development and field application of a nonlinear
TRANSCRIPT
This content has been downloaded from IOPscience. Please scroll down to see the full text.
Download details:
IP Address: 143.248.122.79
This content was downloaded on 17/07/2017 at 03:16
Please note that terms and conditions apply.
Development and field application of a nonlinear ultrasonic modulation technique for fatigue
crack detection without reference data from an intact condition
View the table of contents for this issue, or go to the journal homepage for more
2016 Smart Mater. Struct. 25 095055
(http://iopscience.iop.org/0964-1726/25/9/095055)
Home Search Collections Journals About Contact us My IOPscience
You may also be interested in:
Baseline-free fatigue crack detection based on spectral correlation and nonlinear wave modulation
Peipei Liu, Hoon Sohn, Suyoung Yang et al.
Wireless ultrasonic wavefield imaging via laser for hidden damage detection inside a steel box
girder bridge
Yun-Kyu An, Homin Song and Hoon Sohn
Baseline-free damage visualization using noncontact laser nonlinear ultrasonics and state space
geometrical changes
Peipei Liu, Hoon Sohn and Byeongjin Park
Monitoring of pipelines in nuclear power plants by measuring laser-based mechanical impedance
Hyeonseok Lee, Hoon Sohn, Suyoung Yang et al.
Guided wave based structural health monitoring: A review
Mira Mitra and S Gopalakrishnan
A large-area strain sensing technology for monitoring fatigue cracks in steel bridges
Xiangxiong Kong, Jian Li, William Collins et al.
Instantaneous reference-free crack detection based on polarization characteristics ofpiezoelectric
materials
Seung Bum Kim and Hoon Sohn
Fatigue crack detection in metallic structures with Lamb waves and 3D laser vibrometry
W J Staszewski, B C Lee and R Traynor
Development and field application of anonlinear ultrasonic modulation techniquefor fatigue crack detection without referencedata from an intact condition
Hyung Jin Lim1, Yongtak Kim1, Gunhee Koo1, Suyoung Yang1,Hoon Sohn1, In-hwan Bae2 and Jeong-Hwan Jang3
1Department of Civil and Environmental Engineering, KAIST, Daejeon, 34141, Korea2New Airport Hiway Co., Ltd, Incheon, 22694, Korea3 TM E&C, Seongnam, Gyeonggi-do, 13209, Korea
E-mail: [email protected]
Received 28 December 2015, revised 1 July 2016Accepted for publication 27 July 2016Published 24 August 2016
AbstractIn this study, a fatigue crack detection technique, which detects a fatigue crack without relyingon any reference data obtained from the intact condition of a target structure, is developed usingnonlinear ultrasonic modulation and applied to a real bridge structure. Using two wafer-type leadzirconate titanate (PZT) transducers, ultrasonic excitations at two distinctive frequencies areapplied to a target inspection spot and the corresponding ultrasonic response is measured byanother PZT transducer. Then, the nonlinear modulation components produced by a breathing-crack are extracted from the measured ultrasonic response, and a statistical classifier, which candetermine if the nonlinear modulation components are statistically significant in comparison withthe background noise level, is proposed. The effectiveness of the proposed fatigue crackdetection technique is experimentally validated using the data obtained from aluminum platesand aircraft fitting-lug specimens under varying temperature and loading conditions, and througha field testing of Yeongjong Grand Bridge in South Korea. The uniqueness of this study lies inthat (1) detection of a micro fatigue crack with less than 1 μm width and fatigue cracks in therange of 10–20 μm in width using nonlinear ultrasonic modulation, (2) automated detection offatigue crack formation without using reference data obtained from an intact condition, (3)reliable and robust diagnosis under varying temperature and loading conditions, (4) applicationof a local fatigue crack detection technique to online monitoring of a real bridge.
Keywords: nonlinear ultrasonic modulation, fatigue crack detection, PZT transducers, onlinebridge monitoring
(Some figures may appear in colour only in the online journal)
1. Introduction
A crack is one of the primary culprits for the failure ofmetallic structures and estimated that up to 90% of failures ofin-service metallic structure are the result of fatigue cracks[1]. A fatigue crack is initiated from a damage precursor atunperceivable level (e.g. dislocation or micro crack in mate-rials), when the material is subjected to repeated loading. The
crack often continues to grow to a critical point and leads to asudden failure of the system without providing a sufficientlead time for prevention [2]. For most metallic materials, it isknown that a fatigue crack becomes conspicuous only afterthe crack reaches about 80% of the total fatigue life [3]. Forexample, the middle span of Seongsu Bridge suddenly fellinto Han River in Seoul, South Korea in 1994, claiming thelives of 32 people and injuring 17 people. A later
Smart Materials and Structures
Smart Mater. Struct. 25 (2016) 095055 (14pp) doi:10.1088/0964-1726/25/9/095055
0964-1726/16/095055+14$33.00 © 2016 IOP Publishing Ltd Printed in the UK1
investigation discovered that this bridge collapse resultedfrom failure of welded joints due to repeated traffic load-ings [4].
To detect a fatigue crack at its early stage, several non-destructive testing (NDT) and structural health monitoringtechniques have been developed. Radiographic techniquewhich uses X or Gamma ray is a well-known and widely usedNDT technique not only for fatigue crack detection but alsofor other industrial applications [5]. However, due to strictregulations by radioactive problem, the radiographic techni-que has a difficulty applying to large structure such as bridge.Eddy current technique is particularly well suited for detect-ing surface cracks in conductive materials, and can also beused for checking electrical conductivity and coating thick-ness measurements [6]. Acoustic emission (AE) techniquedetects elastic waves generated when a fatigue crack isinitiated. AE technique has been used for detecting andlocalizing damage in composite, concrete and metallic mate-rials [7]. However, the passive monitoring characteristic AEtechnique makes the sensors always be activated to ‘listen’the elastic waves from crack and the wave from crack also canbe missed due to ambient noise. Thermography technique isalso applied for fatigue crack detection. A hybrid ultrasonic/infrared technique for fatigue crack detection was introducedusing heat occurs at a crack due to friction between the cracksurfaces when an ultrasonic wave propagates to the fatiguecrack [8]. Laser lock-in thermography, which utilizes a con-tinuous wave laser as a heat source for lock-in thermographytechnique is developed [9]. For metallic structure applicationof thermography techniques, the surface of target structureshould be coated using a matt paint avoiding the lightreflected from the structure.
