Professional Documents
Culture Documents
Sahin 2003
Sahin 2003
www.elsevier.com/locate/engstruct
Received 19 May 2003; received in revised form 1 August 2003; accepted 4 August 2003
Abstract
This paper presents a damage detection algorithm using a combination of global (changes in natural frequencies) and local
(curvature mode shapes) vibration-based analysis data as input in artificial neural networks (ANNs) for location and severity predic-
tion of damage in beam-like structures. A finite element analysis tool has been used to obtain the dynamic characteristics of intact
and damaged cantilever steel beams for the first three natural modes. Different damage scenarios have been introduced by reducing
the local thickness of the selected elements at different locations along finite element model (FEM) of the beam structure. The
necessary features for damage detection have been selected by performing sensitivity analyses and different input–output sets have
been introduced to various ANNs. In order to check the robustness of the input used in the analysis and to simulate the experimental
uncertainties, artificial random noise has been generated numerically and added to noise-free data during the training of the ANNs.
In the experimental analysis, two steel beams with eight distributed surface-bonded electrical strain gauges and an accelerometer
mounted at the tip have been used to obtain modal parameters such as resonant frequencies and strain mode shapes. Finally, trained
feed-forward backpropagation ANNs have been tested using the data obtained from the experimental damage case for quantification
and localisation of the damage.
2003 Elsevier Ltd. All rights reserved.
Keywords: Damage identification; Vibration-based analysis; Strain mode shape; Finite element analysis; Artificial neural networks
0141-0296/$ - see front matter 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2003.08.001
1786 M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802
Nomenclature
E elastic modulus
h distance between two collocation points
i number of data points
N number of random entries
vi normalised displacement
v⬙i curvature
g gravitational acceleration
Greek symbols
e strain
r density
n Poisson’s ratio
⌬ noise percentage
Abbreviations
changes in natural frequencies alone is difficult as modal Experimental validation however is quite limited. The
frequencies are global properties of the structure and technique used here aims to test the efficacy of global
hence cannot provide spatial information about structural or local vibration parameters and their different combi-
changes. Therefore, displacement or curvature mode nations as input to feed-forward backpropagation ANNs
shapes that provide spatial information about the damage for accurate prediction of damage location and severity.
are needed in order to overcome this drawback. In The global parameters are changes in natural fre-
addition to this, multiple modes need to be considered quencies, and the local parameters are changes in curva-
for better estimation of severity and location of the dam- ture mode shapes of the structure in the first three natural
age since each natural frequency and corresponding modes. To generate these parameters, free vibration
mode shape are affected to different extents depending dynamic behaviour of intact and damaged beams has
on the location of the damage. As damage detection is been obtained from FEA, and an experimental procedure
an inverse, non-linear and non-unique problem, that gen- has been applied to cantilever steel beams to validate the
erally does not have a feasible algorithmic solution or proposed method.
for which an algorithmic solution is too complicated to
be found and handled, a different approach is required.
Therefore, artificial neural networks (ANNs) [18–22] 2. Adopted approach
offer capabilities such as self-adaptiveness, generalis-
ation, abstraction and suitability for real-time appli- The adopted vibration-based damage detection pro-
cations. They have been widely used to estimate the cedure can be divided into three main parts: numerical
damage severity and its location in monotonic [23–31] analysis based on finite element model (FEM), experi-
and composite [32–37] structures. mental analysis, and application of the designed ANNs
Different vibration-based analysis data have been used in the prediction of damage severity and location.
