Professional Documents
Culture Documents
5.field Verification of A Nondestructive Damage Loca
5.field Verification of A Nondestructive Damage Loca
net/publication/234470341
Article in Proceedings of SPIE - The International Society for Optical Engineering · March 1995
CITATIONS READS
250 1,733
3 authors:
Charles Farrar
Los Alamos National Laboratory
441 PUBLICATIONS 26,193 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Jeong-Tae Kim on 27 December 2013.
210
Each bent consists of a 35 feet beam and two 4-f&-10- 3. DAMAGE LOCALIZATION ALGORITHM IN
inch-diameter hexagonical piers which vary in length from BEAMS
39 feet to 56 feet. Both the beam and the piers are made of
reinforced concrete. 3.1 Theory
The substructure consists of a footing which supports the The damage localization scheme described below
bents and acts simultaneously as a pile cap for a bank of utilizes the paradigm of pattern recognition. In the
piles. The substructure is in a soil environment which problem at band, the dynamic response of the structure in
consists of typically a 3 feet layer of clay at the river bed, a the time domain represents the physical world data and the
5 feet layer of sand, a 10 feet layer of gravel in sand, a 5 modal parameters represent the pattern space. We develop
feet layer of compacted clay, and at least a 20 feet layer of a feature space by forming indicators that are a function of
quick or compacted sand. The footing is typically at an measurable pre-damage and post-damage modal
elevation of 4942 feet in the compacted sand layer with parameters. These indicators are selected in such a manner
dimensions of 36 A x 12 A x 3 ft. Each reinforced concrete that they reflect internal structure in the data. The decision
footing also functions as a pile cap for a typical bank of 35 algorithm is a means by which the data space is
piles, the outer 20 of which are battered. The piles are partitioned into C “decision regions”. In our case C=2 and
typically 16 inch diameter steel pipes filled with concrete. corresponds to the cases: (a) a structore is not damaged at
a given location, and (b) a structure is damaged at a given
location. Each instance of OUT indicator will fall into one
of the categories.
We first develop indicators of damage that represent the
feature space. The structure of interest here will be
modeled as a beam, therefore, for this paper, the theory of
damage localization will be limited to beams. The
approach presented here can be routinely extended to
plates or other arbitrary three-dimensional structures.
Consider an arbitrary homogeneous 1-D beam with NE
members (in the finite element sense) and N nodes.
Assume that the beam behaves linearly. On solving the
eigenvalue problem, the P modal stiffness, K,, of the
beam is given by
211
c=q/K;= F,(l+~&z,+H.O.T.)
I-3
where scalars C; and K; are given by
c; = k;lJ~~(+fx (5)
212
the ““II hypothesis, Ho, assumes that the value of Z, without solving a system of equations, and damage may be
localized in structures containing many elements.
consists of only noise so that Z, = N. in the second
hypothesis, HI, the location is assumed to be damaged so 3.2 Numerical Validation of the Theory
that Z, = N + D. We let Do denote the choice of the Ho The objective here is to evaluate the feasibility of the
proposed theory (i.e., to examine the validity that Eq. 8
hypothesis and D1 denote the choice of the HI hypothesis. derive a consistent indicator of damage localization). We
Four outcomes based on a set of samples of Z, are meet this objective in four steps: firstly, a test structure is
possible: defined; secondly, modal parameters of a” undamaged
1. We correctly identify the presence of damage. The state and two damaged states (with different magnitudes of
probability of this outcome is called the damage at a fixed location) of the test structure are
probability of detection (PD) denoted by P(Dl/Hl); generated “sing the software package ABAQUS[“l; and
2. We incorrectly declare that damage is present when finally, the theory is used to localize damage inflicted in
in fact there is no damage present. This is the the test structure.