Among these techniques, nonlinear ultrasonic techniqueshave gained prominence due to their higher sensitivity to amicro fatigue crack and potential for online monitoring [10–12]. Nonlinear ultrasonic techniques attempt to detect a fati-gue crack by tracing nonlinear phenomena such as harmonicsand modulations created by a source of nonlinearity such as afatigue crack. When an ultrasonic wave at a single inputfrequency propagates through a localized nonlinear sourcelike a fatigue crack, additional components at harmonics ofthe input frequency are generated [11–13]. On the other hand,when two ultrasonic waves at two distinctive frequencies arepropagated, the interaction of the ultrasonic inputs producesnonlinear ultrasonic responses at the sum and difference ofthese two frequencies [14–18].
Nonlinear ultrasonic modulation has been used fordetecting cracks in welded pipe joints and concrete beams[19, 20]. Using a piezoelectric stack actuator for generation ofa low frequency (LF) signal and a surface-mounted lead zir-conate titanate (PZT) transducer for creation of a high fre-quency (HF) signal, a fatigue crack in an aluminum plate isdetected [21]. Bolt-loosening in aluminum plates and dela-mination in composites are detected using two surfacemounted PZTs for generation of both LF and HF signals[22, 23]. Fixed LF and swept HF signals are used to find anoptimal combination of LF and HF signals that can amplifythe modulation [24]. A fatigue crack in an aircraft fitting-lug
mock-up specimen is detected by sweeping the frequencies ofboth LF and HF input signals [25].
In spite of recent developments in fatigue crack detectiontechniques using nonlinear ultrasonic modulation, there arestill technical challenges that need to be overcome for realfield applications. First, the generation of nonlinear modula-tion components is not guaranteed at one specific combina-tion of two input frequencies even at the presence of a fatiguecrack, and the generation of the nonlinear modulation com-ponents heavily depends on the dynamic characteristics of thehost structure. In return, the dynamic characteristics of thestructure constantly vary according to the temperature andloading conditions surrounding the structure [26, 27].Therefore, the fatigue detection cannot rely on one particularcombination of two input frequencies, and the input fre-quencies need to be constantly adjusted or swept. Second, fordamage detection, most nonlinear ultrasonic modulationtechniques use reference data obtained from the intact con-dition of host structure. However, these techniques are sus-ceptible to false alarms due to signal variations unrelated tothe damage such as changing temperature conditions. Anautomated fatigue crack detection technique based on thecombination of nonlinear ultrasonic modulation and sequen-tial outlier analysis has been developed by the authors’ groupand validated through laboratory testing [28]. However, theapplications of this previously developed technique to in-sitebridge structures reveal that (1) the harsh field conditions atthe bridge site can increase the noise level at the modulationfrequency, and this increased noise level can occasionallyproduce a false positive indication of crack, and (2) if mod-ulation occurs over the majority of the investigated frequencycombinations, the outlier analysis may fail to detect the pre-sence of crack (false negative).
In this study, the shortcomings of the previous algorithmare overcome and the reliability of crack detection is sig-nificantly improved by developing a new fatigue crackdetection technique without relying on any reference dataobtained from the intact condition of a target structure. Theproposed technique exploits the statistical properties, such asthe skewness and median values, of the modulation compo-nents to reliably identify the presence of crack. The effec-tiveness of the proposed fatigue crack detection technique isexperimentally tested using data obtained from aluminumplates and aircraft fitting-lug specimens under varying temp-erature and loading conditions. Finally, the proposed techni-que is applied to real periodic monitoring of YeongjongGrand Bridge.
The uniqueness of this study lies in that, (1) detection ofa micro fatigue crack with less than 1 μm width and fatiguecracks in the range of 10–20 μm in width using nonlinearultrasonic modulation, (2) automated detection of fatiguecrack formation without using reference data, (3) improvedreliability and robustness of crack diagnosis, particularlyunder varying temperature and loading conditions, (4) appli-cation to periodic monitoring of a real bridge structure.
This paper is organized as follows. In section 2, thebackgrounds and development of the proposed fatigue crackdetection technique is presented. Section 3 describes the
2
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
fatigue crack detection on aluminum plates and aircraft fit-ting-lug specimens under various temperature and loadingconditions. Next, field application of the fatigue crackdetection technique to Yeongjong Grand Bridge for periodicmonitoring is provided in section 4. Finally, the summary andconclusion are presented in section 5.
2. Development of a fatigue crack detectionalgorithm
2.1. Working principle of crack-induced nonlinear ultrasonicmodulation
When two longitudinal plane waves a and b at distinctivefrequencies wa and wb w w<a b( ) with amplitudes A0 and B0
are applied to an intact (linear) structure in the z direction, thedisplacement of the input waves u0( ) become
k w k w= - + -u A z t B z texp i exp i , 1a a b b0 0 0( ( )) ( ( )) ( )
where ka and kb are the wavenumbers of waves a and b,respectively. Here, intrinsic material nonlinearity of thestructure is omitted because this nonlinear effect is muchsmaller than the nonlinearity produced by a localized fatiguecrack [29]. The stress induced by the input waves s0( ) can bewritten as
s k w
k w
=¶¶
= -
+ -
Eu
zE A z t
B z t
exp i
exp i i,
2L
a a
Lb b
0 00
0 0
0
[ ( ( ))
( ( ))]( )
where E0 the Young’s modulus of the intact structure. A L0
and B L0 are the amplitude of strain due to A0 and B ,0
respectively.When a localized fatigue crack is introduced to the
structure at z0 and it is assumed that the average Young’smodulus is locally reduced from E0 to E and the instanta-neous Young’s modulus at z ,0 E z ,1 0( ) fluctuates around thisreduced Young’s modulus in proportion to the amplitude ofthe applied wave at the crack location [30]:
⎡⎣⎢
⎛⎝⎜
⎞⎠⎟
⎤⎦⎥
g
a a
= +¶
¶
= -¶
¶+
¶¶
E z E zu z
z
u z
zE E
u z
z1 max ,
3
1 0 00 0
0 00 0
0 0
( ) ¯ ( ) ( )
( ) ( )
( )
where γ is the nonlinear elastic constant (g a= E ,0
a <0 1 ,) E z0¯ ( ) is the average Young’s modulus afterfatigue crack formation at z ,0 and α is the nonlinearcoefficient for representing the nonlinearity due to thelocalized fatigue crack. The max operation in equation (3)finds the maximum strain induced by the input waves at thecrack location. The stress induced by the input waves at the
crack location, s ,1 can be written as
⎡⎣⎢
⎛⎝⎜
⎞⎠⎟
⎤⎦⎥
⎛⎝⎜
⎞⎠⎟
s a
a
=¶
¶= -
¶¶
´¶
¶+
¶¶
E zu z
z
u z
z
Eu z
zE
u z
z
1 max
.
