You are on page 1of 28

Fourier and Asymptotic Analysis of Nonlinear Free Vibrations of a Beam with

a Hysteretic Damage Plane

Daniel A. Mendelsohn

Department of Mechanical Engineering

The Ohio State University

(mendelsohn.1@osu.edu)

and

Claudio Pecorari

Hesjakollen 111, 5142 Fyllingsdalen

Bergen, Norway

(claudio.pecorari@hotmail.com)

Submitted For Publication in the Journal of the Acoustical Society of America

Keywords: Hysteresis, nonlinear vibrations, damage detection, damage characterization

1
I. Introduction

Material characterization based on vibration characteristics is an important quantitative

non-destructive evaluation (QNDE) tool. Variations in structural and material compliance due to

cracks and other material defects cause reductions in the natural frequencies and changes in the

mode shapes. Crack geometry may also affect the dynamics of a structure, so that knowledge of

the relationships between the two may be relevant to a correct assessment of the performance

integrity of structures.

The reduction in natural frequencies produced by transverse cracks in linear elastic beams

and similar thin structures has been studied extensively both theoretically and experimentally.

An extensive review can be found in Ref. 1, as well as in some more recent studies 2-13 and the

numerous references cited therein. The majority of this work concerns linear and elastic

behavior and only two types of nonlinear mechanisms have been studied.

The first mechanism is intermittent crack face contact, known as the breathing crack

phenomenon. In this case, partial closure of the crack’s faces during a vibration cycle produces

higher harmonics and couples the bending vibration with the longitudinal motion.1,14,1516,17 The

emphasis in all of these investigations, some experimental and some theoretical, is on the fact

that crack closure can reduce, by almost half, the fundamental frequency shift, which can be used

to locate and size a crack. The higher harmonics and their dependence on the nonlinearity or the

initial amplitude of excitation were not of much interest to these investigators.

A second type of mechanism of interest involves spatially confined nonlinearity of the

material’s properties. References 12 and 13, for instance, investigate the bending response of a

beam with an elastic quadratic nonlinearity distributed across the ligament of a transverse crack

plane. The quadratic nonlinearity is derived from quasi-static fracture mechanics calculations of

2
the moment-slope-discontinuity relations for a crack with a cohesive zone in the ligament ahead

of the crack edge. A tensile static preload was assumed, which prevents crack face contact. The

nonlinear response of this structure to the first order in the perturbation parameter consisted in

the generation of the second harmonic component. Harmonics of higher order were shown to

appear only as corrections proportional to higher powers of the perturbation parameter.

The configuration in which a beam containing an open crack is vibrated lends itself also

to the examination of a second dynamic nonlinear effect. In fact, within the cohesive or

otherwise damaged region ahead of a crack tip, fatigue produces high density distributions of

dislocations, the dynamics of which is known to be characterized by hysteresis 18. A similar

situation may be also met in a fatigued material prior to crack initiation while the coalescence of

dislocations takes place.

In the present analysis, the nonlinear dynamics of a slender beam containing an open

crack with a cohesive hysteretic region around its tip is examined. The specific macroscopic

hysteretic behavior considered here has already been investigated by other authors in connection

to nonlinear wave propagation phenomena in hysteretic materials 19,20, and is characteristic of

rocks and other man-made geomaterials. Mechanisms that may produce this hysteretic behavior

may vary from friction21 between asperities of partially closed micro-cracks, to double-well

potentials controlling the state of micro- or mesoscopic material constituents 22. This type of

hysteresis resembles also that which is caused by the interaction between moving dislocations

and point defects distributed along the glide planes of the former 23. Given the variety and

complexity of mechanisms governing the interaction of dislocations with a material’s

microstructure, the goal of the present work is limited to developing a mathematical framework

3
within which specific contributions to the nonlinearity caused by dislocation dynamics can be

built in.

As in Ref. 13, in view of its limited extension in comparison with the wavelength of the

vibration considered here, the damaged region is modeled as a thin interface with suitable

boundary conditions. Section 2 describes the formulation of the vibration problem with its

associated nonlinear boundary conditions. The solution is found by a perturbation approach that

substantially differs from that adopted by these authors in earlier publications 23,24,13, although it

still relies on the harmonic balance method and on considerations about the order of magnitude

of the nonlinear effects with respect to the perturbation parameter. Section 3 presents numerical

results that illustrate the time evolution and magnitude of the nonlinear dynamics as functions of

the strength of the nonlinearity. A final section summarizes the most significant points of this

investigation.

II. Nonlinear Beam Vibration Formulation


A. The Hysteretic Boundary Condition
Figure 1a shows a simply-supported beam of length L and depth d with a thin plane of

damaged material located at x = c. While the analysis here does not depend on a particular

mechanism of damage, in practical situations a localized damage plane is usually realized by

cracking or notching the beam. Therefore, an edge-crack with an associated damage extending

in the plane of the crack over the whole cross section of the slender bean is envisioned. Together

they cause an average discontinuity in either the deflection and/or the slope of the deflection

across the damage plane as shown schematically in Fig. 1b. The constitutive relations for this

type of damage relate the local shear force plane to the deflection discontinuity, and the local

bending moment at the damage plane to the slope discontinuity at the damage plane. We ignore

4
the actual thickness of the damaged region or interphase between the two beam sections and

assume that these boundary conditions may be modeled as springs which connect the beam

sections on either side of the damage plane to each other 25. This limits the sensitivity of the

model BCs to frequencies distinctly less than that corresponding to the wavelength of the actual

thickness of the damaged interphase. This is not a limitation here since the highest harmonic

frequencies considered correspond to wavelengths on the order of tens of millimeters for a beam

of length 1/3 m, which is much larger than the thickness of most localized damage planes.

