Performance of Diagnosis Methods for IGBT Open Circuit Faults in Three Phase Voltage Source Inverters for AC Variable Speed Drives

Transcription

1 Performane of Dagnoss Methods for IGBT Open Crut Faults n Three Phase Voltage Soure Inverters for AC Varale Speed Drves Ka Rothenhagen, Fredrh W. Fuhs Chrstan-Alrehts-Unversty of Kel Kaserstr.2, Kel, Germany Tel kro@tf.un-kel.de, fwf@tf.un-kel.de Keywords Dagnosts, Varale Speed Drve, Vetor Control, Voltage Soure Inverters. Astrat Varale speed drves have eome ndustral standard n many applatons. Therefore fault dagnoss of voltage soure nverters s eomng more and more mportant. One possle fault wthn the nverter s an open rut transstor fault. An overvew of the dfferent strateges to detet ths fault s gven, nludng the algorthms used to loalze the open transstor. Prevous work showed sgnfant dfferenes among the avalale methods to detet suh a fault for a mans sde atve retfer. Ths paper extends the performane evaluaton for the nverter onneted to the mahne wth varale stator voltage and frequeny. Smulaton results are presented. They show the nfluene of the appled standard feld orented ontrol on the urrents durng a fault. An expermental setup n the laoratory s used to valdate smulaton results. Typal deteton results are presented nludng tme-todeteton measurements. Roust deteton of open transstor faults has een found to e possle. I. Introduton Voltage soure nverter (VSI) fed varale speed drves have eome the standard n ndustral applatons. Whle safety rtal applatons are equpped wth hgh sophstated fault dagnost systems, standard applatons are regularly only equpped wth standard fault deteton. Inreased onverter osts are usually the reason for ths, ut advaned dagnoss features an on the other hand e eonomally reasonale, onsderng operaton osts of the onverter and the whole system. Hgh osts due to standstll and repar, as well as seondary faults aused y unnoted damages make advaned dagnoss methods nterestng. Fg. 1: VSI feedng a Permanent Magnetzed Synhronous Mahne, wth fault n T4. Wthn the varale speed drve, faults an our n the motor, retfer, or the nverter. Whle the dagnoss of eletral mahnes s thoroughly nvestgated, wth an overvew gven y Capolno [1], dagnoss of the retfer and nverter are not as well researhed Fuhs [2]. Wthn the nverter, semondutors are one of the man auses of faults next to eletrolyt apators. A lassfaton of thnkale faults n VSI has for example een pulshed y Kastha [3]. The usual fault mode of semondutors s a short rut, ut an open rut fault an also our.

2 Whle short rut proteton y means of deteton va olletor-emtter voltage has eome a standard feature of today's VSI, makng nverters short rut proof, the open rut fault has not yet reeved so muh attenton. Open rut faults may for example e aused y the lftng of ondng wres due to therm ylng, y a drver falure, or y a short rut fault ndued rupture of the IGBT. An open rut fault wll not neessarly ause the drve to e noperale. It may therefore e undeteted for an extended perod of tme. Sne an open transstor auses pulsatng urrent and torque, t may lead to seondary faults n other semondutors, the nverter, the motor or the load. Other researhers have developed several deteton methods for open transstor faults [4, 5, 6]. These methods have een sujet to a performane omparson for a voltage soure atve retfer [7]. Ths researh has shown major dfferenes n the performane, tunng and omputng effort, and false alarms resstvty. As a result, two of the methods have een modfed. Ths researh shall herewth e extended to a voltage soure nverter feedng a permanent magnet synhronous mahne. Here, espeally the nfluene of varale AC voltage and frequeny has to e nvestgated. The paper s organzed as follows: At frst, ehavour and dfferent dagnoss tehnques of transstor open rut faults wll e presented, nludng the modfatons as pulshed y [7]. Then, smulaton results of these dagnoss methods wll e gven. Measurement results ak up these results, and typal deteton sequenes are shown. Ths work wll e summed up n a onluson. II. Transstor Open Fault Behavor and Dagnoss Methods Transstor Open Faults Behavor The topology of the nverter as ass for the nvestgaton s gven n fgure 1, wth a fault n swth 4 ndated y an open gate onneton. Fgure 2 shows the nverter AC urrents for a healthy nverter n Park's vetor referene frame. In ase of an open rut fault, the mahne urrent n the faulty phase an ether e only negatve or only postve, dependng on whh transstor s damaged. For a fault n swth 1, ths leads to urrents as shown n fgure 3, where the urrent of phase u an e only negatve. It s mportant to remark that all lne urrents ontan a dret omponent, whereas urrents n healthy ondton do not. Usng the Park s Vetor transformaton (1), (2) [5] on the urrents yelds to trajetores as dsplayed n fgure 4. The measurements are explaned later. Fg 2: Inverter AC Current trajetory n Park's Vetor depton wthout fault (Measurement). Fg 3: Inverter AC urrents n no fault and fault ondton (Measurement). I I α a (1) β = I = I I 3 (2)