41 1 0
0 0 0 0
00 0
00 0
2
( ) ( ) ( )
( ) ( )( )
The stress at the crack location induced by the inputwaves is obtained substituting equation (1) into (4)
s s s s= + + , 5L H M1 1 1 1 ( )
where s ,L1 sH
1 and sM1 are the linear components at wa and w ,b
the nonlinear harmonic components at w2 a and w2 b (secondharmonics) and the nonlinear modulation components atw wa b (first sidebands), respectively. Each terms can bewritten as
s k w
k w
= -
+ -
E A z t
B z t
exp i
exp i i,6
L La a
Lb b
1 0 1 0
1 0
[ ( ( ))( ( ))]
( )
s k w
k w
= - -
+ -
E A z t
B z t
exp 2i
exp 2i , 7
H Ha a
Hb b
1 0 1 0
1 0
[ ( ( ))( ( ))] ( )
and
s k k w w= - - E A z texp i , 8M Mb a b a1 0 1 0( [( ) ( ) ]) ( )
where A L1 and B L
1 are the strain amplitude of linearcomponents, A H
1 and B H1 are the strain amplitude of harmonic
components and A M1 is the strain amplitude of modulation
component.Using a more complex crack model with additional
square and cubic terms in equation (3), the contributions ofhigher order harmonics and modulations terms can be con-sidered. However, only the second harmonic and the firstmodulation terms are considered in equation (3) forsimplicity.
2.2. Damage classifier based on skewness–median analysis
Based on nonlinear ultrasonic modulation generated by alocalized fatigue crack, a damage classifier with the followingfour steps is developed.
Step I: Extraction of modulation componentsFirst, both LF and HF sinusoidal inputs are simulta-
neously applied to the structure, and the correspondingultrasonic responses at w wb a w wu
b a( ) are obtained using a
modified discrete Fourier transform at w wb a as shown infigure 1(a).
Step II: Establishment of a threshold value consideringnoise level
Next, as shown in figure 1(b), additional ultrasonicresponses w wn
b a( ) are obtained at w wb a by applying only
the HF input. The reasons of applying only HF input at wb forthreshold establishment are as follows: (1) when a sinusoidalinput is applied at w ,b the response spectrum actually hassome narrow bandwidth and the noise levels at the modula-tion frequencies also increase as shown in figure 2; (2) whenanother sinusoidal input is applied at w ,a a similar increase of
3
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
the noise level may be observed near w .a However, becausethe modulation frequencies are further away from wa
(wa<w ,b ) the noise levels at the modulation frequencies arenot much increased by the application of the LF input aspresented in figure 2; (3) the nonlinear interaction at betweenthe excitation at wa and the noise component at wb (orvice versa) is also negligible. Thus, w wn
b aare the sole out-
come of the measurement noise because nonlinear modulationoccurs only due to the interaction of HF and LF inputs. Then,the measurement of w wn
b ais repeated multiple times, the nth
noise value is denoted as w wn .n,b aFor example, the nth noise
values at w wb a are w wn .n,b aOnce multiple w wn n,b a
valuesare obtained, an exponential distribution is fitted to the
w wn n,b avalues. Here, because the magnitude of the noise
level is always positive in the frequency domain, an expo-nential distribution which has only positive values is selected.Then, a threshold value w wT
b a( ) corresponding to a user
specified one-sided confidence interval is established.Step III: Calculation of a nonlinear index (NI) for various
frequency combinations
Figure 1. Schematic diagram of the damage classifier based on skewness–median analysis by extracting modulation (first sideband) andthresholding considering noise level.
4
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
The NI at w wb a is defined as
= -w w w ww w u TNI . 9b a b ab a ( )
When nonlinear modulation occurs due to crack formation, itis expected that the modulation amplitude w wu
b a( ) become
larger than the threshold w wT b a( ) and the w wNI b a valuebecomes positive as shown in figure 1(c). On the other hand,the w wNI b a
value will remain mostly negative without afatigue crack. As mentioned previously, the temperature andloading conditions of a field structure constantly vary and theconditions for modulation generation heavily depend on thedynamic characteristics of the host structure. Therefore, the
w wNI b a values are estimated at various combinations of waand wb to increase the chance of detecting a fatiguecrack [31].
Step IV: Crack diagnosis using skewness and medianstatistics
For crack diagnosis, skewness and median statistics ofthe NI values obtained from multiple combinations of wa andwb are computed as shown figure 1(d). ‘Skewness’ is the third
standardized moment and defined as
⎜ ⎟⎡⎣⎢
⎛⎝
⎞⎠
⎤⎦⎥
ms
=-
Eskew NINI
,3
[ ] where m and s are the mean
and standard variation of the NI values, respectively, and E isan expectation operator. The skewness represents the asym-metry of the NI distribution. The ‘median’ is the numericalvalue separating the higher half of a data sample (here, NIvalues) from the lower half. When the modulation compo-nents are generated at several frequency combinations, the NIdistribution is skewed to the positive value, implying thepresence of a fatigue crack. Similarly, crack formation alsocan shift the median value of the NI distribution to a positivevalue when the modulation occurs in the majority of testedfrequency combinations. By performing crack diagnosisusing two test statistics as follows, the reliability of crackdiagnosis is improved:
If both ‘skewness’ and ‘median’ are <0,‘intact’ otherwise, ‘damage’.
Note that, in the proposed diagnosis algorithm, both theNI and threshold values are obtained using data only from thecurrent state of the target specimen without relying on anyhistory date from the intact condition.
