CRACK DETECTION IN BEAM-TYPE STRUCTURES USING

FREQUENCY DATA

Jeong-Tae Kim1) and Norris Stubbs2)

1) Associate Professor, Department of Ocean Engineering, Pukyong National University

2) A.P. & Florence Wiley Endowed Professor, Department of Civil Engineering,
Texas A&M University

Address of Corresponding Author:

Prof. Jeong-Tae Kim, Dept. of Ocean Eng., Pukyong National University,

Nam-gu, Pusan 608-737, Korea
(Tel: +82-51-620-6227; Fax: +82-51-62-8146; E-mail: [email protected])

1
 ABSTRACT

A method to nondestructively locate and estimate the size of a crack by using

changes in natural frequencies of a structure is presented. First, a crack detection

algorithm to locate and size cracks in beam-type structures using a few natural

frequencies is outlined. A crack location model and a crack size model are formulated

by relating the fractional changes in the modal energies to the changes in the natural

frequencies due to damage such as cracks or other flaws. Next, the feasibility of the

crack detection scheme are evaluated for several damage scenarios by locating and

sizing cracks in test beams for which a few of the lower natural frequencies are

available. By applying the approach to the test beams, it is observed that crack location

can be estimated accurately. The crack-size can also be estimated accurately.

Keywords: nondestructive, crack detection, modal parameters, beam-type structure,

system identification,

2
1. INTRODUCTION

During the past two decades, a significant amount of research has been conducted in

the area of nondestructive damage evaluation (NDE) via changes in the dynamic modal

responses of a structure. The NDE methods developed to date can be classified into four

levels (Rytter, 1993): (1) Level I Methods, i.e., those methods that only identify if

damage has occurred (Vandiver, 1975; Crohas and Lepert, 1982); (2) Level II Methods,

i.e., those methods that identify if damage has occurred and simultaneously determine

the location of damage (Cawley and Adams, 1979; Pandey et al., 1991; Chance et al.,

1994); (3) Level III Methods, i.e., those methods that identify if damage has occurred,

determine the location of damage as well as estimate the severity of damage (Stubbs and

Osegueda, 1990; Wu et al., 1992; Kaouk and Zimmerman, 1994; Kim and Stubbs,

1995); and (4) Level IV Methods, i.e., those methods that identify if damage has

occurred, determine the location of damage, estimate the severity of damage, and

evaluate the impact of the damage on the safety or useful life of the structure.

Despite these combined research efforts in Levels II and III methods, several

problems remain to be solved before damage assessment in real structures becomes a

routine activity. Among these problems, a need remains to develop practical theories of

damage detection to simultaneously predict the location of damage and estimate the

geometry of the damage (e.g., quantification of crack-depths or extent of corrosive

zones) in structures. A need also remains to circumvent the reality of being capable of

measuring only limited modal information such as only the lower modes with restricted

degrees of freedom.

Research studies to nondestructively detect crack location and magnitude via the

3
change in natural frequency have been performed by many researchers. Attempts have

been made to relate changes in natural frequencies to changes in beam properties such

as cracks, notches, or other geometrical changes (Gudmunson, 1982; Cristides and

Barrs, 1984; Ricos and Aspragathos, 1990) and to identify crack location and magnitude

in a beam from vibration modes (Adams and Cawley, 1979; Ostachowicz and Krawczuk,

1990; Sundermeyer and Weaver, 1993; Kim et al., 1998)

The objective of this paper is to present a practical methodology to nondestructively

localize cracks and estimate the sizes of the cracks in beam-type structures using

changes in frequencies. This objective is achieved via the following approach. First, we

outline a crack detection algorithm to locate and size cracks in beam-type structures

using a few natural frequencies. A crack location model and a crack size estimation

model are formulated by relating fractional changes in modal energy to changes in

natural frequencies due to damage such as cracks or other geometrical changes. Next,

we demonstrate the feasibility and practicality of the crack detection scheme by locating

and sizing cracks in test beams. Finally, we assess the accuracy of the crack detection

results obtained from test beams for which natural frequencies were measured for

several damage scenarios (Silva and Gomes, 1990).

