Wavelet transforms and pattern recognition on ultrasonic guides waves for frozen surface state diagnosis

Accepted Manuscript Wavelet transforms and pattern recognition on ultrasonic guides waves for frozen surface state diagnosis Carlos Quiterio Gómez Muñoz, Alfredo Arcos Jiménez, Fausto Pedro García Márquez PII:

S0960-1481(17)30243-4

DOI:

10.1016/j.renene.2017.03.052

Reference:

RENE 8645

To appear in:

Renewable Energy

Received Date: 2 December 2016 Revised Date:

22 February 2017

Accepted Date: 17 March 2017

Please cite this article as: Gómez Muñoz CQ, Jiménez AA, García Márquez FP, Wavelet transforms and pattern recognition on ultrasonic guides waves for frozen surface state diagnosis, Renewable Energy (2017), doi: 10.1016/j.renene.2017.03.052. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

4 5 6 7

Carlos Quiterio Gómez Muñoz, Alfredo Arcos Jiménez, Fausto Pedro García Márquez

36 37 38

SC

Icing blades require of advanced condition monitoring systems to reduce the failures and downtimes in Wind Turbine Blades (WTB). This paper presents a novel fault detection and diagnosis system that combines ultrasonic techniques with Wavelet transforms for detecting ice on the blades. Lamb waves were generated with Macro Fibre Composites (MFC) and collected with MFC. Ice affects to the normal propagation of the wave through the material of the blade. The changes in the signal are due to the forces that ice exercise on the surface. Three different scenarios were considered according to ISO 12494, 2001 (Atmospheric icing of structures): at room temperature; the frozen blade without accumulation of ice, and; the frozen blade with accumulation of ice on its surface. In order to validate the approach, Morlet wavelet transformation has been used for filtering the signal. The time-frequency analysis has been done by Wigner-Ville distribution. On the other hand, the envelope of the filtered signal by wavelet transforms is done by Hilbert Transform, and the pattern recognition is done by autocorrelations of the Hibert transforms. The approach detects the cases considered in ISO 12494 of unfrozen, frozen without ice, and frozen with ice in the WTB. New scenarios, considering mud, have been considered to test the approach.

M AN U

14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

Abstract

TE D

13

Ingenium Research Group, Castilla-La Mancha University, Spain (CarlosQuiterio.Gomez; FaustoPedro.Garcia)@uclm.es [email protected]

EP

8 9 10 11 12

AC C

2

RI PT

3

Wavelet Transforms and Pattern Recognition on Ultrasonic Guides Waves for Frozen Surface State Diagnosis

1

Keyword: Health Monitoring Systems, Wavelet Transforms, Icing Blades; Wind Turbines; Macro-Fiber Composites; Guided Waves.

39 40 41 42

1. INTRODUCTION The ice on wind turbine blades is being one of the most important issues for the operators to reduce costs and 1

ACCEPTED MANUSCRIPT downtimes. This study is based on research project IncinBlades [1], where windfarm considered generates around 650 MW employing more than 500 wind turbines in Spain. The windfarm presented a reduction of 19 GWh due to the icing blades in two and a half years. Figure shows the main causes of the production energy losses, being the highest the ice on blades. These energy losses involve an increment of the operation and maintenance (O&M) costs. In Spain, with more than 21,000 MW installed, this phenomenon would be equivalent to more than 550 GWh of power losses (about 45 million € every 29 months). These production losses would be equivalent to the energy consumption of 200,000 households and savings of 658,682 tons of CO2 [1].

RI PT

43 44 45 46 47 48 49 50 51 52 53 54

SC 10267

9770 7539

M AN U

MWh lost

18966

5118

58

1866

1630

1609

1133

1046

Figure 1. Power losses due to different alarms in ICINGBLADES [1].

EP

56 57

1995

AC C

55

2020

TE D

2769

59 60 61 62 63 64 65 66

The ice on blades increases the surface roughness and reduce the aerodynamic efficiency, generating an imbalance in the rotor that generates stress in blades and drive train. The wind turbine is stopped under these conditions. During the ice alarm, the workers cannot access to the wind farm until the alarm is deactivated and the ice of the blades disappear. This problem is even greater when there are false alarms that report ice on WTB but there is not.

67 68 69 70

The new advances on technologies and information systems lead to the renewable energy industry to be more competitive in the energy market, reducing the Operation and Maintenance (O&M) costs [2]. It leads to increase the Reliability, Availability, 2

ACCEPTED MANUSCRIPT Maintainability and Safety (RAMS) of the system [3-5]. This paper presents a novel fault detection and diagnosis system (FDD). FDD is compound by a condition monitoring system (CMS) based on ultrasonic guided waves, and simple signal processing method based in wavelet transform.

76 77 78 79 80

CMS employs the technology and the information system to measure the parameters that show the state of a component [6-8]. It supports the predictive/preventive maintenance tasks to reduce the O&M costs [9], especially in off-shore machines [10,11].

81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96

There are sensors that are designed for detecting icing based on direct and indirect techniques [12]. The ice detection is carry out on the surfaces of the WTB using direct techniques, for example: measurement of the resonance frequency [13], damping of ultrasonic waves, measurement of ice amount [14], optical measurement techniques [12,15], measurement of temperature changes [16], measurement of the damping of the vibrations of a diaphragm [17]; or measurement of electrical properties [18]. The indirect techniques include the processing of the data acquired and the historical data [19], e.g. video monitoring, measurement of noise [20], difference in real and expected power output, patterns of heated and unheated anemometers [21], dew point and air temperature, change in the resonance frequency of the WTB, prediction of ice and frost probability maps, and direct measurement of liquid water content and volume of raindrops.

97 98 99 100 101 102

There is not any paper that consider the transducers and sensors macro-fiber composite (MFC) types for icing detection in wind turbine blades (WTB). MFC, composed by piezoceramics unidirectionally aligned fibers, generates and collects Lamb waves [22]. It has been employed in the literature to detect faults, delamination, etc. [23-25], but not ice in WTB.

103 104 105 106 107 108 109

Lamb waves are a type of guided waves that can be easily generated in structures such as plates or shells [26]. It can detect structural changes inside the material or on its surface [27]. Lamb waves propagation is confined between the two surfaces, and the attenuation is lower for this type of geometries. Lamb waves are composed by the symmetric and antisymmetric vibration modes [28].

110 111 112 113 114

Several phenomena appear in the propagation of Lamb waves, such as dispersion, different phase, velocities and vibration modes, etc. In composites, an anisotropic material, the slowness factor is important because of the propagation velocity depends on the propagation direction, i.e. WTBs are composed of layers

AC C

EP

TE D

M AN U

SC

RI PT

71 72 73 74 75

3

ACCEPTED MANUSCRIPT of fibers with a specific orientation, and the wave propagation is sensitive to the fibers direction [29].

