Identification of incipient faults in subsea HVDC systems

Transcription

1 Identification of incipient faults in subsea HVDC systems Antônio P. C. Magalães 1, João P. L. Salvador 1, Antonio C. S. Lima 1, and M. Teresa Correia de Barros 2 1 Universidade Federal do Rio de Janeiro COPPE/UFRJ Rio de Janeiro, RJ, Brazil Abstract Tere is an increase interest in offsore applications involving te connection of renewables and a better usage of subsea oil plants. In bot scenarios tere is te possibility of applying a HVDC transmission system. A DC submarine cable may present lower costs decreasing te initial large investment for te system installment. However, for a suitable operation, one must be aware of possible faults and te associated current/voltage stress. A dielectric failure may cause te wole system to a collapse, given its unattended and virtually inaccessible caracteristics. In tis paper, we investigate te performance of distinct fault location scemes based on travelling waves wen submarine cables are involved. Te first goal is to evaluate te dielectric stress associated wit possible faults and we also analyze te impact of te accuracy associated wit te cable parameters may ave in a fault location sceme. Index Terms HVDC systems, DC grids, fault location. I. INTRODUCTION Nowadays we can note a rapid expansion of offsore wind farms and oter renewable sources as ydrokinetic systems [1] and an ever-growing interest in oil fields far from te coast [2]. In all tese scenarios, complex underwater cable systems are expected. In te case of a subsea oil plant, te system comprises a variable speed drive (VSD), a step-up transformer, an umbilical cable and a motor. Tere are several companies developing subsea power equipment, like transformers, VSDs, circuit breakers, connectors, to name just a few. In fact, tere is a great movement toward developing subsea HVAC equipment, like transformers, switcgears, VSDs. For te transmission of iger power ratings, electric manufacturers are proposing to increase te voltage of te transmission and/or decrease drastically te transmission frequency. Te main purpose is to avoid compensating bulk reactive power [3]. A possibility to diminis te overall installation costs is to use DC cables, wic would use a simpler umbilical cable. Te umbilical cable is te most expensive item of a subsea electrical grid. It comprises te power cables reinforced wit mecanical structure to witstand dynamic and static loads. Suc complex item as been used on point-to-point connections, wit motors rated mainly between 4 and 6 kv. Te low voltages are due to dimension restriction of te penetrators (responsible for te 2 Insitituto Superior Técnico Universidade de Lisboa, IST-UL Lisboa, Portugual electrical feed troug connections at te motors) of motors installed inside te wells. Tis very strict dimension restriction no longer exists wen te motors are installed on te seabed, inside Modules of Boosting (MoBos), allowing te voltage increase of te transmission of point-to-point connections. Furtermore, te possibility of using a subsea HVDC becomes even more attractive given its applicability to long distance transmission [4]. In an oil system, it is expected to operate wit frequencies below 50/60 Hz for a quite long period of time, depending on te oil reservoir caracteristics [5][6]. For te renewables, te use of HVDC systems may be interesting for an easier connection to te AC grid. In eiter configuration tere are several protection issues related to te identification of faults along te cables. Altoug tere are some works dealing wit incipient faults in underground cables [7]-[13], it is not straigtforward to adapt tem to submarine cables given te peculiarities of suc configuration. In fact, to te best of te autor knowledge, te tecnical literature as not addressed te analysis of incipient fault in submarine cables. Te paper is organized as follows: Section II presents te network configuration including details on te submarine cable considered for te analysis. Section III presents te simulation tests used for te fault identification considering distinct converter topologies. Section IV proposes some identification tecniques tat could also be installed to detect te incipient fault in te submarine cables. Te main conclusions of tis paper are presented in Section V. II. NETWORK CONFIGURATION AND FAULT SPECTRUM Te number of application involving subsea DC cables is increasing rater rapidly and tere are some possible configurations, e.g., independent cables for te positive and negative poles or te use of umbilical cables. In tis work, we ave only considered single core cables wit armor. We assume tat eac cable is laid at a distance sufficiently apart from eac oter as mutual coupling between positive and negative poles can be neglected. An off-sore wind-farm is assumed to be connected to a local power grid by means of a subsea HVDC Link. Initially, we assume a wind-farm located 50 km from te sore. Te wind-farm, te rectifier station and te cable are considered off-sore. An on-sore inverter connects te system to te Tis work received a partial financial support from Conselo Nacional de Desenvolvimento Científico e Tecnológico - CNPq, Coordenação de Aperfeiçoamento de Pessoal de Nível Superior CAPES, Instituto Nacional de Energia Elétrica - INERGE and Fundação de Amparo à Pesquisa do estado de Minas Gerais - FAPEMIG. Paper submitted to te Power Systems Computation Conference (PSCC2016) in Genoa, Italy, June 20-24, 2016.