Consider the two types of deformation depicted in Fig 1b and let the generalized load P

represent the bending load M or shear force Q. Let also the generalized displacement 

represent the discontinuity in slope  or in deflection w. Then, considering a wide range of

possible plastic damage mechanisms, we postulate that as the cracked beam element in Fig. 1b is

loaded quasi-statically and monotonically from zero there is an essentially linear range for P

where there is negligible nonlinear deformation in the plane ahead of the crack. As the load is

increased and appreciable plastic behavior occurs, most damage mechanisms exhibit softening,

at least initially. A schematic of this general type of quasi-static monotonic constitutive behavior

is shown in Fig. 2a.

The following general scenario for dynamic interrogation of the damage state is

considered next. A cracked beam is loaded quasi-statically into the nonlinear softening region so

that only one of damage modes, i.e., bending or shear, is activated, say, to the state A in Fig 2.

From state A, the load is decreased by an amount PD which corresponds, for example, to a

variation of the slope discontinuity equal to D. Then, the beam is set into a chosen free

vibration mode by an initial forced excitation at the known linear frequency and amplitude D

of that mode. The choice of mode and, at least in a lab experiment, of the location of the damage

5
plane allows one to eliminate one or other of the damage modes. An example of this strategy for

investigating single modes can be found in Ref. 26, which also discusses the practical advantages

of this approach with respect to a more conventional way to interrogate a material structure by

means of forced vibrations with sweeping frequency. It is assumed that the free-vibration P-

dynamic behavior follows the hysteresis loop in Fig. 2b. Until this model is informed by a

particular damage model and associated full 2D elastic-plastic analysis, which yields the quasi-

static Ptot-tot curve envisioned in Fig 2a, it is not particularly important where the hysteresis

starts or where the origin of the dynamic coordinate system is. Define the dynamic moment and

slope discontinuity (without subscripts) to be the difference of the total moment and slope

discontinuity from the center of the hysteresis loop:

1
P  Ptot   PA  PD   K1 2D ;    tot    A   D 
2

In this work, quadratic unloading and loading curves which are anti-symmetric about the line

connecting the two endpoints are considered

 1 
 P    K 0   K1   2   2D  sgn      D   
 2 
c

Equation links the load to the discontinuity via an effective stiffness of the bow-tie type. The

amplitude of the loop in the  direction, D, is a measure of how much energy goes into the

forced harmonic vibration that sets the beam into motion and is independent of the amount of

damage. Note that the average slope of the loop depends linearly on the amplitude of the loop.

This implies that the position of the center of the loop varies during the free vibration because of

dissipation effects that are later investigated in this work. The final state of the system at rest is

the one defined by the coordinates A - D, PA -PD). This is the state of the bar prior to the
6
beginning of the forced vibration. The amount of damage correlates to the area of the hysteresis

loop, which is controlled by the parameter K1. The previous remark on the average slope of the

loop means also that the average stiffness of the loop is less than the linear stiffness K0 by an

amount proportional to K1 and D. Since no specific damage mechanism is of interest here, the

coefficients K0 and K1 are free parameters in the current model study.

B. Nonlinear Free Vibration


Transverse deflection, w
 ( x , t ) , of a beam with cross-sectional area A, moment of inertia

I, and material of Young’s modulus E and density  is investigated within standard Euler-

Bernoulli beam theory. In anticipation of a non-dimensionalization, the tildes indicate

dimensional quantities. The free vibration problem is posed below for a beam containing a

damage plane at x  c (see Fig.1a). The equation for unforced motion is

 4 w  2 w
EI   A 0 ; 0  x  c and c  x  L , t  0
x 4 t 2

The continuity of bending moment and shear force across the damage plane are enforced by

  2 w    2 w 
 x 2    x 2  : t  0
  x c    x c 
  w 
3
  w 
3

 x 3    x 3  : t  0

  x c    x c 

The damage constitutive relations are

  2 w 
EI  2 
 x  x c 
1
   
 K B 0   K B1   2  D2 sgn   D   : t  0
2  

7
  3 w  1
 EI  3   K S 0 w  K S 1  w 2  w D2  sgn  w   w D w  : t  0
 x  x c  2

where  and w are the jumps (right minus left) in   w x and w , respectively, across the

damage plane, and the subscripts B and S of the stiffnesses refer to the bending and shear

springs, respectively, and subscripts 0 and 1 refer to the first and second order stiffnesses of the

nonlinear springs. The formulation is completed with the four standard support boundary

conditions for the beam. In the present investigation we restrict ourselves to simply supported

conditions (zero deflection and bending moment) at each end: x  0 and x  L .

