Renewable Energy 79 (2015) 209e218
Contents lists available at ScienceDirect
Renewable Energy journal homepage: www.elsevier.com/locate/renene
Transformation algorithm of wind turbine blade moment signals for blade condition monitoring Jae-Kyung Lee a, Joon-Young Park a, Ki-Yong Oh b, *, Seung-Hwan Ju a, Jun-Shin Lee a a b
Division of Offshore Wind Power, KEPCO Research Institute, Daejeon, Republic of Korea Department of Mechanical Engineering, University of Michigan, 2350 Hayward Street, Ann Arbor, MI 48109-2125, USA
a r t i c l e i n f o
a b s t r a c t
Article history: Received 21 March 2014 Accepted 12 November 2014 Available online 2 December 2014
To simplify signal analysis on wind turbine blades and enable their efficient monitoring, this paper presents a novel method of transforming blade moment signals on a horizontal axis 3-blade wind turbine. Instead of processing 3-blade moment signals directly, the proposed algorithm transforms the three sinusoidal signals into two static signals relative to the center of blade rotation through vector synthesis and coordinate transformation, and eliminates frequency components due to blade rotation from the obtained signals. Moreover, as an alternative to a rotational sensor, a blade rotation angle estimator is introduced. Its effectiveness was confirmed through simulations and field tests on an actual wind turbine. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Wind turbine Blade monitoring Health monitoring Signal transformation
1. Introduction Wind energy has taken center stage because of its clean, safe and practical aspects, and plenty of wind farms have been constructed as a means of replacing fossil power plants. Though wind turbines are generally designed to operate for over 20 years, a Condition Monitoring System (CMS) is required to enhance wind turbine reliability and availability. The main function of the CMS is to detect turbine faults at their early stage, and thereby to prevent the risk of turbine accidents. The CMS can be mainly classified into three types according to the kind of used sensor or monitoring object: vibration-based monitoring, oil debris-based monitoring and blade monitoring [2,3]. Among these, the blade monitoring has been generating much interest recently and has been actively studied, because blade failure accounts for a substantial proportion of total failures from the aspects of downtime and costs as shown in Fig. 1. As part of this trend, Germanischer Lloyd (GL) guideline for the certification of CMS recommends monitoring wind turbine blades as well as other main components such as a main bearing, a gearbox and a generator [4]. Several research studies on wind turbine blade monitoring have been in progress to secure wind turbine blade integrity. Papasalouros set up a wind turbine blade health monitoring system using
* Corresponding author. Tel.: þ1 734 353 5191; fax: þ1 734 647 9379. E-mail address:
[email protected] (K.-Y. Oh). http://dx.doi.org/10.1016/j.renene.2014.11.030 0960-1481/© 2014 Elsevier Ltd. All rights reserved.
Acoustic Emission (AE) sensors on a wind turbine of NEG-MICON NW48/750 [5] as shown in Fig. 2. Apart from this method, various monitoring techniques have been introduced to monitor the integrity of wind turbine blades, such as power characteristic monitoring [6,7], spectral analysis and order analysis [7], strain monitoring-based algorithms [7], inertial sensing [8], optical coherence tomography [8], bicoherence based on electrical power [9] and AE based on pattern recognition [10,11]. Especially, a Fiber Bragg grating (FBG) sensor has been recently applied to structural health monitoring of wind turbines, because it has the advantage of a direct physical correlation between the measured Bragg wavelength and strain, and has immunity to lightning and electric shortage, and moreover, uses far fewer cables and channels [12]. However, large amounts of its measured data and much computational load in implementing monitoring algorithms including frequency analysis require a high-speed data transmission unit and a high-speed data processing unit, respectively. To solve the problems aforementioned, the KEPCO Research Institute developed a novel signal transformation algorithm that is very effective in monitoring turbine blades in real time. The proposed algorithm first synthesizes three raw moment signals into two orthogonal signals relative to the rotating blades, and then, transforms the two orthogonal signals into two final transformed signals relative to the center of blade rotation in order to eliminate rotational components with a fundamental frequency from the obtained signals. When performing this transformation, the
210
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
Fig. 1. Failure rate of wind turbine components [1].