117 118 119 120 121 122 123 124

Wavelets transforms are employed to analysed the Lamb waves and to diagnosis the ice condition on the WTB surface. Morlet wavelet transform has been also employed for filtering the signal. Wigner-Ville distribution are employed to time-frequency analysis. Finally, Hilbert transform is employed to obtain the envelope of the filtered signal. The autocorrelation of the Hibert transforms is used in this paper to identify patterns within a signal to validate the results.

RI PT

115 116

125

2. Ice on WTB The appearance of ice in WTB can be due to variables such as temperature, wind speed, relative humidity or air density, but can be also others. There are different ice types, e.g. in-cloud icing, precipitation and hoar frost [30]. They appear in clouds with high humidity and atmospheric temperature below 0ºC. The characteristics of ice (colour, strength…) are influenced by the parameters specified in the ISO 12494, 2001 (Atmospheric icing of structures) [14]. It will be considered in this paper to set the scenarios.

136 137 138 139 140 141 142 143 144

It is estimated that 20% of the wind farms are in areas with high probability of icing [31], and exposed to freezing temperatures [32]. It generates a reduction of wind farm productivity and increases the costs [33]. The Wind Energy in Cold Climate (WECO) project analysed the ice effects, energy generation and icing in wind turbines. It shows that frost is the most common cause of ice appearance in wind turbines, where it can appear in any place of the surface, but mainly in edge of the WTB [34]. Figure 2 shows examples of ice on wind turbines.

145 146

(a) (b) Figure 2 (a) Turbine on Scout Moor. Waterfoot, Great Britain, and; (b) ice on a 150kW WTB in Grenchenberg, Switzerland [35].

AC C

EP

TE D

M AN U

SC

126 127 128 129 130 131 132 133 134 135

4

ACCEPTED MANUSCRIPT The ice on WTB makes loads on the turbine. Icing causes an increase of mass, drag coefficient, imbalance of the rotor and vibrations. The fatigue due to loads reduces the life cycle of the components of the wind turbine such as WTB, hub, gearbox, shafts, etc. The drag coefficient of the WTB increases. The frozen layer modifies the thickness of the leading edge and, therefore, the aerodynamic characteristics of the WTB. The ice fragments can break off and impact with any person, animal or object. The anemometers, temperature sensors and wind vanes are exposed to icing conditions, having measurement errors more than 40% [33]. The wind turbine is stopped in case that the alarm of ice is activated.

159 160 161 162 163 164

In the first stage, pre-icing, the wind turbine works properly because there is not ice growth. Then, the ice is starting, but the wind turbine operates until its operational limit. In the third stage, the ice accumulation continues and the turbine stop to prevent possible damages. In the last stage, post-icing, the turbine continues stopped until the ice disappears [36].

M AN U

SC

RI PT

147 148 149 150 151 152 153 154 155 156 157 158

165

3. EXPERIMENTS This paper considers the guided waves to inspect WTB, i.e. composite materials. In this research, Lamb waves flow over a long distance, even in materials with a high attenuation ratio, e.g. fiber-reinforced composite structures. In anisotropic materials, Lamb wave propagation is even more complex to predict than in isotropic materials, where the properties are strongly dependent on the direction of propagation.

174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192

The CMS employs an ultrasonic system (Figure 3), composed by a data board to acquire and generate signals at 4 MS/s. The channels are isolated and can work simultaneously in both differential and single ended mode. The platform is controlled by a computer, where experiments are programmed and the dataset are processed online. The pulse function is set to the signal generator (Figure 3.1), that is amplified by 50 using a highpower amplifier (Figure 3.2) and sent to a piezoelectric transducer. It generates the Lamb waves and the ultrasonic mechanic waves flow through the composite material (Figure 3.3). The wave is received by two ultrasonic transducers that convert the mechanic waves in analog electrical signals. They are collected by the acquisition module (Figure 3.4), and registered in the computer by an analogic digital converter (Figure 3.5). The temperature affects the propagation velocity of the wave and the ice on the WTB modify its flow [37]. The change in the wave velocities with temperature depends on frequency of the wave. Lower frequencies will be less affected by temperature than higher frequencies [38]. Low frequencies were chosen to reduce

AC C

EP

TE D

166 167 168 169 170 171 172 173

5

ACCEPTED MANUSCRIPT the influence of temperature on the waves, with the aim of seeing more clearly the changes detected by the ice. It has been the main reference for designing the novel signal processing approach presented in this paper.

M AN U

SC

RI PT

193 194 195 196

197 198

Figure 3 Icing detection system.

199

The transducers used are MFC types. It is composed by piezoceramics unidirectionally aligned fibers. The electrodes are interdigitated in a polyamide film, and embedded in an adhesive polymer matrix composite. The use of MFC has been done in areas of research and development [23]. The main advantages of these sensors are the low cost, flexibility and adaptation to the surfaces. MFC is used in this paper in FDD, but they have been employed in the literature to demonstrate the versatility of the composite, together of signal processing using wavelet transform [24,25]. The optimal location and number of the MFC sensors have been studied in the last few years [39-41].

211 212 213 214 215 216

Three scenarios are set in the experiments in order to detect and diagnosis the cases consider in ISO 12494, 2001 (Atmospheric icing of structures) [14]: The first experiment was carried out at room temperature; the second was realized with the frozen WTB but with no ice on the WTB; the third experiment has been performed with the frozen WTB and ice on the WTB.

217 218 219 220 221

The ultrasonic transducers have been aligned, where the transducer (Tx) emits ultrasonic signals that will become into elastic waves having the same frequency. It will be collected by the sensor 1 (S1), and then by the sensor 2 (S2) (see Figure 4). It is expected that the elastic wave flowing through the WTB,

AC C

EP

TE D

200 201 202 203 204 205 206 207 208 209 210

6

ACCEPTED MANUSCRIPT made of fiberglass with sandwich structure, changes its shape, amplitude, energy, phase, etc. An attenuation appears between the sensor 1 and sensor 2 due to the properties of fiberglass that dissipates more energy than other composite materials, such as the carbon fibber. In addition, it is assumed to have the influence of temperature on Lamb wave propagation [42], i.e. when the temperature decreases the speed of the lamb wave propagation increases.