The associated linear problem is obtained by setting L and  equal to zero in the

nonlinear boundary conditions, Eqns. and . The resulting linear eigenvalue problem has been

solved by many researchers, e.g. by Yokoyama and Chen5 and in the present context by

Mendelsohn, et al.11, 13. With the damage plane at the midpoint of the beam there are an infinite

number of distinct bending (symmetric) and shear (anti-symmetric) damage modes, which

exhibit slope and deflection discontinuities and have natural frequencies which are found to be

up to 20% smaller than the corresponding frequency of the intact beam for crack lengths up to

half the beam depth. Figure 3 shows typical linear deflection and slope mode shapes for the

second lowest shear damage mode of the elastically cracked beam (with symbols) and the

uncracked beam (without symbols).

Now introduce the following dimensionless variables (without tildes)

x c L  AL4 w ( x , t )
x ; c ; L   1 ; t  t / t ; t  ; w( x, t ) 
L L L EI C10

8
where t is the usual time constant for beam vibrations which relates the frequency to the wave

number in the usual manner, see Eq. . The constant C1 is the initial value of the one

undetermined deflection amplitude constant in the first order (fundamental) eigenvalue problem

posed below. By choosing the appropriate excitation mode and damage plane location, only one

or the other of the nonlinear springs may be excited; the other one remaining inactive because the

damage plane is located at a node for either the moment or shear force. The calculations will

only be done when this is the case, and we set the inactive spring nonlinearity to zero. C10

therefore represents the amount of energy in the forced excitation which initiates the free

vibration. We introduce the following dimensionless measures of the strength of the nonlinearity

for these two situations:

K S 1C10
Case S  shear spring active: B  0 ; S 
KS 0
K B1C10
Case B  bending spring active :  B  ; S  0 ,
 B0
LK

In the former  S is assumed to be small and in the latter  B is assumed to be small, and they are

used as perturbation parameters. The dimensionless equation of motion and boundary

conditions,

Eqs. - can then be rewritten.

4w 2w
  0 ; 0  x  c and c  x  1 , t  0
x 4 t 2
 2w   2w 
 x 2    x 2  : t  0
  x c    x c 
 3w   3w 
 x3    x3  : t  0
  x c    x c 

9
 2w  KB0L
 x 2   EI f ( B ,  D ,  ,  ) : t  0

  x c 
  3w  K S 0 L3
 x3    f ( S , wD , w, w ) : t  0
  x c  EI

 1  1 
f ( ,  D , ,  )   1    D       2   2D  sgn    
 2  2 

Note that the form of the dimensionless equation of motion results from the non-

dimensionalization of the space and time coordinates and has nothing to do with the non-

dimensionalization with respect to the amplitude of the fundamental eigenmode. Also note that

 and w are normalized the same way the deflection is in Eq. .

C. Combined Fourier and Asymptotic Analysis of the Nonlinear Problem

1. Derivation of the Ordered Boundary Conditions

The discontinuous and nonlinear nature of the hysteresis relation is treated by expanding

all quantities in complex Fourier series. For instance,

w  x, t    n  x  e jn
n

 1
where  t, and  is the unknown corrected dimensional fundamental frequency. All

summations are assumed to go from -∞ to +∞. Substituting eqn. (14) into eqn. yields ordinary

differential equations in the mode shapes with general solution

10
 ( x)  Cn sin  kn x   Dn cos  k n x   En sinh  kn x   Fn cosh  k n x  ; 0  x  c
 n  x    nL
 nR ( x )  Gn sin  kn x   H n cos  k n x   I n sinh  kn x   J n cosh  k n x  ; c  x  1

The dimensional counterpart of C1 in Eq. is C1 , which is the normalization constant for all

deflection quantities, including all of the amplitude constants in Eq. . Note then that

C1
C1   1 . The dimensionless wave number and frequency are defined as
C1

kn  kn L ; n   n t

and Eqs. and require, respectively,

t t
kn2  2  n  2 n1 ; kn2  n  n1
L L

The time constant t is given in Eq. .

The solution method is illustrated with Case S, the situation in which the shear spring is

active and nonlinear and hysteretic, and the bending springs are inactive. Anticipating the

application of each of the BCs at each n = 1, 2, 3, …, we now expand the jumps in deflection at x

= c as well:

w(t )    n e jn     nR (c)   nL (c)  e jn


n n

from which it follows that

    
w2 (t )    q e jq    p e jp     p   qe j p  q       p( r  p )e jr
 q  p  p  q  r p

11
We assume that the velocity of the deflection discontinuity in Eq. can be approximated using the

time dependence of the linear problem, taken to be cos(Lineart), and that the nonlinear effects are

all first order in  at least. In this case, the function f in Eqs. and becomes

 1  1  
f  1   S wD    ne jn   S   n( p n )e jp  wD2   d q e jq
 2  n 2  n p  q
 1  1  
 1   S wD    ne jn   S   d q n( p n )e j ( p  q )  wD2  d q e jq 
 2  n 2  n p q q 

where
 sgn  w   sgn  sin  Linear t     d n e jn
n

0 , n  0, 2, 4,6,...
1  
sgn  sin     e d   2 j
2
dn  
jn

2 0
 , n  1,3,5,... 
 n 

Rename n above as r and set the arbitrary indices p and q so that they sum to the new index n.

