No. of Pages 11, Model 5G

2 December 2014 Ultrasonics xxx (2014) xxx–xxx 1

Contents lists available at ScienceDirect

Ultrasonics journal homepage: www.elsevier.com/locate/ultras 5 6

Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves

3 4 7

Q1

Alireza Farhidzadeh, Salvatore Salamone ⇑

8

Q2

Smart Structures Research Laboratory, Department of Civil, Structural, and Environmental Engineering, University at Buffalo, Buffalo, NY 14228, United States

9 10

a r t i c l e

1 2 2 5 13 14 15 16 17

i n f o

Article history: Received 3 July 2014 Received in revised form 29 September 2014 Accepted 21 November 2014 Available online xxxx

18 19 20 21 22 23 24

Keywords: Corrosion Steel strand Guided ultrasonic wave Wavelet transform Uncertainty analysis

a b s t r a c t This study presents a nondestructive evaluation method based on guided ultrasonic waves (GUW) to quantify corrosion damage of prestressing steel strands. Speciﬁcally, a reference-free algorithm is proposed to estimate the strand’s cross-section loss by using dispersion curves, continuous wavelet transform, and wave velocity measurements. Accelerated corrosion tests are carried out to validate the proposed approach. Furthermore, the propagation of Heisenberg uncertainty to diameter measurement is also investigated. The method can reasonably estimate the wires’ diameter without any baseline as a reference. Ó 2014 Published by Elsevier B.V.

26 27 28 29 30 31 32 33 34

35 36 37

1. Introduction

38

Multiwire steel strands are widely used in civil structures such as cable-stayed bridges, prestressed concrete structures, and re-centering systems. Despite this increase in usage, the corrosion of the multiwire strands has become a concern for designers, owners and regulators. Many of these structures have suffered the failure of strands due to corrosion [1–3]. Extensive inspection and maintenance/repair programs have been established, with attendant direct costs and signiﬁcant indirect costs due to business interruption [4]. Evaluation of strands is technically challenging. In many structures, inaccessibility of steel strands, eventuate in difﬁcult, expensive and often inconclusive evaluation. Several nondestructive evaluation (NDE) techniques for evaluating the condition of strands have been developed to address these issues in the past few years [5]. Half-Cell potential [6], time domain reﬂectometry (TDR) [7], linear polarization resistance (LPR) sensors [8], magnetic ﬂux [9], and acoustic emission [10] are some of the most commonly used NDE methods for corrosion diagnosis. Although these techniques have shown promise, very few if any are capable to quantify the cross-sectional loss. A technique that shows potential to quantify the extent of corrosion (e.g., cross-sectional loss) is based on guided ultrasonic waves (GUWs). As opposed to the waves used in traditional

39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

Q3

⇑ Corresponding author. E-mail addresses: [email protected] (A. Farhidzadeh), [email protected] (S. Salamone).

impact-echo (IE), that propagate in 3-D within the structure, GUWs propagate along the strand itself by exploiting its waveguide geometry [11]. Most previous researches attempted to use GUWs to detect and assess corrosion damage on reinforcing bars in reinforced concrete (RC) structures [12–15] and steel strands in prestressed concrete structures [4,16–20]. For instance, to diagnose corrosion in reinforcing bars, energy and attenuation characteristics of longitudinal and ﬂexural GUW modes were used [13,14]. Other methods based on time of ﬂight (ToF) of the ﬁrst packets were also proposed to investigate various levels of corrosion in reinforcing bars [21]. Longitudinal GUWs were also used to monitor pitting and delamination in steel rebars [15]. To monitor the corrosion process in post-tensioned concrete beams, fractal analysis of GUWs was investigated [4]. In this study, a reference-free algorithm is proposed to quantify the extent of corrosion through estimating the cross-section loss using GUW measurements. An experimental setup was designed to carry out an accelerated corrosion test on a loaded strand. The diameter of the strand’s wires was estimated using the continuous wavelet transform (CWT) [22] of the GUWs. Furthermore, the uncertainty associated with estimated diameter, which originates from the Heisenberg principle in CWT [23], was quantiﬁed. In the next section, the behavior of GUW in rods and the effect of crosssection loss on dispersion curves, followed by a short introduction on CWT are presented. The details of experimental setup are then explained. The results of diameter measurement using GUW are discussed and ﬁnally the conclusion is given.

http://dx.doi.org/10.1016/j.ultras.2014.11.011 0041-624X/Ó 2014 Published by Elsevier B.V.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 2

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx

86

2. Theory

5000

87

2.1. Guided ultrasonic waves in rods

4500

88

106

A guided ultrasonic wave (GUW) is generated whenever an ultrasound propagates into a bounded medium [24]. GUWs are known to be multimode (many vibrating modes can propagate simultaneously) and dispersive (the propagation velocity depends on the wave frequency f ). In cylindrical waveguides, such as rods, three different modes can propagate: longitudinal, ﬂexural, and torsional [24,25]. The dispersive behavior of these modes is represented by the dispersion curves like the ones shown in Fig. 1. These curves describe the relationship between wave velocity and frequency, and can be calculated analytically or they can be computed by approximate solutions derived from numerical methods [24]. In this work a MATLAB open source toolbox, PCdisp (PochhammerChree dispersion) [26,27] was used to generate the dispersion curves (Fig. 1). The longitudinal modes have received signiﬁcant interest in the past few years, for the nondestructive evaluation of cylindrical waveguides, mostly because the ﬂexural and torsional modes experience high attenuation during the propagation phenomena [16,18,28–30].

107

2.2. Effect of corrosion on GUWs

108

Corrosion is usually an electrochemical oxidation–reduction process that converts the major component of steel, iron (Fe), to ferrous hydroxide (Fe(OH)2). A consequence of this process is the reduction in member cross sectional area, which eventually may affects its structural performance. To simulate the effect of corrosion and investigate how the reduction in a rod’s cross sectional area affects the guided wave propagation, the dispersion curves for the ﬁrst longitudinal mode L(0,1), were generated for various diameters, ranging from 5 mm to 2.5 mm. The results are shown in Fig. 2. It can be observed that in a frequency range between 0 and 600 kHz the group velocity increases as the cross-sectional area decreases. Moreover, certain frequencies (e.g., 500 kHz) provide a larger sensitivity to diameter changes than other frequencies (e.g., 300 kHz). It is worth to mention that guided waves thickness measurement methodologies based on group velocity changes have been studied by some researchers for other applications such as plate-like structures [31–35].

