A New Method for Detection and Evaluation of Winding Mechanical Faults in Transformer through Transfer Function Measurements

Transcription

1 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 A New Method for Detecto ad Evaluato of Wdg Mechacal Faults Trasformer through Trasfer Fucto Measuremets Mehd BIGDELI *, Mehd AKILIAN 2, ad Ebrahm RAHIMPOUR 3 Departmet of Electrcal Egeerg, Scece ad Research Brach, Islamc Azad Uversty, Tehra, Ira 2 Departmet of Electrcal Egeerg, Sharf Uversty of Techology, Tehra, Ira 3 ABB AG, Power Products Dvso, Trasformers, Bad Hoef, Germay * Emal: m.bgdel@srbau.ac.r Abstract Trasfer fucto (TF) s a acowledged method for power trasformer mechacal faults detecto. However the past publshed wors mostly dscovered how to specfy the faults levels ad pad less atteto to detecto of the type of faults usg comparso of TFs. whereas, t seems mportat for most of the applcatos to specfy the type of fault wthout opeg the ut. Ths paper presets a ew method based o vector fttg (F) to compare the TFs ad specfy the type, level ad locato of the fault. For developmet of the method, ad ts verfcato the requred measuremets are carred out o four model trasformers; uder tact codto, ad uder dfferet fault codtos (axal dsplacemet, radal deformato, dsc space varato ad short crcut of wdg) ad the TFs are determed. Employg F, the coeffcets of TFs are determed wth the requred accuracy. Usg those coeffcets, a ew dex s troduced to specfy the type, level ad locato of the fault the wdg. Covcgly good results were obtaed. Therefore t s beleved that ths fdg could be helpful fault dagoss actual power trasformer wdgs. Idex Terms Trasformer, Fault Dagoss, Measuremet, Trasfer Fucto, ector Fttg I. INTRODUCTION Power trasformers are amog the most mportat equpmets the electrcal power trasmsso ad dstrbuto systems ad occurrece of ay fault these trasformers wll reduces the power system relablty, other words creases the possblty of power supply terrupto. I addto, t wll mpose extra costs requred for ther mateace ad for ther trasportato to a repar factory. Due to the exstece of a strog competto the electrcal power supply dustry, the mportace of applcato of the trasformer motorg systems has creased over the tme []. Dfferet fault modes trasformers are classfed to two ma groups; the mechacal types, ad the electrcal types. Dfferet methods have bee offered for detecto of each of these faults, such as: a Ol aalyss (Dssolved Gas Aalyss, Furfural) b Partal dscharge measuremets c Delectrc respose aalyss d Trasfer fucto aalyss. Amog these methods, TF method s a well-ow method for detecto of the wdg mechacal faults [2]- [4]. TF method s a comparatve method; ths method a ew measuremet s put besdes a referetal measuremet. If the devatos were remarable, the drecto ad magtude of devatos should be studed ad aalyzed. Mechacal faults such as; axal dsplacemet (AD), radal deformato (RD), dsc space varato (DS) ad short crcut (SC) of the wdg affect the TFs ad cause dslocato of resoace frequeces ad moreover wll decrease or crease the magtude of resoace frequeces the TF. These chages wll cause a relevat shft the poles ad zeros of TF. Therefore, by evaluato ad comparso of the TFs, ay varato the structure of the wdg due to a mechacal damage ca be detected. TF evaluato ad ts comparso ca be doe tme doma or frequecy doma. The detfcato process s ormally carred out frequecy doma through specto of the ratoal fuctos. Some studes [5]-[6] have show that F method s a effectve ad accurate method for estmatg the trasformer TFs. I [7]-[8] have proposed a dex (based o TF estmato usg F) for determato of AD ad RD extet. That dex played ts role oly to specfy the level of AD or RD ad does ot specfy the type of fault. I addto, ther proposed dex oly perceves the varatos the poles of the ratoal fucto where as, durg cdece of wdg mechacal faults both the poles ad the zeros chage. Therefore cotuato of the researches o F applcato for fault detecto s ecessary. I ths paper employg F, the TFs of the trasformer are estmated both durg faulty codto ad durg ormal codto. Itroducg a ew algorthm the TF s detfed ad compared wth the referece oe. Not oly ths algorthm s able to recogze the type of fault, but also t ca specfy the level ad locato of fault. Comparg ths ew method wth oe of the past methods, t s show that the proposed dex s more effcet, eve determato of the fault level. II. LITERATURE REIEW AND PROBLEM DEFINITION TF aalyss s a relable method for wdg falures detecto power trasformers. Dfferet methods are troduced for evaluato of the TFs. Some of these methods are verfed agast the practcal samples of mechacal 23