232 233 234 235 236 237 238 239 240 241 242 243 244 245 246

Figure 4 WTB dimensions and sensors locations

The excitation frequency has been set experimentally by sweep tests [0,100] kHz. Figure 5 shows the spectral amplitudes, or periodograms, for the main scenarios for ice detection: (a) Room temperature; (b) frozen without ice; (c) frozen with ice. The experimental tests show that the optimal frequency is between 26-33 kHz for sensors 1 (s1) and 2 (s2). It has been set in this paper 30 kHz to achieve a compromise between good response and MFC limitations when they work as actuators.

EP

231

AC C

230

TE D

M AN U

SC

RI PT

222 223 224 225 226 227 228 229

The experiments will be done for 20, 30 and 50 kHz to compare the results in the considered scenarios. The emitted ultrasonic signal will be a Hanning pulse at 5 cycles to create a narrow pulse and the sampling frequency of 4,000,000 samples per second.

247

7

y1

y1

ACCEPTED MANUSCRIPT

SC

y1

y1

bs2

y1

y1

M AN U

bs1

256

EP

252 253 254 255

Figure 5. Frequency analysis for sensor 1(s1) and sensor 2 (s2): (a) Room temperature for sensor 1; (b) frozen without ice; (c) frozen with ice.

White noise was also emitted by the actuator and collected by the sensor to analyse the most suitable frequencies. Figure 6 shows the spectral amplitudes, or periodograms. The experimental tests show the same conclusions obtained by Figure 5.

AC C

251

cs2

TE D

cs1

248 249 250

RI PT

as2

as2

257 258

8

y1

y1

ACCEPTED MANUSCRIPT

as2

bs2

y1

y1

M AN U

bs1

269

Figure 6. Frequency analysis employing Gaussian noise sensor 1 (s1) and sensor 2 (s2): (a) Room temperature; frozen without ice; (c) frozen with ice.

for (b)

EP

263 264 265 266 267 268

cs2

The experiments have been repeated three times. The WTB was introduced in a chest freezer for 24 hours with the sensors, and drops of water have been deposited on the surface periodically until the formation of a layer of frost and ice to recreate the conditions that would have a real frozen blade (Figure 7: a.without ice, b.- with ice).

AC C

262

TE D

cs1

259 260 261

RI PT SC

y1

y1

as2

9

RI PT

ACCEPTED MANUSCRIPT

(a)

Figure 7: (a) Wind turbine WTB at room temperature and; (b) frozen WTB with ice on the surface

SC

270 271

(b)

272

292 293 294 295 296 297 298

EP

TE D

M AN U

Recent works for robustness diagnosis for stochastic dynamic are considered to consider the noise and random factors [43]. The signals analysed in this paper present three information sources: vibrations (low frequencies); acoustic emission (medium frequencies), and; ultrasonic signals (high frequencies) [44]. Low frequencies are related to vibration conditions. They are employed to the structure integrity analysis [45,46]. Low frequencies consider the natural frequencies of a blade and the harmonics. Medium frequencies are associated to acoustic emission (In the range of the audible). It is employed to detect micro-breaks in the blades [47]. The material breakage generate elastic waves. This paper is focused in high frequencies of the signal. The signal processing approach employs Wavelets techniques to denoise the signals, i.e. low and medium frequencies are filtered and not considered in this paper, therefore the dynamic effect of the wind turbine and blades are omitted for signal processing. The study of the received frequency is delimited by a small range with respect to the excitation frequency, filtering other frequencies.

AC C

273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291

4.

Signal Processing Approaches

4.1. Wavelet Transforms: Daubechies It has been demonstrated that Wavelet transforms improve the limitations of resolution and the loss of information presented by the Short-Time Fourier Transform or the Fast Fourier Transform [48].

10

ACCEPTED MANUSCRIPT Wavelet transforms are commonly categorized as continuous wavelet transforms (CWT), discrete wavelet transforms (DWT) or wavelet packet transforms (PWT), etc. [49].

302 303 304 305 306 307 308

DWT is employed in this paper to set the coefficients in a wavelet series, i.e. the local energies at certain levels. The most recurrent families of wavelet transforms are Haar, Daubechies, Biortogonal, Coiflets, Morlet or Symlet transforms [50]. The Daubechies wavelet family were employed in this paper. It is demonstrated that they are more sensitive to sudden changes [51].

309

The mother wavelet is given by the equation (1), ∈ , ≠ 0

310

where s is the scale factor, and τ is the translational factor. The wavelet transform Wf(s,τ) of a function  ∈  ℜ is the decomposition of  in a set of functions forming a base with the conjugate of the mother wavelet (ψ*s,τ(t)), where * is the conjugate of the function, defined by equation (2).

316 317



  ! 


∗ 

TE D

  ,  = | | / 

M AN U

311 312 313 314 315

(1)

SC

  ,  


,  = | | /


RI PT

299 300 301

(2)

Wavelet transforms can be also expressed as equation (3),  ,  = √ #  $%
∗   &

(3)

where % is the Fourier transform of the signal x(t),
∗  is the Fourier transform of
∗  and #  is the inverse of the Fourier transform.

321 322 323 324

The information can be set in a hierarchical scheme of nested subspaces called multiresolution analysis in  ℜ, where  ℜ is decomposed into into sum of the subspaces ' , j from −∞ to +∞

AC C

EP

318 319 320

 ℜ = ⋯ , ⨁ ⨁ ⨁. ⨁ ⨁ ⨁, ⨁ ⋯

325 326

(4)

DWT provides the multiresolution decomposition of the original signal S(t) by equation (5) / = ∑23 12 4 − 5 + ∑'3 ∑23 !',2
62'  − 58

(5)

11

ACCEPTED MANUSCRIPT where 12 are the scaling coefficients, 4(·) is the scaling function, !',2 are the wavelet coefficients and
(t) is a set of wavelet function. ∑23 12 4 − 5 provides an approximation to the general trend of the signal, and ∑'3 ∑23 !',2
62'  − 58 shows the local details. The wavelet coefficient !',2 multiplying with the dilated and translated wavelet function is the detail of the original signal at scale j, which is defined by equation (6). 9' 5 = ∑23 !',2
62'  − 58

SC

DWT transforms uses a variable window size by the scaling functions, for large windows, where it is required accuracy in low frequencies, and wavelet functions, for small windows, where the information is done in high frequencies. The resulting signal from low pass filter is the Approximations (Ai) (Figure 8), and from the high pass filter are the details (Di) (Figure 8), where c is the subsampling. It can be applied single-level or multi-level filters for discrete signals. The sum of the approximations and details is the original signal [52].

TE D

344

Figure 8. Wavelet decomposition scheme.

The energy of !',2 is considered as the energy at the different decomposition levels, :' , ; = 1, ⋯ , =, by equation (7).