93 94 95 96 97 98 99 100 101 102 103 104 105

109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124

Group velocity [m/sec]

92

4000

3500

3000 d=5.0mm d=4.5mm d=4.0mm d=3.5mm d=3.0mm d=2.5mm

2500

2000

1500

0

100

200

300

400

500

600

700

800

900

1000

Frequency [kHz] Fig. 2. Dispersion curves for the ﬁrst longitudinal mode L(0,1) in steel rods with various diameters.

Fig. 3 shows the group velocity of the L(0,1) mode as a function of the frequency-diameter product (i.e., v = F(fd)). Therefore, the inverse problem, which is to calculate the diameter given the dispersion curves, can be solved by measuring the velocity of a certain frequency component in the signal. In order to limit the effects of the high order modes the excitation frequency was selected in a frequency range below the ﬁrst cut-off frequency. In this work, a time–frequency analysis (continuous wavelet transform) was used to measure the group velocity of the L(0,1) mode.

125

2.3. Continuous wavelet transform

134

Continuous wavelet transform (CWT) for analyzing non-stationary signals has received signiﬁcant interest in the last few years, due to its ability to extract signal time and frequency information simultaneously [23]. The CWT of a time domain signal f(t) is given by Ref. [23],

135

1 tb dt WTðs; bÞ ¼ pﬃﬃﬃﬃﬃ f ðtÞw s jsj 1

140

Z

ð2:1Þ

6000

6000

5000

L(0,1)

5000

L(0,2) 4000 T(0,1) 3000 F(1,3) F(1,2) 2000

F(1,1)

at constant frequency: d↓ → v↑

4000

(470kHz-5.002mm,3440)

3000

2000

1000

1000 0

0 0

100

200

300

400

500

600

700

800

900

1000

0

500

1000

1500

2000

2500

3000

Frequency-diameter [kHz-mm]

Frequency [kHz] Fig. 1. Group velocity dispersion curves for a 5 mm diameter steel rod.

126 127 128 129 130 131 132 133

136 137 138 139

1

where w⁄(t) denotes the complex conjugate of the mother wavelet w(t), s is the dilation parameter (scale), and b is the translation

Group Velocity [m/sec]

90 91

Group speed [m/s]

89

Fig. 3. Dispersion curve for the ﬁrst longitudinal mode L(0,1) in a steel rod versus product of frequency and diameter.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

142 143

144

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx Table 1 Geometrical and mechanical characteristics of the tested strand. Geometrical characteristics

145 146

147

Mechanical characteristics 5.2 5.0

Young’s modulus E (GPa) Poisson’s ratio (v)

196 0.29

15.2 230

203 1860

Lay angle b (°)

7.9

Yielding load (kN) Ultimate tensile strength (MPa) Linear weight (kg/m)

1.10

parameter (location). The energy density of a signal is represented by the Power Spectral Density (PSD) as [23],

PSD ¼ 10 logðjWTðs; bÞj2 Þ

150

where

151

Gaussian [36]. In this study, a complex Morlet was used as mother wavelet [23], deﬁned as:

Core wire diameter dc (mm) Helical wire diameter dh (mm) Strand diameter D (mm) Pitch of helical wire p (mm)

149

rt rf

ð2:3Þ

f ¼ psﬃﬃﬃﬃﬃ s fb

154

The methods based on wavelet transform have an inherent source of uncertainty that originates from the Heisenberg principle [23]. Based on this principle, the resolution in time and frequency domains is limited according to the following relationship [23]:

157

158 160 161 162 163

r2t r2f

1 P 4

ð2:5Þ

where fb is the bandwidth parameter and fc is the wavelet center frequency. Morlet wavelet is a sine wave multiplied by a Gaussian Window. The Morlet wavelet provides a desirable compromise between time and frequency resolution [22]. The complex Morlet parameters were set by trial and error and ﬁxed (fc = 5 Hz, fb = 2 Hz). For a given signal with sampling frequency fs, the complex Morlet transform yields the following equations for rt and rf [23]:

ð2:2Þ

jWTðs; bÞj2 ¼ WTðs; bÞWT ðs; bÞ

156

2 1 t wðtÞ ¼ pﬃﬃﬃﬃﬃﬃﬃﬃ expð2ipf c tÞ exp fb pf b

pﬃﬃﬃﬃﬃ s fb ¼ 2f s

153

155

3

ð2:6Þ

164 165

166 168 169 170 171 172 173 174 175

176 178 179

ð2:7Þ

in which the dilation parameter (i.e., s) is inversely proportional to the local frequency f according to the following equation:

f f s ¼ s c f

ð2:8Þ

181 182 183

184 186

ð2:4Þ

where r2t and r2f are time and frequency variances, respectively. These quantities represent the local resolution of CWT. This inequality reaches its minimum value of 1/4 when the signal is

3. Experiments

187

An experimental setup was designed to validate the proposed approach. Speciﬁcally, a loading apparatus was designed and built

188

Fig. 4. The loading device designed for the accelerated corrosion test on a steel strand (a) plan view, (b) cross section and side views.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

189

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 4

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx

Fig. 5. Experimental setup overview.

190 191

to apply a tensile load to a seven-wire strand. Then, accelerated corrosion tests were carried out on the axially loaded strand.

192

3.1. Strand characteristics

193

204

In this study, a 15.2 mm Grade 270 steel tendon consisting of six 5.0 mm diameter wires spirally wrapped around a central 5.2 mm diameter wire was considered (see Table 1 for details). The length of each helical wire was assumed equal to the length of the central wire since the lay angle b was very small (the difference was as small as 0.89%) [37,38]. In addition, considering that the pitch length (230 mm) was much longer than the typical wavelength in the signals (smaller than 10 mm [37]), each wire was treated as an individual rod [16]. Beard et al. [16] observed that guided waves in individual wires of a strand behave similar to single wires with minor discrepancies associated to inter-wire contact.

