Fault Detection and Diagnosis in Induction Machines: A Case Study

Transcription

1 Fault Detecton and Dagnoss n Inducton Machnes: A Case Study Mguel Marques, João Martns, V. Fernão Pres 3,4, Ru Das Jorge 5, and Luís Flpe Mendes 5 Department of Electrcal and Computer Engneerng, Faculty of Scences and Technology Unversty New of Lsbon, Qunta da Torre, Caparca, Portugal CTS/UNINOVA, Faculty of Scences and Technology Unversty New of Lsbon, Qunta da Torre, Caparca, Portugal 3 Insttuto Poltécnco de Setúbal, Campus do IPS, Estefanlha, Setúbal, Portugal 4 Center for Innovaton n Electrcal and Energy Engneerng (CIEEE), Lsbon, Portugal 5 EFACEC, Rua da Garagem, nº, Carnaxde, 79-78, Oeras, Portugal Abstract. Nowadays n ndustry there many processes where human nterventon s replaced by electrcal machnes, especally nducton machnes due to hs robustness and performance. Although, nducton machnes are a hgh relable devce, they are also susceptble to faults. Therefore, the study of nducton machne state s essental to reduce human and fnancal costs. It s presented n ths paper an on-lne system for detecton and dagnoss of electrcal faults n nducton machnes based on computer-aded montorng of the supply currents. The man objectve s to detect and dentfy the presence of broken rotor bars and stator short-crcuts n the nducton motor. The presence of faults n the machne causes dfferent dsturbances n the supply currents. Through a statonary reference frame, such as αβ-vector transform t s possble to extract and manpulate the results obtaned from the supply currents usng Prncpal Component Analyss (PCA). Keywords: nducton motor, dagnoss, fault detecton, prncpal component analyss, PCA, egenvalue, egenvector. Introducton Three-phase nducton machnes perform crtcal functons as part of ndustral processes. It s estmated that nducton motors typcally consume 4 % to 5 % of all electrcal energy produced n a country []. Therefore, nducton motors have a specal role n the economy of ndustralzed countres. Due to ts mportance, s urgent to develop ntellgent systems that detect the presence of faults n the machnes n order to reduce mantenance costs. These systems wll allow the possblty of scheduled mantenance and predct the need for mantenance before serous deteroraton or fault occurs, makng t possble to ncrease the relablty of equpment, the mprovement of hs behavor and performance []. L.M. Camarnha-Matos, S. Tomc, and P. Graça (Eds.): DoCEIS 3, IFIP AICT 394, pp , 3. IFIP Internatonal Federaton for Informaton Processng 3

2 8 M. Marques et al. Concernng to mantenance of nducton machnes, preventve mantenance s wdely used currently. It conssts n perodc nspectons wth the objectve of replacng parts that are supposed to break after a certan number of hours. However, even wth perodc nspectons ths type of mantenance presents dsadvantages such as, unneeded mantenance costs and n some stuatons catastrophc falures stll lkely to occur. Thus, the concept of condton-based mantenance (CBM) has emerged as an alternatve to preventve mantenance. Ths type of mantenance s defned as the process of montorng characterstcs or parameters of a machne, n order to verfy early changes and trends that can be used to ndcate a fault stuaton or the need for mantenance. In the last years, FDD n nducton machnes s a research area that had a great evoluton, as seen by the number of proposed methodologes, such as neural networks [3], current space patterns [4], fuzzy logc [5], spectral analyss [6] and vbraton montorng [7]. Contrbuton to Internet of Thngs The paper presents the development of a commercal applcaton for fault detecton and dagnoss of electrcal faults n nducton machnes. Ths analyss, based on PCA methodology [8,9], allows to conclude the practcal feasblty of on-lne montorng through current space pattern analyss usng an ndustral product, such as the Termnal Protecton Unt (TPU) developed by EFACEC. In the future the use of condton-based systems, based on non-nvasve measurements, wll contrbute to connect the nducton machnes wth cyberspace makng t possble to know the state of the machne remotely. Thus, snce TPU allow the use of multple communcaton standards t wll be possble to make a remote montorng and dagnoss of the devces connected to TPU. 3 Prncpal Components Analyss (PCA) Prncpal Component Analyss s a non-parametrc statstcal method used to reduce the number of orgnal varables, whch are correlated, n a set of new uncorrelated varables referred as Prncpal Components (PC). The frst publc descrptons of ths method were gven n 9 by Pearson [] and latter developed n 933 by Hotellng []. PCA can be obtaned through several ways, such as egenvalue decomposton of a matrx or sngle value decomposton (SVD) of a matrx []. In the case of egenvalue decomposton t conssts n the representaton of matrx n terms of ts egenvalues and egenvectors. Through the defnton of egenvectors, ths technque s able to obtan the man drectons of the data sample on a space-vector. It also possble to measure the weght of the sampled data spread through the man drectons defned by the egenvectors. These metrc values are defned as egenvalues []. n m Let X R represents a data matrx, where n denotes the number of T m n measurements and m denotes the number of physcal varables. The R X