3. Experimental validations
3.1. Crack detection in aluminum plates
Two identical aluminum plates (specimen I) with a centerhole were fabricated using 7075-T351 aluminum alloy. Thegeometry and dimensions of the specimen I are shown infigure 3(a). A fatigue crack, which is 35 mm long and lessthan 10 μm wide, was formed in one of the specimens duringa cyclic loading test (figure 3(b)). The fatigue test was per-formed using a MTS machine with a 10 Hz cycle rate, amaximum load of 64.6 kN and a stress ratio R=0.1. Detailson the fatigue test are presented in [32]. Four identical PZTsmanufactured by APC International were installed on each
Figure 2.Noise levels obtained from the aluminum plate specimen infigure 3 due to the application of (1) HF (185 kHz) input only, (2) LF(45 kHz) input only and (3) no input. The noise level at themodulation frequencies (140 and 230 kHz) due to HF input only islarger than the noise level due to LF input only. The noise levelwithout input is similar with that of LF input only.
Figure 3. Aluminum plate specimen (specimen I). (a) The geometry and dimensions of the specimen I. (b) A close-up of the fatigue crack.
5
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
specimen. Each PZT has a diameter of 6.35 mm and athickness of 0.254 mm. Two PZTs labeled as ACT 1 andACT 2 were used for generation of ultrasonic waves, and theother two denoted as SEN 1 and SEN 2 for sensing.
For data acquisition (DAQ), a National Instruments(NIs) PXI system consists of two NI PXI-5421 arbitrarywaveform generators (AWGs), a NI PXI-5122 2-channel highspeed digitizer (DIG). One AWG was used to apply aHF input at wb to ACT 1 and the other AWG to apply a LFinput at wa to ACT 2. The output responses from SEN 1 andSEN 2 were simultaneously measured using DIG. The AWGs
and DIG are synchronized and controlled by LabVIEWsoftware.
Both LF and HF input signals had a peak-to-peak voltageof ±12 V, and they were converted to analog input signals ata conversion rate of 1 MHz with zero-order holding. Theoutput responses were measured with 1MHz sampling ratefor 0.5 s. The responses were measured 10 times and averagedin the time domain to improve the signal-to-noise ratio. Forinvestigating various frequency combinations, wa was steppedfrom 45 to 50 kHz with a 1 kHz increment and wb from 180 to185 kHz with a 1 kHz increment, respectively. As shown in
Figure 4. Frequency domain responses of the raw signals obtained from the intact (a) and damaged (b) specimen I when wa at 45 kHz and wb
at 185 kHz are applied simultaneously.
Figure 5. NI values and distribution for crack diagnosis using the data obtained from the specimen I.
6
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
figure 4, frequency domain responses of the raw signalsobtained from the aluminum plate specimen show thatintrinsic material nonlinearity are negligible and the mod-ulation components are mainly generated by the presence of afatigue crack.
Figures 5(a) and (b) show the NI values obtained fromSEN 1 of the specimen I in intact and damaged conditions,respectively. The NI values are obtained using eqation (9),and a threshold value corresponding to a 99.99% confidenceinterval is estimated by fitting an exponential distribution to 6
w wnb a
values. In figure 5(a), both the skewness and medianhave negative value for the intact case, while only theskewness value is positive for the damage case in figure 5(b).In figures 5(c) and (d), similar diagnosis obtained from SEN 2is shown.
Next, an aluminum (6061-T6) plate specimen (specimenII) with 3 mm thickness was fabricated as shown infigure 6(a). Three identical PZTs with 10 mm diameter and0.5 mm thickness manufactured by Haiying Group wereinstalled on specimen II. Two PZTs labeled as ACT 1 andACT 2 were used for generation of LF and HF inputs, and thethird one denoted as SEN for sensing, respectively. wa wasstepped from 30 to 40 kHz with a 1 kHz increment, and wb
from 181 to 183 kHz with a 1 kHz increment. The other DAQparameters were identical with the previous test. The
specimen was subjected to 4∼40 kN (R=0.1) tensile cyclicloading with 10 Hz cycle rate using a universal testingmachine (INSTRON 8801). After 50 000 cycles, a microcrack was formed near the center hole. A microscopic imageof specimen II was taken after tensile loading tests. As shownin figure 6(b), the length and width of the micro crack near thecenter hole were less than 60 μm and 1 μm, respectively. Anexponential distribution is fitted using 11 w wn
b avalues, and
the threshold value corresponding to a 99.99% confidenceinterval is estimated. Both skewness and median statistics hadnegative values before crack formation, while both becamepositive after crack formation as shown in figures 6(c)and (d).
The experimental results demonstrate that the proposednonlinear ultrasonic modulation is able to detect even a microfatigue crack with less than 1 μm width. Other researchershave also reported the superior sensitivity of the nonlinearultrasonic technique to micro defects over that of the linearones [12, 33–35].
3.2. Crack detection in aircraft fitting-lug specimens
Two mock-up specimens, which represent a fitting-lug con-necting an aircraft wing to a main fuselage frame, were fab-ricated from 6061-T6 aluminum alloy as shown in figure 7. A
Figure 6. Detection of a micro crack in aluminum plate specimen (specimen II) (a) the geometry and dimensions of specimen II, (b)microscopic image of the micro crack near the center hole, (c) diagnosis result before crack formation, and (d) after crack formation.
7
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
40 mm long and 20 μm wide fatigue crack was introduced toone of the specimens through cyclic loading with varyingamplitudes of 0–6.7 kN as shown in figure 7(b). This loadingis equivalent to real operational loading corresponding to1000 flight hours according to current aircraft design speci-fications. Details on the fatigue test are presented in [28].