205

3.2. Loading apparatus

206

Drawings of the loading apparatus are shown in Fig. 4. It consists of two I-shape rigid beams (web: 76.2 cm 7.6 cm 1.9 cm (3000 300 3/400 ), ﬂanges: 50.8 cm 1.3 cm (2000 400 1/200 )) con00 nected with two u 118 all thread steel bars. In the middle of each 00 beam’s web, a 5/8 hole was drilled; the strand passed through the holes and was ﬁxed using wedge anchorages. The load was applied by tightening the nuts, which pushes the rigid beams and stretches the strand; a large wrench with handle extension was used for this purpose. Therefore, the all thread bars are under compressive force. To prevent bucking, two wood beams were connected to the bars using u-shaped hooks (see Fig. 5). Two hollow

195 196 197 198 199 200 201 202 203

207 208 209 210 211 212 213 214 215 216

217

3.3. Accelerated corrosion test

223

An accelerated corrosion test [39] was carried out using the impressed current technique. Speciﬁcally, the strand was immersed in 3.5% sodium chloride (NaCl) solution. A power supply equipped with built-in ammeter and potentiometer was used to impress a direct current (DC) to the strand to induce signiﬁcant corrosion in a short period of time. The direction of the current was adjusted so that the steel strand served as the anode, while another metal, superior than steel in electro-chemical series, served as the cathode [40] (see Fig. 4). A constant voltage of 0.16 V was maintained across the solution and the specimen until the 18th day, and then was increased to 0.32 V to failure. Fig. 5 shows an overview of the experimental setup that consists of loading apparatus, water tank, power supply, and the steel strand. To monitor the corrosion process, a video camera with recording rate of 1 frame per minute was mounted on top of the specimen.

224

3.4. Mass loss measurement

239

The mass loss during the corrosion tests was determined through a Gravimetric analysis in which the iron oxide or rust (Fe2O3) was the analyte. The ferrous ions (Fe2+) precipitate through combination with hydroxyl ions (OH) and form ferrous hydroxide (Fe(OH)2) that are ﬁne solid particles. These particles were collected by ﬁltration, dried, and weighted. When dried, rust (Fe2O3) is produced. The weights of ﬁlters were subtracted to obtain the weight of rust. Finally, to calculate the mass loss of iron, the measured weight was multiplied by a correction factor of 0.7. The correction factor is derived from the mass portion of iron in rust (i.e., (2 56)/(2 56 + 3 16) = 0.7). This process is schematically illustrated in Fig. 6.

240

3.5. Ultrasonic test equipment

252

Ultrasonic tests were performed using broadband piezoelectric transducers in a through-transmission (i.e., pitch-catch) conﬁgura-

253

Fig. 6. The instruction of mass loss measurement.

Low Frequency transmission

194

load cells passed through the all thread bars and were located between the beam’s web and the nuts’ washers. The tensile load on the strands was continuously measured during the corrosion test as the summation of readings from these two load cells. The strand was initially loaded up to 89 kN (=20 kips 40% of yielding force) which was the designed capacity of the loading apparatus.

PICO sensor

Piezo Linear Ampliﬁer

High Frequency Olympus 5077PR Square pulser

NI PXI-5412 AWG

V1091 transducers

transmission

LabVIEW

PICO sensor

Ampliﬁer

PXI-5105 digizer

Oscilloscope Lecroy wave runner 44Xi-A

Inbuilt ampliﬁcaon

V1091 transducers

Fig. 7. Through-transmission testing for low- and high-frequency longitudinal waves.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

218 219 220 221 222

225 226 227 228 229 230 231 232 233 234 235 236 237 238

241 242 243 244 245 246 247 248 249 250 251

254

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 5

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx

S1 and S2: PICO sensor S3 and S4: 5MHz sensor 3.5 cycles

PICO sensor

S3

S2 S4

S1

Magnet holder 5MHz sensor

(a)

(b)

Fig. 8. Instrumenting a multi wire steel strand with transducers attached to the cross-section of helical wires (a) sensor layout, and (b) sensor coupling.

25 Faraday's law Mass measurment

Mass loss [%]

20 15 10 5 0 0

2

4

6

8

10 12 14 16 18 20 22

Time [day]

transducers (Physical Acoustic Corporation PICO), which are sensitive in a frequency range between 200 kHz and 700 kHz, were used to generate and receive ultrasonic waves. LabVIEW software was used to control the sensors, and acquire the data. For the high frequency modes a commercial pulser–receiver (Olympus 5077PR) was used to drive two piezoelectric transducers with a central frequency of 5 MHz (Olympus V1091) with a square wave pulse of selectable width. The received signals were recorded using an oscilloscope (Lecroy Wave Runner 44Xi-A). In both experimental setups, transducers were carefully mounted as perpendicular to the wire’s cross section as possible to measure the longitudinal wave motions. The diameter of the transducers was about the same as the diameter of the single wire so that there was no contact between the transducers and the adjacent wires, as shown in Fig. 8.

267

4. Results

282

Faraday’s law was used to predict the strand mass loss. Faraday’s law provides a relationship between the time of an applied current and the amount of steel weight loss according to the following equation:

283

268 269 270 271 272 273 274 275 276 277 278 279 280 281

Fig. 9. Mass loss versus time during accelerated corrosion test.

100 90 80

Load [kN]

70

A m¼ ZF

60 50

20

1st wire failure

2nd

and 3rd

wires failures

10 0

1 23 4 5 6

7

10

12

14

18

20

22

Time [day] Fig. 10. Load versus time during accelerated corrosion test.