either directly or as an input feature for ANN appli- The numerical analysis concentrates on extraction and
cations for different levels of damage assessment. selection of features from normal mode dynamic analy-
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1787
sis performed using the FEM for cantilever steel beams. 3. Finite element modelling
Different damage scenarios are obtained by changing the
thickness of the selected elements by different extents at 3.1. Background
different locations. The outputs from FEA are the natural
frequencies and corresponding normalised displacement In model-dependent vibration-based analyses, it is
mode shapes in the first three vibration modes. After per- important to have an accurate numerical model before
forming sensitivity analysis, the normalised natural fre- performing experiments. In numerical analysis, the FEM
quencies of the damaged beam with respect to the intact of intact and damaged steel beams are created by using
one, and the maximum absolute difference in curvature a commercial software, ANSYS 6.1 [39]. The structure
mode shapes between the intact and the damaged beams, considered in the analysis is a 450 mm long, 40 mm
are selected as input features. Such features [38] are wide and 3 mm thick steel cantilever beam.
good indicators for damage assessment. A four-node quadrilateral two-dimensional linear
The experimental component of this work examines structural shell element, SHELL63 [40], is selected to
intact and damaged cantilever steel beam specimens, model the beam. Since shell elements have six degrees
with eight surface-bonded electrical strain gauges and of freedom (three translational and three rotational) at
an accelerometer attached at the free end. Experimental each node, cantilever boundary conditions can also be
specimens have the same geometrical and material modelled more realistically by constraining all degrees
properties with the ones used in the numerical analysis. of freedoms of the nodes located at the root of the beam.
The experimental study also aims to use damage sensi- Fig. 2 shows the finite element mesh of the steel beam
tive signal features obtained from the sensory structures having 360 shell elements with 455 nodes; and the
(i.e. measured modal data such as resonant frequencies location of the damage.
and normalised strain mode shapes in the first three The first three bending natural frequencies of the
modes) as an input to selected ANNs for the experi- intact cantilever beam, for different mesh densities, are
mental verification of the proposed method. also obtained in order to check the mesh independency.
The ANNs used in the analysis (Table 5) are selected The variation in natural frequencies is less than 1%
from the previous numerical work [38] performed on
composite beam models; and chosen on the basis of
minimum mean square error and better convergence with
an increase in the number of epochs (i.e. the number of
presentations of the training input set to the ANN).
These ANNs are also trained by using numerically gen-
erated input data from FEM of the intact and damaged
steel beams. After validation of neural networks, exper-
imentally obtained input data are introduced to the selec-
ted ANNs for severity and location prediction of the real
damage case.
The outline of the damage assessment method is sche-
matically shown in Fig. 1. Fig. 2. FEM of cantilever steel beam.
Fig. 3. Percentage reduction in natural frequencies for different damage locations: Damage located at (a) 0.24L (b) 0.3L (c) 0.34L (d) 0.44L
(e) 0.54L (f) 0.6L (g) 0.64L (h) 0.74L.
Fig. 4. Variation of the magnitude of the absolute differences in curvature along the beam with different severities in Mode 1: Damage at
(a) 0.24L (b) 0.3L (c) 0.34L (d) 0.44L (e) 0.54L (f) 0.6L (g) 0.64L (h) 0.74L.
4.1. Test procedure and specimens face of the specimen along the width (40 mm) at 205
mm (0.4556L) away from the cantilever end.
Two steel beams having same geometrical (450 × The dynamic characteristics of the structure are inves-
40 × 3 mm) and material properties tigated by performing two different tests. The first used
(E = 207.109 (Pa), r = 7850 (kg / m3), n = 0.33) are a digital vibration controller to measure and record the
used during the experimental modal analysis. One of the acceleration response of the structure under different
beams is selected as a control specimen (intact structure) excitations, covering a certain range of frequencies,
and the other is used as a damaged beam. The local dam- which can then be used to obtain resonant frequencies
age is introduced to the steel beam as a 2 mm deep of the beam. In the second experiment, a signal-con-
(66.67% reduction in thickness) and 10 mm wide slot, ditioning amplifier is used to measure the dynamic
which is created by removing the material from the sur- strains at eight different locations, from which the strain
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1791
Fig. 5. Variation of the magnitude of the absolute differences in curvature along the beam with different severities in Mode 2: Damage at
(a) 0.24L (b) 0.3L (c) 0.34L (d) 0.44L (e) 0.54L (f) 0.6L (g) 0.64L (h) 0.74L.