Then, there is a common exponential and sum over n in all three terms, and f can be reduced to a

sum of complex exponentials

f   f n e jn
n

 1  1  
f n  1   S wD n   S   d ( n  p )r( p r )  wD2 d n 
 2  2  p ,r 

The higher harmonic amplitudes of the deflection discontinuity ( n , n  2) are due only

to the  terms in f and must be of O(S) or smaller. Further we assume that 1 and  1 are the

only components of O(). Retaining only terms up to O(S) in the expression for fn, the terms

that survive in the double sum are those that contain products of 1 and  1 , which are

12
associated with the following combinations of indices: (r = 1, p = 2), (r = 1, p = 0), (r = -1, p =

0), (r = -1, p = -2). Thus, Eq. (22) simplifies as

 1  1
f n  1   S wD n   S  d n 211  2dn1 1  d n  2 1 1  wD2 d n 
 2  2

Note that d-n = -dn and that 1 and  1 are each other’s conjugates. Thus, adding to Eq. its

conjugate and making use of Eq. , both the 1  1 and the dn term in the second brackets in Eq.

are eliminated and the result is

 1  1
2Re  f n   1   S wD  2Re  n    S  d n  211  d n  2 1 1  d  n  2 1 1  d  n 211 
 2  2
 1  1
  1   S wD  2Re   n    S  d n 2  d  n 2   11   1 1 
 2  2
 1   8 S 
  1   S wD  2Re   n     Re  j11 
 2     n2  4
 

Bearing in mind that only the real part of a complex expression has physical meaning, the full

complex quantity rather than the real part only can be considered in each instance. This yields

the following final form for the Fourier coefficients of the function f in Eq. .

 1  4j 
 1   S wD 1   S11 ; n  1 
 2  3 
 1  4 j 
f n   1   S wD  n   S11 ; n  3,5,7,...
 2    n  4
2

 1 
 1   S wD  n ; n  2,4,6,... 
 2  

13
Making use of Eqns. and and eliminating the common time dependence at each n, the simple

support BCs and damage plane BCs at each base mode m and for each Fourier harmonic, n,

become

 nL (0)   nL (0)   nR (1)   nR


 (1)  0
 (c)   nR
 nL  (c)
 (c)   nR
 nL  (c)
Case S K L
 (c)  BO  nR
 nL  (c ) 
 (c)   nL
EI
K L3
 (c)   S 0 f n
 nL
EI
Note that the boundary equations of the linear problem are obtained from Eq. by replacing fn,

 nL , and  nR with  Linear   LinearR (c)   LinearL (c)  ,  LinearL , and  LinearR , respectively.

The analysis of the problem in which the bending spring is active while the nonlinear and

the shear spring are inactive, heretofore referred to as Case B, is completely analogous to Case S

above. The Fourier coefficients in Eq. of the hysteresis function remain as they are, except that

the coefficients   n are now the Fourier coefficients of the slope discontinuity, rather than the

deflection discontinuity,

 (t )    ne jn     nR  ( c)  e jn
 (c)   nL
n n

The resulting BCs are


 nL (0)   nL (0)   nR (1)   nR  (1)  0
 nL (c)   nR
 (c)
 nL (c)   nR
 (c)
Case B K L
 nL (c)  BO f n
EI
K S 0 L3
 nL (c)    nR (c)  nL (c)
EI

14
where again the coefficients fn of the hysteresis function are given by Eq. .

2. Solution of the Corrected Fundamental Free Vibration Problem

We first solve the problem for the fundamental Fourier contribution (|n|=1). For Case S

(shear spring is active), the shear spring BC in Eq. reads for n  1

K S 0 L3  1  4j 
Case S  1L (c)    1   S wD    S1  1
EI  2  3 

The n  1 term adds no information since  1 is the complex conjugate of 1 . This BC is

nonlinear in 1 , but due to the small magnitude of the nonlinearity, it can be solved by iteration.

The first guess consists of replacing the 1 in the bracket with its known real counterpart from

the associated linear problem, 1 . Using the definition of  n in Eq. and Eq. in Eq. along with

the other n  1 BCs in Eq. yields a linear complex eigenvalue problem for the complex wave

number k1 and the eight complex mode shape constants; C1 , D1 , E1 , ... J1 . After each iteration,

1 in the bracket in Eq. is replaced by the 1 calculated from the previous iteration. The

calculation is stopped when the wave number converges to some tolerance.

Once again the situation is analogous for Case B, in which the bending spring is active.

The nonlinear BC for the fundamental component from Eqn is

K B 0 L  1  4j 
Case B  1L (c)   1   B  D    B1  1
EI  2  3 

15
and 1 is the fundamental Fourier coefficient of the slope discontinuity. The iteration procedure

described above holds except that Eqns and are used instead of Eqns and .