3 Fg. 4: Trajetory of nverter AC urrents n Park s Vetor depton (Measurement, 7 A RMS ). Prevous Researh on Open Transstor Fault Dagnoss Park s Vetor Method As an e seen from fgure 4, an open transstor fault an e deteted y onsderng the "entre of gravty" of the park's vetor trajetory, as suggested y Mendes [5]. The algorthm s ased on averagng over one perod (3) n order to alulate the dret omponent of the lne urrents. Then a Park's Vetor transform (1), (2) s appled to ompute magntude (4) and angle (5) of the AC urrent n the omplex plan. For a system wthout open transstor fault the urrent spae vetor runs n a rle and the mean value s zero. If a fault ours, the magntude of the spae vetor s not zero, wll exeed the threshold and the atual faulty swth an e dentfed y onsderng the argument, as shown n tale I. N 1 µ ν = Iν ( kτ ) (3) N k = (4) µ = µ = µ + µ arg{ µ } = artan f 1 mans = Nτ ν [ α, β ] α µ µ Tale I: Loalsaton of Fault wth Park's Vetor Method β α β Transstor Magntude µ Argument (deg) (5) (6) (7) T1 150 to 210 T2 210 to 270 T3 Exeed 270 to 330 T4 Threshold 330 to 30 T5 30 to 90 T6 90 to 150

4 Normalsed DC Current Method One drawak of the aove mentoned method s the load dependene of the algorthm. The dret omponent alulated y (3) wll e larger the larger the AC urrent s. In order to make the sheme ndependent from the load, Aramk [6] suggested to use a normalsed dret omponent nstead. To aheve ths, the frst order harmon oeffents of the nverter AC urrents are omputed y means of a DFT (9), (10). The dret omponent as alulated y (3) s then dvded y the asolute value of the frst harmon (8). Ths s done for eah of the three phases (13). For dentfaton of the faulty swth, the resultng resdual γ s ompared to thresholds (11), (12). Usng tale II, the faulty swth an e dentfed. The threshold of 0.45 s reported to e a unversal value derved from experene. µ γ = (8) a + 2 1, 2 1, N 2 2πk a1, = I( kτ ) os N k = 1 N N 2 2πk 1, = I( kτ ) sn N k = 1 N 1: γ > 0 d1, = 0 : γ 0 1: γ > 0.45 d2, = 0 : γ 0.45 [ a,, ] (9) (10) (11) (12) (13) Tale II: Loalsaton of fault wth Normalsed DC Current Transstor a a T T T T T T Modfed Normalsed DC Current Method As stated n [7], the Normalsed DC Current Method has some drawaks when mplemented n a losed loop ontrol sheme. Therefore, ts adaptaton for etter usalty has een proposed. The Modfed Normalsed DC Current method uses the same algorthms as the Normalsed DC Current method, ut employs a less restrtve way to loalze the faulty swth, as dsplayed n tale III. Blanks n tale III mean ths state s not relevant. To prevent that more than one ondton s fulflled, only the largest asolute value of γ a, γ and γ as alulated n (8) s onsdered.