mode shapes of the structure can be obtained by using 4.2. Frequency measurement
surface-bonded electrical resistance strain gauges. Fig. 7
shows the schematic view of the experimental set-up for A four-channel digital vibration controller with a data
vibration testing, used to determine frequency response acquisition unit is used to measure acceleration ampli-
functions, resonant frequencies, dynamic strain histories tude histories from two channels including both input
and strain mode shapes of the cantilever intact and dam- (control excitation) and output (dynamic behaviour of
aged beams. the specimen). The software is used to post-process the
data, plot the input and output frequency spectra and
1792 M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802
Fig. 6. Variation of the magnitude of the absolute differences in curvature along the beam with different severities in Mode 3: Damage at
(a) 0.24L (b) 0.3L (c) 0.34L (d) 0.44L (e) 0.54L (f) 0.6L (g) 0.64L (h) 0.74L.
zoom to peak values for more accurate prediction of the 4.3. Strain measurement
resonant frequencies. One of the two accelerometers is
used to control the input excitation amplitude supplied
to electro-dynamic vibration generator and the other is A 10-channel Measurements Group Model 2000 sig-
used to monitor the output acceleration amplitude of the nal-conditioning amplifier is used to amplify the signals
specimen. The accelerometer monitoring the response of from the strain gauges, which are recorded using a
the beam is attached on the centre line at the free end National Instruments VXI A/D acquisition system. The
of the beam. This location gives sufficient responses for excitation level is set to 2 V that is equivalent to
all three modes. 1000me and the bridge is balanced. In order to find cali-
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1793
brated and measured strain values, the following equa- the spatial location of the gauges along the beam is given
tions [42] are used: schematically in Fig. 9.
Rg·106
mecal ⫽ where 4.4. Excitation
K·(Rc ⫹ (Rg / 2)) (2)
Rc ⫽ 59.94 ⍀, Rg ⫽ 120 ⍀, K ⫽ 2.11 Several experimental techniques are available to mea-
sure the dynamic characteristics of the structures. These
Vloaded
memeasured ⫽ me where techniques generally excite the structure externally to
(Vcal⫺Vunloaded) cal (3) acquire a dynamic response. During the experimental
Vunloaded ⫽ 0, Vcal ⫽ 2 V analyses, two different types of excitation are applied
using an electro-dynamic vibration generator controlled
In these equations, K, Rg, Rc, Vcal, Vloaded, ecal and emeasured by a digital vibration control system.
are the gauge factor of strain gauge, strain gauge resist-
ance, calibration resistor, calibration voltage, voltage 4.4.1. Frequency sweep excitation
reading from loaded strain gauge, calibrated strain and A swept sine excitation is used to identify the
measured strain, respectively. Eight single element elec- response of a test piece to vibration across a range of
trical resistance gauges, with gauge length of 3 mm, frequencies. In this way, resonant frequencies of the test
nominal resistance of 120⍀ ± 0.3 and a gauge factor of specimen are investigated. A uni-directional type swept
2.11, are used to measure the direct strain on the surface sine, with linear incremental sweep rate, excites the sys-
of the beams. The distributed surface-bonded strain tem from 3 to 253 Hz, at increments of 0.5 Hz for the
gauges, which are numbered from the free end of the experimental modal analysis used to find the first three
beam, can be seen in Fig. 8. Detailed information about resonant frequencies of the intact and the damaged can-
tilever beams. The range is estimated from theoretical
Table 5
ANNs used in the analyses
Table 4
5. Application of ANN
First three resonant frequencies of intact and damaged beams (Hz)
5.1. Design of ANNs
Mode no. Experimental results
In order to find the most effective ANN that uses
Intact beam Damaged beam fdamaged/fintact vibration-based analysis data for the severity and
location predictions, different neural networks (Table 5)
1st bending 11.8 9.2 0.7796 with one hidden layer are designed by using MATLAB
2nd bending 73.1 50.5 0.6908
3rd bending 205.6 200.8 0.9766
neural network toolbox [43]. Different combinations of
input and output pairs are then introduced to these super-
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1795
Fig. 11. Strain output of eight strain gauges in the first mode (intact beam).