2. CRACK DETECTION METHOD

The scheme shown in Fig. 1 represents a crack detection method that yields

information on location and geometrical size of damage directly from changes in modal

characteristics of a target structure. The modal characteristics of interest here are natural

frequencies and mode shapes. Once, two sets of modal parameters are experimentally

4
measured for the as-built (reference) structure and the corresponding damaged structure,

the crack detection methodology outlined here can be used to predict crack locations

and to estimate the sizes of the located cracks. Note that system identification (SID)

techniques can be adopted to generate plausible baseline modal parameters if there are

no baseline field records for the reference structure. (Kim and Stubbs, 1995)

2.1 Crack-Size Model

With reference to Fig. 2, suppose we are given a pristine (i.e., undamaged) MDOF

structural system with the ith natural frequency  i and the ith mode shape  i . Next,

assume that at some later time the structure is damaged (i.e., the occurrence of flaws

such as the crack shown in Fig. 2) at one or more locations of the structure. The

resulting characteristic equation of the damaged structure yields the ith natural

frequency  *i and the ith mode shape  *i . (Note that the damaged structure is

characterized by the asterisk.)

Assuming no volume or mass changes due to the cracks or other geometrical

changes, Gudmunson (1982) proposed a first order perturbation method that predicts the

changes in natural frequencies of a structure resulting from the inflicted damage. For

small cracks, the fractional changes in modal strain energy can be related to the

fractional changes in frequency as follows:

Wi i
 (1)
Wi i
where Wi is the ith modal strain energy of the initial structure, Wi is the loss in the

ith modal strain energy after damage, and i i is the fractional change in the ith

eigenvalue due to the damage.

5
 In the present study, we limit our discussion of cracks to those systems which can be

modeled by Euler-Bernoulli beams. If the Euler-Bernoulli beam theory is used, the ith

modal strain energy Wi can be written as

L
1
Wi   2 EI  ( x )
2
i dx (2)
0

where E is Young’s modulus, I is the second moment of area, L is the beam span length,

and  i ( x ) is the ith mode shape function. Next, the strain energy for the crack

problem can be computed from the energy release rate by implementing linear elastic

fracture mechanics. On assigning plane strain condition to the cracked beam, the energy

loss rate of the ith modal strain energy is given by

 Wi (1   2 ) 2
t KI (3)
a E
where Wi a is the energy loss rate of the ith modal strain energy with respect to

crack depth, t is the beam thickness,  is Poisson’s ratio, and KI is the stress

intensity factor which depends upon the crack depth a, the applied flexural stress level

 , and the beam dimension (e.g., thickness t, height H and length L as shown in Fig. 1).

For the edge-crack case under going bending deformation (e.g., as shown in Fig. 2),

the stress intensity factor is given by

K I  F  a (4)

The term F is a geometrical factor which depends upon dimensionless crack depth

ratio a H (Anderson, 1995). For a rectangular beam with an edge crack,

F  1.122  1.40(a H )  7.33(a H )  13.08(a H )  14.0(a H )

2 3 4
(5)

Substituting Eqs. (4) and (5) into Eq. (3) and further integrating Eq. (3) over the crack

contour generates

6
   t (1   2 ) 2 2 2 
Wi   F  k a k  (6)
 2E i

in which, for the ith mode, ak  a ( x k ) represents the crack size at location xk and

 k   ( x k ) represents the maximum flexural stress at location xk along the beam’s

longitudinal axis. For the Euler-Bernoulli beam, the ith modal stress level is given by
1
 i ( xk )  E H  i( x k ) (7)
2
On dividing Eq. (6) by Eq. (2), the fractional change in the ith modal strain energy is

given by
Wi  t (1   2 ) H 2
 F 2 S ik a k2 (8)
Wi 4 I
and

Sik  k  i dx o  i dx

2 L 2
(9)

in which Sik represents the sensitivity of the kth location in the ith modal strain energy.

On substituting Eq. (2) - Eq. (8) into Eq. (1), we obtain a relationship between the

crack depth and the fractional changes in the ith eigenvalue as follows:

 i
2
a 
  S ik  k  (10)
i  H i

and for the beam section considered here:

  0.25 t (1  2 ) F 2 H 4 I 1 (11)

in which (ak H )i is the dimensionless crack size at the kth location defined in the ith

mode and  is a constant value depending on beam dimensions, crack types, and

Poisson’s ratio. Equation (10) can be solved to estimate crack sizes if the quantities

i i and Sik are experimentally determined or numerically generated.

