Condition Monitoring and Fault Diagnosis of a Wind Turbine with a Synchronous Generator using Wavelet Transforms

Transcription

1 Condition Monitoring nd Fult Dignosis of Wind Turbine with Synchronous Genertor using Wvelet Trnsforms Wenxin Yng 1, P. J. Tvner 1, Michel Wilkinson 2 1 New & Renewble Energy Group, School of Engineering, Durhm University, Durhm DH1 4RL, UK 2 Grrd Hssn & Prtners Ltd, St Vincent Works, Bristol, BS2 0QD Keywords: wind turbine, condition monitoring, fult dignosis, wvelet trnsforms. Abstrct Some lrge wind turbines use low speed synchronous genertor, directly-coupled to the turbine, nd fully rted converter to trnsform power from the turbine to mins electricity. This pper considers the condition monitoring nd dignosis of mechnicl nd electricl fults in such vrible speed mchine. The ppliction of wvelet trnsforms is investigted becuse of the disdvntges of conventionl spectrl techniques in processing instntneous informtion in turbine signls derived from the wind, which is vrible nd noisy. A new condition monitoring technique is proposed which removes the negtive influence of vrible wind in mchine condition monitoring. The technique hs verstile function to detect mechnicl nd electricl fults in the wind turbine. Its effectiveness is vlidted by experiments on wind turbine condition monitoring test rig using permnent-mgnet synchronous genertor, which cn be driven by erodynmic forces from drive motor controlled by n externl model, representing wind nd turbine rotor behviour. Within the technique wvelet trnsforms re employed for noise cncelltion nd re extended to dignose fults by tking dvntge of their powerful cpbilities in nlysing non-sttionry signls. The dignosis of wind turbine rotor imblnce in the will be used s n illustrtive exmple, herlding the possibility of detecting wind turbine mechnicl fults by power signl nlysis. 1 Introduction The development of wind turbine technology hs benefited from the government decisions fvourble to green or renewble power. Wind turbines re becoming economiclly vible lterntive to conventionl fossil-fuelled power genertion. In some countries, notbly Germny nd Denmrk, wind turbines hve been plying vitl role in the power network, lthough not without some problems [1]. However, wind turbines do experience filures [2], due to their vrible lod condition nd ggressive operting environment, however, turbines re beginning to show relibility tht is better thn other forms of power genertion, for exmple diesel genertors. So developing economic condition monitoring nd fult dignosis techniques for them would be highly desirble nd this will be especilly importnt if they re deployed offshore. SCADA techniques re being pplied widely to wind turbines but the dt rte, once every 5-10mins, is too slow for most rotting mchine fult dignosis. There re mny techniques developed in electric power production, erospce, mrine propulsion, nd other process industries tht could be pplied to wind turbines [3]. However, the results to dte hve not proved stisfctory, due to the peculirities of the wind turbine, tht is slow nd vrible speed, t lest for the lrger types. In recent yers, some efforts hve been mde to improve this sitution [4]. However, the mjority of wind turbine condition monitoring nd fult dignosis techniques proposed hve used the Fourier Trnsform (FT), which is less cpble of solving the problem due to its shortcomings in deling with non-sttionry signls. In view of this, the potentil ppliction of the wvelet trnsform to the condition monitoring nd fult dignosis of wind turbines is investigted in this pper s n extension of the work described in [6].The Discrete Wvelet Trnsform (DWT) is used for noise cncelltion s the signls from the wind turbine contin noise which is difficult to remove by using conventionl filter with fixed cut-off frequencies.