vised feed-forward backpropagation ANNs for training 5.2. Training and validation
and validation. Values (separated with semicolon) used
in the architecture column of Table 5 show the total
5.2.1. Noise-free case
number of neurons in the input, hidden and output lay- In this analysis, 480 different damage scenarios are
ers, respectively. generated by using 32 different reductions in thickness
at 15 different locations throughout the beam. Four hun-
1796 M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802
Fig. 12. Strain output of eight strain gauges in the first mode (damaged beam).
dred and fifty input–output pairs are given to the ANN mode shape data that were used in the noise-free case,
for training and the rest of the input–output pairs are in order to achieve a better generalisation during the
used to check the generalisation of the learning during training of ANNs and to simulate the experimental
the validation process. uncertainties. A random number generator, randn in
MATLAB, generates random numbers with a normal
5.2.2. Addition of artificial noise case distribution having zero mean and a variance and stan-
Artificial random noise (Eq. (4)) is added to normal- dard deviation of one.
ised frequency and maximum differences in curvature
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1797
Fig. 13. Normalised strain mode shapes of intact beam. (a) Mode 1, Fig. 14. Normalised strain mode shapes of damaged beam.
(b) Mode 2, (c) Mode 3. (a) Mode 1, (b) Mode 2, (c) Mode 3.
1798 M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802
Table 6
Addition of different percentage of noise to input data
Fig. 15. Absolute differences in strain mode shapes between the RNF MADC
intact and damaged beams. (a) Mode 1, (b) Mode 2, (c) Mode 3.
RNF 0.5, 1, 2 –
MADC&LOC – 1, 3, 5
RNF&MADC&LOC 0.5 1
RNF&MADC&LOC 1 3
RNF&MADC&LOC 2 5
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1799
Fig. 16. ANN results for severity prediction (input: RNF and output: Fig. 17. ANN results for location prediction (input: MADC and out-
DS). (a) Mean square error with number of epochs, (b) severity predic- put: DL). (a) Mean square error with number of epochs, (b) location
tion. prediction.
run, the severity (Fig. 20a) and location (Fig. 20b) of Table 8 that more accurate results are obtained in
the damage are predicted by introducing inputs, namely, location predictions (MADC&LOC-DL) for the damage
reduction in natural frequencies, maximum absolute dif- than severity predictions (RNF-DS), although the noise
ferences in curvature mode shape and their correspond- levels are comparably higher than the ones used on RNF
ing locations along the beam from the first three natural data. When the gradually increased noise combinations
modes. The results for noise-free (Table 7) and for applied to the normalised frequency and maximum
additional artificial noise (Table 8) cases are tabulated absolute differences in curvature mode shape are used
for performance comparison. It can be concluded from in the prediction of both severity and location of the
Table 7
Predictions of ANNs (noise-free case)
Table 8
Predictions of ANNs (additional artificial noise case)
RNF&MADC&LOC (0.5 and 1) DS&DL 0.6667 0.4556 0.7162 0.4319 0.0495 0.0237
RNF&MADC&LOC (1 and 3) DS&DL 0.6667 0.4556 0.7836 0.4712 0.1169 0.0156
RNF&MADC&LOC (2 and 5) DS&DL 0.6667 0.4556 0.8057 0.4507 0.1390 0.0049
M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802 1801
Acknowledgements
References
damage, better estimations are achieved in location pre-
dictions compared to severity at each level of additional [1] Salawu OS. Detection of structural damage through changes in
artificial noise. frequency: a review. Eng Struct 1997;19(9):718–23.