7
2.2 Crack-Location Methodology

A crack-location methodology is developed by linearly relating the structural

system’s sensitivity of modal characteristics to the eigenfrequency changes due to

geometrical changes as described in Fig. 2. For a MDOF structural system with NE

elements and N nodes, the damage inflicted at predefined locations may be predicted

using the following sensitivity equation (Stubbs and Osegueda, 1990)

NE

F 
j 1
ij j  Zi (12)

in which  j ( 1   j  0 ) is the measure of damage inflicted at the jth location (i.e., the

fractional reduction in jth stiffness parameter). The term Z i is the fractional change in

the ith eigenvalue and (by neglecting changes in mass due to damage) is given by

Z i   i2  i2 (13)

where  i2 (  i*2   i2 ) is the change in the ith damped natural frequency before and

after damage. If NM modes are measured, Eq. (12) provides NM equations. The term

Fij is the modal sensitivity of the ith modal stiffness with respect to the jth element:

Fij  K ij K i (14)

where K i is the ith modal stiffness ( K i   Ti C i ) and K ij is the contribution of the

jth element to the ith modal stiffness ( K ij   Ti C j  i ) where  i is the ith modal

vector, C is the system stiffness matrix for the structure, and C j is the contribution

of jth element to the system stiffness.

Once the quantity Z i is experimentally determined, Eq. (12) can be solved to

locate and size damage in the system. However, the inverse solution to determine the

8
 j is possible only if the number of damage parameters is close to the number of

modes (i.e., NE  NM ) (Stubbs and Kim, 1996). In the case when NE  NM , the

system becomes ill-conditioned and alternate methods to estimate the damage

parameters should be sought. In an effort to overcome this difficulty, Stubbs et al.

(1990) proposed a sensitivity ratio concept based on earlier works presented by Cawley

and Adams (1979).

Let us consider the structural system of NE elements ( j  1,2,  , q,  , NE ) with a

measured set of NM vibrational modes ( i  1,  , m, n,  , NM ). Equation (12) is

rewritten for any two modes m and n ( m  n ), respectively. On dividing Eq. (12)

for mode m by the other for mode n , we obtain

NE

Zm
F
j 1
mj j
Fm1 1  Fm 2 2      Fmq q      FmNE NE
  (15)
NE
Fn1 1  Fn 2 2      Fnq q      FnNE NE
F 
Zn
nj j
j 1

Assuming that the structure is damaged at element, j, such that  j  0 when j  q

but  j  0 when j  q , Eq. (15) reduced to

Z m Fmq
 (16)
Zn Fnq

in which Z m Z n is the measured ratio of the fractional changes in frequency for two

modes, m and n. Also, Fmq Fnq is the ratio of the theoretically or numerically

derived sensitivities for those modes and the element q. Thus the damage inflicted at

location, q, exists when the L.H.S of Eq. (16) equals to the R.H.S.

For all measured NM modes, summing the denominators of both sides of Eq. (16)

leads to

9
 Zm Fmq
NM
 NM
(17)
Z
k 1
k F
k 1
kq

Equation (17) is true only if element q is damaged. Thus we introduce an error index

into Eq. (17) as follows:

Zm Fmq
eiq  NM
 NM
(18)
Z k 1
k F
k 1
kq

where eiq represents localization error for the ith mode and the qth location, and

eiq  0 indicates that the damage is located at qth location using the ith modal

information. To account for all available modes we form a single damage indicator (DI)

for the qth member as

1 2
 NM 
DI q   eiq2  (19)
 i 1 

where 0  DI q   and the damage is located at element q if DI q approaches the

local maximum point.

3. EXPERIMENTAL VERIFICATION

3.1 Description of Test Structure

The crack detection model will be validated using comprehensive experimental data

collection provided by Silva and Gomes (1990). Those researchers performed extensive

sets of modal analysis experiments on free-free uniform beams with the goal of

providing objective data to validate proposed techniques for damage detection. Test

specimens were steel beams with 0.032-m x 0.016-m rectangular cross-section and

10
0.72-m long. The corresponding material properties were: E  206 GPa ,   0.29 ,

and   7650 kg / m 3 .