3. Calculation of The Higher Harmonics (|n|=2,3,4,…)

The shear damage BC in Eq. reads for even and odd harmonics

Case S

 1  4j 
  1   w       , n  3,5,7,...
K S 0 L3  2
S D

n
  n  4
2 S 1 1

 nL (c)    
EI  1 
1   w  , n  2,4,6,.. 
 2 S D  n 

Since the Fourier coefficient of the fundamental component of the deflection discontinuity, 1 , is

known up to a free constant, this and the other BCs in Eq. for each odd higher harmonics

consist of a “forced” linear system for the amplitude constants Cn , Dn , En , ... J n which

completely define the nth harmonic. For all even n these equations are homogeneous in the same

unknown constants, for which the determinant of the coefficient matrix is not zero. Hence there

are no even harmonics. For the case of the bending spring being active, the higher harmonics are

generated by the bending spring BC from Eq.

 1  4j 
  1           , n  3,5,7,...
  n2  4 
B D n B 1 1
K B 0 L  2  
Case B  nL (c)   
EI  1 
1     , n  2,4,6,.. 
 2 B D  n 

16
Recall that in this case 1 is the fundamental Fourier coefficient of the slope discontinuity, and

there are no even harmonics because the equations are homogeneous and the determinant of the

coefficient matrix is not zero.

III. Numerical Results and Discussion


The damage plane is taken to be in the center of the beam, c = L/2 = 0.167 m, and the

material is taken to be aluminum with Young’s modulus E = 72.8 GPa, Poisson’s ratio  = 0.3,

and mass density  = 2700 kg/m3. The depth of the beam is d = 0.025 m. The linear stiffness per

unit thickness K 0 is taken from a calculation for a cracked beam with crack length of about half

the beam depth. Recall that the thickness of the beam is unspecified and the moment and

stiffness are assumed to be the moment per unit thickness. We confine ourselves here to the

nonlinear correction of the fourth linear mode (Case A) or the third linear mode (Case B). For

brevity we show only the Case A results. Case B results are quite similar.

Figure 4 illustrates the imaginary parts of the corrected fundamental dimensionless

frequency versus time (Fig. 4a), and versus S (Fig. 4b). The normalized time used in Figs. 4a, 5a,

6a and 7a-f is the number m of elapsed vibration cycles based on the linear dimensionless period

[TL = 0.04035 (Case A, 2nd shear mode) and TL = 0.009157 (Case B, 2nd bending mode)], which

is the dimensional period divided by the time constant t  0.002965 s . The dimensional frequency

is the dimensionless frequency divided by t . The time evolution of the beam’s vibration is

followed by updating at the end of each cycle the amplitude, wavenumber, frequency of the

vibration with a new value which accounts for the attenuation predicted by the imaginary

component of the complex frequency by assuming that the amplitude constant C1 decays as

exp(-Im(1) t). Also note that the definition of the nonlinearity parameter S, Eq. , implies that

17
keeping the amplitude of the initial excitation, C10 , fixed and varying S by varying the nonlinear

stiffness ratio, KS1/KS0 , is the same as keeping KS1/KS0 fixed and varying C10 . This should be

kept in mind when interpreting the dependence of all of the results on S.

Figure 4b illustrates that the imaginary part of the frequency, which determines the

damping of the oscillation and is due solely to the nonlinearity, increases monotonically with

increasing nonlinearity, and decreases monotonically with increasing time. The dependence on

S is only linear at the initial time and becomes increasingly nonlinear for small S, as time

increases, but becomes approximately linear for larger S with decreasing slope as time increases.

As expected, the rate of decrease in time of the imaginary part of the frequency increases

dramatically with increasing S. because the larger the imaginary part of the fundamental

frequency the more damping there is.

Figure 5 shows the real part of the corrected fundamental dimensionless frequency versus

time (Fig. 5a) and versus the nonlinearity parameter S (Fig. 5b). The values are normalized with

respect to the real linear natural dimensional frequency (8358 Hz-Case A, 3883 Hz-Case B).

The correction to the real part of the frequency increases monotonically, though not linearly,

with increasing nonlinearity at all times considered and at it’s largest is only 0.084 % as large as

the linear frequency. Also as time goes on, the real part of the frequency increases as the spring

stiffens, approaching its linear value (S = 0) as the extent of hysteresis decays. This is consistent

with the observations of Van den Abeele, et al. 26 who, however, conducted their investigation on

composite materials with damage extended over the whole structure. The effect occurs very

slowly for small nonlinearity, but as the nonlinearity increases the stiffening becomes more

18
pronounced at earlier times and the return to linear is increasingly more rapid. The effect on the

real part of the frequency is only slightly stronger for Case B than for Case A.

Figure 6 shows the dynamic amplitude of the dimensionless deflection discontinuity

versus time (Fig. 6a) and versus the nonlinearity parameter S (Fig. 6b). The values are

normalized with respect to the unspecified initial amplitude constant C10 , which is a measure of

the energy of the forced excitation which initiates the free vibration and equals one half of the

maximum fundamental deflection along the beam. Comparing the dependence of this quantity

on the extent of nonlinearity to the behavior of the frequency in Figs. 4 and 5, it is seen that there

is a much stronger nonlinear dependence on S at small values of S .at later times, where the

value drops extremely sharply as S increases only slightly from zero. These curves contain data

for several values of S between zero and 0.05, which is the smallest value used in Figs 4 and 5.