[2] Zou Y, Tong L, Steven GP. Vibration-based model-dependent
damage (delamination) identification and health monitoring for
6. Conclusions composite structures—a review. J Sound Vib 2000;230(2):357–
78.
[3] Doebling SW, Farrar CR, Prime MB, Shevitz DW. Damage
An experimentally validated damage detection algor- identification and health monitoring of structural and mechanical
ithm using features extracted from vibration-based systems from changes in their vibration characteristics: a litera-
analysis data as input for ANNs, for location and sever- ture review. Technical Report No.: LA-13070-MS, 1996.
ity prediction of damage in beam-like structures was [4] Cawley P, Adams RD. The location of defects in structures from
measurements of natural frequencies. J Strain Anal
presented. Different damage scenarios have been created 1979;14(2):49–57.
by reducing the local thickness of the selected elements [5] Yuen MMF. A numerical study of the eigenparameters of a dam-
at different locations, and simulated vibration responses aged cantilever. J Sound Vib 1985;103(3):301–10.
have been introduced to ANNs with and without arti- [6] Inada T, Shimamura Y, Todoroki A, Kobayashi H, Nakamura
ficial noise during training. Sensitivity analysis has also H. Damage identification method for smart composite cantilever
beams with piezoelectric materials. In: Proceeding of the 2nd
been performed on extracted features by using different International Workshop on Structural Health Monitoring, Stan-
vibration modes considering the effect of damage ford, 1999. Technomic Publishing Co; 2000. p. 986–94.
location and severity before introducing them to ANNs. [7] Fox CHJ. The location of defects in structures: a comparison of
1802 M. Sahin, R.A. Shenoi / Engineering Structures 25 (2003) 1785–1802
the use of natural frequency and mode shape data. In: Proceeding [26] Yun CB, Bahng EY. Substructural identification using neural net-
of the International Modal Analysis Conference. 1992. p. 522–8. works. Comput Struct 2000;77(1):41–52.
[8] Hu N, Wang X, Fukunaga H, Yao ZH, Zhang HX, Wu ZS. Dam- [27] Mukherjee A, Ravindra V. A fuzzy-neuro health monitoring tool
age assessment of structures using modal test data. Int J Solids transient vibration response. Adv Vib Eng 2002;1(1):100–16.
Struct 2001;38(18):3111–26. [28] Worden K, Ball AD. Modal analysis-based condition assessment
[9] Cornwell P, Doebling SW, Farrar CR. Application of the strain of structures using neural networks. In: Proceedings of 6th Inter-
energy damage detection method to plate-like structures. J Sound national Congress on Condition Monitoring and Diagnostic
Vib 1999;224(2):359–74. Engineering Management. 1994. p. 139–46.
[10] Pandey AK, Biswas M, Samman MM. Damage detection from [29] Barai SV, Pandey PC. Vibration signature analysis using artificial
changes in curvature mode shapes. J Sound Vib neural networks. J Comput Civil Eng ASCE 1995;9(4):259–65.
1991;145(2):321–32. [30] Elkordy MF, Chang KC, Lee GC. Neural networks trained by
[11] Ratcliffe CP. Damage detection using a modified Laplacian oper- analytically simulated damage states. J Comput Civil Eng ASCE
ator on mode shape data. J Sound Vib 1997;204(3):505–17. 1993;7(2):130–45.
[12] Ho YK, Ewins DJ. Numerical evaluation of the damage index. [31] Kudva JN, Munir N, Tan PW. Damage detection in smart struc-
In: Proceeding of the 2nd International Workshop on Structural tures using neural networks and finite-element analyses. Smart
Health Monitoring, Stanford, 1999. Technomic Publishing Co; Mater Struct 1992;1(2):108–12.