AC C

347 348

EP

345 346

(6)

M AN U

334 335 336 337 338 339 340 341 342 343

RI PT

327 328 329 330 331 332 333



:' = ∑2>!',2 > ; = 1, ⋯ , =

349 350

It can be assumed the energy at decomposition 1, :? , is the energy of 12 , given by equation (8). :? = ∑2|12 |

351

N

(7) level

N

+

(8)

The total energy, :@AB , is done by equation (9)

12

ACCEPTED MANUSCRIPT :@AB = ∑? C3 :'

being the relative wavelet energy (RWE), D' , givenby equation (10) ; = 1, ⋯ , = + 1

(10)

where ∑' D' = 1, and the distribution {D' } can be considered as a time-scale density. D' is employed in this paper to characterize the signal energy distribution at different frequency bands. The optimal level was at seven. It was used a multi-signal analysis to study the three different cases together, and obtained the energy and the percentage of information of the decompositions for each signal. Figure 9 shows the seven levels of decompositions with their frequencies.

SC

354 355 356 357 358 359 360 361

EF

EGHGIJ

RI PT

D' =

S

S

D1

1 - 2 MHz

D2

M AN U

352 353

(9)

A1

500 Hz-1 MHz

0.5-1 MHz

A2

500 Hz-0.5 MHz

D3

A3

250- 500 kHz

TE D

500 Hz-250 kHz

D4

A4

500 Hz-125 kHz

D5

62.5-125 kHz

A5

500 Hz- 62.5 kHz

D6

31.25-62.5 kHz

D7

A6

500 Hz- 31.2 kHz

15.6-31.25 kHz

A7

500 Hz-15.6 kHz

AC C

362

EP

125-500 kHz

363

Figure 9 Wavelet Decompositions Levels employing Daubichies.

364 365 366 367 368

Figure 10 shows the denoised signals received in sensor 1 at 30 kHz for each experiment employing Daubichies family. The signal amplitude is greater at 20 and 30 kHz because of the MFC works better as actuator at low frequencies. Sensor 1 receives a signal with more voltage because it is closer to the actuator.

13

ACCEPTED MANUSCRIPT 0.2 Room Temp Frozen blade without ice Frozen blade with Ice

0.15

0.1

Volts

0.05

0

RI PT

-0.05

-0.1

-0.15

-0.2 0

1000

2000

3000

4000

5000

6000

Samples

Figure 10 Signal in sensor 1 at 30 kHz in the three different scenarios.

M AN U

370 371 372

The signal obtained with the sensors on the frozen WTB without ice has bigger amplitude than the signals at room temperature. The amplitude is lower when the WTB has ice on its surface. This is because of the ice opposes the free displacement of the waves. Figure 11 presents the signals received at the sensor 2 in the same cases.

TE D

373 374 375 376 377 378

8000

SC

369

7000

0.02

0.015

0

AC C

Volts

0.005

EP

0.01

Room Temp Frozen blade without ice Frozen blade with Ice

-0.005

-0.01

-0.015

-0.02 0

1000

2000

3000

4000

5000

6000

7000

8000

Samples

379 380 381

Figure 11 Signal in sensor 2 at 30 kHz in the three different scenarios.

382 383 14

ACCEPTED MANUSCRIPT 4.2. Wavelet Transforms: Morlet Morlet wavelet has been employed for denoising in order to validate obtained with Daubichies wavelet. The Morlet wavelet is given by equation (11) as
 = K  /L M 

391

cosR. 

(11)

The mother wavelet is given by the real part of the Morlet function for wavelet transform of a real signal according to equation (12).
 = K  /L M 

N /

cos5

(12)

SC

388 389 390

N /

RI PT

384 385 386 387

The scale factor has been optimized by cross validation method (CVM) according to references [45,46], i.e.: It is set the scale range [m,n] and then is determined the scale factor m+ki that minimise the cost function smallest by CVM. Then the scale range is reduced to TU + 5 − 1V, U + 5 + 1VW consideting the step to i. It is done until is found the best scale factor.

398 399 400 401 402 403

Figure 12(a) shows the original signal, and the signal filtered employing Morlet wavelet with scale ranges [1,100] (Figure 12(b)) and [1,50] (Figure 12(c)). The optimal range found was at [147,152]. The results where compared by previous filter and there were not present significant differences in the results. They were employed to validate the filters.

TE D

M AN U

392 393 394 395 396 397

405 406

AC C

Amplitude

EP

404

(a)

15

407 408

M AN U

Amplitude

SC

(b)

409 410

415 416 417 418 419 420 421 422 423 424

425 426

TE D

Figure 12 (a) Signal in sensor 1 at 30 kHz at room temperature. (b) Signal filtered with Morlet wavelet and scale window 100. (c) Signal filtered with Morlet wavelet and scale window 150.

EP

414

(c)

4.3. Time-Frequency Analysis: Wigner-Ville distribution Wigner-Ville distribution (WVD) has been employed as timefrequency analysis. Auto terms window (ATW) function is used to suppress the cross terms in WVD according to reference [53], describing the suitable window function. It suggests that the auto terms and cross terms have different features in nature, and they can be omitted by cross-correlation. The cross terms can be suppressed by using ideal as a template to do twodimension cross-correlation when both terms are not overlap, then it can be obtained equation (13)

AC C

411 412 413

RI PT

Amplitude

ACCEPTED MANUSCRIPT

, R = X , R ∙ X,AZ@ , R being X , R the cross term and X

(13) ,AZ@ , R

the auto term.

16

ACCEPTED MANUSCRIPT 427

436

4.4. Hilbert Transform and autocorrelations Hiltbert transform has been employed to validate the approach together with autocorrelations analysis. Hilbert Transform is used to obtain the envelope of the filtered signal [22]. The Hilbert transform is an approach to study the energy distribution of a Shear wave in the time domain [55]. The energy envelope was employed to identify local characteristics of the signal.

RI PT

428 429 430 431 432 433 434 435

The correlation coefficient is given by equation (14) ? ? = ∑? ^3 x^ y^ − ∑^3 x^ ∑^3 y^

? ? ?    `= ∑^3 x^  − ∑? ^3 x^  ∙ `= ∑^3 y^ − ∑^3 y^ 

(1)

SC

λX\ =

where λxy is the correlation coefficient of two Lamb wave signals, x and y, being the length of the discrete signals N samples [36,37].

440 441 442 443

When a signal is correlated with itself it is called autocorrelation [38,39]. The autocorrelation of the Hilbert transforms is used in this paper to identify patterns within a signal, i.e. the periodicity hidden by the noise.