5 Tale III: Loalsaton of fault wth Normalsed DC Current Method Transstor a a T1 1 1 T2 1 1 T3 1 1 T4 0 1 T5 0 1 T6 0 1 The Slope Method In another method suggested y Peuget [4], the slope of the trajetory n the omplex plan an e used for fault deteton and dentfaton. The trajetory s alulated y usng the Park's Vetor Transform (1), (2) of the lne urrents. As one an see from fgure 4, the trajetory omprses of a semrle and a lnear seton. The lnear seton has a haraterst slope, dependng on the faulty nverter rdge. By alulaton of the slope aordng to (14), the faulty rdge an e dentfed usng tale IV. It s reported that a quarter perod of unhanged suessve values are a vald ndator for a fault. α σ = (14) β In order to dentfy the faulty swth wthn a faulty rdge, t s neessary to detet whether the urrent n the faulty phase s postve or negatve. Ths an for nstane e done y a Shmtt-trgger that montors the lne urrent, as depted n fgure 5. Ths way, the faulty transstor an e dentfed. Tale IV: Loalsaton of fault wth Slope Method Faulty phase Phase u Phase v Phase w σ Fg. 5: Hysteress to detet urrent polarty for Slope method Smple Dret Current Method A straght forward and easy to mplement fault deteton sheme s the Smple DC Current method, as desred n [7]. Ths method uses the dret omponent alulated y (3) for eah of the three phases. No transformatons or normalsaton s needed. For dentfaton, the largest dret omponent of the three phases s ompared to a threshold aordng to tale V. Ths way, the faulty swth an e deteted. Tale V: Loalsaton of fault wth Smple Dret Current Method Transstor µ a µ µ T1 >δ T2 >δ T3 >δ T4 <-δ T5 <-δ T6 <-δ

6 III. Analyss of Transstor Open Fault Dagnoss Methods y Smulaton Smulatons onernng the ehavour of a voltage soure nverter, onneted to a permanent magnet synhronous mahne wth feld orented ontrol [8], wth an open transstor have een arred out usng MATLAB-Smulnk and the PLECS -Toolox. The used ontrol parameters and struure where the same for all examned dagnoss methods. The smulatons were arred out for output referene frequenes of 50 Hz, 25 Hz, and 10 Hz. The frequeny s ontrolled y a speed ontroller and therefore sujet to flutuaton. Also, the nternal torque of the PMSM wll flutuate eause of the faulty swth. The load was adjusted for a urrent of 6.8, 3.4, and 0.7 Ampere rms. The used topology s dsplayed n fgure 6. The very same topology s also used for the later expermental analyss, exept that a real nverter and mahne s used nstead of the models. Fgures 7 and 13 ompare smulaton and experment results and show good omplane. Fg. 6. Topology of drve system and ontrol used n smulaton and experment Evaluaton of Performane The performane was evaluated and s summarsed n tale VI. An "X" means the fault deteton was suessful, a "-" means the deteton was not suessful, whereas "(X)" means that the deteton was amguous. It an e sad that most deteton methods had prolems wth small urrents, as already mentoned n [7]. Only the Modfed Normalsed DC Current method s ale to detet the faulty swth n every smulated ase. Therefore t was hosen for further testng y experment. The other methods that were realsed n experment are Smple Dret Current method and the Slope Method. Tale VI: Overvew of Smulaton Results for Open Transstor Deteton Methods Frequeny Current [A rms ] Hz X X X Park s Vetor 25 Hz X X - 10 Hz X X - 50 Hz X - - Slope 25 Hz X (X) - 10 Hz X X - 50 Hz X X - 25 Hz X X - Normalzed DC Current Smple DC Current Modfed Normalzed DC 10 Hz Hz X X X 25 Hz X X - 10 Hz X X - 50 Hz X X X 25 Hz X X X Current 10 Hz X X X

7 Fg. 7: Inverter AC Currents n Park's Vetor frame at varous frequenes, (lokwse rotaton). Left: Smulaton at 6.8 A rms. Rght: Measurement at 7 A rms. Colour ode: Red: 10 Hz (top), Blue: 20 Hz, Green: 25 Hz, Cyan: 30 Hz, Yellow: 40 Hz, Blak: 50 Hz (ottom) Influene of the ontrol loops Sne a losed loop feld orented ontrol s used, the ontrol loops wll nfluene the ehavour of the varale speed drve durng a fault. The faulty swth wll not ondut urrent, ausng a ontrol devaton upon whh the urrent ontrol wll reat wth a modfed voltage set value. Also, dependng on the load, the torque rpple may ause a devaton n speed upon whh the speed ontroller reats. It s assumed and onfrmed y smulaton and measurement, that ths nfluene s larger at a slower output frequeny. The ontrol effet on the urrents an e seen n fgure 7, where the trajetory wth the lowest frequeny s most "strethed" and "shfted", whle the trajetory at 50 Hz almost remans a semrle, wth zero as entre. Espeally n measurement, further effets are to e seen. Ths ehavour wll most lkely have an nfluene on the deteton, sne all deteton algorthms employ urrent measurements. Smulatons show, however, that a hange n frequeny s muh more tolerale than small urrents, as an e seen n tale VI. IV. Analyss of Transstor Open Fault Dagnoss y Experment The setup used for expermental analyss of the aove mentoned fault dagnoss shemes nludes, as already ntrodued n fgure 1, a urrent measurement, a mroontroller to exeute the ntrodued algorthms, and a small dsplay. The varale speed drve s a permanent magnet synhronous mahne fed y a voltage soure nverter. Any of the sx swthes of the nverter may e dsaled y swthng the respetve PWM sgnal to zero permanently. A torque ontrolled DC drve s used to load the PMSM. Feld orented ontrol s mplemented y means of a dspae -Controlsystem [8]. Fg. 8: Measurement of step response of rotatonal speed Fg. 9: Measurement of step response of Torque-uldng urrent I q