2 The Continuous Wvelet Trnsform (CWT) is used for feture extrction. A new technique, inspired by the torque-speed dt obtined from series of full nd hlf-lod experiments on wind turbine, is proposed for ssessing the running condition of the wind turbine. The effectiveness of the technique is vlidted by the detection of genertor winding nd rotor imblnce fults on the test rig. Experimentl results show tht the ppliction of the DWT drmticlly enhnces the vibility of this technique for wind turbines. In order to further simplify nd reduce the cost of wind turbine condition monitoring nd fult dignosis, the possibility of detecting wind turbine mechnicl fults by power signl nlysis is lso investigted with the id of the CWT. 2 Test Rig In order to simulte the effects of wind turbines working under different conditions nd develop the new condition monitoring nd fult dignosis techniques, wind turbine test rig ws built, s shown in Fig.1. The test rig comprises 50kW DC vrible speed drive controlled motor, two-stge gerbox nd three-phse synchronous permnent-mgnet genertor. The genertor hs 84 coils on the sttor, 108 permnent-mgnets on the rotor nd three phse rectified output fed to resistnce lod bnk. The system is instrumented using Lbview so tht vriety of wind speed inputs cn be pplied nd the relevnt signls cn be collected from the drive trin nd terminls of the genertor. In the experiments, the speed of the DC motor is controlled by n externl model, in which both the properties of nturl wind nd the mechnicl behviour of turbine rotor re incorported. Both genertor electricl nd wind turbine mechnicl fults were simulted on the test rig s follows: Fig.1: Wind turbine test rig. A genertor sttor winding fult ws simulted by simultneously shorting 3 coils on the sttor; A full short circuit fult ws simulted by connecting the phse terminls of the genertor to ground; A rotor imblnce fult ws simulted by ttching mss to the outer surfce of the genertor rotor. 3 Condition monitoring technique In order to develop n effective technique for monitoring the running condition of synchronous genertor wind turbines, series of full nd hlflod tests were conducted on the test rig. Different lods nd fult conditions were pplied to the rig during the experiments. The shft torque nd rottionl speed dt mesured in the different cses re plotted in Fig.2. The polynomil equtions fitting to these dt re lso derived with the id of the polynomil curve fitting technique [7]. In Fig.2, the fitting curves nd the corresponding polynomil equtions re given for fcilitting nlysis. Fig.2: The shft torque nd speed dt mesured in the full nd hlf-lod tests. From Fig.2, it cn be seen tht the genertor exhibits different torque-speed chrcteristics under different lod nd fult conditions. Most interestingly, the genertor shows significnt chnge in torque-speed chrcteristic when fults occur, regrdless of the lod condition. This suggests tht the torque-speed curve could be sensitive indictor of the running condition of synchronous genertor wind turbine. Inspired by this ide, new condition monitoring technique will be developed in this pper. According to [8], the following reltions exist between the torque nd speed of synchronous genertor wind turbine:

3 ω s = ωr rg T ωs X (1) T pm T where ω r represents the rottionl speed of the DC motor, ω s the rottionl speed of the synchronous genertor nd r g the ger rtio, X the synchronous rectnce of the genertor, T pm the mechnicl torque creted by wind force, nd T the torque creted by genertor. With the id of Eq.(1) the proposed technique will use criterion C s verstile function for monitoring the running condition of the wind turbine, i.e. C = T ωs (2) where both torquet nd speed ωs re mesured on the genertor shft. Criterion C cn be used not only to monitor the presence of drive trin mechnicl fult, but lso to detect the occurrence of genertor electricl fult becuse drive trin mechnicl fult will hve response in T pm nd genertor electricl fult will hve response in X. One more dvntge of criterion C is tht it is independent of the vrible wind demonstrted by the liner reltionship between the T pm nd ω r. Fig.3: The torque nd speed signls when the coils re well connected nd shorted periodiclly. From Fig.3, it cn be seen tht the shft torque nd rottionl speed re indeed very noisy s. So de-noising process is essentil before using them to clculte criterion C. By using the DWT, the signls re de-noised nd the results re lso plotted in Fig.3. Obviously, the noise contined in the originl signls hs been successfully removed. Then, C is clculted nd shown in Fig.4, in which the results obtined from the originl signls re lso shown for comprison. 4 Condition monitoring of the wind turbine In order to verify the effectiveness of C in wind turbine condition monitoring, two illustrtive exmples re given in the following Sections. 4.1 Sttor winding fult in the genertor A sttor winding fult in the genertor ws simulted on the test rig by simultneously shorting 3 coils instlled on the sttor of the genertor. When the connection stte of the coils ws chnged periodiclly, the torque nd rottionl speed signls were collected from the genertor shft. The time-wveforms of the signls re shown in Fig.3. Fig.4: Criterion C derived in the cse of genertor winding fult. Fig.4 shows tht the criterion increses sensibly when the coils re shorted, nd returns to norml when the coils re well-connected. Clerly, the sttor winding fult in the genertor hs been successfully detected using the proposed technique. The corresponding decrese of the synchronous rectnce X of the genertor when the coils re shorted ccounts for the increse of C. 4.2 Rotor Imblnce Fult One more proof of the effectiveness of C is given by the detection of rotor imblnce fult. It