Here the results of 32 separate experiments reported by Silva and Gomes (1990) are

utilized. Sixteen of the experiments were performed on undamaged beams and sixteen

experiments were performed on damaged beams. The following procedures were

utilized for the experiments. The first four bending frequencies were measured for each

of 16 undamaged free-free beams. Then a cut was introduced into each beam and the

same four bending frequencies were measured. The crack in each beam was simulated

by a cut normal to the beams’ longitudinal axis, with a controlled depth (as listed in

Table 1). The thickness of the cut was carefully defined taking into account that both

sides of the crack were not supposed to make contact during the dynamic bending of the

beam.

Table 1 presents the total 16 damage scenarios that include four different crack-

locations and four crack-depth levels at each location. The two sets of bending

frequencies measured before and after the damage episodes are listed in Table 1.

3.2 Crack Detection Practice

For crack detection, the free-free beam was modeled as an Euler-Bernoulli beam.

For analysis purposes, we divided the 72-cm beam into 72 elements of equal size. Each

element then becomes a potential damage location with a length of 1-cm or 1.38 percent

(i.e., 1/72 x 100) of the beam span. The crack detection task was performed in five steps.

In the first step, a theoretical modal analysis was performed and the system

identification technique proposed by Kim and Stubbs (1995) was used to identify a set

of baseline modal parameters for the beam. The average frequencies of the sixteen

11
undamaged beams and the generated frequencies of an identified baseline model are

listed in Table 2. Note that the coefficients of variation (COVs) are less than 0.01 for all

modes.

In the second step, the modal sensitivities, Fij , (i.e., the equivalent expression of Eq.

(14)) of mode i and element j (between two locations x j and x j 1 ) were computed

using
x j 1 dx l
Fij   EI { i( x)}2 ; K i   EI { i( x)}2 dx (20)
xj Ki 0

The flexural rigidity EI is assumed constant over the span of the beam. The

curvatures of the mode shapes were generated at the 73 nodes of the DDM. The

curvatures were obtained as follows: 1) modal amplitudes,  (x) , corresponding to

nodes 1-73 were generated from the baseline model; 2) a modal displacement function

 (x) was generated for the entire beam using a third-order interpolation spline

function; and 3) the curvatures (i.e.,  (x) ) were determined at the 73 nodes. Since four

measured frequencies are available, the sensitivities are defined for 4 modes and 72

elements. Fig. 3 shows the modal sensitivities of the test beam that were computed

along the beam’s longitudinal axis.

In the third step, potential crack locations were predicted. The fractional changes in

frequencies (i.e., Eq. (15)) were computed using the measured frequency results listed in

Table 1. By assuming that EI is constant over the beam span, the sensitivity ratio (i.e.,

the right-hand side of Eq. (17)) for an element q for any two modes m and n is given by

 {  } dx  { ( x)} dx
2 l
2
Fmq m n
 
q 0
(21)
 { } dx  {  ( x)} dx
2 l
Fnq n
2
q m
0

Next, localization errors were computed using Eq. (18) for 4 modes and 72 locations

12
(i.e., e1 j , e2 j , e3 j , and e4 j , j  1,72 ) by implementing the sensitivity ratios and the

fractional changes in frequencies. For example, error indices are plotted in Figs. 4-7

for the following four cases: Crack Case 2 ( a h  0.125 and x L  0.125 ), Crack Case

6 ( a h  0.125 and x L  0.25 ), Crack Case 10 ( a h  0.125 and x L  0.375 ), and

Crack Case 14 ( a h  0.125 and x L  0.5 ). Note for each case along the x-

coordinate that each point where error equals to zero indicates that a crack is located at

that location. Finally, we computed the damage index (i.e., given by Eq. (19)) to

decide potential crack locations. Damage indices are plotted in Figs. 8-11 for Crack

Cases 2, 6, 10, and 14, respectively. Note for each crack case that because of the

symmetry of the test arrangement, two symmetrical predictions are always made thus

one location is false-alarmed. If the test specimen was a cantilever beam, this ambiguity

would not occur. The crack localization results for all the sixteen damage cases are

summarized in Table 3.