444 445 446 447 448 449 450

The pattern recognition approach is based on the autocorrelation of both signals. Then the autocorrelation of one of the signals is divided by the autocorrelation of the other signal in order to emphasize the differences and to obtain the ratio curve between them. The singularity caused by different conditions in the WTB is analysed in the ratio curve, and its corresponding to the scenario considered.

EP

TE D

M AN U

437 438 439

AC C

451 452

5. Results

453 454 455 456 457 458 459 460 461

5.1. Wavelet Transform: Daubechies Table 1 shows the results for all scenarios, showing the energy for each signal according to the wavelet decomposition, and the energy percent for approximations and details. D7 decomposition contains the highest percentage of energy of the original signal in most of the cases. It is associated with the frequency range from 125.6 kHz to 31.25 kHz. It is consistent with the excitation frequencies of the Hanning pulse used in the actuator.

17

ACCEPTED MANUSCRIPT The composite material is highly dispersive, where the dispersion depends on the direction of wave propagation, the orientation and arrangement of the layers of fibre glass. Therefore, the waves received by the sensors contain frequencies that deviate from the centre frequency of excitation.

467 468

Table 1 Wavelet Decomposition of the received signal in the sensor 1. Sensor 1

Frozen Frozen Ice

30 kHz

Room T Frozen Frozen Ice

50 kHz

Room T Frozen Frozen Ice

47.4 8 52.4 7 25.7 7 45.3 8 49.0 8 28.3 5 47.8 6 49.0 8 45.9 6

D6 %

D5 %

19.5 0 21.5 0 12.7 2 19.1 6 22.0 9 16.0 2 20.2 1 22.0 9 21.5 4

D4%

Frozen Ice

30 kHz

Room T Frozen Frozen Ice

AC C

50 kHz

Room T Frozen

Frozen Ice

469 470 471 472 473 474 475 476 477

48.5 7 49.1 8 47.3 8 47.3 7 49.6 5 38.5 1 46.7 5 49.8 0 47.0 7

D6 %

1.42

0.36

0.09

6.14

1.59

0.40

0.10

3.48

0.89

0.23

0.06

5.43

1.40

0.36

0.10

6.32

1.63

0.41

0.10

4.77

1.24

0.34

0.12

5.73

1.49

0.38

0.11

6.32

1.63

0.41

0.10

6.25

1.64

0.43

0.12

20.1 7 21.5 4 16.7 0 20.0 7 23.4 9 22.2 0 19.3 9 23.5 6 23.2 2

D5 %

TE D

Frozen

22.9 8 20.8 2 29.2 9 23.3 4 17.2 3 28.9 2 24.0 7 17.1 8 15.6 5

EP

20 kHz

Room T

D7 %

D2 %

5.51

Sensor

A7 %

D3 %

SC

20 kHz

Room T

D7 %

25.6 0 17.7 8 56.8 3 28.1 4 20.3 4 49.0 6 24.1 8 20.3 4 24.0 0

M AN U

A7 %

RI PT

462 463 464 465 466

D %

0.0 3 0.0 3 0.0 2 0.0 4 0.0 3 0.1 0 0.0 5 0.0 3 0.0 7

Energy 16,49 41,77 4

5,22 16 0,85 0,69 16 0,48

2

D4 %

D3 %

D2 %

5.76

1.53

0.45

0.23

6.16

1.61

0.43

0.15

4.72

1.24

0.35

0.15

5.79

1.57

0.56

0.49

6.90

1.80

0.50

0.21

6.46

1.82

0.65

0.54

5.62

1.55

0.63

0.71

6.92

1.80

0.47

0.16

6.90

2.10

1.09

1.38

D % 0.3 1 0.1 2 0.1 7 0.8 0 0.2 2 0.9 0 1.2 7 0.1 3 2.5 8

Energy 0.271 2 0.400 3 0.266 1 0.099 54 0.196 5 0.043 2 0.016 3 0.196 0 0.007 4

The three scenarios can be identified by the energies shown in Table 1. Analysing the energy of the WTB at room temperature, the frozen WTB with ice has a higher energy (more than doubled to the rest of scenarios). However, when the WTB has ice on the surface, and the energy received by the sensor was about 25% of the energy received at room temperature. Figure 13 shows differences between the energies of each state at 20 and 30 kHz. Considering 50 kHz, the energy of the signal at room temperature

18

ACCEPTED MANUSCRIPT 478 479

is similar when the WTB has ice, therefore, this frequency is not suitable for determining the state of the WTB.

45 40 35 30 25 20 15 10 5 0 30 kHz

Room Temperature

Frozen without Ice

Frozen with Ice

480

M AN U

483 484 485 486 487 488

Figure 13 Energy of each state and frequency received in sensor 1 In sensor 2 (Figure 14), the energy is smaller due to the distance. The energy in each state of the WTB has a similar relationship to the results given by sensor 1, but at 20 kHz it is difficult to distinguish between the cases at room temperature and with the frozen WTB with ice. The optimal frequency was at 30 kHz for this experiment.

TE D

481 482

50 kHz

SC

20 kHz

RI PT

Energy received in sensor 1

Energy received in Sensor 2

489 490 491

AC C

EP

0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

20 kHz Room Temperature

30 kHz

50 kHz

Frozen without Ice

Frozen with Ice

Figure 14 Energy of each state and frequency received in sensor 2

492 493 494 495

Figures 15 and 16 show the energies of the original signal for each level of decomposition in Sensor 1 and 2. The levels with more energy are: A7, D7 and D6. The signal has two peaks when 19

ACCEPTED MANUSCRIPT 496 497 498 499 500 501

that WTB has ice, at 20 and 30 kHz, while D7 and D6 have less energy. This could be due by presence of ice on the surface of the WTB. Other decompositions also present less energy, but the percentage of signal information is low. D7 has a period without peaks at 20 and 30 kHz using sensor 2 (Figure 16), while the decomposition A7 has peaks in these cases.

RI PT

Wavelet decomposition: Sensor 1 60.00% 50.00% 40.00%

20 kHz

A7

D7

Frozen with ice

Frozen with ice

50 kHz D6

Figure 15 Wavelet decompositions (approximations and details) from sensor 1

EP

503 504

30 kHz

AC C

502

Frozen without ice

Room Temp

Frozen with ice

TE D

Frozen without ice

Room Temp

0.00%

M AN U

10.00%

Frozen without ice

20.00%

Room Temp

SC

30.00%

20

ACCEPTED MANUSCRIPT 505

Wavelet decomposition: Sensor 2 60.00% 50.00%

514 515 516 517 518 519 520

SC

Figure 16 Wavelet decompositions (approximations and details) from sensor 2

TE D

510 511 512 513

D6

The optimal frequency to determine differences in the state of the WTB is 30 kHz. Sensors 1 and 2 have been able to identify clearly the scenarios, where the decompositions A7, D7 and D6 provide more information about the condition of the scenarios.