255 256 257 258 259 260 261 262 263 264 265 266

IðtÞdt

ð4:1Þ

where m is the mass in grams, A is the atomic mass of iron (56 g), Z is the valence of the reacting electrode (2 for iron), F is the Faraday’s constant (96,500 A s), t is time in seconds, and I(t) is the current in ampere. Therefore the theoretical percentage of mass loss, Dm, was calculated as:

40 30

Z

tion. A schematic drawing of the experimental setup is shown in Fig. 7. Two through-transmission setup were used to investigate the low frequency modes (below the ﬁrst cut-off frequency), and the higher frequency modes (above the ﬁrst cut-off frequency). For the low frequency modes the measurement equipment used for signal generation and data acquisition was mainly constituted by a National Instruments (NI), modular PXI 1042 unit. The unit included an arbitrary waveform generator card (PXI 5412) and one, 20GS/s 12-bit multi-channel digitizers (PXI 5105). A high voltage ampliﬁer was used to amplify the excitation to the ultrasonic transmitters, while a preampliﬁer (Olympus 5660C) was used to amplify the ultrasonic receivers. Two ultra-mini broadband

m Dm ¼ 100% M

ð4:2Þ

where M is the total mass of the strand. Fig. 9 shows the measured results based on gravimetric mass loss and the corresponding theoretical mass loss based on Faraday’s law. Faraday’s law predicted well the degrees of corrosion (mass loss) with the impressed current technique. Fig. 10 shows the strand axial load measurements through the load cells during the accelerated corrosion test. The failure of three wires was recorded at day 21st. It is noteworthy that load drop after the failure of the 2nd and 3rd wire was almost twice the load drop after the 1st wire breakage. Therefore the load was equally distributed on each wire. The inﬂuence of the impressed current on the load drop rate can be observed at the 18th day, when impressed current was increased from 0.5 A to 1.5 A. Fig. 11 illustrates the strand condition at different levels of corrosion. The measured mass loss is also reported for each level of corrosion. The location of the wire breakage is indicated with arrows.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

284 285 286

287 289 290 291 292 293 294

295 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 6

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx

No corrosion (Mass loss = 0.0%)

Light corrosion (Mass loss = 0.19%)

Pitting (Mass loss = 1.26%)

Heavy Pitting (Mass loss = 2.72%)

Cross section loss (Mass loss = 8.38%)

Fracture (Mass loss = 21.34%)

Fig. 11. Visual inspection of corrosion progress in a steel strand.

315

4.1. Attenuation measurements

316

Attenuation coefﬁcients were calculated using the following equation:

317

318 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341

ai ½dB=m ¼

20 A0 log L Ai

ð4:3Þ

where ai is the attenuation coefﬁcient in dB/m, L is the length of strand, A0 is the amplitude of the signal at pristine state, and Ai is the amplitude at ith day. Two PICO sensors in a through-transmission conﬁguration were used to generate and receive GUWs on the same peripheral helical wire. A sweep frequency test between 300 kHz and 700 kHz was performed at different levels of corrosion. Fig. 12a shows the reduction in amplitude of the signals versus mass loss and excitation frequency. Fig. 12b illustrates the attenuation coefﬁcients. At about 20% mass loss, the attenuation coefﬁcient increased to 30 dB/m. It is interesting to note that the attenuation coefﬁcient is quite constant for all frequencies below 15% mass loss. To investigate the attenuation of the higher longitudinal modes during the corrosion process, two piezoelectric broadband transducers with a central frequency of 5 MHz were used in a through-transmission-conﬁguration, as described in Section 3.4. Fig. 13 shows the CWT of the received signals in day 1, 3, 4, and 5. For the ﬁrst day, the CWT was estimated and compared with numerical results of dispersion curves obtained from the PCdisp toolbox. There is a clear correspondence between theory (solid lines) and experimental results (CWT). After ﬁve days (less than 3% mass loss) only the L(0,1) mode was left.

4.2. Reference-free diameter estimation

342

A 3.5-cycle, Hanning-modulated toneburst was used as the actuating signal. For the selection of the excitation frequency a swept of frequencies were performed between 300 kHz and 700 kHz. The dominant L(0,1) mode was received and analyzed. Fig. 14 shows the signal strengths (RMS) during the corrosion test. The highest RMS was obtained at 440 kHz. Therefore, the frequency of 440 kHz was selected as excitation frequency. A typical received signal and corresponding CWT of the signal before the corrosion test (i.e., pristine condition) is shown in Fig. 15. CWT was compared with the dispersion curves resulting from the numerical simulation described in Section 2.1. It is worth noting a clear correspondence between theory (dashed line) and experimental results (CWT map), which indicate that the ﬁrst longitudinal mode is the predominant mode of displacement in the received signal. To visualize the Heisenberg’s uncertainty principle discussed in Section 2.3, the time and frequency standard deviations (i.e., rt and rf in Eqs. (2.6) and (2.7)) are also indicated on the Heisenberg box [23]. To estimate the wire diameter at the pristine state (i.e., 5 mm), the maximum value of CWT, called the peak ridge, was identiﬁed. The projection of the peak ridge on frequency and time axes corresponds to the dominant frequency f and timeof-ﬂight (ToF) t, respectively. At the pristine state, the peak ridge was at t ¼ 357 ls and f ¼ 470 kHz (see Fig. 15). Given the strand length of 1.23 m, the velocity of 3440 m/s was calculated. Therefore, the dispersion curves shown in Fig. 3 were used to estimate the diameter (i.e., 5.002 mm). There is a clear correspondence between the actual value (5 mm) and the estimated value (5.002 mm). Fig. 16 depicts the time waveforms during the corrosion test at days 3, 6, 10, and 18. As expected, the energy was signiﬁcantly attenuated due to the corrosion process as well as the inter-wire contacts. However, the wave velocities of the ﬁrst wave packets, did not shows signiﬁcantly changes (see the left zoomed-in outset plot) while the time shift was evident on the high energy part of the waveforms (see the right zoomed-in outset plot). This shift is mainly caused by the diameter reduction and its impact on wave velocity at certain frequencies. Fig. 17 shows a typical waveform at day 6th along with its corresponding CWT scalogram. The ﬁrst packet corresponds to an approximate frequency range of 320– 360 kHz at which the velocity is slightly sensitive to diameter (see Fig. 2). On the contrary, the high energy packet corresponds to a higher frequency component (i.e., 470 kHz). This part of the signal is used for diameter estimation. The reference-free algorithm was repeated every day to estimate the average cross-section loss during the corrosion process. To take into account the fact that, the experimental setup was designed to corrode just a portion of strand, some modiﬁcations were needed to ﬁnd the average velocity in the corroded part. At this aims the strand was divided in three regions: (1) left end side of the salt water tank with length l1, (2) inside the salt water tank with length l2, (3) right end side of the salt water tank with length l3. The middle part (i.e., inside the salt water tank) was the corroding part of the strand, while the other two regions with length l1 and l3 remained pristine during the test. Therefore the arrival time t was estimated as:

343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397

398 3 X t ¼ t i ¼ l1 =v 1 þ l2 =v 2 þ l3 =v 1

ð4:4Þ

i¼1

where t i is the ToF at each region, and l1, l2, and l3, are constant (i.e., l1 = l3 = 31 cm, l2 = 61 cm). The wave velocity v 1 considered for the pristine parts was obtained from the dispersion curves (Fig. 3) at d = 5 mm. Since the corrosion was not perfectly uniform, we estimated an average velocity v 2 in the middle part of the strand as:

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

400 401 402 403 404 405

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 7

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx

40 30

0.15 0.1 0.05 0

0 800

5 700

Frequency [kHz]

500

15

400 300

20 10 0 -10 300

10

600

Attenuation [db/m]

Amplitude [v]

0.2

20 15

400 500

Mass loss [%]

20

10

600

Frequency [kHz]

(a)

5

700 800

0

Mass loss [%]

(b)

Fig. 12. Effect of corrosion on (a) signal amplitude, and (b) attenuation coefﬁcient.

Fig. 13. Effect of corrosion on higher longitudinal modes.

406

408 409 410 411 412 413

l2 t 2l1 =v 1

v 2 ¼

ð4:5Þ

2 and frequency f , the average diameter of Having the velocity v a wire in the corroding part was estimated from the L(0,1) mode dispersion curve in Fig. 3. This procedure was repeated at different stages of corrosion and the results are presented in Fig. 18. This ﬁgure shows a reasonable correspondence between the estimated

and measured diameter. The estimation error increases after day 14th when signiﬁcant mass loss was observed. The larger error (i.e., 10%) was recorded at the 20th day.

414

5. Uncertainty quantiﬁcation

417

The most critical challenge here is to provide a quantitative assessment of how closely our estimates (i.e., diameter) reﬂect

418

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

415 416

419

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 8

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx 0.04 Day Day Day Day Day Day Day

0.035 0.03

RMS [V]

0.025

0 3 7 10 14 18 20

0.02 0.015 0.01 0.005 0 300

350

400

450

500

550

600

650

700

Frequency [kHz] Fig. 14. RMS of the received waveforms in sweep frequency test during the accelerated corrosion test.

Fig. 16. Ultrasonic signals at different corrosion states; the velocity does not change considerably for the ﬁrst packets (bottom left) while it clearly changes at the high energy part of the signal (bottom right).

Q8

Fig. 15. (a) Time-domain signal and (b) spectrogram of signal taken from a 1.23 length rod (5 mm diameter) through-transmission. The dispersion curve for L(0,1), time and frequency of the ridge as well as Heisenberg uncertainties are superimposed. Fig. 17. Guided wave signal during corrosion at day 6 (top) and the corresponding CWT scalogram (bottom).

420 421 422 423 424 425 426 427 428 429 430 431 432 433

reality in the presence of measurement uncertainty. In general, uncertainties can be caused by random and systematic errors. The random errors are caused by unknown and unpredictable changes in measurements (e.g., ToF), including instrumentation noise, temperature changes, etc. [41]. Systematic errors are mostly caused by the digital signal processing technique (e.g., CWT) used for analyzing the time waveforms [41]. This section aims at quantifying the systematic uncertainty, caused by the time–frequency analysis (i.e., CWT), which arises from the Heisenberg principle, as described in Section 2.3. Thus, both ToF (t) and frequency (f) were treated as Gaussian random variables with variance r2t and r2f deﬁned according to Eqs. (2.6) and (2.7), respectively. Also the group velocity (v) was treated as a Gaussian random variable with and variance r2v deﬁned as (see Appendix A): mean v

434

v ¼ tl 2

r2v ¼ tl4 r2t

ð5:1Þ

The two random variables f and v are related by the Pochhammer-Chree equation v ¼ FðfdÞ (i.e., dispersion curves). Therefore, given the pdf of f and v, the systematic uncertainty associated to the estimated diameter was evaluated by determining its pdf. Several techniques exist in the literature to approximate the pdf of random variables [42], including Monte Carlo (MC) methods [43], Unscented Transformation (UT) [44], and Delta method [45]. Monte Carlo method, which is the most popular, require extensive computational resources and effort, and become increasingly

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

436 437 438 439 440 441 442 443 444 445

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 9

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx 0.7

5.2

Estimated 5

Measured

Histogram Monte Carlo Unscented transform Delta method

0.6

0.5 4.6 0.4

4.4

Diameter [mm]

4.8

4.2

0.3 4 0.2 3.8 0

2

4

6

8

10

12

14

16

18

20

Time [day]

0.1

Fig. 18. Diameter of a helical wire estimated using the reference-free approach and measured through visual inspection during accelerated corrosion test.

0

2

3

4

5

6

7

8

9

10

Diameter [mm] Fig. 20. Probability density functions of diameter at the 10th day with different approaches.

Measured Estimated

5.5

Diameter [mm]

5

4.5

4

Fig. 19. Heisenberg uncertainty propagation through the nonlinear dispersion function of L(0,1) analyzed by Monte Carlo (MC), Unscented Transform (UT) and Delta methods.