In the fourth step, the crack size at each predicted location was estimated by using

the crack size model (i.e., the equivalent expression of Eq. (10)). Assuming a crack is

located in element k between two locations ( x j , x j 1 ), a solution of crack size is given

by

 ak  i
   (22)
 H i   S ik  i

where (a k / H ) i is the dimensionless crack size estimated at location xk

(  ( x j  x j 1 ) / 2 ) by using the ith modal data. As stated previously, the vertical crack-

depth of the opening fracture mode is our primary attention. The modal sensitivity of

13
mode i and location k was computed using Eq. (20). The constant  was obtained

from Eq. (11) by implementing H  0.032m , L  0.72m , Poisson’s ratio of 0.29, and

the geometrical factor F  1.12 (i.e., an approximate value of Eq. (5) for a small edge-

crack). The fractional changes in the eigenvalues were computed from Eq. (15) by

implementing the measured frequencies listed in Table 1. The crack sizing results for

the sixteen damage cases are summarized in Table 3.

3.3 Accuracy Measures of Crack Detection Results

The accuracy of the crack localization and sizing scheme presented here is evaluated

by measuring the so-called localization error and the so-called size error. The

localization error le, which represents the metrical difference between real crack

location and predicted location, is quantified using the expression:

le  (x / L)  100 (23)

in which x is the spacing between the inflicted and predicted location and L is the

reference span. Also, the size error se, which represents the difference between the real

and predicted crack-depth, is quantified using the expression:

se  {(a / H ) r  (a / H ) p } /(a / H ) r (24)

in which (a / H ) r is the inflicted crack-depth and (a / H ) p is the predicted crack-

depth.

For each crack case, the localization error was measured by Eq. (23). A comparison

between the inflicted and the predicted localization of the crack is plotted in Fig. 12 and

the results are also summarized in Table 3. The minimum localization error is 0.1

percent. This result indicates that the predicted location falls within less than 0.1-cm

14
of the correct location in the test beam (note that L = 72-cm). The maximum location

error is 4.8 percent (damage case 16). That is, the predicted crack is 3.4-cm away from

the correct location in the test beam. The average localization error is 1.3 percent. By

excluding Damage Case 16 that exceeds two standard deviations, the localization errors

range from 0.1 percent to 2.1 percent.

The size error was evaluated using Eq. (24). A comparison between the inflicted and

predicted crack sizes is plotted in Fig. 13 and the results are also summarized in Table 3.

The size errors range from 1.1 percent to 24.4 percent. For example, a one-percent size

error indicates a 0.1-mm difference in the estimation of a 10-mm crack-depth. The

average size-error is 8.6 percent. By excluding Damage Case 16 which exceeds two

standard deviations, size errors range from 1.1 percent to 13.1 percent.

4. DISCUSSION OF RESULTS

On the basis of the results presented here several fundamental questions are

anticipated. The first question that may be raised is: “How do the number of natural

frequencies measured influence the preciseness of the identification process?” The

answer to this question follows directly from a study of Figs. 4 to 7. First keep in

mind that the location of damage for a given mode is given by Eq. (18). Potential

Damage exist if eij=0 or eij is a minimum. Suppose that only one frequency is

measured and take Damage Case 2, Fig. 4, to illustrate the argument. From Fig. 4, if

only the first mode is available, damage exist at approximately x/L=0.125 and

x/L=0.875. Accounting for symmetry, this answer is correct.

If the second mode above was used, the criteria in Eq. (18) and Eq. (19) would

15
predict that damage existed at x/L=0.125, x/L0.438, x/L=0.563, and x/L=0.875.

However, if the first mode and the second mode were used simultaneously, Eq. (19)

would only predict damage at x/L=0.125 and x/L=0.875. Thus, it appears that the

number of natural frequencies measured does influence the preciseness of the

identification process.

The second question that may be raised is: “Is it possible to determine the location

and size of a crack using two or three natural frequencies?” From the above discussion,

the determination of the location from two or more frequencies is affirmative. From Eq.

(22), the determination of the size of the crack is also affirmative.

The third question that may be raised is: “Can the proposed method be applied for

the detection of multiple cracks in a beam?” The answer is affirmative because there

is no reason why the patterns produced in Figs. 4-7 would not replicate for cases of

multiple cracks.

The fourth questions that may be raised is: “What is the smallest crack that can be

detected using this methodology?” The answer to this question lies not in the

methodology but in the resolution of the frequency measurements during field

investigations. In the author’s experience, field measurements using state-of-the-art

instrumentation is in the order of 0.03 Hz. This field capability defines the limits of

any theoretical endeavor.