4 is known tht wind turbine rotors re prone to imblnce due to dmge to bldes or unequl icing. Once the rotor is imblnced strong vibrtion t the shft rottionl frequency will be introduced into the system. It trvels long the drive trin, nd finlly rrives t the genertor nd dynmic ir-gp eccentricity is introduced. The chrcteristic frequency of this type of fult will be the rottionl frequency of the shft where the imblnce hppens. A rotor imblnce fult ws simulted by directly ttching 1.027kg weight mss to the outer surfce of the genertor rotor. The shft torque nd rottionl speed signls collected before nd fter the ttchment of the unblnce mss re shown in Fig.5. Fig.5: The torque nd speed signls in the cse of rotor imblnce fult. The DWT ws pplied to remove the noise before performing the clcultion of C. The denoised signls re lso plotted in Fig.5, from which significnt fluctution of shft torque due to the unblnce mss is clerly observed. By using the de-noised signls, C is clculted nd the results re shown in Fig.6. Fig.6: C in the cse of rotor imblnce fult. From Fig.6, it is found tht criterion C shows lrge fluctution when the unblnce mss is ttched, becoming stble gin s soon s the unblnce mss is removed. Thus, it cn be concluded tht rotor imblnce cn be successfully detected using the proposed technique. 5 Dignosing mechnicl fults The vibro-coustic method is trditionlly populr for dignosing the mechnicl fults in rotting mchines. However, this pproch involves the use of considerble number of trnsducers with consequent cost. For exmple system recently developed by Bently Nevd [9] includes 8 ccelerometers (2 for ncelle, 1 for min bering, 3 for drive trin, nd 2 for genertor), 2 displcement trnsducers nd one keyphsor trnsducer. These trnsducers re connected to dedicted dt cquisition nd processing device known s Dynmic Scnning Module (DSM). This sort of condition monitoring systems is sophisticted nd costly nd probbly cnnot be justified on wind turbine except for the most high risk loctions. Furthermore, the compct nture of the wind turbine ncelle mens tht trnsducer instlltion is difficult. Reducing the number of trnsducers required for condition monitoring wind turbine is strongly recommended. To meet this im, the possibility of detecting wind turbine mechnicl fults using the genertor power signl nlysis s proposed in [3] is investigted. If this works, the number of trnsducers fitted to wind turbine could be significntly reduced. To prove the fesibility of this ide, rotor imblnce fult is dignosed using the terminl power signl of the genertor. Before nd fter the unblnce mss ws ttched, the phse current, voltge nd power signls were monitored nd their time-wveforms re shown in Fig.7. In order to detect the fulty feture CWT is pplied to the power signl, nd the results re shown in Fig.8. From Fig.8, it is clerly seen tht when the imblnce occurs chrcteristic frequency ppers t bout 0.44Hz, corresponding to the rottionl speed of 26 rev/min (see Fig.5). The feture extrcted by the CWT from the power signl is in ccordnce with the theoreticl prediction thus demonstrting tht the detection of mechnicl fults by nlyzing genertor terminl power signl is fesible for wind turbine.

5 Consortium, EP/D034566/1 nd the EPSRC Engineering Doctorte Scheme, GR/R99737/01. The uthors re grteful for the dvice nd collbortion of Supergen prtners. Fig.7: Electricl signls in the cse of rotor imblnce fult. Fig.8: The CWT mp of the power signl. 6 Conclusions From these preliminry investigtions, it hs been concluded tht: When the condition monitoring test rig suffered genertor electricl or mechnicl fult, the chnge of running condition ws correctly detected using the condition monitoring technique proposed, bsed upon the criterion C, rther thn by using trditionl vibrtion signls. The DWT hs powerful de-noising cpbility for wind turbine signls. The ppliction of the DWT strengthens the vibility of the proposed technique in detecting chnges of wind turbine running condition. The fesibility of detecting wind turbine mechnicl fult through nlyzing the genertor power signl using the CWT technique hs been demonstrted. It should lso be possible, using shft torque, wind velocity nd power trnsducers, to develop simple nd chep wind turbine condition monitoring nd fult dignosis system, without resorting to costly vibrocoustic trnsducers. References [1]. E.ON Netz. Wind report Technicl Report, [2]. P.J. Tvner, J. Xing nd F. Spinto, Relibility nlysis for wind turbines, Wind Energy, 10 pp.1-18 (2007). [3]. W. Q. Jeffries, J.A. Chmbers nd D.G. Infield, Experience with bicoherence of electricl power for condition monitoring of wind turbine bldes, IEE Proceedings, Vision, Imge nd Signl Processing, 45(3) (1998) [4]. P. Cselitz nd J. Giebhrdt, Rotor condition monitoring for improved opertionl sfety of offshore wind energy converters, ASME Trnsctions, Journl of Solr Energy Engineering, 127 (2005) [5]. Z. Hmeed, Y.S. Hong, Y.M. Cho, S.H. Ahn nd C.K. Song, Condition monitoring nd fult detection of wind turbines nd relted lgorithms: A review, Renewble & Sustinble Energy Reviews, In Press (2007). [6]. M R. Wilkinson, F Spinto, P J. Tvner, Condition monitoring of genertors & other subssemblies in wind turbine drive trins. IEEE Int Conf SDEMPED, Crcow, Sept 2007 [7]. W.X. Yng, J.B. Hull nd M.D. Seymour, A contribution to the pplicbility of complex wvelet nlysis of ultrsonic signls, NDT & E Interntionl, 37 pp (2004). [8]. M. S. Srm, Synchronous mchines (Their theory, stbility, nd excittion systems), Gordon nd Brech Science Publisher, New York, [9]. T.J. Clrk, R.F. Buer nd J.R. Rsmussen, Wind power comes of ge, Orbit, pp (2004). Acknowledgements The work presented in the pper ws funded by the EPSRC Supergen Wind Energy Technologies