3 Fault Detecton and Dagnoss n Inducton Machnes: A Case Study 8 represents the transposed matrx of X, where m and n have the same meanng as n the T X matrx. From the product of the two matrces X and X s obtaned a square matrx n n E R called correlaton matrx. T E = X X () After establshng the correlaton matrx the egenvectors and the respectve egenvalues, of E are calculated. There are several ways to defne egenvectors and egenvalues, the most common approach defnes an egenvector of the matrx E as a vector that satsfes the followng equaton: ( E - λ I)u = () In fact, ths concept of reducng the number of varables through PCA s useful n energy systems, partcularly three-phase systems, such as three-phase nducton machnes. In three-phase energy systems wthout neutral connecton t s usual to use the αβ-vector transformaton to reduce the number of orgnal varables. Ths transformaton converts the three-phase currents or voltages nto an equvalent twophase system. So n deal condtons, the three-phase currents lead to αβ-vector wth the followng components: α β = = 3 a b 6 b c 6 c α β = = 6 6 I sn ωt π I sn ωt Under normal condtons and wth a balanced and constant frequency power supply, a pure snusodal sgnal makes a crcular pattern centered at the orgn of the αβ coordnates. In Fg. there s the representaton of a healthy motor nput current n the αβ-vector pattern. However under abnormal condtons and consderng a constant frequency power supply the prevous condtons are no longer vald and the αβ pattern loses ts crcular (3) Iα(pu) Fg.. Healthy motor nput current αβ vector pattern

4 8 M. Marques et al. shape. For a stuaton where occurs a stator wndng fault the nput current αβ pattern becomes an ellpse because there s an ampltude varaton n the current of the wndng that s n a fault stuaton. The patterns related to a stator wndng fault are presented n the Fg Iα(pu) (A) Iα(pu) (B) Iα(pu) (C) Fg.. Stator fault nput current αβ-vector patterns. (A) stator fault n phase A (B) stator fault n phase B (C) stator fault n phase C When the motor presents a rotor fault stuaton the αβ-vector pattern presents a crcular shape but the egenvalues are not constant. It s possble to observe (Fg. 3) the appearance of a thck rng and the thckness of the rng ncreases wth the severty of the fault. Cardoso et al. [3] concluded that the severty of the fault s proportonal to the number of the rotor bars, but there s a moment where severty of the factor decreases as the number of broken bars ncreases Iα(pu) Fg. 3. Rotor fault nput current αβ-vector pattern 4 Expermental Results 4. Expermental Set Up The expermental set up used s depcted n Fg. 4. A seres of tests were conducted on three squrrel-cage nducton motors wth a mechancal power (P mec ) of Hp, 3/4