2 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 faults [9]. Also, some aalytcal vestgatos have bee performed usg the equvalet crcut models [0]. O the other had, trasformer modelg usg ts teral geometry ad the materal propertes provded the possblty of evaluatg the sestvty of the TF parameters agast the type of faults []-[2]. I ths paper, the mportat wdg faults, whch are most lely to be detected usg the TF aalyss, have bee studed. These faults ca be classfed as follows: a- Axal Dsplacemet []-[4], b- Radal Deformato [8], [2] c- Dsc Space arato [5] ad d- Short crcut [4], [6] It should be oted that SC s ot mechacal faults per se, but usually t do stem from them, because t ca be detected usg TF method. The above faults have bee studed aalytcally ad expermetally [7]-[7]. These researches have studed the sestvty of TFs parameters agast those faults separately ad have ot dscussed that how a specfc mechacal fault ca be foud ad ca be classfed as to be oe of the AD, RD, DS or SC categores. I ths aalyss, the TF method s aalyzed by a comparatve method. However, geeral the comparatve algorthms troduced the lterature, ca be classfed to four major categores as follows. A. Algorthms Based o Artfcal Itellgece Methods The methods such as eural etwors, geetc algorthm, fuzzy logc, ad so o ca be cluded ths group [8]- [20]. Coclusos these wors emphasze the ecessty for further vestgatos to classfy the type ad locato of the fault. Moreover a lmtato exsts these methods whch made them applcable oly for trasformers le the oe used as test trasformer the developmet of method. B. Algorthms Based o Mathematcal ad Statstcal Methods Some of the methods of ths category have bee used for fault dagoss [2]-[24]. To establsh the ecessary measures for detecto of the faults some parameters have bee defed, amely: correlato coeffcet (CC), stadard devato (SD). Dsadvatages of these methods dscussed [2], [25]. Meawhle all of the algorthms that employ specfc equatos wth sgle-valued solutos for the evaluato of the measured TFs belog to ths category. The vestgatos carred out [26]-[30] troduce some addtoal parameters to evaluate the fault level. However, those wors do ot determe the type ad locato of the fault. Sce ths class of algorthms the phase respose s ot evaluated, therefore some of them cota urelable comparso parameter [25]. C. Algorthms Based o Electrc Crcut Models Some vestgatos have bee doe to get formato about the faults from the measured TFs by descrbg the TFs usg a equvalet electrc crcut model [3]. The parameters of the model represetg the TFs ca be calculated by the meas of the detals of the trasformer desg [2] or ca be estmated usg termal measuremets [32]. Determg the parameters of such models, whch usually cludes errors [2] ad s more dffcult for real wdgs ad trasformers, s ruled out by avodg electrc crcut models. D. Algorthms Based o Estmato Methods TFs of trasformers ca be approxmated by ratoal fuctos cosstg of two polyomals wth real coeffcets [5]. The estmato of such a ratoal fucto approxmato to descrbe the measured termal characterstc s employed for comparg TFs [7]-[8]. However, ths wors oly the AD ad RD are cosdered; thus, specfyg the type of fault by comparg dfferet types of faults wth each other s left as a subject to be studed. F s a wdely applcable ad geeral methodology for estmato of TF parameters from frequecy respose measuremet results [5]-[6]. So, ths wor F s used for estmato of TFs. The algorthms based o F methods to aalyze the process of fault detecto are preferred ths wor, because: a- Applyg these algorthms s very smple ad the results have a excellet match, b- The estmated fuctos are comparable a easer way, c- As t s dscussed [25], to have a effcet dagostc tool, the magtude ad the phase resposes must be tae to accout for the aalyss, ad ths s possble wth F method, ad d- Sce the employed ratoal fuctos are ot lmted to have ay specfed order; t allows fdg a fucto havg