EP

509

D7

50 kHz

AC C

507 508

Frozen with ice

30 kHz

A7

506

Frozen without ice

Room Temp

M AN U

20 kHz

Frozen with ice

Frozen without ice

Room Temp

0.00%

Frozen with ice

10.00%

Frozen without ice

20.00%

Room Temp

30.00%

RI PT

40.00%

5.2. Hilbert transform and autocorrelations Analysis Hiltbert transform has been employed together with correlations analysis has been employed in this paper to validate the approach. Then, the enveloped, done by Hilbert transform, of the filtered signal (Figure 17), obtained by Wavelet transforms, is used for correlations analysis.

521 522

21

ACCEPTED MANUSCRIPT 1.2 envelope Room Temperature envelope Frozen without Ice

1

RI PT

Amplitude

0.8 0.6 0.4

0

1000

2000

3000

4000

5000

6000

7000

M AN U

0

SC

0.2

Samples

523

Figure 17. Envelopes of the signals employing Hilbert Transform (blue) and Wavelet denoising (brown).

526 527 528 529 530 531 532 533

Pattern recognition is based in this section by the autocorrelation coefficient analysis, i.e. autocorrelation of two signals (Figure 18). Then the autocorrelation of one of the signals is divided by the autocorrelation of the other signal in order to emphasize the differences and to obtain the ratio curve between them. The singularity caused by different conditions in the WTB is analysed in the ratio curve and its corresponding to the scenario considered.

EP AC C

534

TE D

524 525

22

ACCEPTED MANUSCRIPT 1 Room Temp Frozen without Ice

0.6

RI PT

Amplitude

0.8

0.4

0

1000

2000

6000

7000

EP

TE D

The correlation of the autocorrelations are close to 1 when the same conditions are the same: Room Tª- Room Tª; Frozen without ice – Frozen without ice; Frozen with ice – Frozen with ice; and it is close to 0 or negatives when the scenarios are different: Room Tª- Frozen without ice; Room Tª – Frozen with ice; Frozen without air- Frozen with ice. Figure 19 shows the results obtained by this approach employed in sensors 1 and 2 at 30 kHz. The results obtained are the same to the results obtained studying the energies from the Wavelet transforms, which is considered to validate the FDD approach presented in this paper.

AC C

550

5000

Figure 18 Autocorrelation of the signals at room temperature (blue) and frozen without ice (brown).

538 539 540 541 542 543 544 545 546 547 548 549

4000

Samples

535 536 537

3000

M AN U

0

SC

0.2

23

ACCEPTED MANUSCRIPT

Correlation coefficient of the autocorrelations 1.2 1 0.8

RI PT

0.6 0.4 0.2 0

-0.4

Room Tº Frozen Frozen with Room Tº Room Tº Frozen Room Tº without Ice Ice Frozen Frozen with without Ice - Frozen Frozen with without Ice Ice - Frozen without Ice Ice with Ice

SC

-0.2

-0.6

M AN U

-0.8 Sensor 1

551 552 553

Figure19. Correlation coefficients of the autocorrelations for sensor 1 and 2 at 30 kHz for all scenarios.

554

567

EP

TE D

5.3. Time-Frequency analysis by Validation of the approach by Wigner-Ville distribution Figure 20 shows the time-frequency analysis for the cases of frozen with ice (Figure 20(a)), at room temperature (Figure 20(b)), and frozen without ice (Figure 20(c)) for sensor 1. The results show that there is a minimum of cross terms, i.e. it presents an accuracy time-frequency resolution and energy aggregation. Figure 20(a) presents a different pattern to Figures 20(b) and 20(c) due to the presence of ice. It is the same conclusion obtained in Figures 13. Figures 20(b) and 20(c) present similar patterns, being of less impact in Figure 20(c), as it was found with the approach (see Figures 13).

AC C

555 556 557 558 559 560 561 562 563 564 565 566

Sensor 2

24

568 569

570

(b)

AC C

EP

571

TE D

M AN U

(a)

SC

RI PT

ACCEPTED MANUSCRIPT

572 573

(c)

25

ACCEPTED MANUSCRIPT 574 575 576

Figure 20. Time-frequency graphs of WVD with de-noising by Morlet wavelet:(a) frozen with ice;(b) room temperature; (c) frozen without ice.

577

SC

RI PT

5.4. Scenarios considering Mud on the surface of the wind turbine blades Tests without prior information about the icing blade status of the WTB has been considered. A layer of mud was applied to the surface of WTB section to simulate dirt and soiling on the wind turbine blade, composed of a mixture of sand and a solution of water and glue, to simulate the soiling. The first thickness of the mud layer has been set between 5 and 6.5 mm. Subsequently, the thickness has been reduced gradually. It has been considered 3 different levels of mud on the blade, being the last one without mud. The signal was generated as the previous scenarios in order to study the response of the approach.

M AN U

578 579 580 581 582 583 584 585 586 587 588 589 590

Energy 0.8 0.7 0.6

0.3 0.2 0.1 0

592 593 594 595 596 597 598 599 600 601

No mud

Mud 3cm

Mud 5cm

AC C

591

EP

0.4

TE D

0.5

Figure 21. Energy for cases with mud: 0 cm; 3 cm; 5 cm

Figure 21 shows a similar pattern to Figures 13 and 14. The presence of ice or mud on the blade could be therefore identify clearly when the temperature is known or the system can be detected the mud previously than the temperature are under cero. Supervisory Control and Data Acquisition (SCADA) system of the wind turbine provides several environment temperatures that can been considered together with the approach presented in this paper.

602 26

ACCEPTED MANUSCRIPT 603

6.

Conclusions

Ice on Wind Turbine Blade (WTB) causes an increase of mass, makes loads on the turbine drag coefficient, imbalance of the rotor and vibrations, etc. The wind turbine is stopped in case that the alarm of ice is activated. The icing WTB require of advanced condition monitoring systems to be detected, and then reduce the failures and downtimes in the WTB and wind turbines.

610 611 612 613 614 615 616 617

The condition monitoring system presented in this paper employs non-destructive techniques based on ultrasonic waves. The transducers used were macro-fiber composites. It was generated Lamb waves that flows through the material of the WTB and are sensitive to elements such as ice on the surface. Different scenarios were simulated to determine the state of the WTB. The scenarios were: WTB at room temperature, WTB frozen without ice accumulation and WTB frozen with ice accumulation.