proposed method uses the rotation angle estimated by the Phase Locked Loop (PLL) instead of using a sensor, thereby achieving additional cost reduction. Through the proposed algorithm, three sinusoidal signals can be finally simplified into two static signals, which can reduce computational burden in implementing monitoring algorithms and signal analysis. The effectiveness of the signal transformation algorithm is confirmed through simulations and field tests with an actual wind turbine.
estimator using the PLL estimates the angular velocity of the turbine blades that will be used for eliminating rotational components from the two orthogonal signals. Finally, the transformed signals for blade monitoring are calculated through the coordinate transformation using a rotation matrix. Each procedure will be explained in more detail in the following subsections.
2. Transformation algorithm of blade moment signals The proposed algorithm presents the effective algorithm of moment signal transformation for a 3-blade horizontal axis wind turbine, which can be used for validating the integrity of its blades. The basic idea of this algorithm is to transform the 3-blade raw moment signals into 2-orthogonal signals, thereby making a blade monitoring process more simplified and efficient. Fig. 3 shows the overall procedures of the moment signal transformation algorithm. First, the 3-blade raw moment signals are obtained by fiber optic sensors from the leading edge sides of the blade roots, which will be described in detail in Section 4. Second, instead of processing the 3-blade moment data directly, the signals are transformed into a reference signal and its orthogonal signal, which are attached to a rotating blade, by using the directquadrature transformation [13]. Third, the blade rotation angle
Fig. 3. Overall transformation algorithm of blade moment signals for blade monitoring.
Fig. 2. Schematic illustration of structural condition monitoring using AE sensors and power characteristics.
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
211
Fig. 6. Estimated blade rotation angle.
estimated rotational angle. Fig. 6 shows an example of the estimated rotation angle. 2.3. Coordinate transformation using rotation matrix Fig. 4. Schematic of transforming 3-blade raw signals into 2-axis transformed signals.
2.1. Transformation of 3-phase raw signals to 2-orthogonal signals Fig. 4 shows the schematic of transforming 3-blade moment signals into 2-axis transformed signals. First, three blade raw moment signals are calculated relative to the axes of J1, J2, J3 attached rigidly to three rotating blades at the center of blade rotation. If the axes of Ra, Rb are assigned parallel and orthogonal to J1 at the center of blade rotation, respectively, then the three raw signals can be transformed into two orthogonal signals relative to Ra, Rb through the equation (1) and (2)
The algorithm described above simplifies three blade moment signals into two orthogonal signals relative to a rotating blade. However, the processed signals are still difficult to analyze, because the signal observer stands outside the two rotating orthogonal coordinates. If the rotational components caused by the blade rotation can be eliminated, the 2-axis synthetic signals can be further simplified. To this end, the final transformed signals are obtained through the following coordinate transformation using a rotation matrix, as can be seen in Fig. 4:
JT ¼ RðQÞJR ;
(3)
where
Ta Ra cosðqÞ ; JR ¼ ; RðQÞ ¼ Tb Rb sinðqÞ
sinðqÞ : cosðqÞ
Ra ¼ J1 þ aJ2 þ bJ3
(1)
JT ¼
Rb ¼ gJ2 þ dJ3
(2)
Therefore, through this transformation equation (3), the frequency components of the 2-axis orthogonal signals due to rotational speed can be eliminated. To conclude, the proposed signal transformation algorithm can reduce the number of signals to be processed for evaluating blade integrity, thereby simplifying their analysis and enabling more efficient monitoring of wind turbine blades.
Here, the constants of a, b, g, d can be simply obtained from the geometry between three blades.