8 Fgures 8 and 9 show the dynam performane of the drve. Fgure 8 shows the step response of the rotatonal speed. The referene steps up from 500 rpm to 800 rpm. To haraterse the urrent ontrol loop, the speed ontrol loop was dsaled. Fgure 9 shows the step response of the quadrature urrent, whh s responsle for the torque, when the referene s swthed from 1 Ampere to 5 Ampere. Fgure 10 shows the phase urrents for varous frequenes. The tme s saled per yle duraton, n order to ompare the ontroller nfluene. As an e seen, the magntude of all urrents s kept the same for all frequenes. The urrents ft well onto eah other. Fgure 10 also shows the urrent waveforms for a fault n swth 1, loated n phase u. Beause of the fault, the urrent n phase u annot eome postve. The urrents of phases v and w are also shown. Eah frequeny s plotted n a dfferent olour. The full olour odes are lsted eneath fgure 7. As an e seen, the urrents dffer for dfferent frequenes. Generally, the reaton of the 10 Hz urrents s largest among all urrents towards the end of the perod, durng whh the faulty swth was supposed to ondut urrent. Ths s reasonale, eause ths perod lasts longest for the 10 Hz referene. Fg. 10: Measurement of nverter AC urrents at varous frequenes. Tme axs s saled per unt. Left: Healthy ondton. Rght: Fault n swth 1. Red: 10 Hz, Blak: 50 Hz Performane of the Deteton Algorthms Three algorthms were tested n experment: The Smple Dret Current method, the Modfed Normalsed DC Current method and the Slope Method. As an e seen from tale VII, the Slope Method showed poor performane. Tale VII: Tme to Deteton Measurements for Open Transstor Deteton Tme to Deteton Measurements Frequeny Current Average Standard Devaton Mn. Max. Slope 50 Hz NA NA NA NA 50 Hz ms 4.28 ms 6.6 ms 25 ms Modfed Normalzed DC Current 25 Hz ms 9.84 ms 9 ms 42 ms 10 Hz 7 A RMS ms ms 17 ms 111 ms 50 Hz 10.2 ms 4.17 ms 3.4 ms 17.6 ms Smple DC 25 Hz ms 8.27 ms 7.5 ms 36.5 ms 10 Hz ms ms 20 sm 86 ms Measurements were amguous and dd not show steady results. If a fault was deteted, the deteton tme was usually long. Typal performane an e seen n fgures 11 and 12. Fgure 11 shows a nofault state. It an e seen that the slope s a typal artan-funton. Fgure 12 shows the ehavour for a fault n phase w, were the lnear porton of the trajetory has a slope of 3. As an e seen, the alulated slope shows a hgher porton of values wthn proxmty of the expeted slope. But no lear, steady value an e measured. The fault s deteted after 35 ms, ut the drawaks of ths method

9 eome lear. Detetng a fault y the slope method s more dffult, eause the slope of a healthy trajetory, e.g. the tangent funton, already has some values wthn the tolerane for a faulty state. Tunng of the thresholds for 3, 3 and zero and tunng of how many deteted values are requred for a fault flag s neessary. The Modfed Normalsed DC Current Method (fgure 13) and Smple Dret Current method (fgure 14) show muh etter and more relale performane at all tested frequenes. The good omplane of experment and smulaton s also demonstrated y fgure 13. Tme to deteton measurements for all methods are lsted n tale VII. Fgure 15 shows the dagnoss setup. Fg 11:Typal no fault ondton for deteton y Slope Method. (50 Hz, 7 A) Fg 12: Deteton of a Fault n phase w, y Slope Method. (50 Hz, 7 A) Fg 13: Typal Deteton of faulty swth y Modfed Normalzed DC Current sheme. Left: Measurement, (50 Hz, 7 A RMS ) Rght: Smulaton, (50 Hz, 6.8 A RMS ). Fg. 14: Typal deteton of faulty swth y Smple DC Current Method (25 Hz, 7 A RMS ). Fg. 15: Setup nludng PMSM, urrent measurement, and mroontroller C167.