a perfect ft wth the measuremets. A comprehesve wdgs (mechacal) fault detecto procedure, whch etals the smultaeous detfcato of all the three faults characterstcs (type, level ad locato), ca be carred out by the algorthms based o applcato of F. The mportat wdgs (mechacal) faults, classfed as: AD, RD, DS ad SC are smulated by applcato of those faults o four dfferet model trasformers of a smlar sze. I the frst step, troducg a algorthm based o F, the type of fault amog the four classfed mechacal faults, (usg a ew preseted dex) s detected. Next, the level ad locato of the mechacal fault are determed. III. TF IDENTIFICATION BY ECTOR FITTING I prcple, a TF approxmato (of a gve order) ca be foud by fttg t wth a fucto whch s made of the rato of the two polyomals [5] as show equato (): f s ams as a0 () b s b s b m 0 Where m ad are the umber of umerator ad th deomator coeffcets, a ad b are the values of coeffcet for the umerator ad deomator, respectvely. However, geerally the TF of a passve system ca be approxmated by a ratoal fucto f(s) the form [5] show by equato (2): 24

3 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 f s r s p d. s e (2) The resdues r ad poles p are ether real quattes or appear complex cojugate pars, whereas d ad e are real. The problem at had s to estmate all of the coeffcets (r, p, d ad e) equato (2), so that a least squares approxmato of f(s) s obtaed over a gve frequecy terval. It should be oted that equato (2), represets a olear problem terms of the uows, because the uows p appear the deomator. F solves the problem of equato 2, sequetally as a lear problem sx stages, as follows [5]: Stage- Specfyg the startg poles p (2) The startg poles should be complex cojugate wth magary parts learly dstrbuted over the frequecy rage of terest. Each par s chose as follows: p=-α±jβ The real part of the poles should be small eough to avod ay ll codtog problem,.e., α=0.0β ca be a good choce. Stage2- Multplyg f(s) by a uow fucto of σ(s) havg poles smlar wth f(s), as show the followg: s f s s r d. s e s p R s p Multplyg the secod row (3) wth f(s) ad usg frst row yelds the followg relato: r d. s e s p Or f s s f s R f ( s) s p ft ft (5) Equato (4) s lear ts uows r, d, e ad R. Dervg (4) for several sgle frequeces, leads to a lear relato of equato (6): A X B (6) A X Where: s p s p r r d e R R B f s s s f s f s p s p Stage 3- Solvg (6) by least square method ad detfyg the coeffcets r, d ad e of (2) Stage 4- Calculato of ew poles If the fuctos equato (5) are reformed as show equato (7): (3) (4) f s s z s s p s ~ z d, (7) ft s p The usg equatos (5), & (7) the equato (8), s derved as follows: f s f ft s s ft d s ~ z s z Equato (8) s employed for calculatg the poles of f(s), t s suffcet to calculate the zeros of σ ft (s). Stage 5- Resolvg equato (6) wth ew poles ad repeatg the operatos utl the best approxmato s obtaed. Stage 6- After achevg a good approxmato of f(s), coeffcets of a ad b equato () ca be calculated usg r, p, d ad e. I. TEST OBJECTS (THE FAULTS MODEL TRANSFORMERS) AND MEASUREMENTS I the preset vestgato, all measuremets were executed the tme doma the usg FFT acqured the frequecy respose of TF whch s llustrated [2]. Four model trasformers were cosdered ths study ad some tests were performed o them to developmet of the proposed method for fdg type, level ad locato of the faults, as follow: A. Study of AD As a model trasformer for the study of AD a hgh voltage wdg wth 3 double verted dscs, where 6 turs are preset each dsc, ad a four layer cocetrc low voltage wdg, where 99 turs are preset each layer, were used. These partcular wdgs were maufactured for the specal expermetal purposes, ad they correspod to wdgs of a trasformer wth a rated voltage of about 0 ad a rated output of.3 MA. Its specfc costructo permts a gradual axal movemet of the teral layer wdg wth respect to the outer wdg. The test object has 82.7 cm heght ad therefore a cm axal dsplacemet s equvalet of.2% dsplacemet. B. Study of RD As a model trasformer for the study of RD a hgh voltage wdg wth 30 double verted dscs, where turs exsts each dsc, ad a oe layer low voltage wdg, havg 23 turs are used. The double dsc wdg has a rated voltage of 0 ad a rated output of.2 MA. The deformato has occurred o the double dsc wdg four degrees, as follow: Degree : The sxth up to the 54-th dscs were all radally deformed o oe sde. Deformato was aroud 7% of the dsc radus Degree 2: The sxth up to the 54-th dscs were all radally deformed o two opposte sdes. Deformato was aroud 7% of the dsc radus ft (8) 25