2.2. Blade rotation angle estimator To eliminating rotational components from the two orthogonal signals, the blade rotation angle should be required. In this paper, the blade rotation angle estimator of Fig. 5 is applied to estimate the rotation angle instead of measuring it directly through a sensor such as a rotary encoder, which is called the PLL [13] and is another advantage of the proposed algorithm. The angular velocity of the turbine blades can be calculated by controlling the error between which is obtained by using the Ra and its estimated signal Ra,
3. Simulation results To verity the effectiveness of the proposed algorithm, several simulations have been performed. First, the transformation of ideal sinusoidal signals was simulated to evaluate its basic performance. Second, the transformation of differently biased signals was performed to evaluate the effect of the imposed bias on each blade
Fig. 5. Blade rotation angle estimation algorithm.
212
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
Fig. 7. Simulation results of ideal sinusoidal signals.
transformed signal. In this case, a bias of 15%, 10%, and 20% of each signal was included in J1, J2 and J3, respectively. Third, a bias was suddenly applied to a blade moment signal at 0.5 s to simulate a situation where the weight of a blade dramatically changes such as rotor blade icing. Finally, a biased signal with an impulse at 0.5 s was simulated to show the blade behavior when an obstacle impacts with a blade, for example, a collision with a bird.
3.1. Ideal sinusoidal signals The ideal sinusoidal signals were generated and applied to evaluate the effectiveness of the algorithm. This condition means that a wind turbine operates at a constant wind speed. The applied signals have zero bias and a frequency of 9.54 Hz with a 0, þ2/3p, and 2/3p phase difference, respectively. Fig. 7 shows the simulation results with the ideal sinusoidal signals. The three raw signals were well synthesized into two orthogonal signals and the blade rotation angle estimator tracked the blade rotation well. In the early stage of the blade rotation angle estimator, there existed estimation errors due to its initial value errors. As the blades went on, however, the PI controller in the rotation angle estimator minimized the error between the estimated rotation angle and its actual value, and thereby the transformed signals were stabilized to certain values. These simulation results exactly show a phenomenon when an observer watches the two orthogonal signals at the outside of the rotating coordinate frame {J1, J2, J3} or {Ja, Jb}. If the observer stands at the center of the rotating coordinate frame, the two orthogonal signals can be regarded as stationary ones as shown in Fig. 7. Therefore, it is confirmed that the proposed
algorithm can eliminate rotational components, a major obstacle to signal processing, from the blade moment signals.
3.2. Differently biased signals In the real world, the moment signals for three blades do not have an identical amplitude with each other and have biases caused by their own weight and initial calibration. To evaluate the effect of the imposed bias on each transformed signal, three sinusoidal signals, which have a bias of 15%, 10%, and 20% of each signal, were simulated as in Fig. 8. In spite of the differently imposed biases, the rotation angle estimator showed good performance, and the transformed signals Ta and Tb had only small fluctuations of 0.14% and 0.71% in magnitude, respectively, which may not be critical for signal processing.
3.3. Biased signals with step change A bias suddenly applied to a blade moment signal is also simulated to describe a situation where the weight of a blade dramatically changes such as rotor blade icing. Fig. 9 shows the simulation results when the amplitude of one signal jumped 10% at 0.5 s. As seen in this figure, the transformed signals Ta and Tb showed considerable oscillations of 8.3% and 4.7% in magnitude, respectively, and their FFT results showed a peak frequency shift from 9.54 Hz to 14 Hz. Therefore, such an added-mass problem can be easily detected through further signal processing of the transformed signals.
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
Fig. 8. Simulation results of differently biased signals.
Fig. 9. Simulation results of biased signals with step change.
213
214
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
Fig. 10. Simulation results of biased signals with impulse.
Fig. 11. Overall configuration of blade monitoring system and its implementation.
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
215
Fig. 12. Field test results when wind turbine starts up.
3.4. Biased signals with impulse To simulate the impulse response such as the impact on a turbine blade caused by a collision with a bird or hail, an impulse was applied to a raw signal at 0.5 s. From the results of Fig. 10, it can be observed that the impulse affected the transformed signals and the estimated rotation angle for only a short-term period immediately after its application, which can be also detected by signal processing on the transformed signals.