2 AC grid via a step-up transformer; see Fig. 1. We assume a constant cable dept of 25 m trougout te wole distance, for te scope of tis work. Figure 2 details te subsea cable used ere and Table 1 sows te cable data. (a) cable layout Fig. 1. Proposed System (a) treeing equivalent circuit Fig. 3. Scematic representation of te treeing process. Te wole system was simulated in a frequency domain program built in Matematica. We ave considered a more general approac including te effect of te external media admittance suc as in [17]. However, based on te values of te sea conductivity and its relative permittivity it was possible to neglect te latter, leading to an external media wit only a return impedance. Tis also allowed us to simplify te return impedance expressions as in [18]. See Appendix A for furter details for te evaluation of te cable parameters and te system modeling in te frequency domain. cable configuration Fig. 2. Subsea cable considered in te simulations Table 1: SC Cable Data Ω Ω Ω Steel Tere are multiple possible causes of failures for a HVDC system. However, tey can be grouped in mainly two types : (i) cables faults; (ii) converter and transformer misoperations. For te former, wen we consider underground or submarine cables, te electro-cemo-mecanical penomenon known as water treeing [14]-[16] is te main cause of insulation failure. Altoug it is very ard to control and probably impossible to avoid, te water treeing process must be monitored and detected. In its final stages, a water treeing will cause te insulation to fail, as a low impedance pat between core conductors and seat is establised. In its early stages, it represents a decrease of te insulation level. Figure 3 scematically represents te treeing effect in te cable insulation, were R d is a variable resistance used to represent te treeing process in te simulations. We postulate to approximate te penomenon by a varying resistor, tus, small resistance connecting core conductor and seat was included to represent an almost complete treeing process and a ig value of resistance to take into account te early stages of treeing. Te approximation of te treeing penomenon by a variable resistance allow us to implement a rater simple circuit to determine te fault spectrum based on its location and/or on te caracteristics of te fault, i.e. weter it is an incipient fault or an insulation failure. In fact, R d can be understood as a fault impedance tat by varying its value and location we can assess te voltage/current spectrum associated wit te fault. Fig. 4. Proposed system for te fault spectrum assessment Unlike underground cable systems, in subsea configuration te seat and armor are usually bolted togeter at te receiving end and are ungrounded. In te onsore part, i.e., sending end, tere are cases were tere is only one grounding point and oters were te wole system remains ungrounded. Terefore, we ave decided to take advantage of tis scenario to identify weter a more suitable connection point acquire more insigt about te cable insulation ealt. Te circuit configuration used for te assessment of te fault current/voltage spectrum is depicted in Fig. 4. Te lengt x corresponds to te fault location and it was varied from 500 m up to 45 km. An ideal impulse voltage source, i.e. a Dirac s delta function is connected at te cable sending end (terminal 1 in Fig.4). Resistor R c is used to represent sort-circuit and open-circuit scenarios. We ave considered 0.01 mω for sort-circuit and 10 MΩ for open circuit. Te sending-end is assumed to be on-sore, tus R 2 wit a value of 0.01 Ω was assumed, wile R 1 =1 kω was used to separate seat and armor. A value of 10 MΩ was for R 3 to represent an ungrounded system. Te fault was assumed to occur at several locations along te cable wit te values of R d varying from 1 mω, representing insulation failure scenario, up to 1 MΩ, in te case of an incipient fault. In tese tests we measured te injected currents and te voltage induced in all nodes. A second test consists of replacing te impulse voltage for an impulse current and aving all te voltages measured.