They also contain data for much later times at these small values of S. This was necessary in

order to capture the full range of behavior and the transients in the results at small S. The

dashed curve in Fig. 6a corresponds to S = 0.05 and serves to illustrate the wide range in

magnitudes over the small change in S from zero to 0 .05 that is not seen in the frequency results

in Figs. 4 and 5. This is to be expected because wD is a direct measure of the extent of

hysteresis, while the frequency is a global measure of the stiffness of the entire beam. Also note

that the defelction discontinuity in Fig. 6 is a slightly increasing function of S up to a value of m

a little larger than 10, at which time it is equal to it’s linear value no matter what S is. For m >

10 the trend is reversed because the decay rate increases with increasing S. The behavior at

early times is consistent with the form of the real part of the stiffness, which decreases with

increasing S. As a final note on the computations it ws found that as time increases it takes

19
more iterations to converge and at a certain time which gets smaller the larger S is, the iteration

fails to converge.

Note in Fig. 3 that the nodes of the acceleration of the cracked beam are shifted slightly

towards the damage plane at the mid-span of the beam compared to the nodes of the intact beam.

The nodes of the corrected fundamental are shifted even further towards the damage plane.

Numerical simulations show that tracking the positions of the nodes by means of a point detector

after the beam is set into free vibration could be used as a comparative indicator of the damage.

However, it is felt that the most valuable information is in the higher harmonics which better

encode the dependence of their amplitude on the specific mechanism that generates them. To

that end, choosing a measurement point close to a node of the fundamental greatly increases the

ratio between the amplitudes of the higher harmonics versus the fundamental. For a beam of

length 1/3 m the total range of nodal positions is about 100 m for the range of strength of

nonlinearity considered here. We choose a point a fixed distance x = 0.0002 to the right of the

initial fundamental deflection nodal position in the left half of the beam (Fig. 3), which for a

beam of length 1/3 m corresponds to 67 m.

The Case A dimensionless fundamental acceleration amplitudes are plotted versus

normalized time m in Fig. 7a for various values of the nonlinearity parameter. The normalization

of the amplitude is with respect to the initial value of the free amplitude constant C10 . The

dimensionless odd harmonic amplitudes for n = 3 to 11 are shown in Figs. 7b-f, respectively.

The harmonic amplitudes are normalized with respect to the value of the instantaneous

fundamental amplitude shown in Fig. 7a, and as such are subject to competing trends. First,

there is energy dissipation during each hysteretic cycle causing both the higher harmonics and

the fundamental to decay, although the latter decay more rapidly. Note that this decay back

20
feeds on the area of the hysteretic cycle. On the other hand, since the fundamental decays on its

own, the normalized higher harmonic spectra decay less slowly than the actual spectra. The

results are both qualitatively and quantitatively similar for observation points up to two times

farther away from the initial fundamental nodal position than the observation point chosen. Note

also that the nodal position changes in time, but the measurement point is kept fixed for each S.

Also note that as can be seen in Eqs. and , the forcing term and the amplitudes of the higher

harmonics are proportional to the square of the amplitude of the fundamental. The initial

harmonic amplitudes are approximately proportional to S, but as is easily seen in Figs 7b-f, since

the time-scale of the decay depends on S, it is impossible to compare responses at the same time

for different values of S. Note that the dimensional harmonic amplitudes are proportional to the

square of C10 , and hence to the square of the amplitude of the initial excitation, while the

fundamental dimensional amplitude is propoartional to C10 .

In general, the higher the harmonic, the higher the decay rate of it’s amplitude. Also, in

all cases the decay of the fundamental and the harmonics are greater the greater the initial

nonlinearity. This leads to the distinction between three time regimes: (i) an initial regime

during which the magnitude of a harmonic component increases with increasing nonlinearity, (ii)

a late time regime in which the opposite is true, and (iii) an intermediate regime during which the

various curves are crossing each other and the dependence of amplitude on nonlinearity is not

monotonic. Note that all of the harmonics are most sensitive to the nonlinearity in the initial

time regime and in the late time regime.

IV. Conclusions

21
In this work, a mathematical framework has been presented which allows the nonlinear

hysteretic dynamics of a slender beam to be modeled. The cause of this nonlinear behavior

resides in a thin region confining hysteretic damage. In this investigation, the damage is

modeled by means of an imperfect interface with a real, bow-tie stiffness. The model can be

generalized to include other, and perhaps more realistic, constitutive relations. In addition to a

piecewise quadratic hysteretic bending spring, a linear spring has been used to account for

discontinuities in slope and transverse displacement, respectively, across the damage plane.

The coefficients in the hysteretic boundary conditions are purely real. This simplicity

notwithstanding, Eqns and contain imaginary terms which describe irreversible energy loss of

the fundamental harmonic component. A similar effect has been recently investigated in

connection to longitudinal vibrations of a hysteretic bar27. There, imaginary coefficients were

introduced in the constitutive relation of the material to account for the finite work performed by

the vibrating system itself or by an external source driving the system through the hysteresis

loop. These latter effects have not been accounted for here in order to focus the present analysis

on a novel source of attenuation that seems to be specific of vibrations of a beam with localized

damage. In fact, the mathematical derivation of the imaginary terms in Eqns (29) and (30) would

indicate their origin in a combination of hysteresis and nonlinear coupling of the type produced

also by a non-dissipative, classic nonlinearity. The complexity of the present mechanism can be

understood also by considering that its irreversible nature is a feature which clearly distinguishes

it from those responsible for the classical generation of higher harmonics. Similarly, the lack of

imaginary coefficients in the boundary conditions states an equally clear distinction from the

dissipative mechanisms discussed in Ref. 27.