The fifth question to be raised is: “How do measurement errors influence the results

of the damage localization process?” Measurement errors can significantly influence

the results of the proposed method. Errors in the measured natural frequencies will

affect the predictions of the potential damage locations.

The sixth question is: “What is the difference between this work and previous works

16
on damage detection?” As stated in the introduction of this work, there exist few

Level III damage detection methods (i.e., detect, locate, and size damage). Six damage

detection methods that are frequently cited in the literature and that have been utilized

independently of the original authors include the frequency ratio method (Cawley and

Adams, 1979), the S-0 method (Stubbs and Osegueda, 1990), the damage index method

(Stubbs et al., 1992), the mode shape curvature method (Pandey, 1991), the change in

flexibility method (Pandey and Biswas, 1997), the change in uniform flexibility mode

shape curvature method (Zhang and Aktan, 1995) and the change in stiffness method

(Zimmerman and Kasuk, 1994). The method proposed adds to the list of Level III

methods that have been corroborated using experimental data. There are several major

differences between the method presented here and the methods cited above. First,

although in concept the localization method proposed here is similar to the method

proposed by Cawley and Adams (1979), the approach proposed here is based on a

strong analytical background instead of an ad hoc hypothesis. Secondly, the S-0

method (Stubbs and Osegueda, 1990) requires the solution of a system of linear

equation that is rank deficient when the number of measured modes (NM) are less than

the number of damage locations (NE). In the method proposed here this situation does

not exist. Third, compared to the other methods cited above, the method proposed

here uses only measured frequency information. Experimental mode shapes are not

necessary thus rendering the method potentially efficient in certain applications.

17
5. SUMMARY AND CONCLUSIONS

This paper presented a methodology to nondestructively locate and estimate the size

of a crack in structures for which only a few natural frequencies are available. The

proposed methodology was presented in two parts. The first part of the paper outlined a

theory of crack detection that yielded information on the location and size of crack

directly from changes in frequencies of the structures. A crack location model and a

crack size estimation model were formulated by relating fractional changes in modal

energy to changes in natural frequencies. The second part of the paper demonstrated the

feasibility of the crack detection scheme by accurately locating and sizing cracks in test

beams for which four natural frequencies were available for several damage scenarios.

By applying the approach to the test beam, it was observed that the crack can be

confidently located with a relatively small localization error (1.3%). It is also

observed that the size of crack can be estimated with a relatively small size error (8.1%).

We conclude that it is possible to localize a crack and estimate the crack size in a beam-

type structure with knowledge of the natural frequencies measured before and after

damage of only a few of the lower modes. Research to improve the crack detection

algorithm presented is continuing along three lines of inquires. First, we are developing

algorithms to more accurately estimate the size of damage. Second, we are extending

the algorithm to more complicated structures such as three-dimensional frames. Third,

we are now in advanced stages of demonstrating the practicality of the approach in

crack detection in full-scale structures.

18
 REFERENCES

Anderson, T. L., Fracture Mechanics, CRC Press, London, 1995.

Cawley, P., and Adams, R. D., The Location of Defects in Structures from
Measurements of Natural Frequencies, J. Strain Anal., Vol. 14, No. 2, pp. 49-57, 1979.

Chance, J., Tomlinson, G. R., and Worden, K., A Simplified Approach to the Numerical
and Experimental Modeling of the Dynamics of a Cracked Beam, Proc. of the 12th Int.
Modal Analysis Conference, Honolulu, Hawaii, Vol. 1, 778-785, 1994.

Crohas, H., and Lepert, P., Damage Detection Monitoring Method for Offshore
Platforms Is Field Tested, Oil and Gas J., Vol. 80, No. 8, 1982.

Gounaris, G., and Dimarogonas, A.D., A Finite Element of A Cracked Prismatic Beam
for Structural Analysis, Computers & Structures, Vol. 28, No. 3, pp. 309-313, 1988.

Kaouk, M., and Zimmerman, D. C., Structural Damage Assessment Using a Generalized
Minimum Rank Perturbation Theory, AIAA J., Vol. 32, No. 4, pp. 836-842, 1994.

Kim, J. T., and Stubbs, N., Model Uncertainty and Damage Detection Accuracy in
Plate-Girder Bridges, Journal of Structural Engineering, ASCE, Vol. 121, No. 10, pp.
1409-1417, 1995.