3.5

0

2

4

6

8

10

12

14

16

18

20

22

Time [day] 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466

infeasible for portable NDT devices with small processing units. The UT and Delta method, are both used to estimate the statistics of a random variable which undergoes a nonlinear transformation as in F 1 ðv Þ [44]. In the UT method, instead of randomly selecting multiple data points from a pdf, a few data points (called sigma points) are intelligently selected. The Delta method is motivated by the assumption that in this study the dispersion curve can be approximated by a linear function in the interested velocity range. More details on the application of these techniques (MC, UT and Delta method) are provided in Appendix A. Fig. 19 illustrates the application of the aforementioned methods for uncertainty propagation analysis. It focuses the dispersion . The curve (shown by dashed-dot line) at the velocity mean v velocity pdf is shown by a thick solid line. The sampled points from the velocity pdf are shown by narrow gray lines, representing the MC simulation. These points are projected to fd axis through the dispersion curve (i.e., ðfdÞi ¼ F 1 ðv i Þ). The UT sigma points are also Q4 superimposed using three red1 dotted lines, signifying the signiﬁcant reduction in required sampling points. The Delta method is represented by a dashed tangent line on the dispersion curve at v ¼ v . These methods yield the pdf for product of frequency and diameter

1

For interpretation of color in Fig. 19, the reader is referred to the web version of this article.

Fig. 21. Estimated diameters and their associated uncertainty.

(i.e., fd Nðfd; r2fd Þ) that is shown by a black narrow line. Given the parameters of fd pdf, the diameter’s mean and variance are calculated via the following equations:

467 468 469

470

¼ fd=f d

r2d ¼

r2fd d2 r2f f 2 þr2 f

ð5:2Þ

Fig. 20 shows the estimated pdf of diameter using the aforementioned methods at the 10th day as an example. The histogram and pdf obtained using MC method are shown by vertical bars and a dashed line. The solid line represents the pdf of the diameter using UT, and the dotted line is that of Delta method. There is a clear correspondence between the pdf obtained with all three methods. However, UT and Delta method provides signiﬁcant computational cost reduction compared to Monte Carlo method. Finally Fig. 21 illustrates the estimated diameter’s mean and standard deviation as well as the measured diameter during the corrosion test. It is remarkable to note that the difference between measured and estimated diameter is always less than one standard deviation. The standard deviations are approximately 0.7 ± 0.1 mm. The total diameter reduction is less than 1.5 mm and this evidence the high resolution of the proposed velocity-based approach to estimate the diameter.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 10

A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx

489

6. Conclusions

490

This study aims at designing a reference-free nondestructive evaluation technique for corrosion damage estimation in prestressed seven wires steel strands using guided ultrasonic waves. A small-scale experiment was designed to investigate corrosion process and behavior of ultrasonic waves in under accelerated corrosion. Axial load and mass loss were measured during the test. Attenuation coefﬁcients were monitored and reached to 35 dB at critical stage of corrosion. A new method based on the velocity of certain velocity components was proposed and validated to estimate the cross-section loss. Continuous wavelet transform was used to localize the highest energy content in time–frequency domain. Having the frequency and ToF (that yields the velocity), the diameter was estimated using the numerically derived dispersion function for the ﬁrst longitudinal mode. This function correlates velocity to multiplication of frequency and diameter. In addition, an uncertainty analysis was carried out to study propagation of time and frequency uncertainties from Heisenberg principle to the diameter calculation. The proposed algorithm does not need a baseline as a reference and directly measure the diameter via through-transmission conﬁguration in ultrasonic testing. The algorithm was validated experimentally and the results conformed very well to the diameter measurements during visual inspection.

491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511

Acknowledgments

513

Funding provided by the Research Foundation of SUNY through the research collaboration fund. Any opinions, ﬁndings, conclusions or recommendations expressed are those of the author(s) and do not necessarily reﬂect the views of the RF/SUNY.

515 516

517

Appendix A

518

A.1. Velocity pdf

519

In general, the mean and variance of the reciprocal of a random variable can be obtained using its equivalent Taylor series at mean value as follows: denoting the expected value (mean) of a random variable t as E½t ¼ t and deﬁning a random variable v as a function of t as v ðtÞ ¼ tl , we have:

520 521 522 523 524

l Taylor series : v ðt ¼ tÞ ¼ t t¼t

2 1 t t ðt tÞ ¼ l 2 þ 3 þ . . . t t t

526 527

528 530

531 532 533

534

Thus the expected value of

! ðA:1Þ

ðA:2Þ

In our case, since the variance in time, r is very small in comparison to the measured time t, we can eliminate the second term in Eq. (A.2) and simply deﬁne the velocity mean as: 2 t;

l t

537

and the velocity variance as,

ðA:3Þ

542 543

The Unscented Transform is a method for estimating the statistics of a random variable which undergoes a nonlinear transformation as in F 1 ðv Þ [44]. Assuming the velocity as a random variable v and variance r2v ; only three weighted sigma with the mean v points v0, v1, and v2 are needed to calculate the statistics of v. These points are given by Ref. [44]:

560

v0 ¼ v ; W0 ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 v1 ¼ v þ ð1 þ jÞrv ; W 1 ¼ 2ð1þ1 jÞ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ v2 ¼ v ð1 þ jÞr2v ; W 1 ¼ 2ð1þ1 jÞ

545 546 547 548 549 550 551 552 553 554 555 556 557 558

561 562 563 564 565

566

ðA:5Þ 568

where Wi is the weight associated with the ith sigma point and j is an arbitrary number providing 1 + j – 0; j = 0 is chosen in this work. Given the set of sigma points calculated by Eq. (A.5), the transformation of velocity pdf to the approximated fd pdf is summarized in the following steps [44]:

569

1. Calculate the fd corresponding to each sigma point using dispersion function:

574

ðfdÞi ¼ F 1 ðvi Þ

ðA:6Þ

ðA:7Þ

i¼0

r2fd ¼

ðA:4Þ

A.2. Monte Carlo simulation The Monte Carlo simulation samples a large number of data from the underlying probability space to generate a family of test