Three identical dual PZT modules manufactured byMetis Design were installed to each specimen near the crackprone location as shown in figure 7. Here, each dual PZTmodule consists of two concentric inner circle and outer ringPZTs, and they are packaged by a Kapton tape with printedcircuit and two SMA connecters. The outer and inner dia-meters of the ring PZT, the diameter of the inner circular PZT,and the thickness of the PZTs are 18 mm, 10 mm, 8 mm and0.3 mm, respectively. One dual PZT module was used as anactuator (ACT), and the others as sensors (SEN 1 and SEN 2).Unlike the aluminum plates, a single dual PZT module wasused for exerting both LF and HF inputs. LF input wasapplied to the outer ring PZT, and HF input to the inner circlePZT of ACT, respectively. Corresponding ultrasonicresponses were measured using the inner circle PZTs of SEN1 and SEN 2. Note that a larger PZT size is preferred forexcitation while a smaller size is more advantageous forsensing [36]. For DAQ, the identical DAQ system describedin section 3.1 was used. For investigating various frequencycombinations, wa was stepped from 45 to 50 kHz with a 1 kHzincrement and wb from 190 to 195 kHz with a 1 kHz incre-ment, respectively. The other DAQ parameters were identicalwith the aluminum plate test.
Figures 8(a) and (b) show the NI distributions, theskewness and median statistics obtained from SEN 1 of theintact and damage aircraft fitting-lug specimens, respectively.The threshold value corresponding to a 99.99% confidenceinterval is estimated by fitting an exponential distribution to 6w wn
b avalues. Both skewness and median statistics have
negative values in the intact case, while both are positive inthe damage case. Similar results obtained from SEN 2 shown
in figures 8(c) and (d) also indicate that the damage case issuccessfully detected.
3.3. Crack detection under varying temperature and loadingconditions
The performance of the developed fatigue crack classifier wasvalidated under various temperature and ambient vibrationconditions for the aluminum plates (specimen I) and the air-craft fitting-lug specimens. For the temperature test, the spe-cimens were placed inside a temperature chamber and signalswere measured under five different temperature conditions(−15 °C, 0 °C, 15 °C, 30 °C and 45 °C). The temperature ofthe chamber was maintained within 1 °C accuracy for eachtemperature condition. For the ambient vibration tests, ran-dom excitations with different maximum peak amplitudeswere exerted to the specimens using a mechanical shaker. Thefrequency range of the random excitation was 0–50 Hz. Thepeak amplitudes were measured using an accelerometerinstalled on the specimens, and the temperature was main-tained at room temperature (20 °C). The fatigue crack diag-noses are summarized in tables 1–4. For all the casesinvestigated in the tables, there was no false-positive indica-tion of crack for the intact cases, and the cracks were suc-cessful detected even at the presence of varying temperatureand ambient loading conditions.
The outlier analysis based fatigue crack diagnosis tech-nique developed by the author’s group in [28] was applied tothe data obtained from various temperature and loadingconditions. Figure 9 shows the false alarms of the outlieranalysis using the data obtained from the fitting-lug speci-mens. Figure 9(a) shows a false positive indication of crackdue to occasionally produced high noise level and (b) shows afalse negative indication with no outliers even at the presenceof crack due to the majority generation of modulation over theinvestigated frequency combinations. Comparing the results
Figure 7. A mock-up specimen representing a fitting-lug connecting an aircraft wing to a main fuselage frame. (a) The geometry anddimensions. (b) A close-up of the fatigue crack.
8
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
presented in table 4, the proposed technique in this studyshows improved reliability and robustness of crack diagnosis.
4. Applicationh to Yeongjong Grand Bridgemonitoring
4.1. Introduction to Yeongjong Grand Bridge
Yeongjong Grand Bridge (figure 10(a)) is the world first 3Dself-anchored suspension bridge which links betweenYeongjong Island (Incheon Airport, ICN) and Incheon
(Seoul) in South Korea. The bridge is part of the IncheonInternational Airport Highway and was completed in 2000.The total length of the bridge is 4420 m, and the length andwidth of the main suspension bridge are 550 m and 35 m,respectively. The bridge carries both highway traffics andrailway trains as shown in figure 10(b). The double decksystem has six lanes of highways on the upper deck and fourlanes of highways and two lanes of railways on the lowerdeck. The bridge is maintained by New Airport Hiway Co.Currently, Yeongjong Grand Bridge is equipped with about400 sensors including GPS, strain gauges, anemometers andaccelerometers [37].
Figure 8. The distribution of NI values and fatigue crack detection results obtained from the aircraft fitting-lug specimens.
Table 1. Crack detection in the specimen I under varying temperatures.
Intact Damage
Temp. Sensor Skewness Median Skewness Median
−15 °C SEN 1 −3.22 −2.14×10−6 0.20 −1.65×10−6
SEN 2 −2.71 −9.88×10−7 −2.22 7.40×10−8
0 °C SEN 1 −6.06 −2.09×10−6 2.58 −1.85×10−6
SEN 2 −4.36 −6.83×10−7 −2.46 9.01×10−7
15 °C SEN 1 −0.29 −2.57×10−6 0.79 −2.11×10−6
SEN 2 −0.80 −1.24×10−6 0.69 1.05×10−7
30 °C SEN 1 −1.00 −2.28×10−6 0.49 −2.24×10−6
SEN 2 −0.18 −1.45×10−6 0.81 1.32×10−7
45 °C SEN 1 −0.54 −2.48×10−6 0.02 −1.71×10−6
SEN 2 −0.89 −1.97×10−6 0.24 −1.54×10−6
Note. Positive values are marked with bold characters.
9
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
In June, 2005, the lower deck of the bridge was modifiedto carry two railways and Incheon Airport Railroad Express(A’REX) has been in operation since 23rd March 2007. Inaddition, Korean Train Express (KTX) has been passingthrough the bridge since 30th June 2014. Note that the typicallength and weight of KTX trains are 338 m and 69 tons,respectively, and KTX trains are 3 times longer and 3.5 timesheavier than A’REX trains as shown in table 5. These addi-tional loadings, which were not considered during the initial
design stage of the bridge, are expected to accelerate thedeterioration of the bridge, in particular, the fatigue crackinitiation and growth. In response to the expected deteriora-tion of the bridge, New Airport Hiway Co. decided to pur-chase and install our proposed monitoring system specificallytargeted for online fatigue crack detection. Due to the con-fidentiality agreement with New Airport Hiway Co., onlylimited outcomes from the field measurement tests are dis-closed in this study in spite of successful applications of the
Table 2. Crack detection in the specimen I under varying loadings.