618 619 620 621 622 623 624

The pattern recognition employs Daubechies wavelet transform with seven levels of decomposition to diagnosis the experimental scenarios. It is analysed the signal energies and percentage for the decompositions to determine the conditions. A difference between the percentage of information of D7, D6 and A7 decomposition is observed in both sensor when the excitation is at 30 kHz.

625 626 627 628 629 630

Morlet wavelet transformation has been also employed for filtering the signal. Wigner-Ville distribution are employed to time-frequency analysis. Finally, Hilbert Transform is employed to obtain the envelope of the filtered signal. The autocorrelation of the Hibert transforms is used in this paper to identify patterns within a signal validate the results.

631 632 633 634

The approach detects the cases considered in ISO 12494 of unfrozen, frozen without ice, and frozen with ice in the WTB. New scenarios considering mud have been considered to test the approach.

636 637 638 639

SC

M AN U

TE D

EP

AC C

635

RI PT

604 605 606 607 608 609

Acknowledgements The work reported herewith has been financially supported by the Spanish Ministerio de Economía y Competitividad, under Research IcingBlades, Ref.: DPI2015-67264-P.

640 641 642

27

ACCEPTED MANUSCRIPT 643

References

644 1.

2.

RI PT

3.

Competitividad, M.d.E.y. Investigación de técnicas avanzadas de deshielo y prevención de la acumulación de nieve en palas de un aerogenerador (icingblades) Ref.: IPT-2012-0563-120000 2012. Márquez, F.P.G.; Tobias, A.M.; Pérez, J.M.P.; Papaelias, M. Condition monitoring of wind turbines: Techniques and methods. Renewable Energy 2012, 46, 169-178. Pérez, J.M.P.; Márquez, F.P.G.; Tobias, A.; Papaelias, M. Wind turbine reliability analysis. Renewable and Sustainable Energy Reviews 2013, 23, 463-472. Marugán, A.P.; Márquez, F.P.G.; Papaelias, M. In Multivariable analysis for advanced analytics of wind turbine management, Proceedings of the Tenth International Conference on Management Science and Engineering Management, 2017; Springer: pp 319-328. Gómez, C.Q.; Villegas, M.A.; García, F.P.; Pedregal, D.J. Big data and web intelligence for condition monitoring. 2015. Papaelias, M.; Cheng, L.; Kogia, M.; Mohimi, A.; Kappatos, V.; Selcuk, C.; Constantinou, L.; Muñoz, C.Q.G.; Marquez, F.P.G.; Gan, T.-H. Inspection and structural health monitoring techniques for concentrated solar power plants. Renewable Energy 2016, 85, 1178-1191. Gómez Muñoz, C.Q.; García Márquez, F.P. A new fault location approach for acoustic emission techniques in wind turbines. Energies 2016, 9, 40. Lane, C. The development of a 2d ultrasonic array inspection for single crystal turbine blades. Springer Science & Business Media: 2013. Marquez, F.P.G. In An approach to remote condition monitoring systems management, Railway Condition Monitoring, 2006. The Institution of Engineering and Technology International Conference on, 2006; IET: pp 156-160. Caselitz, P.; Giebhardt, J.; Kewitsch, R. In Advanced condition monitoring system for wind energy converters, EWEC-CONFERENCE-, 1999; pp 63-66. Pliego Marugán, A.; García Márquez, F.P.; Pinar Pérez, J.M. Optimal maintenance management of offshore wind farms. Energies 2016, 9, 46. Carlsson, V. Measuring routines of ice accretion for wind turbine applications. Master Thesis, Umeaa Univesitet, 2009. Luukkala, M. Detector for indicating ice formation on the wing of an aircraft. Google Patents: 1995. Foder, M.H. In Iso 12494" atmospheric icing of structures" and how to use it, The Eleventh International Offshore and Polar Engineering Conference, 2001; International Society of Offshore and Polar Engineers. Federow, H.L.; Silverman, J.H. Laser ice detector. Google Patents: 1994. Muñoz, C.Q.G.; Márquez, F.P.G.; Tomás, J.M.S. Ice detection using thermal infrared radiometry on wind turbine blades. Measurement 2016, 93, 157-163. DeAnna, R. Ice detection sensor. Google Patents: 1999.

SC

4.

5.

8.

9.

10.

11.

EP

7.

TE D

M AN U

6.

AC C

645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694

12. 13. 14.

15. 16.

17.

28

ACCEPTED MANUSCRIPT

23.

24.

25.

26.

27.

28.

29.

RI PT

22.

SC

21.

M AN U

20.

TE D

19.

Gerardi, J.J.; Dahl, P.R.; Hickman, G.A. Piezoelectric sensor. Google Patents: 1993. Márquez, F.P.G.; Muñoz, J.M.C. A pattern recognition and data analysis method for maintenance management. International Journal of Systems Science 2012, 43, 1014-1028. Seifert, H. In Technical requirements for rotor blades operating in cold climate, VI BOREAS Conference, Pyhatunturi, Finland, 2003. Tammelin, B.; Morgan, C.; Hyvoenen, R.; Peltomaa, A. In Ice free anemometers, EWEC-CONFERENCE-, 1997; BOOKSHOP FOR SCIENTIFIC PUBLICATIONS: pp 389-392. de la Hermosa González, R.R.; Márquez, F.P.G.; Dimlaye, V. Maintenance management of wind turbines structures via mfcs and wavelet transforms. Renewable and Sustainable Energy Reviews 2015, 48, 472-482. Collet, M.; Ruzzene, M.; Cunefare, K. Generation of lamb waves through surface mounted macro-fiber composite transducers. Smart Materials and Structures 2011, 20, 025020. Kessler, S.S.; Spearing, S.M.; Soutis, C. Damage detection in composite materials using lamb wave methods. Smart Materials and Structures 2002, 11, 269. Thien, A.B.; Chiamori, H.C.; Ching, J.T.; Wait, J.R.; Park, G. The use of macro‐fibre composites for pipeline structural health assessment. Structural control and health monitoring 2008, 15, 43-63. Castaings, M.; Hosten, B. Guided waves propagating in sandwich structures made of anisotropic, viscoelastic, composite materials. The Journal of the Acoustical Society of America 2003, 113, 2622-2634. Su, Z.; Wang, X.; Chen, Z.; Ye, L.; Wang, D. A built-in active sensor network for health monitoring of composite structures. Smart Materials and Structures 2006, 15, 1939. Zhao, X.; Gao, H.; Zhang, G.; Ayhan, B.; Yan, F.; Kwan, C.; Rose, J.L. Active health monitoring of an aircraft wing with embedded piezoelectric sensor/actuator network: I. Defect detection, localization and growth monitoring. Smart materials and structures 2007, 16, 1208. Su, Z.; Ye, L.; Lu, Y. Guided lamb waves for identification of damage in composite structures: A review. Journal of sound and vibration 2006, 295, 753-780. Parent, O.; Ilinca, A. Anti-icing and de-icing techniques for wind turbines: Critical review. Cold regions science and technology 2011, 65, 88-96. Tammelin, B.; Holttinen, H.; Morgan, C.; Richert, F.; Seifert, H.; Säntti, K.; Vølund, P. Wind energy production in cold climate. 2000. Fortin, G.; Perron, J.; Ilinca, A. In A study of icing events at murdochville: Conclusions for the wind power industry, International Symposium “Wind Energy in Remote Regions”, Magdalen’s Island, 2005. Shajiee, S.; Pao, L.Y.; Wagner, P.N.; Moore, E.D.; McLeod, R.R. In Direct ice sensing and localized closed-loop heating for