probability of a false alarm (Pfa) denoted by A finite element (FE) model of a continuous beam was
my&J; selected as the test stmcture. As shown in Fig. 2, the FE
3 . We incorrectly declare that there is no damage model consisted of 50 beam members and three linear
present. This is denoted by P(D&l); and axial springs modeling supports of the beam. In this
4 . We correctly declare that the location is hypothetical example we assumed that only vertical
undamaged. This case is denoted by P@&). motion is measured at each nodal point. Values for the
In our damage detection examples, we Qo not have &@x?.L! material properties of the beam elements and springs were
probabilities P(Ho) and P(H1) neither do we assign costs to assigned as follows: (1) the elastic modulus E = 10 Y 106psi
the possible four wtcomes of the experiment. Therefore if (70 Gpo); (2) Poisson’s ratio u=O.33; and (3) the linear
we assume a knowledge of the density functions of the
mass density p=2.536x10“lb~s1/in’ (2710 kg/m’). Vahes
““damaged a” damaged Z,,, we can use the Neyman-
for the geometric properties were assigned as follows: (1)
Pearson (NP) criterion as a classification algorithm[131. for beam elements, the cross-sectional area A = 1.625 in’
Note that this criterion maximizes PD for a specified Pfa (l.OSxlO”m’) and the second moment of area Z=1.2in4
The NP detector uses a likelihood ratio test of the (5.0x 10.‘m’); (2) for Spring 1 member, A =7.7x10Jin’
followi”g form:
(4.96x10dm’) and 1~0; and (3) for Spring 2 member,
,4=1.3x10-*in” (8.4xiOdm~)a”d Z=O.
A =A(l)ty (15)
r P,(Z)
where ,L, is the test statistic; z is the single observation of
damage; p,(t) is the pdfof z given that H1 is tme; p,(r) is
the pdf of z given that Ho is true; and y is a number
which depends upon the significance level of the test. The
decision role for the NP detector is
ChwscH1:whenL,>r
Choose Ho: otherwise.
A” equivalent decision rule is
ChooseHI: whenz>K
Choose Ho: otherwise Fig. 2. Schematic of the Test Structure
Thus damage localization is accomplished in five steps:
(1) Compute fractional modal stiffness for each member; Next, we measured, via numerical simulation, modal
(2) Compute p,; (3) Compute p,; (4) Compute Z,; (5) parameters of an ““damaged state and three damaged
states. Here the three damage cases were limited to the FE
Select K; and (6) Classify location j. model damaged at a fixed location (i.e., Element 39 shown
As will be demonstrated below, this method has several
in Fig. 2) with different magnitudes of damage (i.e., 1
appealing features which include the following items: only percent (Case 1) and 25 percent (Case 2) reductions in the
mode shapes are needed in the analysis (See e.g., Eq. 1 l), second moment of area of Element 39). Typical
damage may be detected “sing few modes, frequency and
damping information are not needed, damage is detected
213
nunwically generated mode shapes and frequencies of the reliable damage localization indicator for beams.
first three modes are shown in Fig. 3 and Table 1. especially when the damage is small
214
4. EXPERlMENTAL DESIGN 1. The data gathering and modal analysis were
performed by the Advanced Engineering Division,
Here we demonstrate, for the I-40 bridge, the Los Alamos National Laboratory;
practicality and effectiveness of utilizing the referenced 2. The tests were performed on the bridge from 31
damage detection scheme. To meet this objective we (i.e., August to 2 September, 1993. The modal analysis
TAMU & NMSU) agreed upon the following strategy: was performed subsequent to the test;
1. TAMU will specify what modal parameters are 3. Pre-damage and post-damage mode shapes were
required as input to the damage detection; provided for three modes (two bending modes and
2. NMSU will instrument the bridge, take predamage a torsional mode);
and post-damage dynamic measurements, and 4. The location and arrangements of the
determine the mode shapes and resonant accelerometers are shown in Fig. 6;
frequencies; 5. Four levels of damage were inflicted upon the
3. NMSU will transmit the modal information to structure near X-6 of the girder shown in Fig. 6:
TAMU; a) A two-foot-long cut in the web (approximately
4. TAhKl will perform the damage localization and 318-i” wide) centered at midheight of the girder;
severity estimation; and b) The cut was extended down 4 feet to the bottom
5. TAMU will report the results to Nh4SU. flange;
To meet the first subobjective above, we specified the c) The bottom flange was cut halfway in from either
following data to be provided. side directly below the cut in the web; and
1. Mode shapes and their corresponding frequencies tillally,
for at least three modes are required; d) The bottom flange was cut completely through
2. The above modal information is required before leaving only the top 4 ft. of the web and top
damage is inflicted and after at least two instances flange to carry the load.