Ostachowicz, W.M. and Krawczuk, M., Vibration Analysis of a Cracked Beam,

Computers & Structures, Vol. 36, No. 2, pp. 245-250, 1990.

Pandey, A. K., Biswas, M., and Samman, M. M., Damage Detection from Changes in
Curvature Mode Shapes, J. Sound and Vibration, Vol. 145, No. 2, pp. 321-332, 1991.

Ricos, P.F., Aspragathos, N., and Dimarogonas, A.D., Identification of Crack Location

19
and Magnitude in a Cantilever Beam from the Vibration Modes, Journal of Sound and
Vibration, Vol. 138, No. 3, pp. 381-388, 1990.

Rytter, A., Vibration Based Inspection of Civil Engineering, Ph.D. Dissertation,

University of Aalborg, Denmark, 1993.

Silva, j. M., and Gomes, A. J. L., Experimental Dynamic Analysis of Cracked Free-Free
Beams, Experimental Mechanics, Vol. 20, No. 3, pp. 20-25, 1990.

Stubbs, N., and Osegueda, R., Global Nondestructive Damage Evaluation in Solids, Int
J. Analytical and Experimental Modal Analysis, Vol. 5, No. 2, pp. 67-79, 1990.

Stubbs, N., Kim, J.T., and Topole, K., The Effect of Model Uncertainty on the Accuracy
of Global Nondestructive Damage Detection in Structures, Computational Stochastic
Mechanics, eds. P.D. Spanos and C.A. Brebbia, Elsevier Applied Science, London, pp.
125-136, 1991.

Stubbs, N., and Kim, J. T., Damage Localization in Structures Without Baseline Modal
Parameters, AIAA J., Vol. 34, No. 8, pp. 1644-1649, 1996.

Sundermeyer, J. N., and Weaver, R.L., On Crack Identification and Characterization in a

Beam by Nonlinear Vibration Analysis, TAM Report No. 74, UILU-ENG-93-604, Univ.
of Illinois, 1993.

Vandiver, J. K., Detection of Structural Failures on Fixed Platforms by Measurement of

Dynamic Responses, Proc. Of 7th Annual Offshore Technology Conference, Houston,
Texas, Paper 2267, 1975.

Wu, X., Ghaboussi, J., and Garrett, J. H., Use of Neural Networks in Detection of
Structural Damage, Computers and Structures, Vol. 42, No. 4, pp. 649-659, 1992.

20
Table 1. Damage Scenarios and Resonance Frequencies (Hz) of Free-Free Beams
(Silva and Gomez, 1990)

Frequency (Hz)
Crack Inflicted Crack Mode 1 Mode 2 Mode 3 Mode 4
Case Loc(x/L) Size(a/H) Initial Cracked Initial Cracked Initial Cracked Initial Cracked
1 0.125 0.125 315.9 316.0 860.2 859.4 1654.5 1649.0 2668.0 2653.0
2 0.125 0.25 316.3 316.1 862.6 857.8 1659.0 1632.5 2674.0 2608.0
3 0.125 0.375 317.6 316.6 864.6 851.4 1663.0 1593.5 2682.0 2520.0
4 0.125 0.5 314.7 313.0 856.8 826.6 1647.0 1515.0 2657.0 2378.0
5 0.25 0.125 316.8 315.9 861.6 855.2 1657.5 1647.5 2673.0 2665.0
6 0.25 0.25 317.7 314.1 864.4 840.6 1662.0 1626.5 2676.0 2666.0
7 0.25 0.375 317.8 308.8 864.8 805.2 1662.5 1580.5 2675.0 2660.0
8 0.25 0.5 323.8 305.4 878.8 870.4 1689.5 1534.0 2721.0 2685.0
9 0.375 0.125 313.5 311.7 855.0 853.8 1646.0 1646.5 2657.0 2652.0
10 0.375 0.25 315.4 307.1 858.6 842.4 1653.0 1651.5 2665.0 2604.0
11 0.375 0.375 316.6 296.2 862.4 825.0 1659.5 1655.5 2675.0 2532.0
12 0.375 0.5 328.8 279.0 873.4 805.2 1679.5 1672.0 2707.0 2439.0
13 0.5 0.125 316.7 313.0 862.8 863.2 1658.0 1645.5 2675.0 2676.0
14 0.5 0.25 315.6 303.0 859.4 859.8 1652.0 1606.5 2667.0 2665.0
15 0.5 0.375 317.8 291.5 865.8 866.0 1664.5 1574.0 2687.0 2683.0
16 0.5 0.5 320.6 265.1 873.0 873.2 1678.0 1498.0 2701.0 2701.0

