Elsevier

Renewable Energy

Volume 116, Part B, February 2018, Pages 42-54
Renewable Energy

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

https://doi.org/10.1016/j.renene.2017.03.052Get rights and content

Highlights

  • A new method for ice detection employing guided waves and Wavelet transform based on the energy decomposition of the signals.

  • The low computational cost would facilitate its implementation in Condition Monitoring systems and the online analisys.

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

Abstract

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.

Introduction

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

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.

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, Maintainability and Safety (RAMS) of the system [3], [4], [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.

CMS employs the technology and the information system to measure the parameters that show the state of a component [6], [7], [8]. It supports the predictive/preventive maintenance tasks to reduce the O&M costs [9], especially in off-shore machines [10], [11].

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.

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], [24], [25], but not ice in WTB.

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 anti-symmetric vibration modes [28].

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

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.

Section snippets

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 0 °C. 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

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.

The CMS employs an ultrasonic system (Fig. 3), composed by a data board to acquire

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].

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

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

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.

The composite material is highly dispersive, where the

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.

The condition monitoring system presented in this paper employs non-destructive techniques based on ultrasonic waves. The

Acknowledgements

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

References (54)

  • F.P.G. Márquez et al.

    Condition monitoring of wind turbines: techniques and methods

    Renew. Energy

    (2012)
  • J.M.P. Pérez et al.

    Wind turbine reliability analysis

    Renew. Sustain. Energy Rev.

    (2013)
  • A.P. Marugán et al.

    Multivariable analysis for advanced analytics of wind turbine management

  • C.Q. Gómez et al.

    Big Data and Web Intelligence for Condition Monitoring

    (2015)
  • C.Q. Gómez Muñoz et al.

    A new fault location approach for acoustic emission techniques in wind turbines

    Energies

    (2016)
  • C. Lane

    The Development of a 2d Ultrasonic Array Inspection for Single Crystal Turbine Blades

    (2013)
  • F.P.G. Marquez

    An approach to remote condition monitoring systems management

  • P. Caselitz et al.

    Advanced condition monitoring system for wind energy converters

  • A. Pliego Marugán et al.

    Optimal maintenance management of offshore wind farms

    Energies

    (2016)
  • V. Carlsson

    Measuring Routines of Ice Accretion for Wind Turbine Applications

    (2009)
  • M. Luukkala

    Detector for Indicating Ice Formation on the Wing of an Aircraft

    (1995)
  • M.H. Foder

    Iso 12494“ atmospheric icing of structures” and how to use it

  • H.L. Federow et al.

    Laser Ice Detector

    (1994)
  • C.Q.G. Muñoz et al.

    Ice detection using thermal infrared radiometry on wind turbine blades

    Measurement

    (2016)
  • R. DeAnna

    Ice Detection Sensor

    (1999)
  • J.J. Gerardi et al.

    Piezoelectric Sensor

    (1993)
  • F.P.G. Márquez et al.

    A pattern recognition and data analysis method for maintenance management

    Int. J. Syst. Sci.

    (2012)
  • Cited by (80)

    • Parametric identification of ultrasonic guided wave aliasing modes based on dispersion effect

      2023, Measurement: Journal of the International Measurement Confederation
    • Robust and quantitative characterization of aircraft icing with mode and frequency selective ultrasonic guided wave

      2023, Ultrasonics
      Citation Excerpt :

      For example, Zhao et al. [24] used Pearson correlation coefficient to compare the signals before and after ice formation. Munoz et al. [25] tried a variety of algorithms to extract signal features, such as energy ratio, autocorrelation, etc., to identify ice forming. Jimenez et al. [26] developed an approach to detecting and classifying ice thickness with UGW signals based on mode recognition.

    • Prediction of operating parameters and output power of ducted wind turbine using artificial neural networks

      2022, Energy Reports
      Citation Excerpt :

      Data acquisition setups evaluate all parameters to establish the behavior of the system (García Márquez et al., 2012). Data processing needs robust techniques (Jiménez et al., 2019) that activate as much detail as possible from the given data (Gómez Muñoz et al., 2018). Machine learning techniques are frequently used according to their potential to handle a huge quantity of data, ANNs optimization algorithms being one of the most utilized techniques (Jiménez et al., 2018).

    View all citing articles on Scopus
    View full text