22
A perturbation solution has been presented for the free vibration of the beam. For any

given mode of free vibration of the associated linear problem, a Fourier series representation of

the beam deflections is assumed. Retaining terms up to in the nonlinear boundary condition

leads to uncoupled problems for the corrected fundamental and each of the higher harmonics.

The corrected fundamental boundary condition is still nonlinear and the eigenvalue problem is

solved by iteration to obtain the corrected fundamental frequency. Due to the assumed time-

dependence of the hysteresis loop only odd higher harmonics are generated, and each of these are

found from a forced vibration problem with forcing term proportional to the small perturbation

parameter and the square of the amplitude of the fundamental.

A key feature of the model is the complex frequency due to the hysteretic part of the

nonlinearity. Worthy of notice is that the dependence of the imaginary part of the complex

frequency and the magnitude of the correction to the real part of the frequency on the vibration’s

initial amplitude is only quasi-linear at the initial time and becomes incresingly nonlinear as time

proceeds and damping occurs. Even initially the dependence departs from linear at higher values

of S. This implies that the coefficients of S in the stiffness start to show a small but measurable

dependence on S as well. The imaginary part of the corrected fundamental frequency increases

with the nonlinearity and decreases in time. The real part of the fundamental frequency

decreases with increasing nonlinearity and increases in time. In other words, for a given initial

value of the nonlinearity, as time proceeds, the real part of the frequency increases tending

towards the linear frequency as the nonlinearity itself decays in time. The same dependence is

seen for a shift in nodal position of the corrected fundamental compared to the linear mode

shape. The initial shift increases with initial nonlinearity and then decreases as time goes on

approaching the linear nodal position.

23
Choosing an observation point near a node of the corrected fundamental improves the

ratio between the amplitude of the higher harmonics versus that of the corrected fundamental.

The first five higher harmonics that are generated in both shear and bending, with particular

emphasis on the third one, show sensitivity to the nonlinear parameter that may be exploited for

characterization purposes. Two major trends may also be observed. The higher harmonics

decay faster as the nonlinearity increases. In conclusion, the time evolution of these spectra and

their dependence on the perturbation parameter show promise for identification of the hysteresis

loop parameters from free vibration measurements of a damaged beam.

REFERENCES
1
A. D. Dimarogonas, “Vibration of cracked structures: A state of the art review,” Engng Frac.

Mech. 55, 831-857 (1996).


2
T. G. Chondros and A. D. Dimarogonas, “Vibration of a cracked cantilevered beam,” J. Vib.

Acoust. - Trans. ASME 18, 221-228 (1997).


3
G. D. Gounaris and C. A. Papadopoulos, “Analytical and experimental identification of beam

structures in air or fluid,” Comp. and Struct. 65, 633-639 (1997).


4
T. G. Chondros, A. D. Dimarogonas and J. Yao, “A continuous cracked beam vibration theory,”

J. Sound. and Vib. 215, 17-34 (1998).


5
T. Yokoyama and M.-C. Chen, “Vibration analysis of edge-cracked beams using a line-spring

model,” Engng. Frac. Mech. 59, 403-409 (1998).


6
E. I. Shifrin and R. Ruotolo, “Natural frequencies of a beam with an arbitrary number of

cracks,” J. Sound and Vib. 222, 409-423 (1999).


7
M. A. Mahmoud, M. Abu Zaid and S. Al Harashani, “Numerical frequency analysis of uniform

beams with a transverse crack,” Comm. Num. Meth. Engng 15, 709-715 (1999).

24
8
Q. S. Li, “Dynamic behavior of multistep cracked beams with varying cross section,” J. Acoust.

Soc. Am 109, 3072-3075 (2001).


9
T. G. Chondros, “The continuous crack flexibility model for crack identification,” Fat. Frac.

Engng. Mat. and Struct. 24, 643-650 (2001).


10
S. S. Kessler, S. M. Spearing, M. J. Atalla, C. E. S. Cesnik, and C Soutis, “Damage detection in

composite materials using frequency response methods,” Comp. Part B-Engng. 33 87-95 (2002).
11
D. A. Mendelsohn, “Free vibration of an edge-cracked beam with a Dugdale-Barenblatt

cohesive zone,” J. Sound Vib. 292, 59-81 (2006).


12
P. S. Mokashi and D. A. Mendelsohn, “Nonlinear vibration of an edge-cracked beam with a

cohesive zone, I: Nonlinear bending load-displacement relations for a linear softening cohesive

law,” J. Mech. Matls. Struct. 3, 1573-1588 (2008).