21
Table 2. Comparison of Frequencies: Undamaged Beams Vs Baseline Model

Mode Average Frequency (Hz) Frequency (Hz) Coefficient

No. of 16 Uncracked Beams of Baseline Model of Variation

1 317.85 318.07 0.010

2 864.01 863.02 0.008

3 1661.59 1657.96 0.008

4 2678.75 2675.75 0.006

22
 Table 3. Crack Prediction and Accuracy Assessment Results of Test Beams

Inflicted Crack(s) Predicted Crack(s) Prediction Accuracy

Crack
Case Location Size Location Size Loc. Error Sizing Error
(x/L) (a/H ) . (x/L) . (a/H ) ) . (% ) (% )
% )(
1 0.125 0.125 0.110 0.126 1.5 1.1

2 0.125 0.25 0.121 0.234 0.4 6.0

3 0.125 0.375 0.140 0.335 1.5 10.4

4 0.125 0.5 0.146 0.440 2.1 11.8

5 0.25 0.125 0.243 0.131 0.7 5.1

6 0.25 0.25 0.257 0.244 0.7 2.2

7 0.25 0.375 0.264 0.395 1.4 5.5

8 0.25 0.5 0.256 0.434 0.6 13.1

9 0.375 0.125 0.326 0.093 4.8 25.4

10 0.375 0.25 0.376 0.226 0.1 9.3

11 0.375 0.375 0.378 0.345 0.3 7.9

12 0.375 0.5 0.390 0.439 1.5 12.1

13 0.5 0.125 0.511 0.132 1.1 6.0

14 0.5 0.25 0.514 0.232 1.4 6.8

15 0.5 0.375 0.486 0.328 1.4 12.4

16 0.5 0.5 0.511 0.488 1.1 2.3

23
Captions of Figures

Fig. 1. Nondestructive Crack Detection Scheme

Fig. 2. Geometry of Free-Free Beam With A Crack

Fig. 3. Modal Sensitivities of Free-Free Beam
Fig. 4. Localization Error Indices for Individual Modes for Crack Case 2

Fig. 5. Localization Error Indices for Individual Modes for Crack Case 6

Fig. 6. Localization Error Indices for Individual Modes for Crack Case 10

Fig. 7. Localization Error Indices for Individual Modes for Crack Case 14

Fig. 8. Crack Localization Results of Crack Case 2

Fig. 9. Crack Localization Results of Crack Case 6

Fig. 10. Crack Localization Results of Crack Case 10

Fig. 11. Crack Localization Results of Crack Case 14

Fig. 12. Comparison of Inflicted Crack Location and Predicted Crack Location
Fig. 13. Comparison of Inflicted Crack-Depth and Predicted Crack-Depth

24
 Experimental Crack Localization
MODAL Modal Parameters - Natural Model
TESTING Frequencies
- Mode Shapes Predicted Crack Location(s)
- Model
FE MODEL Baseline Modal Parameters Crack Detection
OF Parameters . Model
STRUCTURE
Geometric Crack Size(s)

Fig. 1. Nondestructive Crack Detection Scheme

25
 t

X H
a a

Y L

Fig. 2. Geometry of Free-Free Beam With A Crack

26
Fig. 3. Modal Sensitivities of Free-Free Beam

27
Fig. 4. Localization Error Indices for Individual Modes for Crack Case 2

28
Fig. 5. Localization Error Indices for Individual Modes for Crack Case 6

29
Fig. 6. Localization Error Indices for Individual Modes for Crack Case 10

30
Fig. 7. Localization Error Indices for Individual Modes for Crack Case 14

31
Fig. 8. Crack Localization Results of Crack Case 2

32
Fig. 9. Crack Localization Results of Crack Case 6

33
Fig. 10. Crack Localization Results of Crack Case 10

34
Fig. 11. Crack Localization Results of Crack Case 14

35
Fig. 12. Comparison of Inflicted Crack Location and Predicted Crack Location

36
Fig. 13. Comparison of Inflicted Crack-Depth and Predicted Crack-Depth

37