of damage; 6. Pm-damage and post-damage accelerometer
3. Nh4SU should utilize appropriate averaging readings were taken for each damaging episode.
techniques to lessen the occurrence of noise in the Each mode shape reported in the next section represents
modal information; the average of 30 readines! In summary, the
4. Sensor readings normal to the longitudinal axis of measurements by Los Alamos surpassed the requirements
the girder should be taken on the girder to be specified by TAMU.
damaged;
5. A sensor reading should be taken directly over each 6. RESULTS OF MODAL ANALYSIS
support; and
6. Subsequent readings should be taken at intervals of The results of the modal analyses performed by Los
at least one-tenth the span. Alamos personnel are summarized in Tables 2-6. The
tables are derived from the more complete data provided
5. SUMMARY OF DATA GATHERING by the Los Alamos Team. Here we neglected the phase
angle, which was typically near zero or *180”, and
The following discussion summarizes the data-gathering retained only the magnitude of the mode shapes. Note also
at the project site: that the tables do not represent the sequence of damage
described in the last &on. Note that accelerometer
locations X-l to X-l 1 are indicated in Fig. 6.
Table 3. Modal Amplitudes for Damage Care 1 Table 6. Modal Amplitudes for Damage Case 4
216
and the clement modal amplitude values (normalized) utilized the following approach: first, we described the
from Tables 24. Using the interpolated modal coordinates bridge under study; second, we provided a summary of
for the beam for each of the pm-damage and post-damage aspects of the localization theory that related to steel girder
mode shapes, we generate functions K(X), where x is the bridges; third, we summarized the design of the entire
coordinate along the axis of the beam. From K(x) we damage localization experiment; fourth, we described the
actual collection of the data and summarized the results of
estimate the instantaneous curvature F(X).
the modal analysis; finally, we used the scheme and the
We next establish the criterion for the damage
field data to localize damage in the bridge. Using only
localization indicator as follows: select Ho (i.e., no damage
three modes of vibration and with no knowledge of the
at location j) if Z, < 2 or (2) select the alternate Hl if Z, 2 material properties of the bridge, we accurately located
2 (i.e., Pfa = 0.0228). The resulting damage localization damage in the structure.
for the 4 damage cases arc shown in Fig. 8(a) to 8(d). Note
that the damage was inflicted in clement 82 which is 9. ACKNOWLEDGMENTS
highlighted on Fig. 8. Note also that the combined damage
index is greater than 2 in the region near to and including The research reported herein was partially supported by
clement 82. In the Figures, about 10 damage locations are New Mexico State University under Agreement No. 01-3-
effected (i.e., approximately locations 73 to 84). This 4439x1 and was conducted at the Center for Mechanics
smeared damage behavior is expected since the strain and Materials at Texas A&M University The authors
energy in the region adjacent to the cut is very small or would like to thank especially Drs. Larry1 Matthews and
zero. According to our damage algorithm, there is no Ken White of NMSU for the invitation to participate in
damage in any other part of the span. this project.