13
D. A. Mendelsohn, S. Vedachalam, C. Pecorari, and P. S. Mokashi, “Nonlinear vibration of an

edge-cracked beam with a cohesive zone, II: Perturbation analysis of Euler-Bernoulli beam

vibration using a nonlinear spring for damage characterization,” J. Mech. Matls. Struct. 3, 1589-

1604 (2008).
14
T. G. Chondros, A. D. Dimarogonas, and J. Yao, “Vibration of a beam with a breathing crack,”

J. Sound Vib. 239, 57-67 (2001).


15
J. A. Brandon, E. M. O. Lopez, and A. E. Stephens, “Spectral indicators in structural damage

identification: a case study,” J. Mech. Engng. Sci.-Proc.Inst. Mech. Engnrs. C 213, 411-415

(1999).
16
A. S. Sekhar and P. Balaji Prasad, “Crack identification in a cantilever beam using coupled

response measurements,” J. Engng. Gas Turb. Power 120, 775-777 (1998).

25
17
R. Ruotolo, C. Surace, P. Crespo, and D. Storer, “Harmonic analysis of the vibrations of a

cantilevered beam with a closing crack,” Comp. Struct. 61, 1057-1074 (1996).
18
A. Granato, K. Lücke, Theory of mechanical damping due to dislocations, J. Appl. Phys. 27,

583-593 (1956).
19
V. Gusev, “Theory of non-collinear interactions of acoustic waves in an isotropic material with

hysteretic quadratic nonlinearity,” J. Acoust. Soc. Am. 111, 80-94 (2002).


20
V. Gusev, “Propagation of acoustic pulses in material with hysteretic nonlinearity,” J. Acoust.

Soc. Am. 107, 3047-3058 (2000).


21
V.A. Aleshin, K. Van Den Abeele, Friction in unconforming grain contacts as a

mechanism for tensorial stress-strain hysteresis, Mech. Phys. Solids 55 (2007) 765-787.
22
V.A.Aleshin, K. Van Den Abeele, Micro-potential model for stress-strain hysteresis

of micro-cracked materials, Mech. Phys. Solids 53, (2005) 795-824.


23
C. Pecorari and D. A. Mendelsohn, “Hysteretic constitutive relations including dislocations-

point defects interactions”, unpublished work.


23
C. Pecorari, “Nonlinear interaction of plane ultrasonic waves with an interface between rough

surfaces in contact,” J. Acoust. Soc. Am. 113, 3065-3072 (2003).


24
C. Pecorari, “Adhesion and nonlinear scattering by rough surfaces in contact: beyond the

phenomenology of the Preisach-Mayergoyz framework,” J. Acoust. Soc. Am. 116, 1938-1947

(2004).
25
J. R. Rice and N. Levy, “The part-through surface crack in an elastic plate,” J. App. Mech.-

Trans. ASME 39, 185-194 (1972).

26
26
K. Van Den Abeele, P. Y. Le Bas, B. Van Damme and Tomasz Katkowski, “Quantification of

material nonlinearity in relation to micro damage density using nonlinear reverberation

spectroscopy: Experimental and theoretical study,” J. Acoust. Soc. Am. 126, 963–972 (2009).
27
C. Pecorari, D.A. Mendelsohn, Nonlinear longitudinal forced vibration of a hysteretic bar: an

analytical solution, to appear in Wave Motion.

List of Figure Captions

Figure 1. (a) Simply supported beam with a damage plane at x = c. (b) Schematic of the

transverse shear and bending loading and resulting kinematic discontinuities of a section of beam

containing the damage plane.

Figure 2. (a) A generic damage plane softening nonlinear load versus kinematic discontinuity

relationship with (b) a schematic of the assumed dynamic hysteresis loop shown at some

statically held applied load Ps.

Figure 3. Deflections and slopes of the fourth base mode of vibration (2 nd shear mode) of the

associated cracked (line with symbols) and uncracked (solid line) linear problems.

Figure 4. Imaginary part of the corrected fundamental dimensionless frequency (a) versus

normalized time at various values of S and (b) versus S at various dimensionless times.

Normalized time is the number m of elapsed hysteretic cycles. Dimensional frequencies are

dimensionless frequencies times the time constant t  2.965  10-3 s .

Figure 5. Real part of the corrected fundamental frequency normalized with respect to the linear

frequency (a) versus normalized time at various values of S and (b) versus S at various

normalized times.

27
Figure 6. Amplitude of the dimensionless deflection discontinuity (a) versus normalized time at

various values of S and (b) versus S at various normalized times. The dimensionless deflection

is normalized with respect to the unspecified amplitude constant C10 , which is a measure of the

energy of the forced excitation which initiates the free vibration.

Figure 7. Dimensionless fundamental and harmonic acceleration amplitudes versus

dimensionless time for various values of S. (a) fundamental dimensionless amplitude, (b) 3rd

harmonic, (c) 5th harmonic, (d) normalized 7th harmonic, (d) 9th harmonic, and (e) 11th harmonic.

The fundamental amplitude is normalized with respect to the unspecified amplitude constant C10

, which is a measure of the energy of the forced excitation which initiates the free vibration. The

harmonics are normalized with respect to the instantaneous value of the dimensionless

fundamental in (a), which is equivalent to normalizing the dimensional harmonic amplitudes

with respect to the dimensional fundamental amplitude.

28

You might also like