10 V. Conluson Fault Deteton and Identfaton s eomng more and more mportant for ndustral applatons. Sne Varale Speed Drves play a key role n automaton, mprovng ther fault dagnoss apaltes s a neessary task. In ths paper, the up to now less nvestgated dagnoss of open rut faults n three phase voltage soure nverters onneted to AC mahnes wth feld orented ontrol s analysed. Here, a permanent magnet synhronous mahne s used. Several deteton methods for open transstors are evaluated and ompared for losed loop ontrol. The nfluene of the ontrol loops s llustrated y means of urrent waveforms shown for varous urrent frequenes. Even though ampltude and frequeny of the nverter AC urrents are varale due to mahne operaton and the ontrol loop hanges the waveforms of the urrents, two of the ompared dagnoss algorthms are roust enough to aheve a stale deteton for all tested frequenes. The deteton tme s relatvely long and dependng on the urrent frequeny, the omputng effort s low and tunng s easy. No addtonal hardware s needed, f urrent ontrol loops are used. Exeutng the algorthms an e performed y the nherent proessor of the drve. Inludng ths dagnoss sheme wll lose a gap n today's dagnoss methods, eause an open transstor fault wll not neessarly lead to a trpped fuse or an overurrent fault deteton y means of olletor-emtter voltage. The fault may therefore reman undeteted and ould ause seondary faults due to torque pulsaton or nreased urrent n the healthy transstors. Appendx Tale VIII: Varale Speed Drve Data Referenes Type Permanent Magnet Synhronous Nomnal Frequeny 50 Hz Nomnal Power 15.5 kw Stator Resstane 0.08 Ω Stator Indutane 3.1 mh Inerta kg m² Inverter Nomnal Power 22 kw Inverter DC Voltage 800 V Inverter DC Capator 2200µF Inverter Pulse Frequeny 3000 Hz [1] G.A. Capolno. A Comprehensve Analyss of the Current Status n Low Voltage Induton Motor Dagnoss. ICEM 2000, Internatonal Conferene on Eletral Mahnes, Vol.2, pp , Espoo, Fnland, [2] F.W. Fuhs. Some Dagnoss Methods for Voltage Soure Inverters n Varale Speed Drves wth Induton Mahnes - A Survey. IECON03, 29th Annual Conferene of the IEEE Undustral Eletrons Soety,Roanoke, USA, 2003; Proeedngs on CD. [3] D. Kastha, B.K. Bose. Investgaton of Fault Modes of Voltage-Fed Inverter System for Induton Motor Drve. IEEE Transatons on Industry Applatons, Vol. 30, No. 4, pp , [4] R. Peuget, S. Courtne, J.P. Rognon. Fault Deteton and Isolaton on a PWM Inverter y Knowledge-Based Model, IEEE Transatons on Industry Applatons, vol. 34, No. 6, pp , [5] A.M.S. Mendes, A.J.M. Cardoso. Fault Dagnoss n a Retfer-Inverter System Used n Varale Speed AC Drves, y the Average Current Park's Vetor Approah, EPE'99. 8th European Conferene on Power Eletrons and Applatons, pp.1-9, Lausanne, [6] S. Aramk, W. Sleszynsk, J. Neznansk, H. Pquet. A Dagnost Method for On-Lne Fault Deteton and Loalzaton n VSI-Fed AC Drves, EPE 2003, 10th European Conferene on Power Eletrons and Applatons,Toulouse, Frane. CD-ROM paper. [7] K. Rothenhagen, F. W. Fuhs. Performane of dagnoss methods for IGBT open rut faults n voltage soure atve retfers. PESC '04 Power Eletrons Spealsts Conferene, Vol. 6, pp , Aahen 2004 [8] S. Hünemörder, M. Berhoff, F.W. Fuhs. Drve wth Permanent Magnet Synhronous Mahne and Voltage Soure Inverter for Wnd Power Applaton. NORPIE 2002,Nord Workshop on Power and Industral Eletrons, Stokholm, 2002; Pro. on CD.