A comparison between the inflicted and predicted
location of damage is presented in Table 7. The accuracy 10. REFERENCES
of the localization scheme presented here is evaluated by
the so called localization error e: [l] Gudmunson, P. Eigenfrequency Changes of Structures
e=(Ax/L)loo (16) Due to Cracks. Notches or Other Geometrical Changes, J.
in which Ax is the spacing behwco the inflicted and Mech. Phys. Solids, Vol. 30, No. 5, pp. 339-353, 1982.
predicted location and L is the reference span. These [2] Gudmunson, P. The Dynamic Behavior of Slender
values are listed in the last column of Table 7. Note that Structures with Cross-Sectional Cracks, J. Mech. Phys.
the maximum localization error is 2.5 percent of the span Solids, Vol. 31, No. 4, pp. 329-345, 1983.
(i.e., 4 ft in a span of 163 ft.) and the minirimm [3] Adams, RD., Cawley, P., Pye, C.J. and Stone, B.J.
localization error is 0.6 percent of the span (i.e. just less A Vibration Technique for Non-Deshuctively Assessing
than one foot). the lntegriry of Structures, J. Mcch. Engr. Science, Vol.
20, pp. 93-100, 1978.
8. SUMMARY AND CONCLUSION [4] Cristides, S., and Barrs, A.D.S. On-Orbit Damage
Assessment fir Large Space Stmctwes, AMA Journal,
The objective. of this paper was to demonstrate the Vol. 26,No. 9, pp. 1119-1126, 1984.
practicality of a nondestructive damage localization
scheme on a full-scale stnxture. To meet this objective, we
217
[5] Stubbs, N . , aad Osegueda, R G l o b a l Non-
Destructive Damage Evaluation in sOli& ht. J. Anal.
Exp. Modal Analysis, Vol. 5, No. 2, pp. 67-19, 1990.
[6] Natke, E.G., aad Yao, J.T.P. System Identification
Methods for Fault Detection and Diagnosis, Int. Conf on
Stroctoml Safety and Reliability, ASCE, New York, pp.
1387-1393,199O
[7] Kenley, RM., and Dodd& C.J. West Sole WE
Platfom: Detection of Damage by SYructural Response
A4eamrements, offshore Tech. Goof., Houston, Texas, pp.
111-118, 1980.
[8] Kummer, E., Yang, J.C.S., and Dagalakis, N.G.
Detection of Fatigue Cracks in Structural Members, 2nd
American Society of Civil Engineerin@EMD Specialty
Conference, Atlanta, Georgiq pp. 445-460,1981.
[9] Biswas, M., Pandey, AX., and Samman, M.M.
Model Technology for Damage Detection of Bridges,
NATO Advanced Research Workshop on Bridge
Evaluation, Repair and Rehabilitation, ed. A. Nowak,
Kluwer Academic Publishers, Maryland, pp. 161-174,
1990.
[lo] Flesch, RG., aad Keraichler, K Bridge Inspection
by Dynamic Tests and Calculations Dynamic
Investigations of Lovent Bridge, Workshop on Structural
Safety Evaluation Based on System Identification @) Damage Case 2
Approaches, eds. H.G. Natke and J.T.P. Yao, Vieweg &
Sons, LambrechtiPfalz, Germany, pp. 433-459, 1988.
Ill] Maaarek, D.F., and DeWolf, J.T. Experimental
Study of Bridge Monitoring Technique, J. of Structural
Engineering, ASCE, Vol. 116, No. 9, pp. 2532-2549,
1990.
[12] Cbeo, J, and Garba, J.A. On-Orbit Damage
Assessment for Large Space Structures, AIAA Journal,
Vol. 26, No. 9, pp. 1119-l 126, 1988.
1131 Gibson, J.D., and Melsa, J.L. Introduction to
Nonporametn’c Detection with Applications, Academic
Press, New York, 1975.
[14] Nielson, RO. Sonor Signa/ Processing, Artech
House, London, 1991.
[15] Kosko, B. Neural Networks for Signal Processing,
Prentice hall, New Jersey, 1992.
[16] Zimmermann, E.J. Fuq Set Theory and its
Applications, 2nd Ed., Kluwer Academic Publishers,
Boston, 1992.
[17] ABAQUS User Manual, Hibbitt, Karlsson &
Sorensen, Inc., 1987.
218