Monitoring of Wind Turbine Blades Based on Dual?Tree Complex Wavelet Transform

LIU Rongmei，ZHOU Keyin，YAO Entao

1.College of Aerospace Engineering，Nanjing University of Aeronautics and Astronautics，Nanjing 210016，P.R.China；2.College of Automation Engineering，Nanjing University of Aeronautics and Astronautics，Nanjing 210016，P.R.China

Abstract: Structural health monitoring（SHM）in-service is very important for wind turbine system. Because the central wavelength of a fiber Bragg grating（FBG）sensor changes linearly with strain or temperature，FBG-based sensors are easily applied to structural tests. Therefore，the monitoring of wind turbine blades by FBG sensors is proposed. The method is experimentally proved to be feasible. Five FBG sensors were set along the blade length in order to measure distributed strain. However，environmental or measurement noise may cover the structural signals.Dual-tree complex wavelet transform（DT-CWT）is suggested to wipe off the noise. The experimental studies indicate that the tested strain fluctuate distinctly as one of the blades is broken. The rotation period is about 1 s at the given working condition. However，the period is about 0.3 s if all the wind blades are in good conditions. Therefore，strain monitoring by FBG sensors could predict damage of a wind turbine blade system. Moreover，the studies indicate that monitoring of one blade is adequate to diagnose the status of a wind generator.

Key words：wind turbine blade；structural health monitoring（SHM）；fiber Bragg grating（FBG）；dual-tree complex wavelet transform（DT-CWT）

0 Introduction

Wind power is becoming one of the fastest growing energy technologies in the world［1］. Howev?er，generation cost may be extra increased due to an?nual operating costs，such as maintenance on the generation systems［2］. Therefore，a reliable structur?al health monitoring （SHM） system is pressing needed for the successful implementation of wind turbine systems［3］.Du et al. reviewed damage detec?tion techniques for blades［4］.

Wind turbine blades are among the most easily damaged parts of a turbine system because of the working environment. Blade failure would be cata?strophic. Therefore，it is critical to detect blade damages at their early stage. Periodic inspections are proposed to reduce the probability of complete blade failures［5］.

Lee et al. monitored the blade deflection based on strain gauge sensors［6］. Dilek et al. studied moni?toring of blades by an automated laser scanning sys?tem［7］. Procházka et al. used magneto-resistive sen?sors to study strain and damage of steam turbine blade［8］. It was pointed out that monitoring of dy?namic characteristics，such as amplitudes and fre?quencies of vibrations，could represent a fundamen?tal damage assessment for turbine blades［9］.The pro?posed method was based on analysis of structural resonance［10］. Finite element analyses were suggest?ed to study the behavior of wind turbine blade［11-12］.The analyses and supported tests on designed wind turbine blades implicated that the lamination of the outer skin was the initial destruction. It followed by lamination buckling which led to breakdown of the generation.

Non-destructive testing （NDT） techniques were proposed to detect damage in composite blades. These techniques，including visual inspec?tion or acoustic emission［13］，were labor intensive，or difficult to be used because of testing noise during operation. Besides，extended time was required to access the blade［14］. However，the above mentioned studies indicated that surface dynamic characters monitoring could help detect blade fault.

Optical fiber possesses the advantages of immu?nity to electromagnetic interference，corrosion resis?tance，and so on［15］. Therefore，optical fiber based sensors have increasingly applied to SHM in their to?tal life cycles［16-18］. Among the optical fiber sensors，fiber Bragg grating（FBG）sensors are proposed in?creasingly for strain monitoring in structures be?cause of their real time response，accurate perfor?mance［19-22］.

Mieloszyk et al. used a SHM system based on FBG sensors on a wind turbine model［23］. According to the results by Ref.［11］，monitoring of the outer skin of a blade is essential. Therefore，FBG sensors were assigned on the blade surfaces of a wind tur?bine. The dynamic behavior of wind turbine system was monitored. Measured noise caused by back?ground and wind turbines noise would indistinct the data［24-25］. Therefore，an algorithm based on dualtree complex wavelet transform（DT-CWT）is pro?posed to remove the system noise. The methodolo?gy is verified by experiments.

1 Methodology

Not just the wind turbine performance but its safety would be affected by wind evidently［26］. Ac?cordingly，wind loaded turbine systems are studied.The structural responses are monitored by FBG sen?sors.

Before the blade is fixed on a hub，five FBG sensors are glued on its surface. All the sensors are joined together and connected to broadband source（BBS） and demodulator. The sense principle is sketched in Fig.1.

To simplify the analysis，the working blade on the hub is considered as a cantilever. The blowing wind on the turbine can be simplified as uniform load at any moment. The computing model for a ro?tating blade at any instance is sketched in Fig.2.

Fig.1 Sketch of the sensing principle

Fig.2 Schematic diagram of a rotating blade at any instance

Therefore，the strain along the blade，i.e.ε（x）distributes as follows

where the parameters E and W（x）are Young’s modulus and section modulus in bending，respec?tively.

