analysiso f damped structuresTh. e approach Rather,
begins with the molecular theory of polymer solids, proceeds
to generate accurate mathematical modelsfo r viscoelastic st st
behavior, results in well-posed equations of motion, and e [X(s)}ds+ \ e {X(s)}ds
concludes with closed-form solutions for structures of 3 J5

engineering interest.
Appendix: Calculation of the Response {X(e-i*)}e~rtdr\ (A6)

of Impulsive Loading
Th e final stepi n determiningth Th e arresidues [bj] e beval
e impulse responsethf e
structurei st o calculateth e inverse transformothf e Laplace techniques

transform of the response to impulsive loading, bj] li=m ( s - \ J ' ) ( X ( s ) } e st (Al)

Fothr e impulsive load,th e resultis

(Al)

(A8)
The inverse transform integral
Sincethe integralso n contours2 4, an, d6 ar e zero, Eqs. (A
(A2) an d (A8) enableth e evaluationoEqf . (A5),

is evaluated usingth e residue theorem fromth - Im/r (X(re-i*)e-rtdr\

e calculuso f a complex variable. 7T \J0 /
Th e closed contouro f integration, usedin conjunction with
th e residue theorem,i s given n FigAl.Th . (A9)
e contouri s
divided into six segments and the direction of integration
along each segmenti s indicatedb thy e arrows. Segments-3
5 thof e contourar e required, becauseth e branchcu an t d POSITIVE IMAGINARY AXIS
1/m
branch pointthf e function sar e takent bo e alongth e
negative real axisandath e originofthes plane, respectively. S PLANE
The residue theorem states that the integral along the
closed contour, divided by 27r/, is equal to the sum of
the residues of the poles of the integrand. In this case,
the statement of the residue theorem translates into POSITIVE
REAL AXIS
{X(s)}e+s<ds
2-Kl
(A3)
whereth e index indicatesth e contour over whichth ine -
tegration st ob e performed,an d {arbj } the e residuesothf e
poleso f (X(s)} enclosedb yth e contour. Fig. Al Integration contour for inversion transform.
 748 R.L. BAGLEYAN JDP.. TORVIK AIAA JOURNAL

3
Thus, we have obtained the impulse response of the system. Nutting, P. G., "A New General Law of Deformation," Journal
Th e summation overth e indexji e impulse response
thn of the Franklin Institute, Vol. 191,May 1921, pp. 679-685.
4
[Eq. (A9)] differs from the sum over the index n in Eq. (Al). Gemant, A". , A Method of Analyzing Experimental Results
The poles of the Laplace transform shown in Eq. (Al) \n are Obtained from Elasto-Viscous Bodies," Physics, Vol. 7, 1936, pp.
]/m 311-317.
originally found in the s plane. Since the inverse Laplace 5
Gemant, A., "On Fractional Differentials," Philosophical
transformi s evaluatedi n thes plane,th e transformation \m
Magazine, Vol.25, 1938, pp. 540-549.
6
must be applied to bring the poles into the s plane. Under the Graham, A., "The Phenomenological Method in Rheology,"
transformation Xm many of the poles in the Xplane map onto Research, London, 1953', pp. 92-96.
7
Riemann surfaces not included within the contour of in- Caputo,.M , Elasticitae Dissipazione, Zanichelli, Bologna, Italy,
tegrationthn esTh1969plane. . e oresidues f
8
contribute to the response of the ,system. The index / in the Caputo, M., "Vibrationso af n Infinite Plate witha Frequency
summation is therefore allowed to assume only those values Independent Q," Journalthf e Acoustical Society of America, Vol.
correspondingt o poles within e closed contourithn es •60,
9
No. 3, Sept. 1976,p. 637.
Caputo, M. and Minardi, F., Pure and Applied Geophysics, Vol.
plane. 91, 1971,pp. 134-147.
10
Bagley, R. L., "Fractional Derivative Models for a Family of
Acknowledgments Corning Glasses," Technical report to be published by Air Force
This work was sponsored by the Metals Behavior Branch, Wright Aeronautical Laboratories.
n
Metalsan d Ceramics Divisionoth f e Materials Laboratory, Bagley,R L. an. d Torvik,P . J"., A Theoretical Basisfo thr e
Wright Aeronautical Laboratories, Wright-Patterson AFB, Application f Fractional Calculust o Viscoelasticity," o be
Ohio. In particular, we are indebted to Drs. Jack Henderson, published in Journal of Rheology.
12
David Jones, and Ted Nicholas of the Metals Behavior Graves, G., Cannon, C., and Kumar, B., "A Study to Determine
the Effecto f Glass Compositional Variationson Vibration Damping
Branchfo r their encouragementan d suggestions. Partso f this
Downloaded by AFRL D'Azzo Wright-Patterson on July 8, 2016 | http://arc.aiaa.org | DOI: 10.2514/3.8142