Intact Damage
Max acc. Sensor Skewness Median Skewness Median
2.0 g SEN 1 −0.56 −2.51×10−6 0.76 −2.27×10−6
SEN 2 −6.47 −1.13×10−6 0.90 1.41×10−7
2.5 g SEN 1 −5.23 −2.76×10−6 0.30 −1.78×10−6
SEN 2 −3.68 −2.91×10−6 0.18 1.41×10−6
3.0 g SEN 1 −1.17 −2.81×10−6 1.97 −2.47×10−6
SEN 2 −0.63 −2.52×10−6 1.50 −1.95×10−6
3.5 g SEN 1 −0.64 −3.14×10−6 0.23 1.69×10−6
SEN 2 −0.20 −3.02×10−6 1.03 −1.72×10−6
4.0 g SEN 1 −0.39 −3.13×10−6 1.78 −2.45×10−6
SEN 2 −0.39 −2.92×10−6 1.83 −2.10×10−6
Note. Positive values are marked with bold characters.
Table 3. Crack detection in the aircraft fitting-lug under varying temperatures.
Intact Damage
Temp. Sensor Skewness Median Skewness Median
−15 °C SEN 1 −7.02 −1.13×10−6 −0.84 3.71×10−7
SEN 2 −3.37 −9.51×10−7 −6.29 6.06×10−7
0 °C SEN 1 −6.44 −2.42×10−7 −2.22 1.51×10−6
SEN 2 −3.26 −3.37×10−7 −3.73 9.22×10−7
15 °C SEN 1 −0.48 −1.40×10−6 0.27 2.50×10−6
SEN 2 −3.59 −3.65×10−7 −2.87 8.06×10−7
30 °C SEN 1 −0.49 −6.69×10−7 −4.62 1.98×10−6
SEN 2 −3.85 −9.38×10−7 −3.26 5.98×10−7
45 °C SEN 1 −1.33 −8.93×10−7 1.90 1.84×10−6
SEN 2 −2.86 −4.05×10−7 1.30 1.32×10−6
Note. Positive values are marked with bold characters.
Table 4. Crack detection in the aircraft fitting-lug under varying loadings.
Intact Damage
Max acc. Sensor Skewness Median Skewness Median
2.0 g SEN 1 −5.50 −1.84×10−6 −2.27 3.17×10−7
SEN 2 −4.54 −1.49×10−6 −3.56 3.44×10−7
2.5 g SEN 1 −5.23 −2.25×10−7 1.84 3.76×10−6
SEN 2 −4.61 −1.67×10−7 −3.38 3.13×10−6
3.0 g SEN 1 −5.76 −3.33×10−7 2.18 7.00×10−6
SEN 2 −5.26 −7.47×10−7 0.72 4.42×10−6
3.5 g SEN 1 −5.65 −5.92×10−7 1.44 7.23×10−6
SEN 2 −4.93 −6.82×10−7 1.45 4.47×10−6
4.0 g SEN 1 −4.45 −8.76×10−8 2.21 4.83×10−6
SEN 2 −4.16 −1.70×10−7 −1.53 3.40×10−6
Note. Positive values are marked with bold characters.
10
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
proposed monitoring system to fatigue crack detection inthe field.
4.2. Monitoring system deployment
For Yeongjong Grand Bridge monitoring, a triple PZTmodule is designed by the authors and manufactured by MetisDesign as shown in figure 11. The triple PZT module consistsof three identical circular PZTs, and they are packaged by aKapton tape with printed circuit and three SMA connecters.The diameter and the thickness of each circular PZT are25 mm and 0.5 mm, respectively. LF and HF inputs areapplied to two circular PZTs (PZT A and B), and thecorresponding response are obtained by the third PZT(PZT C).
The triple PZT modules were installed on the west end ofthe suspension portion of Yeongjong Grand Bridge as shownin figure 12(a). These installation locations are determinedbecause the welded connections in these areas are discoveredto be prone to fatigue cracks during periodic visual inspec-tions (once every 5 years) required by a special act of Koreangovernment [38]. 6 triple PZT modules were installed insideof the upper box girder (figure 11(b)). Detailed configurationsof the DAQ system and the triple PZT modules installedinside the box girder are shown in figure 12(c).
For DAQ, the identical NI PXI system described insection 3.1 was used. Both LF and HF input signals had apeak-to-peak voltage of ± 12 V, and they were converted toanalog input signals at a conversion rate of 1 MHz with zero-order holding. The output responses were measured with1MHz sampling rate for 1 s. The responses were measured 10times and averaged in the time domain to improve the signal-to-noise ratio. The sweeping ranges for LF and HF inputs aredetermined considering the local resonance characteristics at
Figure 9. False alarms of the outliner analysis based crack diagnosis using the data obtained from the fitting-lug specimens.
Figure 10. Overview of Yeongjong Grand Bridge.
Figure 11. A triple PZT module composed of two circular PZTs forLF and HF ultrasonic excitation and third circular PZT for sensing.
11
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
each measurement location as summarized in table 6. Forexample, the LF and HF ranges for Loc. 1 were matched withthe local resonance frequency bands at Loc. 1 (40–50 kHz and197–199 kHz), respectively, so that crack motion necessaryfor nonlinear modulation can be maximized [18, 31].
4.3. Results of field tests
For periodic fatigue crack monitoring, data were obtained 6times in 2015 and fatigue crack diagnosis results are sum-marized in table 7. Similar to the laboratory tests shown infigure 13, intrinsic material nonlinearity and nonlinear inter-action between the inputs and the noise component are neg-ligible. NI values are obtained by subtracting the modulation
amplitude at w wb a when LF and HF inputs are appliedsimultaneously from a threshold value corresponding to a99.999% confidence interval by fitting an exponential dis-tribution using 11 noise values at w wb a when only HFinput is applied. It was concluded that there is no fatiguecrack on the presented locations and the findings were vali-dated through visual inspection. The data on 14th October2015 was obtained at night without traffics while the otherdata was obtained during the normal operation conditions ofthe bridge with traffics including trains. The results show thatthe proposed fatigue crack classifier shows no false-positiveindication of crack under the normal operational conditions.The maximum temperatures were 15 °C and 3 °C on 28thOctober and 4th December, respectively, which were lowerthan that of the other dates (22 °C). The results also show thatthe crack diagnosis was not affected by ambient temperaturevariations. Although fatigue cracks were not detected in thepresented locations, the proposed technique is possible todetect fatigue cracks in real bridge structure. However, thedetailed outcomes are not reported here because of the con-fidentiality. This fact gives reasons to expect that the eventualdevelopment of fatigue cracks in real bridge structures will besuccessfully detected using the proposed fatigue crackdetection technique.