4 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 Degree 3: The sxth up to the 54-th dscs were all radally deformed o three sdes wth 90º wth respect to each aother. Deformato was aroud 7% of the dsc radus Degree 4: The sxth up to the 54-th dscs were all radally deformed o four sdes wth 90º wth respect to each aother. Deformato was aroud 7% of the dsc radus C. Study of DS For studyg DS the same model wdg that s used secto 4-2 expermeted. For ths purpose aother tact wdg that space betwee ts dscs s 5mm, s selected. For research about the effect of DS o TFs, the space of dscs s chagg to 7.5, 0, 5, 20 ad 25mm, the TF of them s measured. To better studyg the followg states are expermeted: State of DS: space varato betwee dsc 2 ad dsc 3 State 2 of DS: space varato betwee dsc 4 ad dsc 5 State 3 of DS: space varato betwee dsc 8 ad dsc 9 State 4 of DS: space varato betwee dsc 2 ad dsc 3 State 5 of DS: space varato betwee dsc 6 ad dsc 7 Although some of the DS faults gve above seem to be exaggerated ad mpossble to occur o real trasformers, they have bee executed ths wor order to accumulate more data about DS ad fd out precse formato cocerg TF varatos. Aother mportat pot s that the hgher dstaces betwee dscs mplemeted here are compesable wth less crcumferece of wdg covered wth the DS, so that the DSs carred out ths wor could be more smlar to real faults trasformers. D. Study of SC For the study of SC a hgh voltage wdg wth 30 double verted dscs, where 9 turs exsts each dsc s used. The double dsc wdg has a rated voltage of 0 ad a rated output of.2 MA. All dscs of the model wdg are provded wth a tap, whereby the recordg of the mpulse voltage dstrbuto s eabled alog the wdg. The put termal of the wdg s subjected wth the mpulse voltage, as respose sgals s the earth curret at a selected tap alog the wdg recorded. A short crcut betwee two dscs s appled to dfferet places alog the wdg for the expermetal determato of the effect of short crcuts ad ther spatal arragemet o the TF. For each dvdual posto of the SC the TF s measured. To vestgate the sestvty of trasfer TF measuremets, for AD, RD, DS ad SC, dfferet termal codtos have bee studed, as show Fgure. L H L H L L2 H H2 I H2 L I H 2 c) TF3 d) TF4 H e) TF 5 put Ioutput I output put L L2 H H2 L H Fgure. Dfferet termals coectos, for sestvty aalyss of trasfer fuctos measuremets whe examg: a) AD, b) AD, c) RD ad DS, d) RD, ad e) SC The measured results show that the mechacal faults (AD, RD, DS ad SC) affect the TFs ad modfy them. For example, the measured TFs (TF, TF 3 ad TF 4 ) are showed Fgures 2 to 4. Admttace [ma/] cm Dsplacemet cm Dsplacemet 2cm Dsplacemet 3cm Dsplacemet 4cm Dsplacemet 5cm Dsplacemet 6cm Dsplacemet 7cm Dsplacemet 8cm Dsplacemet Frequecy [MHz] Fgure 2. The measured TF of trasformer wdg whe AD occurred (setup TF ) Admttace [ma/] Wthout Deformato Degree of Deformato Degree2 of Deformato Degree3 of Deformato Degree4 of Deformato L2 H2 I H2 L2 H2 I H 2 a) TF b) TF2 H L H Frequecy [MHz] Fgure 3. The measured TF of trasformer wdg whe RD occurred (setup TF 3 ) 26

5 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 Admttace [ma/] Frequecy [MHz] Wthout SC SC betwee uts, 2 SC betwee uts 2, 3 Fgure 4. The measured TF of trasformer wdg whe SC occurred (setup TF 5 ). PROPOSED METHOD FOR FAULT DIAGNOSIS I ths paper, fault dagoss trasformer wll be doe wth udereath assumptos: Measuremets s carred out o a frequecy rage of Hz Measuremets s lmted to layer- dsc type wdg of trasformer Impact of other two phases o the TF of spected phase s cosdered eglgble Usg the three assumptos troduced ths secto, there ca be some msleadg results oly f those assumptos were ot applcable specal cases. However to evaluate ad detty those cases we pla to exted our verfcato wor to cover some specal type trasformers future. Remdg that F method s oe of the best methods for estmatg TF, ths paper usg the measured data, the TFs are estmated by F method. A 40-th order approxmato has bee ftted to these fuctos. Ths s employed to