4. Field tests For the field tests of the developed algorithm, a blade monitoring system was installed on a commercial wind turbine of Doosan Heavy Industries' WinDS3000 in the YeongHeung wind farm of Korea. Fig. 11 shows the overall configuration of the blade monitoring system and its actual implementation, respectively [12]. FBG strain gauges with 1 mε accuracy were installed at the roots of the blades to monitor the moment exerted on the blade roots. FBG temperature sensors with ±0.5 C accuracy were also attached at each strain gauge location to compensate the strain due to temperature variations. Three raw moment signals were obtained from the strain and the temperature values, which were measured from the leading edge sides of the blade roots at a sampling rate of 100 Hz in real time [12]. A wireless network system was constructed independently to prevent signal interference
between the blade monitoring system and the control system of the wind turbine. The data measured from the rotor were transmitted to servers in the control room of the YeongHeung wind farm and the KEPCO Research Institute. The proposed algorithm was tested under four operation modes of the WinDS3000 as follows: 1) Start-up Mode where the wind turbine starts up 2) Power Generation Mode where the wind turbine is generating near the cut-in wind speed 3) Power Generation Mode where the wind turbine is generating above the cut-in wind speed 4) Stop Mode where the wind turbine is stopping.
4.1. Start-up mode The field test results of the proposed algorithm when the wind turbine started up are given in Fig. 12, where the graph legend of ‘Blade a’ denotes signals of J1, J2, and J3; ‘Rot a’ those of Ra and Rb; ‘Tran a’ those of Ta and Tb; ‘Theta’ q; and all the data were 10min average values. It can be observed in this figure that the proposed algorithm worked successfully and the rotation angle estimator tracked the blade angle well. Here, just before the 400th sample, the turbine operation changed from the STOP mode to the START-UP mode and adjusted the blade pitch angle from 90 to zero.
216
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
Fig. 13. Field test results when wind turbine is generating near cut-in wind speed.
Then, its blades started to rotate in accordance with this pitch motion, and consequently, the blade moments were changing. After the wind turbine started to generate electric power, the transformed blade moments settled down and showed repetitive characteristics.
4.2. Power generation mode near cut-in wind speed Fig. 13 shows the field test results of the proposed algorithm while the wind turbine was generating electric power near the cut-in wind speed. The rotation angle estimator tracked the blade rotation angle successfully and the transformed data showed identical characteristics. Fig. 14 shows the FFT results of blade signals to check the similarity between the original and the transformed data, and Table 1 shows their peak frequencies and amplitudes. The FFT results of the blade raw moment signal J1 show a peak frequency of 0.133833 Hz and its harmonic frequencies. It can be also observed that the transformed data, Ta and Tb, showed the same peak frequencies as those of the raw signal, and their amplitudes were much smaller than those of the original data. This comparison confirms that by decoupling the blade rotating frequency from the blade raw data, the data analysis using the transformed data can be more simple and efficient.
Table 1 Peak frequencies and magnitudes of original and transformed data.
Fig. 14. FFT results of blade signals when wind turbine is generating near cut-in wind speed.
Peak frequency of ƩJi, Ta, Tb (Hz) Amplitude of ƩJi Amplitude of Ta Amplitude of Tb
0.133833 2618 0.08265 62.5
0.27166 129.99 0.4235 6.72
0.41499 19.15 1.4736 23.59
0.54332 13.47 1.37125 0.13416
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
217
Fig. 15. Field test results when wind turbine is generating above cut-in wind speed.
4.3. Power generation mode above cut-in wind speed Figs. 15 and 16 show another field test results of the proposed algorithm while the wind turbine was generating electric power above the cut-in wind speed. In a similar way to the results near the cut-in speed, the transformed signals reflected the characteristics
of their original ones well and the rotation angle estimator also showed an excellent performance. Especially, the FFT results of Fig. 16 show their fundamental frequency movement corresponding to the change of the rotation speed. 4.4. Stop mode Fig. 17 shows the field test results where the wind turbine performed its STOP Mode operation. While the blade angle was changed to stop the turbine, the proposed algorithm gave a consistent performance of maintaining its identical characteristics, and worked successfully independently of the pitch angle movement of the wind turbine. 5. Conclusion
Fig. 16. FFT results when wind turbine is generating above cut-in wind speed.