2 W i ðfdÞi fd

571 572 573

575

576 578 581

2 X fd ¼ W i ðfdÞi

2 X

570

579 580

2. The mean of fd is calculated from:

583 584 585

586

ðA:8Þ

i¼0

588 589

Since f and d are statistically independent, the following equation holds for estimating fd

590

fd ¼ E½fd ¼ E½f E½d ¼ f d

594

ðA:9Þ

is calculated as: and thus d

¼ fd=f d

2

l r2v ¼ E½ðv v Þ ¼ 4 r2t t 2

541

559

v can be found as:

1 r2 E½v ¼ l þ 3t þ . . . t t

v ¼ E½v ¼

540

A.3. Unscented Transformation

3. The variance of random variable fd is determined by:

536 538

544

j 1þj

512

514

points random, feeds forward the samples individually through the exact nonlinear function to ﬁnd the output data deterministically, and ﬁnally evaluates the pdf of outputs [43]. In this study, a large number of data points (>10,000) were sampled from the ; r2v Þ; v i ; i ¼ 1; . . . ; n), and were fed into the velocity pdf (v Nðv inverse of dispersion function to ﬁnd (fd)i deterministically (i.e., ðfdÞi ¼ F 1 ðv i Þ). Next, another set of n data points are randomly sampled from the frequency pdf (i.e., f Nðf ; r2f Þ; f i ; i ¼ 1; . . . ; n). Assuming statistical independency between frequency and diameter, the diameter di is found (di ¼ ðfdÞi =f i ). Therefore, the diameter and r2 ; are computed having the set of d ’s. It is pdf parameters, d i d worth to mention that the outliers are eliminated from calculation > 3r and recalcuof mean and variance (i.e., remove di if jdi dj d and r2 ) since the number of sampled points were limited late d d and outliers could cause error.

591

592

595

596

ðA:10Þ

598

one can calculate r2 using the variance relationship Given r2fd and d, d for the product of independent variables (i.e., var½xy ¼ ðE½xÞ2 var½y þðE½yÞ2 var½x þ var½xvar½y) as follows:

599

r2 d2 r2f r2d ¼ fd2 f þ r2f

600 601

602

ðA:11Þ

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

604

ULTRAS 4965

No. of Pages 11, Model 5G

2 December 2014 A. Farhidzadeh, S. Salamone / Ultrasonics xxx (2014) xxx–xxx 605

A.4. Delta method

606

609

Delta method uses the second-order Taylor expansions to ﬁnd the variance of a function of a random variable. Given a random variable x and a function g(x), the variance of this function is obtained as follows [46],

612

var½gðxÞ ðg 0 ðE½xÞÞ var½x

613 615

where g 0 is the derivative of the function g with respect to x. In case of this study, velocity v is the variable and F 1 ðv Þ is the function. Therefore, we have:

618

r2fd ¼ var½F 1 ðv Þ F 10 ðv Þ r2v

607 608

610

614

616

2

ðA:12Þ

2

ðA:13Þ

620

Given r2fd , the variance of diameter r2d is calculated via Eq. (A.11) is given through Eq. (A.10). and d

621

References

622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666

[1] G.M. Clark, Post-tensioned structures – improved standards based on lessons learnt historical background of post-tensioning, Proc. Inst. Civ. Eng. 166 (4) (2013) 171–179. [2] C. Burgoyne, R. Scantlebury, Lessons learned from the bridge collapse in Palau, Proc. ICE – Civ. Eng. 161 (6) (2008) 28–34. [3] R. Woodward, F. Williams, Collapse of Ynys-y-Gwas Bridge, West Glamorgan, Proc. Inst. Civ. Eng. 84 (1998) 1177–1191. [4] A. Moustafa, E. Dehghan Niri, A. Farhidzadeh, S. Salamone, Corrosion monitoring of post-tensioned concrete structures using fractal analysis of guided ultrasonic waves, Struct. Control Health Monit. 21 (3) (2014) 438–448. [5] P. Virmani, Corrosion Monitoring Research of New York City Bridges FHWA Publication No.: FHWA-HRT-14-024 FHWA, Federal Highway Administration, McLean, VA, USA, 2013. [6] H. Song, V. Saraswathy, Corrosion monitoring of reinforced concrete structures – a review, Int. J. Electrochem. 2 (2007) 1–28. [7] W. Liu, R.G. Hunsperger, M.J. Chajes, K.J. Folliard, E. Kunz, Corrosion detection of steel cables using time domain reﬂectometer, J. Mater. Civ. Eng. 14 (3) (2002) 217–223. [8] S.G. Millard, D. Law, J.H. Bungey, J. Cairns, Environmental inﬂuences on linear polarisation corrosion rate measurement in reinforced concrete, NDT E Int. 34 (2001) 409–417. [9] H.A. Kishawy, H.A. Gabbar, Review of pipeline integrity management practices, Int. J. Pressure Vessels Piping 87 (7) (2010) 373–380. [10] J. Mangual, M. Elbatanouny, P. Ziehl, F. Matta, Corrosion damage quantiﬁcation of prestressing strands using acoustic emission, J. Mater. Civ. Eng. 25 (9) (2013) 1326–1334. [11] S. Salamone, I. Bartoli, R. Phillips, C. Nucera, F. Lanza di Scalea, Health monitoring of prestressing tendons in posttensioned concrete bridges, Trans. Res. Rec. J. Trans. Res. Board 2220 (2011) 21–27. [12] B.L. Ervin, D.A. Kuchma, J.T. Bernhard, H. Reis, Monitoring corrosion of rebar embedded in mortar using high-frequency guided ultrasonic waves, J. Eng. Mech. (2009). January. [13] H. Reis, B.L. Ervin, D.A. Kuchma, J.T. Bernhard, Estimation of corrosion damage in steel reinforced mortar using guided waves, J. Pressure Vessel Technol. 127 (3) (2005) 255. [14] B.L. Ervin, H. Reis, Longitudinal guided waves for monitoring corrosion in reinforced mortar, Meas. Sci. Technol. 19 (5) (2008) 055702. [15] S. Sharma, A. Mukherjee, Longitudinal guided waves for monitoring chloride corrosion in reinforcing bars in concrete, Struct. Health Monit. 9 (6) (2010) 555–567. [16] M.D. Beard, M.J.S. Lowe, P. Cawley, Ultrasonic guided waves for inspection of grouted tendons and bolts, J. Mater. Civ. Eng. 15 (3) (2003) 212–218. [17] S. Chaki, G. Bourse, Guided ultrasonic waves for non-destructive monitoring of the stress levels in prestressed steel strands, Ultrasonics 49 (2) (2009) 162– 171.