When a FBG sensor is connected to a broad?band source，a light travels through the FBG sen?sors. The light will interact with FBG sensors. Af?terwards，a narrow-band spectral output at the sen?sor，which is known as the Bragg wavelength，will be reflected. The reflected Bragg wavelength is de?termined by the well-known formula［27］

where the parameters λB，neffand Λ are the Bragg grating wavelength，the effective refractive index and the average grating period of the FBG，respec?tively.

As the physical parameter on the blade chang?es，the parameters neffand Λ will change. At the as?sumption of no temperature change，the Bragg wavelength will change with axial strain. The axial strain at position x is time variable and denoted as ε（x，t）. The wavelength shift ΔλBcan be expressed in the form［28］

where Sεis the relative strain sensitivity of a FBG.For the common used FBG，the core is made of sili?con oxide.Hence，the sensitivity Sεis 0.784.

As the wind turbine works，the strain on the blade will change with time. For a given detected point on the blade，the time variable strain ε（t）is captured by FBG sensor. Consequently，the moni?tored strain value can be obtained by

where the parameters λB（ε0，t0）and λB（ε，t），are the original and changed FBG wavelengths，respec?tively.

However，noise would disguise the strain varia?tion during blade vibration. Wavelet transform was applied to condition monitoring and denoising diag?nostics［29］.

Many different techniques，such as discrete wavelet transform （DWT）， second-generation wavelet transform（SGWT），empirical mode de?composition（EMD），etc.，have been proposed.Nevertheless，DT-CWT consistently outperforms others［30］. This method enjoys many attractive properties including nearly shift invariance and re?duced aliasing. The properties may be favorable to both the surveillance and diagnosis of rotating ma?chinery.

Therefore，an algorithm based on DT-CWT was proposed to reduce the noise and to extract the dynamic signals. By using dual-tree of wavelet fil?ters，the real and imaginary parts of the signal could be obtained. Therefore，limited redundancy was in?troduced and computational efficiency was pre?served［31］.

The captured signal ε（t）is discretized by a set of wavelets［32］

where ?（t）and ψ（t）represent the scaling functions and the band-pass wavelets，respectively. c（n）and d（j，n）are the associated scaling coefficients and the wavelet coefficients，and they can be obtained by the following equations

The wavelets and scaling functions，i.e. ψ（t）and ?（t），are complex-valued. The real and imagi?nary parts of the wavelets ψ（t）form a Hilbert trans?form pair. Therefore，the complex-valued wavelet is an analytic signal.

The above mentioned transform is approximate shift invariant. This feature is very important in pat?tern recognition and signal analysis［33］. In the trans?formation，two separate real discrete bases ψr（t）and ψi（t），i.e. the real and imaginary parts of the wavelets ψ（t），are used. They are two individually orthonormal wavelet transforms.

Consequently，the complex wavelet coefficient d（j，n），can be obtained by projecting the signal ε（t）onto 2j/2ψ（2jt ?n）［34］.

The coefficients，i.e. real and imaginary parts of d（j，n），are filtered individually. The approach is based on two filter bank trees and thus on two bas?es. Afterwards，a new complex wavelet coefficient dcnew（j，n）could be obtained by using the original phase and new wavelet magnitude

where mdand α are the filtered magnitude and the original phase，respectively.

2 Experimental Setup

An experimental program is designed to evalu?ate the efficiency of the proposed methodology. The wind turbine （type of NE-100S），as shown in Fig.3（a），is developed by Jiangsu Naier Wind Pow?er Technology Development Co.，Ltd. The length of each blade is 550 mm. Three blades are mounted on a hub. The diameter of the hub is 200 mm. FBG sensors are glued on blade surfaces. A commercial FBG demodulator from Micron Optics （model# SM130），as shown in Fig.3（b），is used in experi?ments to capture strain data.

Fig.3 Schematic of experimental setup for wind blade monitoring

Behaviors of two blade systems are studied ex?perimentally. For the first kind，all the blades are in good conditions. Five FBG sensors are glued on one of the blades. The monitored blade is denoted as Blade A. For the second kind，one of the good blades，except Blade A，is replaced by Blade B.For Blade B，the free end is broken. Blade A and Blade B are shown in Fig.3. The details of the two blades are shown in Table 1.

Table 1 Original details of FBG sensors

The original spectrums of the FBG sensors on Blade A is shown in Fig.4.

Fig.4 Original spectrums of the FBG sensors on Blade A

After the blades are fixed on the hub，the whole system is wind loaded by an electric fan. The blowing velocity is 2.4 m/s.

For a real scale wind turbine，it may not rotate if the wind velocity is too small. In order to well un?derstand a real scale wind turbine，two working con?ditions are considered. Firstly，no rotation of the blades is allowed by fixing the hub as it is subjected to wind. Secondly，the blades are set free and they can spin under wind load.