5 Fault Detecton and Dagnoss n Inducton Machnes: A Case Study 83 V nomnal voltage (V nom ), a rated speed (N) of 3 rpm, all wth same parameters. One motor was consdered a healthy motor and tested. The other two motors were tested wth stator short-crcuts and broken rotor bars faults. Fg. 4. Expermental apparatus used n ths work In the performed tests the motors {5} were fed wth a three-phase auto-transformer wth an apparent power (S) of 4 kva and -4 V lnl (lne-to-lne). The mechancal load was appled to the nducton motor by connectng the shaft to a DC generator {6} (controlled by a dc current source{3}) of.75 kw rated power (P el ). The output of the DC generator was connected to a varable resstve load. In order to allow tests to be performed at dfferent load levels, the DC exctaton current and the load resstor were both adjustable. The data acquston, sgnal condtonng and data processng are performed by the TPU S {4} developed by EFACEC. For the laboratory tests, a broken rotor bar fault was ntroduced by drllng a hole nto a bar, the hole dameter s slghtly larger than the bar wdth. In the case of short-crcuts n the stator wndngs they were appled by ntroducng an external varable resstor n seres wth the wndngs of each phase. 4. Expermental Results The motor was ntally tested n a healthy stuaton, wth the stator wndngs and ts cage ntact, n order to verfy the current αβ-vector reference pattern. In the conducted tests were appled varous levels torque, n order to verfy the robustness of the algorthm at dfferent load levels. The start-up condton was dscarded snce t s not consdered by the algorthm. The temporal evoluton of the motor stator currents s represented n the Fg. 5 (A). The Fg. 5 (B) does not present the αβ-vector wth a crcular shape due to expermental lmtaton, such as harmonc dstorton n the supply voltage and the magnetc feld dstrbuton n the machne s not perfectly snusodal. However the machne was consdered n a healthy condton. For stator faults were used three

6 84 M. Marques et al. varable resstors wth. Ω/ 5A n seres wth the mpedance of each phase of the machne (Z = 4.8 Ω). In the case of a fault severty factor of 6% the resstance value s. Ω. The αβ-vector (Fg. 6) no longer presents a crcular shape and exhbt an ellptcal shape. Ampltude (A) 4 - Stator Currents X:.9 Y: X:.7 Y: Current A Current B Current C X:.34 Y: Tme (s) (A) Frequency content of stator current X: 49.8 Y: Magntude (db) Iα(pu) (B) Frequency (Hz) (C) Fg. 5. (A) Stator currents of the machne n nomnal operaton (B) Expermental αβ-vector (C) Current A spectrum Ampltude (A) 5 Stator Currents X:.95 Y: 4.77 X:. Y: Current A Current B Current C X:.37 Y: Tme (s) (A) Iα(pu) (B) Fg. 6. Expermental results obtaned for a stator fault stuaton n nomnal operaton wth a SF = 6 % n the phase A (A) Stator currents of the machne (B) Expermental αβ-vector The results obtaned for 6 broken rotor bars (major fault) are shown n the Fg. 8. The appearance of a thck rng (Fg. 8 (A)) n a rotor fault stuaton can be detected through the varaton of egenvalues (Fg. 8 (B)). In the expermental results the egenvalues does not present a snusodal behavor, but show a perodc varaton over tme.

7 Fault Detecton and Dagnoss n Inducton Machnes: A Case Study 85 The results obtaned for 6 broken rotor bars (major fault) are shown n the Fg. 8. The appearance of a thck rng (Fg. 8 (A)) n a rotor fault stuaton can be detected through the varaton of egenvalues (Fg. 8 (B)). In the expermental results the egenvalues does not present a snusodal behavor, but show a perodc varaton over tme Iα(pu) (A) - - Iα(pu) (B) Fg. 7. Expermental αβ-vector pattern for a SF = 6 % (A) Phase B (B) Phase C Iα(pu) (A) Egenvalue Computaton Cycles (B) Fg. 8. Expermental results obtaned for the machne wth 6 broken rotor bars (A) Expermental αβ-vector pattern (B) Varaton of the egenvalues over the computaton cycles 5 Conclusons and Future Work The research n the feld of fault detecton and dagnoss could provde a reducton n operatonal costs and the ncrease of the machne s relablty that assocated wth remote systems wll enable the nterconnecton of the nducton machnes wth the cyberspace. In ths paper s presented an on-lne fault detecton and dagnoss system for threephase nducton motors based on PCA. The proposed system uses the egenvalue decomposton to detect changes n the αβ-vector pattern.