EP

18.

AC C

695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746

30.

31.

32.

33.

29

ACCEPTED MANUSCRIPT

38.

39.

40.

41.

42.

43.

44.

RI PT

SC

37.

M AN U

36.

TE D

35.

EP

34.

active de-icing of wind turbine blades, 2013 American Control Conference, 2013; IEEE: pp 634-639. Morgan, C.; Bossanyi, E.; Seifert, H. In Assessment of safety risks arising from wind turbine icing, EWEC-CONFERENCE-, 1997; BOOKSHOP FOR SCIENTIFIC PUBLICATIONS: pp 141-144. Laakso, T.; Holttinen, H.; Ronsten, G.; Tallhaug, L.; Horbaty, R.; Baring-Gould, I.; Lacroix, A.; Peltola, E.; Tammelin, B. State-of-the-art of wind energy in cold climates. IEA annex XIX 2003, 24. Kraj, A.G.; Bibeau, E.L. Phases of icing on wind turbine blades characterized by ice accumulation. Renewable Energy 2010, 35, 966-972. de la Hermosa González, R.R.; Márquez, F.P.G.; Dimlaye, V.; Ruiz-Hernández, D. Pattern recognition by wavelet transforms using macro fibre composites transducers. Mechanical Systems and Signal Processing 2014, 48, 339-350. Dan, C.A.; Kudela, P.; Ostachowicz, W. In Compensation of temperature effects on guided wave based structural health monitoring systems, EWSHM-7th European Workshop on Structural Health Monitoring, 2014. Lee, B.; Staszewski, W. Sensor location studies for damage detection with lamb waves. Smart materials and structures 2007, 16, 399. Tongpadungrod, P.; Rhys, T.; Brett, P. An approach to optimise the critical sensor locations in one-dimensional novel distributive tactile surface to maximise performance. Sensors and Actuators A: Physical 2003, 105, 47-54. Shi, Z.; Law, S.; Zhang, L. Optimum sensor placement for structuraldamage detection. Journal of Engineering Mechanics 2000, 126, 1173-1179. di Scalea, F.L.; Salamone, S. Temperature effects in ultrasonic lamb wave structural health monitoring systems. The Journal of the Acoustical Society of America 2008, 124, 161-174. Liu, X.; Gao, Z. Robust finite-time fault estimation for stochastic nonlinear systems with brownian motions. Journal of the Franklin Institute 2016. Gómez, C.Q.; Villegas, M.A.; García, F.P.; Pedregal, D.J. Big data and web intelligence for condition monitoring: A case study on wind turbines. In Big data: Concepts, methodologies, tools, and applications, IGI Global: 2016; pp 1295-1308. Abouhnik, A.; Albarbar, A. Wind turbine blades condition assessment based on vibration measurements and the level of an empirically decomposed feature. Energy Conversion and Management 2012, 64, 606-613. Chen, J.; Pan, J.; Li, Z.; Zi, Y.; Chen, X. Generator bearing fault diagnosis for wind turbine via empirical wavelet transform using measured vibration signals. Renewable Energy 2016, 89, 8092. Beattie, A.; Beattie, A. In Acoustic emission monitoring of a wind turbine blade during a fatigue test, 35th Aerospace Sciences Meeting and Exhibit, 1997; p 958.

AC C

747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797

45.

46.

47.

30

ACCEPTED MANUSCRIPT

52.

53.

54.

TE D

55.

RI PT

51.

SC

50.

M AN U

49.

Eristi, H. Fault diagnosis system for series compensated transmission line based on wavelet transform and adaptive neurofuzzy inference system. Measurement 2013, 46, 393-401. Dong, Y.; Shi, H.; Luo, J.; Fan, G.; Zhang, C. Application of wavelet transform in mcg-signal denoising. Modern Applied Science 2010, 4, 20. Patil, M.S.M.; Kasturiwala, S.B.; Dahad, S.O.; Jadhav, M.C. Daubechies wavelet tool: Application for human face recognition. International Journal of Engineering Science and Technology 2011, 1, 2392-2398. Jain, B.; Jain, S.; Nema, R. Investigations on power quality disturbances using discrete wavelet transform. International Journal of Electrical, Electronics and Computer Engineering 2013, 2, 47-53. Gomez Munoz, C.; De la Hermosa Gonzalez-Carrato, R.; Trapero Arenas, J.; Garcia Marquez, F. A novel approach to fault detection and diagnosis on wind turbines. GLOBAL NEST JOURNAL 2014, 16, 1029-1037. Tang, B.; Liu, W.; Song, T. Wind turbine fault diagnosis based on morlet wavelet transformation and wigner-ville distribution. Renewable Energy 2010, 35, 2862-2866. Coverley, P.; Staszewski, W. Impact damage location in composite structures using optimized sensor triangulation procedure. Smart materials and structures 2003, 12, 795. Chen, H.; Yan, Y.; Chen, W.; Jiang, J.; Yu, L.; Wu, Z. Early damage detection in composite wingbox structures using hilberthuang transform and genetic algorithm. Structural Health Monitoring 2007, 6, 281-297.

EP

826

48.

AC C

798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825

31

ACCEPTED MANUSCRIPT A new method for ice detection employing guided waves and Wavelet transform based on the energy decomposition of the signals is shown. The low computational cost of this approach would facilitate its implementation in Condition Monitoring systems and the online analisys.

AC C

EP

TE D

M AN U

SC

RI PT

A real case study is shown where it is possible to determine if the blade is unfrozen, frozen without ice and frozen with ice.