2000. p. 995–1011. [32] Islam AS, Craig KC. Damage detection in composite structures
[13] Wahab MMA, Roeck DD. Damage detection in bridges using using piezoelectric materials. Smart Mater Struct
modal curvatures: application to a real damage scenario. J Sound 1994;3(3):318–28.
Vib 1999;226(2):217–35. [33] Okafor AC, Chandrashekhara K, Jiang YP. Delamination predic-
[14] Lakshmi K, Jebaraj C. Sensitivity analysis of local/global modal tion in composite beams with built-in piezoelectric devices using
parameters for identification of a crack in a beam. J Sound Vib modal analysis and neural network. Smart Mater Struct
1999;228(5):977–94. 1996;5(3):338–47.
[15] Pai PF, Young LG. Damage detection of beams using operational [34] Staszewski WJ, Worden K, Wardle R, Tomlinson GR. Fail-safe
deflection shapes. Int J Solids Struct 2001;38(18):3161–92. sensor distributions for impact detection in composite material.
[16] Pai PF, Jin S. Locating structural damage by detecting boundary Smart Mater Struct 2000;9(3):298–303.
effects. J Sound Vib 2000;231(4):1079–110. [35] Zapico JL, Worden K, Molina FJ. Vibration-based damage
[17] Waldron K, Ghoshal A, Schulz MJ, Sundaresan MJ, Ferguson F, assessment in steel frames using neural networks. Smart Mater
Pai PF et al. Damage detection using finite element and laser Struct 2001;10(3):553–9.
operational deflection shapes. Finite Elements Anal Des [36] Xu YG, Liu GR, Wu ZP, Huang XM. Adaptive multilayer per-
2002;38(3):193–226. ceptron networks for detection of cracks in anisotropic laminated
[18] Beale R, Jackson T. Neural computing: an introduction. Bristol: plates. Int J Solids Struct 2001;38(32-33):5625–45.
The Institute of Physics, 1990. [37] Sensburg O, Worden K, Tomlinson GR. Detection and location
[19] Hassoun MH. Fundamentals of artificial neural networks. The of structural damage using dynamic test data. In: Proceedings of
M.I.T. Press, 1995. the 19th ICAS Congress/AIAA Aircraft Systems Conference.
[20] Mehrotra K. Elements of artificial neural networks. The M.I.T 1994. p. 2834–44.
Press, 1997. [38] Sahin M, Shenoi RA. Vibration-based damage identification in
[21] Patterson DW. Artificial neural networks: theory and appli- beam-like composite laminates by using artificial neural net-
cations. Prentice Hall, 1996. works. Proc Inst Mech Eng, Part C, J Mech Eng Sci
[22] Bishop CM. Neural networks for pattern recognition. Clarendon 2003;217(6):661–76.
Press, 1995. [39] ANSYS. User Manual. ANSYS, Inc, 2002.
[23] Worden K, Tomlinson GR. Damage location and quantification [40] ANSYS Help System. Elements Manual. ANSYS, Inc, 2002.
using neural networks. In: Edwards JH, Kerr J, Stanley P, editors. [41] Blevins RD. Formulas for natural frequency and mode shape.
Engineering integrity assessment. Sheffield: Engineering Publi- Robert E. Krieger Publishing Co., Inc, 1979.
cations; 1994. p. 11–31. [42] Cunningham PR, White RG. A new measurement technique for
[24] Worden K, Burrows AP. Optimal sensor placement for fault the estimation of core shear strain in closed sandwich structures.
detection. Eng Struct 2001;23(8):885–901. Composite Struct 2001;51(3):319–34.
[25] Worden K, Ball AD, Tomlinson GR. Fault location in structures [43] Demuth H, Beale M. Neural network toolbox, user’s guide, ver-
using neural networks. In: Proceeding of 75th Meeting of the sion 4. The Math Works, Inc, 2000.
AGARD Structures and Materials Panel on Smart Structures for
Aircraft and Spacecraft. 1992. p. 1-9.