619

11

[18] F. Lanza, P. Rizzo, F. Seible, M. Asce, Stress measurement and defect detection in steel strands by guided stress waves (2003) 219–227. [19] R. Raišutis, R. Kazˇys, L. Mazˇeika, E. Zˇukauskas, V. Samaitis, A. Jankauskas, Ultrasonic guided wave-based testing technique for inspection of multi-wire rope structures, NDT E Int. 62 (2014) 40–49. [20] P. Rizzo, F.L. Di Scalea, Ultrasonic inspection of multi-wire steel strands with the aid of the wavelet transform, Smart Mater. Struct. 14 (4) (2005) 685–695. [21] T.H. Miller, T. Kundu, J. Huang, J.Y. Grill, A new guided wave-based technique for corrosion monitoring in reinforced concrete, Struct. Health Monit. 12 (1) (2012) 35–47. [22] W.C. Lang, K. Forinash, Time-frequency analysis with the continuous wavelet transform, June 1997 (1998) 794–797. [23] C. Pereyra, Wavelets, their friends, and what they can do for you (2006). [24] J.L. Rose, Ultrasonic waves in solid media (2004). [25] J. Zemanek, An experimental and theoretical investigation of elastic wave propagation in a cylinder, J. Acoust. Soc. Am. 51 (1B) (1971) 265–283. [26] F.S. Granja, PCdisp (2011). [27] F. Seco, J.M. Martín, A. Jiménez, J.L. Pons, PCdisp : a tool for the simulation of wave propagation in cylindrical waveguides, in: Ninth International Congress on Sound and Vibration, vol. 1, 2002, pp. 1–8. [28] C. Nucera, F. Lanza di Scalea, Monitoring load levels in multi-wire strands by nonlinear ultrasonic waves, Struct. Health Monit. 10 (6) (2011) 617–629. [29] S. Salamone, M.J. Veletzos, F. Lanza di Scalea, J.I. Restrepo, Detection of initial yield and onset of failure in bonded post-tensioned concrete beams, J. Bridge Eng. 17(6) 966–974 (Special issue: nondestructive evaluation and testing for bridge inspection and evaluation). [30] P.J. Shull, M. Dekker, Nondestructive Evaluation Theory, Techniques, and Applications, 106, Marcel Dekker Inc., 2002. 10. [31] P. Belanger, Feasibility of Thickness Mapping Using Ultrasonic Guided Waves, Imperial College London, 2009. [32] K.R. Leonard, E.V. Malyarenko, M.K. Hinders, Ultrasonic Lamb wave tomography, Inverse Probl. 18 (6) (2002) 1795–1808. [33] J.C.P. Mckeon, Parallel projection and crosshole Lamb wave contact scanning tomography, Acoust. Soc. Am. 106 (5) (1999) 2568–2577. [34] F. Jenot, O. Uaftouh, M. Duquennoy, M. Ourak, Corrosion thickness gauging in plates using lamb wave group velocity, Meas. Sci. Technol. 12 (2001) 1287–1293. [35] T.R. Hay, R.L. Royer, H. Gao, X. Zhao, J.L. Rose, A comparison of embedded sensor lamb wave ultrasonic tomography approaches for material loss detection, Smart Mater. Struct. 15 (4) (2006) 946–951. [36] P.S. Addison, The Illustrated Wavelet Transform Handbook: Introductory Theory and Applications in Science, Engineering, Medicine and Finance, IOP Publishing Ltd, London UK, 2002. [37] A. Farhidzadeh, Application of Pattern Recognition Algorithms and Nondestructive Evaluation Techniques for the Structural Health Monitoring of Civil Structures, University at Buffalo, 2014. [38] I. Bartoli, Structural Health Monitoring by Ultrasonic Guided Waves, University of California, San Diego, 2007. [39] ASTM Standard F2832-11, Standard Guide for Accelerated Corrosion Testing for Mechanical Fasteners. ASTM International, West Conshohocken, PA, 2011. [40] S. Sharma, A. Mukherjee, Nondestructive evaluation of corrosion in varying environments using guided waves, Res. Nondestruct. Eval. 24 (2) (2013) 63–88. [41] E. Dehghan Niri, A. Farhidzadeh, S. Salamone, Adaptive multisensor data fusion for acoustic emission source localization in noisy environment, Struct. Health Monit. 12 (1) (2013) 59–77. [42] M. Kumar, S. Chakravorty, P. Singla, J.L. Junkins, The partition of unity ﬁnite element approach with hp-reﬁnement for the stationary Fokker-Planck equation, J. Sound Vib. 327 (1–2) (2009) 144–162. [43] D.P. Kroese, T. Taimre, Z.I. Botev, Handbook of Monte Carlo Methods, Wiley, 2011. [44] S.J. Julier, J.K. Uhlmann, Unscented ﬁltering and nonlinear estimation, Proc. IEEE 92 (3) (2004) 401–422. [45] G. Casella, R.L. Berger, Statistical Inference, second ed., BROOKS/COLE CENGAGE Learning, 2002. [46] A.A. Clifford, Multivariate Error Analysis: A Handbook of Error Propagation and Calculation in Many-Parameter Systems, Wiley, 1973.

Please cite this article in press as: A. Farhidzadeh, S. Salamone, Reference-free corrosion damage diagnosis in steel strands using guided ultrasonic waves, Ultrasonics (2014), http://dx.doi.org/10.1016/j.ultras.2014.11.011

Q5

Q6

Q7

667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734