3 (a) Fig. 5. Open-circuit Current Spectrum at Terminal 1 for distinct values of (a) Fig. 6. Sort-circuit Current Spectrum at Terminal 1 for distinct values of Due to te paper lengt restriction, we present ere only te results related to te core conductor. Figure 5 depicts te results for te open-circuit results at node 1 (core cable sending end) for a frequency spectrum varying from 1 μhz up to 10 khz wen an impulse voltage is connected at node 1. It can be seen tat in te case of te incipient faults (ig values of resistance R d ) tere is a rater very noticeable pattern of ig frequency oscillations. For te sort-circuit current spectrum it was possible to identify an anti-resonance occurs around 100 Hz for te case of incipient faults. It is interesting to note tat similar to te open-circuit tests, te case of insulation failure would not be easily identified. Te results for various values of te R d and fault location are presented in Fig. 6. An important aspect found in tose tests is tat te incipient faults presented a similar current profile wit te resonances or anti-resonances occurring at basically te same frequencies regardless of faulted point in te cable lengt. Altoug not sown ere, a similar beavior was found for distinct cable lengts. In tis case, te difference lies in te frequency were te resonances (or anti-resonances) occur. Te results for te voltage spectrum considering open and sort circuits conditions are presented in Figs. 7 and 8. A similar beavior was found. Te incipient faults presented a more concise response wit te resonances in bot sort and open circuits tests occurring at some given frequencies around 1 khz. It is interesting to note tat te case of insulation failure presented very distinct beavior depending on te fault location. III. FAULT IDENTIFICATION Te results in te previous section indicate tat early stages in te treeing process can be observed and detected in te frequency domain by te canges in te voltage and current spectral density. Given tat te main idea is to identify te incipient fault by exciting some of te fault resonance frequencies, canges in te measured voltage, bot amplitude and pase, may be of use for tis detection. It is a known fact from complex variable analysis tat a resonance peak will also contain an abrupt cange in te pase of te complex variable under analysis. Te circuit used for te identification of te incipient faults in te HVDC cable is sown in Fig. 9.

4 (a) Fig. 7. Open-circuit Voltage Spectrum at Terminal 1 for distinct values of (a) Fig. 8. Sort-circuit Voltage Spectrum at Terminal 1 for distinct values of Initially, we consider te case were te conventional LCI-HVDC (based on tyristors) is used. We used a very simple modeling for te converter as in tis case te oter main effects associated wit a non-ideally DC-current source are of second order in te analysis ere proposed. In Fig. 9 te current source are step functions in te frequency domain. We assumed tat te injected current is te same as te current in te receiving end. Te voltage source V s injected in te cable seat is a simple sinusoidal voltage at 100 Hz and wit a low amplitude, 0.01 V is used for te amplitude of te inject voltage source. To improve te observability we ad to use a small capacitor, 15.9 connecting te voltage source and te cable seat, te resistor used to separate te cable seat and armor is 1 Ω, te grounding resistor is now 10 Ω and is te same as before. Tis topology allows us to use te voltage at te cable armor as te reference for te treeing inception. Te frequency of 100 Hz was considered in order to verify weter previously found resonances would be excited or not. Tis is an important aspect as tere is not guarantee tat a previous evaluation of a given cable can be determined to allow a better definition of te injected voltage at te cable seat Fig. 9. Proposed Configuration for Fault Identification To test te proposed identification sceme, we considered initially te case of a fault occurring near te on-sore inverter station, i.e. at 1 km from te inverter station, and near te oter end close to te off-sore rectifier station, i.e., at 49 km. Figures 10 and 11 sow te voltage response at terminal 3, i.e., cable armor for a fault at 1 km considering different values for R d. It can be seen tat in tis configuration regardless of te value used to represent te