Table 5. Length and weight of A’REX and KTX.
Train Operation Length (no. of cars) Weight w/o passenger
Airport Railroad Express (A’REX) From 2007.03.23 117 m (6) 203 tKorean Train Express (KTX) From 2014.06.30 388 m (20) 692 t
Figure 12. The installation locations of the DAQ system and triple PZT modules on Yeongjong Grand Bridge.
Table 6. Input frequency combinations (1 kHz increment).
Location LF wa( ) HF wb( )
Loc. 1 40–50 kHz 197–199 kHzLoc. 2 40–50 kHz 197–199 kHzLoc. 3 40–50 kHz 197–199 kHzLoc. 4 40–50 kHz 197–199 kHzLoc. 5 30–40 kHz 181–183 kHzLoc. 6 40–50 kHz 181–183 kHz
12
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
5. Conclusions
In this study, a new fatigue crack detection technique withoutrelying on any reference data obtained from the intact con-dition of a host structure is developed based on nonlinearultrasonic modulation so that fatigue cracks in in situ metalstructures can be automatically and reliably detected undertemperature and loading variations. When two ultrasonicinputs at distinctive frequencies are applied to the hoststructure, nonlinear modulation components are generateddue to fatigue crack. By analyzing skewness and medianstatistics of nonlinear modulation obtained from various inputfrequency combinations, crack diagnosis is performed.
The performance of the proposed fatigue crack detectiontechnique is experimentally validated using data from alu-minum plates and aircraft fitting-lug specimens and Yeong-jong Grand Bridge. The experimental results demonstrate that(1) a micro fatigue crack with less than 1 μm width andfatigue cracks in the range of 10–20 μm in width can besuccessfully detected using nonlinear ultrasonic modulation,(2) fatigue crack can be detected without relying on thereference data obtained from the pristine condition of themeasurement location, and (3) reliable and robust diagnosiscan be performed even under harsh varying temperature and
loading conditions in the field. From the experimental vali-dations on Yeongjong Grand Bridge, New Airport HiwayCo., in charge of the bridge maintenance, decided theextended application of our proposed system for online fati-gue crack detection from 2016.
For continuous transition of the proposed technique tocommercialization, the following issues need to be addressed.First, field engineers are indeed concerned about theremaining useful life once a fatigue crack is detected.Therefore, the development of fatigue prognosis as well asdiagnosis techniques is crucial for timing repair and main-tenance. Second, cabling for DAQ and power supply makesup about 50% of the total monitoring system cost. Formonitoring of large-scale structures, the cost of cablingincreases drastically and wireless sensing can be a goodalternative. Third, logistics for optimal sensor placement isnecessary so that sensors can be installed only to criticalpoints.
Acknowledgments
This work was supported by the Smart Civil InfrastructureResearch Program (13SCIPA01) funded by Ministry of Land,
Figure 13. Frequency domain responses of the raw signals obtained from Yeongjong Grand Bridge when wa at 45 kHz and wb at 199 kHz areapplied simultaneously.
Table 7. Periodic monitoring results.
Date 15.10.05 15.10.13 15.10.14 15.10.22 15.10.28 15.12.04
Loc. 1 Skewness −0.04 −2.83 −3.63 −2.37 −2.31 −3.60Median −2.14×10−6 −2.31×10−6 −3.75×10−6 −1.92×10−6 −2.60×10−6 −2.27×10−6
Loc. 2 Skewness −3.37 −3.01 −2.75 −2.69 −2.43 −2.39Median −3.95×10−7 −8.50×10−7 −2.51×10−6 −1.19×10−6 −1.64×10−6 −2.09×10−6
Loc. 3 Skewness −2.31 −2.69 −2.59 −2.93 −3.85 −3.30Median −1.10×10−6 −9.22×10−7 −1.54×10−6 −8.03×10−7 −3.44×10−6 −3.92×10−6
Loc. 4 Skewness −4.57 −3.25 −2.72 −2.30 −0.69 −0.23Median −2.07×10−6 −1.96×10−6 −2.45×10−6 −1.86×10−6 −2.44×10−6 −1.06×10−6
Loc. 5 Skewness −1.19 −1.17 −1.18 −1.29 −1.44 −2.05Median −5.18×10−6 −4.52×10−6 −4.95×10−6 −6.03×10−6 −5.18×10−6 −5.43×10−6
Loc. 6 Skewness −2.39 −1.38 −0.86 −1.27 −0.57 −0.74Median −2.17×10−6 −2.65×10−6 −2.94×10−6 −3.03×10−6 −1.27×10−6 −1.80×10−6
13
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al
Infrastructure and Transport (MOLIT) of Korea governmentand Korea Agency for Infrastructure Technology Advance-ment (KAIA), and New Airport Hiway Co., Ltd.