8 86 M. Marques et al. The expermental results ndcate that t s possble to detect the presence of shortcrcuts n the stator wndngs and broken rotor bars n nducton machnes wth ths FDD method. Through the analyss of the egenvalues from the αβ vector components t s possble to nfer f the machne s n a fault stuaton or not. For stator fault stuatons usng the egenvectors, t s also possble to dentfy n whch phase the fault occurred. Concernng to future works the stator faults were appled through the addton of varable resstors n seres wth the stator wndngs. Ths set up can be consdered as an approxmaton to the fault stuaton. In the future should be short-crcuted some stator wndngs and test the algorthm to observe hs behavor. Acknowledgements. Authors would lke to thank to CTS of UNINOVA for the fnancal support and to EFACEC for provdng ts facltes and equpment for ths work. References. Ahmed, I., Supangat, R., Greger, J., Ertugrul, N., Soong, W.L.: A Baselne Study for On- Lne Condton Montorng of Inducton Machnes. In: Australasan Unverstes Power Engneerng Conference, pp. 6 9 (4). Syambalaptya, D., Mclaren, P.: Relablty Improvement and Economc Benefts of On- Lne Montorng Systems for Large Inducton Machnes. IEEE Transactons on Industry Applcatons 6(6), 8 5 (99) 3. Chow, M., Mangum, P.: Neural Network Approach to Real-Tme Condton Montorng of Inducton Motors. IEEE Transactons on Industry Electroncs 38(6), (99) 4. Cardoso, A., Sarava, E.: Computer Aded Detecton of Argap Eccentrcty n Operatng Three-phase Inducton Motors, by Park s Vector Approach. IEEE Transactons on Industry Applcatons, (993) 5. Nejjar, H., Benbouzd, M.: Applcaton of Fuzzy Logc to Inducton Motors Condton Montorng. IEEE Power Engneerng Revew, 5 54 (999) 6. Thomson, W.: On-lne MCSA to Dagnose Shorted Turns n Low Voltage Stator Wndngs of 3-Phase Inducton Motors pror to Falure. In: IEEE Internatonal Electrc Machnes and Drves Conference, pp () 7. Tsypkn, M.: Inducton Motor Condton Montorng: Vbraton Analyss Technque - a Practcal Implementaton. In: IEEE Internatonal Electrc Machnes & Drves Conference, pp () 8. Martns, J.F., Pres, V.F., Amaral, T.: Inducton motor fault detecton and dagnoss usng a current state space pattern recognton. Pattern Recognton Letters 3(), 3 38 () 9. Pres, V., Martns, J., Pres, A.: On-Lne Dagnoss of Three-Phase Inducton Motor Usng an Egenvalue αβ-vector approach. In: IEEE Conference on Power Electroncs and Drve Systems, pp (7). Pearson, K.: On Lnes and Planes of Closest Ft to Systems of Ponts n Space. Phlosophcal Maganze, (9). Hotellng, H.: Analyss of a complex of statstcal varables nto prncpal components. Journal of Educatonal Psychology 4, (933). Jollffe, I.: Prncpal Component Analyss. Sprnger, New York () 3. Cardoso, A., Cruz, S., Carvalho, J., Sarava, E.: Rotor Cage Fault Dagnoss n Three- Phase Inducton Motors, by Park s Vector Approach. In: IEEE Industry Applcatons Conference, vol., pp (995)