5 treeing penomenon tere is a sensible gain in te voltage around 100 Hz. Tere is also an abrupt cange in te pase of te armor voltage around 100 Hz as sown in Fig. 11. Appendix B we sow te results for 75 km cable lengt wit fault in 74km. Fig. 10. Magnitude of cable armor voltage for a fault at 1 km. Fig. 14. Magnitude of cable armor voltage for a fault at a random selected position. Fig. 11. Pase of cable armor voltage for a fault at 1 km. In te case of a fault at 49 km we observe a cange in te magnitude of te cable armor voltage for te incipient fault; see Fig. 12. Again tere is a sensible cange around 100 Hz for most of te cases, wit te exception for te value of 100 Ω, were a cange around 50 Hz is observed. As expected, canges in te amplitude of te cable armor voltage sow a noticeable cange in te pase for te same voltage. Fig. 12. Magnitude of cable armor voltage for fault at 49km Fig. 13. Pase of cable armor voltage for fault at 49km As a final test for tis topology, we consider te case were a random number generation is used to define te point were te fault occurs. In tis case, te selected fault position was at km. Te results for te voltage at te cable armor are sown in Figs. 14 and 15. Again tere is a sensible voltage oscillation around 100 Hz wit only one case were tere is a resonance sligtly below 100 Hz. Neverteless, te abrupt cange in bot voltage magnitude and pase around 100 Hz serves as an indication tat regardless of te fault position an identification of an incipient fault is possible. In Fig. 15. Pase of cable armor voltage for a fault at a random selected position. Anoter possible configuration is associated wit voltage source converters. Altoug tis configuration is not so easily extended to large power transmission, it is possible tat in a near future tis configuration will be more widely used. To test weter te identification is possible, we consider a sligtly modified version of te circuit in Fig. 9. Instead of te current sources, DC voltage sources are considered, a voltage drop of 15% was considered as tis seems to be found in some oter topologies of conventional HVDC systems. In Figs. 16 and 17 te armor voltage in magnitude and pase for a fault at 1 km are depicted. Given te cange in te source caracteristic tere is a sensible cange in te voltage profile, owever, again a steep increase in te magnitude can be found around 100 Hz and an abrupt cange in pase around te same value of frequency. We observe a similar beavior wen te fault occurs at distinct lengts. As an illustration, consider now te case were te fault occurred at km, again using a random number generator. Te results are presented in Figs. 18 and 19. It can be seen tat tere is a similar beavior of te voltage profile around 100 Hz for te sort fault (at 1 km) and te randomly selected fault, at 26.3 km. Altoug not sown ere, tis type of beavior did also appen for faults close to te end of te cable. IV. DISCUSSIONS Te results presented in te previous section indicated tat independently of te type of converter used in te HVDC configuration, i.e., current source or voltage source, it is possible to identify an incipient fault as long as associated wit treeing in te cable insulation.

6 are encapsulated in te same pipe. Anoter future work perspective is related to te beavior of subsea cables in deep waters aiming a true subsea HVDC system as in te case of te oil exploitation at great depts, i.e., over 2 km below sea level. Fig. 16. Magnitude of cable armor voltage for a fault at 1km considering a voltage source HVDC converter. Fig. 17. Pase of cable armor voltage for a fault at 1km considering a voltage source HVDC converter. V. CONCLUSIONS Tis work as addressed te topic of identifying incipient faults in submarine cables used in off-sore wind-farm topologies. Some simple configurations were considered to identify te frequency response of te cable in te case a fault occurs along te cable lengt. Tis type of fault is associated wit te treeing penomenon occurring at te cable insulation between te core conductor and its seat. Te results indicate tat tere are some noticeable resonances regardless of te value of te resistance used to represent te treeing. Furtermore, tese results indicated tat noticeable differences occur in te voltage/current profile even for incipient faults, i.e., ig impedance faults wic are caracteristic of te early stages of te treeing effect. All te tests presented ere ave focused only on te positive pole, owever, similar results were found wen considering te negative pole. Fig. 18. Magnitude of cable armor voltage for at a randomly selected point considering a voltage source HVDC converter. Fig. 19. Pase of cable armor voltage for at a randomly selected point considering a voltage source HVDC converter. Te ypotesis of