develop TFs for possble form of faults ad to compare them wth the TFs of the related ormal codtos. Besdes detectg the poles ad resdues of TFs, the coeffcets of TFs polyomals umerator ad deomator are determed. A maxmum error of less tha 0-6 all cases s the proof of a good ft. Refereces [7-8], troduce oe dex for comparso of TF usg F, as follows: Pd Pw SDP (9) P d Where s the umber of poles, P d ad P w are the values of th pole for dsplaced (deformed) ad tact cases, respectvely. SDP s the sum of absolute dffereces of poles ad does t perceve the effect of zeros. However, Fgure 5 shows that the mechacal faults cause dslocato of both poles ad zeros. Therefore to compare the TFs, the effect of both of them (pole ad zero) s cosdered ths method. For ths purpose, a ew dex (FI: Faulted-Itact relato) s troduced as follows: FI m m a f b f (0) a b Where m ad are the umber of umerator ad deomator coeffcets ( equato ()), a ad b are the values of th coeffcet for the faulted codtos, ad b are the values of th f f a coeffcet for the ormal codtos, respectvely. Ths ew dex s the rato of sum of the umerator coeffcets faulted codto to sum of the umerator coeffcets ormal codto multply by the rato of sum of the deomator coeffcets faulted codto to sum of the deomator coeffcets ormal codto. The proposed dex s used for detecto of mechacal faults as follow: Imagary Axs Pole-Zero Map Poles of tact case Zeros of tact case Poles of degree of deformato Zeros of degree of deformato Poles of degree2 of deformato Zeros of degree2 of deformato Poles of degree3 of deformato Zeros of degree3 of deformato Poles of degree4 of deformato Zeros of degree4 of deformato Real Axs Fgure 5. Poles ad zeros of the estmated TF 4 for study of RD A. Developmet of Wdg Model for a Layer-Dsc Type Wdg The test objects that s troduced secto 4, s employed for developmet of ths wdg model. Here t wll be show that all of the four fault types (AD, RD, DS ad SC) ca be detected by applcato of the ew method. Based o our expermets as t wll be show ext secto, for each of the four fault types the FI wll preserve a value that s dstgushable from the other type easly. Although ths fdg ca be used fault detecto of smlar wdgs real trasformers, however ths method wll be extedg to cover all types of wdgs future. A.. Determg the Type of Fault The related results demostrate that the value of FI f AD or RD occurs s less tha however, f DS or SC occurs s more tha. A careful specto of Fgures 2, 3 mples that the chages TFs do ot spread equally over the etre frequecy rage, ad those are ofte bouded to several frequecy rages. O the other had t s clear that by subdvdg the etre frequecy rage to smaller rages, t ca crease the sestvty of the comparso dex, such as FI. I other words, exstece of ay smlarty betwee the faulted TF ad the tact TF depeds o the frequecy rage of terest ad the type of fault. Hece, evaluato of the TFs a set of smaller frequecy rages ca be more helpful determg the type of fault. Examg the smulato results, t has bee show [2] that the varato of TF characterstcs due to RD the 27

6 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 wdg occurs the etre frequecy rage; whereas the varatos due to occurrece of AD are located the frequecy rages (approxmately) above 00 KHz. Cosequetly, to dstgush the occurrece of a AD fault from a RD fault, FI dex should be determed wth the low frequecy rage of (0-00 KHz), whch s amed FI lf. I tables I ad II, FI lf results are determed uder dfferet possble faults types of AD ad RD, ad uder dfferet termal coectos. Studyg the tables shows that the maxmum varato due to occurrece of AD s 3.2% whch s less tha the mmum varato uder occurrece of RD whch s 6.5%; moreover these tables show that f the measured TF be admttace fucto, the the expected mmum varato of RD s 9.2%. Therefore t would be smple to dstgush betwee AD ad RD. Meawhle as t s show Table III ad Fgure 6, due to occurrece of some other faults, such as; DS ad SC, the value of FI dex wll be more tha. Remdg that SC causes a substatal varato, whle DS maes a lttle varato the measured TF, t s easy to dstgush betwee them. Also they ca be dstgushed by examato of FI dex, sce the mmum value of FI durg SC s.550, whereas the maxmum of FI durg DS s.553. It meas that durg a SC, the mmum varato FI s 