Properties,"Ai r Force Wright Aeronautical Laboratories, TR-80-

paper are extracted from a Ph.D. dissertation submitted by 4061,Ma y 1980.
13
th e first authorttho Aie r Force Instituteo f Technology. Jones,D I . . G., "Viscoelastic Materialsfo r DampingAp -
plications," Damping Applications for Vibration Control, edited by
P. J. Torvik, ASME, New York, 1980, pp. 27-51.
References 14
Zienkiwicz,O . CTh., e Finite Element Methodi n Structuralan d
iRagley, . L., "Applications f Generalized Derivatives o Continuum Mechanics, McGraw-Hill Book Co., London, 1967, p.
Viscoelasticity," Air Force Materials Laboratory, TR-79-4103, Nov. 220.
1979. 15Christensen,R . M., Theory f Viscoelasticity,A n Introduction,
2
Crandall,S.H. , "Dynamic Responsesof Systems with Structural Academic Press,Ne w York, 1971,pp . 207-218.
16
Damping, "Air, Spacean d Instruments, Draper Anniversary Volume, Foss,K . A., "Co-ordinates Which Uncoupleth e Equationso f
editedb Hy S . . Lee, McGraw-Hill Book CoNe., w York, 1963,pp . Motion of Damped Linear Dynamic Systems," Journal of Applied
183-193. Mechanics, Vol. 25, Sept. 1958, p. 361.

AIAA Meetings of Interest to Journal Readers*

Meeting Call for
Date (Issueo f X\/X\yA Bulletinn which program will appear) ' Location Papersf
1983
May 2-4 AIAA/ASME/ASCE/AHS 24th Structures, Structural Sahara Hotel
Dynamics& Materials Conference (Mar.) LakeTahoe, Nev. June 82
May 10-12 AIAA Annual Meetingan d Technical Display Long Beach
(Aerospace Engineering Show)(Apr.) Convention Center
Long Beach, Calif.
June 1-3 AIAA 18th Thermophysics Conference (Apr.) The Queen Elizabeth Sept. 82
Hotel, Montreal,
Quebec, Canada
June 27-29 AIAA/SAE/ASME 19th Joint Propulsion Conference Westin Hotel ' Sept8. 2
and Technical Display (Apr.) Seattle, Wash.
July 12-14 AIAA 16th Fluidan d Plasma Dynamics Conference (May) Radisson Ferncroft Oct. 82
Hotel and Country
Club, Danvers, Mass.
July 13-15 AIAA 6th Computational Fluid Dynamics Conference (May) Radisson Ferncroft Oct.28
Hotel and Country
Club, Danvers, Mass.
July 13-15 AIAA Applied Aerodynamics Conference (May) Radisson Ferncroft Oct. 82
Hotelan d Country
Club, Danvers, Mass.
*Fora complete listingo f AIAA meetings,se eth e current issueothf e AIAA Bulletin.
tissue of AIAA Bulletin in which Call for Papers appeared.
View publication stats