3 Testing Results

As the blades are wind loaded，they will de?form. The central wavelengths of the glued FBG sensors will change and be recorded by the demodu?lator. The captured strain data can be obtained ac?cording to Eq.（4）.

3.1 Blowing with fixed hub

As the wind turbine does not spin around the hub，the monitored results by FBG sensors are pre?sented in Figs.5 and 6.

As shown in Fig.5，if all the blades are in good conditions，the data oscillate evenly with wind. For the values of FBG5，both the peak value and the valley one are 1 micro-strain smaller than those of other FBGs. The strain curve offset down integral?ly. The testing error may be caused by the instru?ment. The testing error for the FBG demodulator is about 1 micro-strain.

If one of the blades is broken，the testing re?sults on different blades are displayed in Fig.6. The data goes up and down obviously even if the hub is fixed.

Fig.5 Results on Blade A as all the blades are in good conditions as the hub is fixed

3.2 Blowing freely

At the same blowing velocity，the monitored results when the turbine rotates about the hub are shown in Figs.7 and 8.

Shown in Fig.7 are strain data monitored by the FBG sensors on Blade A，as all the blades are in good conditions. Obviously，the stain of FBG5 is smaller than others. The reason lies in the position of FBG5. FBG5 is glued near to the free end of the cantilever. Consequently，the structural strain is smaller.

The strain data on Blade A and Blade B，as Blade B is broken，are displayed in Fig.8. The da?ta on Blade A shares the same change with Blade B at the same position. Nevertheless，the value is different at the same time. The initial variation trend is different during the testing for each blade.It is caused by different deformation on different blades. In the first 0.5 s，the testing surface of Blade A is undergoing increasing compressed.However，the surface of Blade B is gradually elongated.

Fig.6 Results as Blade B is broken with fixed hub

Fig.7 Monitored strain data as all rotating blades are in good conditions

The strain data on Blade A change with time apparently when all the blades are in good condi?tions. As one of the blades is broken，the data varia?tion is more obvious as shown in Fig.8（a）compared with Fig.7.

In order to study the strain character further more，the timely strain data is discretized，filtered and recomposed.

3.3 Data processing

The captured strain data is processed in order to extract the feature during turbine rotation.

Fig.8 Strain data on rotating Blades A & B as Blade B is broken

Firstly，the strain data is discretized into four layers of wavelets after Hilbert transformation. The real and imaginary parts of the wavelet coefficients，as all the blades are in good conditions，are indicated in Figs.9（a）and（b），respectively. In Figs.9（a）and（b），the horizontal and vertical axes represent num?bers and amplitudes for four layers，respectively.

Secondly，the real and imaginary parts of the coefficients are filtered，respectively. Afterwards，the wavelet coefficients are obtained according to Eq.（8）and the result is sketched in Fig.10.The ver?tical axes，d1—d4，represent amplitudes in four dif?ferent subbands.

Fig.9 Wavelet coefficients after Hilbert transformation

Fig.10 Mode of the filtered wavelet coefficient

Finally，the signal is reconstructed according to Eqs.（5—8）.

The reconfigured strain at each monitoring point as all the blades are good is illustrated in Fig.11.

It is obviously in Fig.11 that strain on each site varies similarly. It is all but periodical changes dur?ing the wind turbine blade rotation. The period for individual point is about 0.3 s. The strain values are almost the same except for the points detected by FBG4 and FBG5. The reason lies in that the bend?ing moment is very small near the free end of the blade，whose value is zero at the free end. The equal strain illustrates that the blade is designed with an uniform strength.

The strain data for another blade system is pro?cessed and illustrated in Fig.12.

Fig.11 Processed strain data on Blade A as all blades are good

Fig.12 Processed strain on Blades A & B as Blade B is broken

Apparently，the strain data fluctuate as one of the blades is broken. The strain shares similar vari?ety for each point on the same blade.

For Blade A and Blade B，the turn periods are all about 1 s.

Obviously，the good blade shares similar dy?namic character with the broken one. Therefore，it could be concluded that detection of one blade is suf?ficient for health monitoring of the whole wind tur?bine.

4 Conclusions

A non-destructive evaluation technique of wind blade SHM was investigated. The technique in?cludes application of FBG sensors.

When the wind turbine rotates，the strain on blade changes periodically. The gravity of the blades may affect the strain value. However，it would not cover the variation trend. If one of the blades is bro?ken，the strain fluctuates apparently. Furthermore，periods of strain change will be different. The rea?son lies in the structural inherent frequency change caused by broken blade. Therefore，variation of strain period or frequency could be a reference for SHM.

According to the tests，no matter whether the turbine revolves or not，the strain difference could be monitored by the proposed method.

The experimental testing results indicate that sensing of one blade is adequate for monitoring the whole wind turbine in service. On the blade being monitored，one FBG sensor is enough for the job.However，the sensor should be placed far away from the free end.