55%, whereas the varato FI durg a DS s 5.5% at most. Therefore, to specfy the type of a fault, the FI dex s determed, f ths dex s more tha, that fault s SC or DS ad otherwse the fault s AD or RD. Addtoally, f FI s less tha, the FI dex low frequecy rage should be determed (amed FI lf ) to specfy the type of fault. However, f t s more tha, by determg the value of FI the type of fault ca be specfed. TABLE I. FI LF WITHIN THE RANGE OF LOW FREQUENCY (0-00 KHZ) FOR AD Dsplacemet FI lf (cm) TF TF TABLE II. FI LF WITHIN THE RANGE OF LOW FREQUENCY (0-00 KHZ) FOR RD FI lf Degree of Deformato TF 3 TF 4 Wthout RD Degree Degree Degree Degree TABLE III. CALCULATED FI USING TF 3 FOR DIFFERENT DISC SPACES IN FIE STATES State of Dsc Space (mm) DS State State State State State Amout of FI SC betwee dscs 2 ad 3 SC betwee dscs 4 ad 5 SC betwee dscs ad 2 SC betwee dscs 3 ad 4 SC betwee 2 dscs whch 2 turs short crcuted SC betwee 2 dscs whch 8 turs short crcuted Number of Dsc Fgure 6. Calculated FI for dfferet locatos of SC betwee dscs A.2. Detecto of Fault Level To specfyg the fault level, the suggested coeffcets are calculated both faulted ad tact codtos wth the whole frequecy rage. Usg the value of FI dex, the amout of AD, RD, DS or SC occurred a wdg ca be determed. As t s show the tables I, ad ay decrease FI dex, correspods to a crease the degree of deformato, or axal dsplacemet of the wdg. Whereas, Table III shows a crease FI dex s related to a crease DS. Meawhle, Table I also shows that the values of SDP, for larger dsplacemets (such as 6, 7 ad 8 cm), are very closer to each other, ad t s very hard to dstgush betwee these dsplacemets. Whle, the value of FI dex these codtos are more desrable to be employed. Not oly the FI dex ca specfy the type of fault, but also t s a better measure determato of the fault level tha the other dex. Smlarly, the level of SC ca be detected usg FI dex. As show Fgure 6, occurrece of SC betwee 2 dscs chages the proposed dex. However f SC occurs betwee 2 dscs where 2 turs are short crcuted, the FI dex chages less tha whe 8 turs are short crcuted. Moreover for all states of SC, the FI dex has a uque ad specfc value for ay SC betwee dscs. Therefore f the value of FI s betwee.550 ad.9543, the SC betwee 2 dscs s occurred across two turs ad f the value of FI s betwee ad 5.30, 8 turs are short crcuted. Therefore, the level of the SC ca be determed wth the help of the FI as show Fgure 6. TABLE I. FI AND SDP WITHIN THE WHOLE MEASURED FREQUENCY RANGE (0-000 KHZ) FOR AD Dsplacemet TF TF 2 (cm) SDP FI SDP FI TABLE. FI AND SDP WITHIN THE WHOLE MEASURED FREQUENCY RANGE (0-000 KHZ) FOR RD TF 3 TF 4 Degree of Deformato SDP FI SDP FI 28

7 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 Wthout RD 0 0 Degree Degree Degree Degree A.3. Fault Localzato To demostrate the ablty of FI dex (to recogze locato of fault) Table III, ts value s gve for dfferet locatos ad levels of DS. Table III demostrates FI has a uque ad specfc value for ay level ad locato of the fault. If the value of FI s (for example) betwee.0743 ad.088, the dstace betwee two successve dscs of the wdg s 5 mm. If the fault level s ow, the locato of the fault ca be determed wth the help of the FI dex ad oe colum of the Table III for the ow fault levels. It meas the proposed dex, addto to detecto of the type ad the level of the fault, ca be employed to detect the locato of fault. B. erfcato of Wdg Model I order to evaluate ad valdate the proposed method, aother test object of smlar type (layer-dsc wdg) wth a dfferet ratg s examed. Specfcatos of ths test object are: - hgh voltage wdg wth 40 double verted dscs, where 2 turs are preset each dsc - low voltage wdg wth four layer cocetrc, where 88 turs are preset each layer A short crcut s appled betwee two dscs of the wdg, two dfferet locatos alog the wdg separately, ad the TFs of them (accordg to fg. (e)) s acqured. The for fault dagoss, the proposed algorthm s used. The proposed method determes the values of FI for these places as.6602 ad 5.75, respectvely. For these values of FI, the proposed algorthm determes the fault to be Short Crcut, correctly. Moreover these result show precsely that the frst case (FI=.6602) s related to a SC occurred betwee 2 dscs where 2 turs are short crcuted, ad