References
[1] Campbell F C 2008 Elements of Metallurgy and EngineeringAlloys (Materials Park, OH: ASM International)
[2] Zhou C, Hong M, Su Z, Wang Q and Cheng L 2013 Evaluationof fatigue cracks using nonlinearities of acousto-ultrasonicwaves acquired by an active sensor network Smart Mater.Struct. 22 015018
[3] Kim J Y, Jacobs L J and Qu J 2011 Nonlinear ultrasonictechniques for nondestructive damage assessment in metallicmaterials Proc. 8th IWSHM pp 531–8
[4] Kunishima M 1994 Collapse of the Korea seoul seongsu bridgeJST Failure Knowledge Database/100 Selected Cases
[5] Williams J J, Yazzie K E, Padilla E, Chawla N, Xiao X andDe Carlo F 2013 Understanding fatigue crack growth inaluminum alloys by in situ x-ray synchrotron tomographyInt. J. Fatigue 57 79–85
[6] Zilberstein V, Schlicker D, Walrath K, Weiss V andGoldfine N 2001 MWM eddy current sensors for monitoringof crack initiation and growth during fatigue tests and inservice Int. J. Fatigue 23 477–85
[7] Roberts T M and Talebzadeh M 2003 Acoustic emissionmonitoring of fatigue crack propagation J Conster. Steel.Res. 59 695–712
[8] Favro L D, Han X, Ouyang Z, Sun G and Thomas R L 2000Infrared imaging of defects heated by a sonic pulse Rev. Sci.Instrum. 71 2418
[9] An Y-K, Kim J M and Sohn H 2014 Laser lock-inthermography for detection of surface breaking fatiguecracks on uncoated steel structures NDT E Int. 65 54–63
[10] Zaitsev V, Nazarov V, Gusev V and Castagnede B 2006 Novelnonlinear-modulation acoustic technique for crack detectionNDT E Int. 39 184–94
[11] Cantrell J H and Yost W T 1994 Acoustic harmonicsgeneration from fatigue-induced dislocation dipoles Phil.Mag. A 69 315–26
[12] Jhang K-Y 2009 Nonlinear ultrasonic techniques fornondestructive assessment of micro damage in material: areview Int. J. Precis. Eng. Manuf. 10 123–35
[13] Cantrell J H and Yost W T 2001 Nonlinear ultrasoniccharacterization of fatigue microstructures Int. J. Fatigue 23S487–90
[14] Sutin A M and Nazarov V E 1995 Nonlinear acoustic methodsof crack diagnostics Radiophys. Quantum Electron. 38109–20
[15] Van Den Abeele K E-A, Johnson P A and Sutin A 2000Nonlinear elastic wave spectroscopy (NEWS) techniques todiscern material damage, part i: nonlinear wave modulationspectroscopy (NWMS) Res. Nondestruct. Eval. 12 17–30
[16] Duffour P, Morbidini M and Cawley P 2006 A study of thevibro-acoustic modulation technique for the detection ofcracks in metals J. Acoust. Soc. Am. 119 1463–75
[17] Croxford A J, Wilcox P D, Drinkwater B W and Nagy P B2009 The use of non-collinear mixing for nonlinearultrasonic detection of plasticity and fatigue J. Acoust. Soc.Am. Electron. 126 EL117–22
[18] Klepa A, Staszewski W J, Jenal R B, Szwedo M and Iwaniec J2011 Nonlinear acoustics for fatigue crack detection—experimental investigations of vibro-acoustic wavemodulations Struct. Health Monit. 11 197–211
[19] Sutin A M and Donskoy D M 1998 Vibro-acoustic modulationnondestructive evaluation technique Proc. SPIE 3397226–37
[20] Didenkulov I N, Sutin A M, Ekmov A E and Kazakov V V1999 Interaction of sound and vibrations in concrete withcracks Proc. 15th AIP Conf. pp 279–82
[21] Parsons Z and Staszewski W J 2006 Nonlinear acoustics withlow-profile piezoceramic excitation for crack detection inmetallic structures Smart Mater. Struct. 15 1110–8
[22] Amerini F and Meo M 2011 Structural health monitoring ofbolted joints using linear and nonlinear acoustic/ultrasoundmethods Struct. Health Monit. 10 659–72
[23] Chrysochoidis N A, Barouni A K and Saravanos D A 2011Delamination detection in composites using wavemodulation spectroscopy with a novel active nonlinearacousto-ultrasonic piezoelectric sensor J. Intel. Mater. Syst.Struct. 22 2193–206
[24] Yoder N C and Adams D E 2010 Vibro-acoustic modulationusing a swept probing signal for robust crack detectionStruct. Health Monit. 9 257–67
[25] Sohn H, Lim H J, DeSimio M P, Brown K and Derriso M 2014Nonlinear ultrasonic wave modulation for online fatiguecrack detection J. Sound Vib. 333 1473–84
[26] Sohn H 2007 Effects of environmental and operationalvariability on structural health monitoring Phil. Trans. R.Soc. A 365 539–60
[27] Kess H R and Adams D E 2007 Investigation of operationaland environmental variability effects on damage detectionalgorithms in a woven composite plate Mech. Syst. SignalProcess. 21 2394–405
[28] Lim H J, Sohn H, DeSimio M P and Brown K 2014 Reference-free fatigue crack detection using nonlinear ultrasonicmodulation under various temperature and loadingconditions Mech. Syst. Signal Process. 45 468–78
[29] Hong M, Su Z, Wang Q, Cheng L and Qing X 2014 Modelingnonlinearities of ultrasonic waves for fatigue damagecharacterization: theory, simulation and experimentalvalidation Ultrasonics 54 770–8
[30] Donskoy D, Sutin A and Ekimov A 2001 Nonlinear acousticinteraction on contact interfaces and its use fornondestructive testing NDT E Int. 34 231–8
[31] Zaitsev V Y, Matveev L A and Matveyev A L 2009 On theultimate sensitivity of nonlinear-modulation method of crackdetection NDT E Int. 42 622–9
[32] Derriso M M, Little J E II, Vehorn K A, Davies M J andDeSimio M P 2011 Crack detection using combinations ofacoustic emission and guided wave signals from bondedpiezoelectric transducers Proc. 8th IWSHM pp 1986–93
[33] Cantrell J H 2004 Substructural organization, dislocationplasticity and harmonic generation in cyclically stressedwavy slip metals Proc. R. Soc. A 460 757–80
[34] Solodov I and Busse G 2011 New opportunities for NDT usingnon-linear interaction of elastic waves with defects J. Mech.Eng. 57 169–82
[35] Su Z, Zhou C, Hong M, Cheng L, Wang Q and Qing X 2014Acousto-ultrasonic-based fatigue damage characterization:linear versus nonlinear signal features Mech. Syst. SignalProcess. 45 225–39
[36] Yeum C M, Sohn H and Ihn J B 2011 Lamb wave modedecomposition using concentric ring and circular PZTtransducers Wave Motion 48 358–70
[37] Oh C K, Sohn H and Bae I–H 2009 Statistical noveltydetection within the Yeongjong suspension bridge underenvironmental and operational variations Smart Mater.Struct. 18 125022
[38] Ministry of Land, Infrastructure and Transport (MOLIT) ofKorea government 2015 Special act for the safety control ofpublic facilities and multi-use facilities 12981
14
Smart Mater. Struct. 25 (2016) 095055 H J Lim et al