In this study, a new signal transformation method was introduced for blade monitoring of a horizontal axis 3-blade wind turbine. The newly developed algorithm is designed to simplify the 3blade moment signals into two static signals, which leads to enhanced signal processing efficiency and reliability. In addition, the algorithm considers the viewpoint of an observer standing outside the rotating blades, and as a result, significantly reduces the effect of a blade rotational frequency, thereby ensuring more precise blade monitoring. The usability of the developed algorithm was successfully verified by simulations and field tests through wind
218
J.-K. Lee et al. / Renewable Energy 79 (2015) 209e218
Fig. 17. Field test results when wind turbine stopping.
turbine operations with actual turbine data. The developed algorithm in this study will be used for further development of blade monitoring techniques. Acknowledgments This work was supported by the New & Renewable Energy Program of the Korea Institute of Energy Technology Evaluation and Planning (KETEP) grant funded by the Korea government Ministry of Knowledge Economy (Project No. 2011T100100307). References [1] Oh KY, Lee JK, Park JY, Kim BJ, Lee JS. Development of blade root monitoring system with fiber optical sensors for multi-MW class wind turbine. EWEA (European Wind Energy Association); 2011. Annual event. [2] Hyers RW, McGowan JG, Sullivan KL, Manwell JF, Syrett BC. Condition monitoring and prognosis of utility scale wind turbines. Energy Mater 2006;1(3): 187e203. [3] Crabtree CJ. Survey of commercially available condition monitoring systems for wind turbines. Durham University; 2010. [4] Chap. 5 Guideline for the certification of condition monitoring systems for wind turbines. Germanischer Lloyd; 2013. p. 3e5.
[5] Papasalouros D, Tsopelas N, Ladis I, Kourousis D, Anastasopoulous A, Lekou DJ, et al. Health monitoring of NEG-MICON NM48/750 wind turbine blades with acoustic emission. In: 30th European conference on acoustic emission testing & 7th international conference on acoustic emission; 2012. [6] Caselitz P, Giebhardt J. Rotor condition monitoring for improved operational safety of offshore wind energy converters. J Sol Energy Eng 2005;127:253e61. [7] Hameed Z, Hong YS, Ahn YM, Song CK. Condition monitoring and fault detection of wind turbines and related algorithms: a review. Renew Sustain Energy Rev 2009;13:1e39. €nder J, Rusborg J. Fundamentals for [8] Lading L, McGugan M, Sendrup P, Rheinla remote structural health monitoring of wind turbine bladesda preproject. Denmark: Risø National Laboratory; 2002. [9] Jeffries WQ, Chambers JA, Infield DG. Experience with bicoherence of electrical power for condition monitoring of wind turbine blades. IEE Proc Vis Image Signal Process 1998;145(No. 3):141e8. [10] Dutton AG, Blanch MJ, Vionis P, Lekou D, van Delft DRV, Joosse PA, et al. Acoustic emission condition monitoring of wind turbine rotor blades: laboratory certification testing to large scale in-service deployment. In: Proceedings of the 2001 European wind energy conference; 2001. [11] Schulz MJ, Sundaresan MJ. Smart sensor system for structural condition monitoring of wind turbines. subcontract report. Colorado: NREL; 2006. SR500e40089. [12] Oh KY, Park JY, Lee JS, Epureanu BI, Lee JK. A novel method and its field tests for monitoring and diagnosing blade health for wind turbines. IEEE Trans Instrum Meas ; [accepted]. [13] Oh JK. Maximum power tracking control of a wind turbine with permanent magnet synchronous generator. Master's Thesis. Postech; 2011.