the secod case (FI=5.75) s related to 24 turs beg short crcuted. It meas the proposed algorthm ca be employed to detect the type ad level of fault dfferet trasformer wdgs. C. Proposed Algorthm I ths paper to specfy the type, level ad locato of the mechacal fault trasformer wdgs, the followg algorthm has bee proposed: - Obtag TF of trasformer both tact ad faulted codtos usg the results of measuremets 2- Estmatg coeffcets of TF both tact ad faulted codtos usg F method 3- Calculato of FI 4- f FI> the GOTO step 5 5- f percetage of chages FI s more tha 30% the Fault s SC 6- else fault s DS 7- ed f 8- else Estmatg coeffcets of TF both tact ad faulted codtos usg F method wth the rage of low frequecy (0-00 KHz) 9- Calculato of FI lf 0- f percetage of chages FI lf s more tha 5% the fault s RD - else fault s AD 2- ed f 3- ed f 4- Specfyg the level of AD, RD, DS ad SC usg FI dex 5- Localzato of DS usg FI dex I. CONCLUSION The past methods suggested by other researcher oly focus o detecto of the level of mechacal fault trasformer ad there s o well defed method to specfy the type of fault, by comparso of the TFs. Therefore a ew method s preseted ths wor to specfy the type, level ad locato of mechacal fault wdgs of trasformers wth a relatvely well defed ad accurate procedure. For ths purpose, the requred measuremets have bee carred out o four proper Models both faulted ad tact codtos wth dfferet degrees of AD, RD, DS ad SC of wdg. The results of measuremets clearly demostrate how the mechacal faults modfy the TFs. Therefore, usg vector fttg method, approprate coeffcets of TFs has bee estmated tact ad faulted codtos. Subsequetly by troducg a ew dex based o the calculated coeffcets amed FI, the TFs are compared. The obtaed results show that: a) The value of FI s less tha durg AD or RD faults, ad more tha durg DS or SC faults, b) The dfferece betwee the FI dexes, for dfferet degrees of faults, durg DS or SC faults s large ad ca easly dstgush betwee them, c) arato TF the low frequecy rage (0-00 KHz) durg AD s less tha durg RD. Thus, to dstgush AD from RD, the FI dex s calculated the low frequecy rage. The measured results showed that t s easy to dstgush easly betwee AD ad RD faults, d) FI dex has show ts effectveess, eve the determato of the fault level, ad e) Results of the applcato of the proposed dex for DS fault show that the FI dex ca be used for detecto of fault locato. I order to evaluate ad valdate the proposed method, aother test object of smlar type (layer-dsc wdg) wth a dfferet ratg s examed. The related result showed the proposed method s vald for wdgs whch have the same desg. Applcablty of the proposed method o fault detecto of dfferet type of trasformers (cludg several wdgs of dfferet types) wll be studed the future wors, too. ACKNOWLEDGMENT Authors would le to tha Mr. Kharaz ad Mr. Frooz for provso of the secod sample wdg, permsso ad assstace performg the requred measuremets o the wdgs. REFERENCES [] C. Begtsso, Status ad treds trasformer motorg, IEEE Tras. Power Delvery, vol., o.3, pp ,

8 [Dowloaded from o Wedesday, Jue 0, 20 at 6:04:09 (UTC) by Redstrbuto subject to AECE lcese or copyrght. Ole dstrbuto s expressly prohbted.] Advaces Electrcal ad Computer Egeerg olume, Number 2, 20 [2] J. Chrsta, ad K. Feser, Procedures for detectg wdg dsplacemets power trasformers by the trasfer fucto methods, IEEE Tras. Power Delvery, vol. 9, o., pp , [3] T. Lebfred, ad K. Feser, Motorg of power trasformers usg the trasfer fucto method, IEEE Tras. Power Delvery, vol. 4, pp , 999. [4] K. Feser, J. Chrsta, C. Neuma, U. Suderma, T. Lebfred, A. Kachler, ad M. Loppacher, The trasfer fucto method for detecto of wdg dsplacemets o power trasformers after trasport, Proc. Short Crcut or 30 Years of Servce, CIGRE 2/33-04, [5] B. Gustavse, ad A. Semlye, Ratoal approxmato of frequecy doma resposes by vector fttg, IEEE Tras. Power Delvery, vol. 4, pp ,999. [6] B. Gustavse, Wde bad modelg of power trasformers, IEEE Tras. Power Delvery, vol. 9, pp , [7] P. Karmfard, G. B. Gharehpeta, ad S. Tebohle, Determato of axal dsplacemet extet based o trasformer wdg trasfer fucto estmato usg vector-fttg method, Europea Tras. Electrcal Power (ETEP), vol. 8, pp , [8] P. Karmfard, G. B. Gharehpeta, ad S. Tebohle, Localzato of wdg radal deformato ad determato of deformato extet usg vector fttg-based estmated trasfer fucto, Europea Tras. Electrcal Power (ETEP), vol. 9, pp , [9] Cgre Worg Group A2.26, Mechacal-Codto assessmet of trasformer wdgs usg frequecy respose aalyss (FRA), 2007 [0] L. Satsh, ad S. K. Sahoo, A effort to uderstad what factors affect the trasfer fucto of a two-wdg trasformer, IEEE Tras. Power Delvery, vol. 20, o.2, pp , [] A. Sgh, F. Castellaos, J. R. Mart, ad K. D. Srvastava, A comparso of tras-admttace ad characterstc mpedace as metrcs for detecto of wdg dsplacemets power trasformers, Electrc Power Systems Research, vol. 79, o., pp , [2] E. Rahmpour, J. Chrsta, K. Feser, ad H. Mohse, Trasfer fucto method to dagose axal dsplacemet ad radal deformato of trasformer wdg, IEEE Tras. Power Delvery, vol. 8, o.2, pp , [3] M. Florows, ad J. Furgał, Modellg of wdg falures detfcato usg the frequecy respose aalyss (FRA) method, Electrc Power Systems Research, vol. 79, o., pp , [4] J. Jayasghe, Z. D. Wag, P. N. Jar, ad A. W. Darw, Ivestgatos o sestvty of FRA techque dagoss of trasformer wdg deformatos, Proc. It. Symp. Electrcal Isulato, Idaapols, USA, September [5] E. Rahmpour, ad S. Tebohle, A mathematcal model to vestgate dsc space varato power trasformer usg trasfer fucto aalyss, Proc. It. Symp. Hgh oltage Egeerg, Ljubljaa, Slovea, August [6] S. M. Islam, ad G. Ledwch, Locatg trasformer faults through sestvty aalyss of hgh frequecy modelg usg trasfer fucto approach, Proc. It. Symp. Electrcal Isulato, Motreal, Caada, Jue 996. [7] A. Shtemrov, W. H. Tag, ad Q. H. Wu, Trasformer wdg codto assessmet usg frequecy respose aalyss ad evdetal reasog, IET Elec. Pow. Applcatos, vol. 4, o.3, pp , 200. [8] D. K. Xu, C. Z. Fu, ad Y. M. L, Applcato of artfcal eural etwor to the detecto of the trasformer wdg deformato, Proc. IET Cof. It. Symp. Hgh oltage Egeerg, August 999. pp [9] J. Zhja, L. Jgtao, ad Z. Zshu, Dagoss of trasformer wdg deformato o the bass of artfcal eural etwor, Proc. It. Cof. Propertes ad Applcatos of Delectrc Materals, Jue 2000, pp [20] S. Brlaseara, Y. Xgzhou, F. Fetherstoe, R. Abell, ad R. Mddleto, Dagoss ad detfcato of trasformer faults from frequecy respose data, Proc. IEEE Cof. Power Egeerg Socety Wter Meetg, Ja. 2000, pp [2] K. Jog-Woo, P. ByugKoo, J. Seug, K. S. Woo, ad P. PooGyeo, Fault dagoss of a power trasformer usg a mproved frequecy-respose aalyss, IEEE Tras. Power Delvery, vol. 20, o., pp , [22] S. A. Ryder, Trasformer dagoss usg frequecy respose aalyss: results from fault smulatos, Proc. IEEE Cof. Power Egeerg Socety Summer Meetg, July 2002, pp [23] S. A. Ryder, Dagosg trasformer faults usg frequecy respose aalyss, IEEE Electrcal Isulato Magaze, vol. 9, o.2, pp. 6-22, [24] P. M. Nrgude, D. Ashoraju, A. D. Rajumar, ad B. P. Sgh, Applcato of umercal evaluato techques for terpretg frequecy respose measuremets power trasformers, IET Sc. Meas. Techology, vol. 2, o.2, pp , [25] J. R. Secue, ad E. Mombello, Sweep frequecy respose aalyss (SFRA) for the assessmet of wdg dsplacemets ad deformato power trasformers, Electrc Power Systems Research, vol. 78, o.6, pp. 9-28, [26] J. Gu, W. Gao, K. Ta, ad S. Gao, Deformato aalyss of trasformer wdg by structure parameter, Proc. It. Cof. Propertes ad Applcatos of Delectrc Materals, paper o. P2-47, [27] P. Karmfard, ad G. B. Gharehpeta, A ew algorthm for localzato of radal deformato ad determato of deformato extet trasformer wdgs, Electrc Power Systems Research, vol. 78, o.0, pp. 70-7, [28] R. Wmmer, S. Tebohle, M. Hedl, A. Kraetge, M. Krüger, ad J. Chrsta, Developmet of algorthms to assess the FRA, Proc. It. Symp. Hgh oltage Egeerg, paper o. T7-523, [29] E. Rahmpour, ad D. Gorz, A ew method for comparg the trasfer fucto of trasformers order to detect the locato ad amout of wdg faults, Electrcal Egeerg, vol. 88, o.5, pp. 4-46, [30] M. Wag, A. J. adermaar, ad K. D. Srvastara, Evaluato of frequecy respose aalyss data, Proc. It. Symp. Hgh oltage Egeerg, 200, pp [3] S. K. Sahoo, ad L. Satsh, Dscrmatg chages troduced the model for the wdg of a trasformer based o measuremets, Electrc Power Systems Research, vol. 77, o.7, pp , [32] K. Ragava, ad L. Satsh, Localzato of chages a model wdg based o termal measuremets: expermetal study, IEEE Tras. Power Delvery, vol. 22, o.3, pp ,