assuming te treeing beavior as an equivalent resistance causing a sort circuit between te cable core conductor and its seat is based on preliminary studies on te nature of te cable insulation at lower frequencies, around a few ertz. Naturally, furter work is needed to assess weter or not tis assumption olds for te frequency band considered. Te usage of a simple detection circuit allows to detect te fault regardless te lengt were it migt occur and independently of te converter type considered. Naturally, studies regarding te electromagnetic compatibility of suc circuit are needed. Te same can be said about te protection of suc circuit during outages due to converter malfunctioning. Future work will deal wit te analysis of tis system using a more detailed representation of te converter station possibly in an EMT-type of program and wit cases dealing wit umbilical cables were bot positive and negative poles APPENDIX A CABLE PARAMETERS CALCULATION AND MODELING To obtain rigorously te parameters impedance and admittance per unit lengt (p.u.l.) of an overead line or an underground cable, we must solve te so-called full-wave equation to initially derive te beavior of te propagation constant in te frequency spectrum of interest. For most of te application in power systems as pointed out in [17], we can resort to te quasi-tem approximation, witout great loss. Neverteless, even in tis case, te p.u.l. parameters involve te evaluation of infinite integrals wit igly oscillatory kernels wic tends to be a numerical callenge mainly wen underground or subsea cable are considered. Tis scenario led to a searc of simplified expressions. As mentioned before, we opt to use te return expressions based on asymptotic approximation of te infinite integrals used in [18] tat even toug were developed for a ground return impedance can be implemented as a sea return impedance for cables in sallow waters. For a single core cable wit armor, te expressions of impedance and admittance of te cable can be formulated as in [19]. Te impedance matrix of te cable is given below Z= Zi + Z ext (1) were Z is te internal matrix impedance of te cable and Zext i is te impedance matrix of te external media, given by ze ze ze Z ext = ze ze z e ( ze ze z e wit eac element given by (3) as indicated in [18]

7 jωμ0 Zext = K 0 η r6 + (1+ 2π jωμ 0 (1+ r6 (3) + K 2 η r6 + (1+ 2 π (1+ + r 6 jωμ 0 2 exp[ (1+ ](1 + (1+ η) 2 2 π η ( r6 + (1+ ) For tis cable, te admittance matrix is given by (6). yc yc 0 Y = yc yc + ys y (6) s 0 ys ys + y a being yc, ysand y a respectively te admittances of te core, seat and armor 2π jωε 2π jωε 2π jωε yc = ys = ya = (7) r ln 2 r ln 4 r ln 6 r1 r3 r5 A. Caracteristic Admittance and Propagation Function Te relation between te voltage and current waves is 1 given by te caracteristic admittance Yc = Z ZY. and te delay and distortion of te tese waves propagating between te line ends of lengt is given by te propagation function H = exp ( YZ. ) were Z and Y calculated by (1) and (6) respectively. For te nodal admittance matrix formulation, as described in [20], using te expression of (8) and (9), is given by A B Yn = (10) B A wit A = Y ( I+ H )( I H ) 1 2 and B = 2Y ( I H ) 1 c were I is te identity matrix. Te system is evaluated as 1 V = Yn I (1 being V te nodal voltage vector and I is te current vector. REFERENCES [1] M. Kuscke, K. Strunz, "Energy-Efficient Dynamic Drive Control for Wind Power Conversion Wit PMSG: Modeling and Application of Transfer Function Analysis," IEEE Journal of Emerging and Selected Topics in Power Electronics,, vol.2, no.1, pp.35,46, Marc 2014 [2] TOTAL. EOR maximizing recovery factors [Online]. Available: ttp:// pedagogique/total-eorbis-gb2.pdf [3] G. Asplund, K.Eriksson, H. Jiang, J. Lindberg, R. Pålsson and K. Svensson. DC Transmission Based on Voltage Source Converters, CIGRÉ SC14 Colloquium, Sout Africa, [4] S. J. Sao and V. G. Agelidis, Review of DC System Tecnologies for Large Scale Integration of Wind Energy Systems wit Electricity Grids, Energies, n. 3, ISSN , pp , [5] A. Mæland and R. S. Cokawala, Powering Oil & Gas Offsore Operations from Mainland Electrical Grid, World Energy Congress, Montreal, c [6] F. Wang, L. Bertling, T. Le, A. Mannikoff and A. Bergman. An Overview Introduction of VSC-HVDC: State-of-art and Potential Applications in Electric Power Systems, CIGRÉ Session, Bologna, [7] W. E. Anderson, J. D. Ramboz and A. R. Ondrejka, "Te Detection of Incipient Faults in Transmission Cables using Time Domain Reflectometry Tecniques: Tecnical Callenges". In: IEEE Transactions on Power Apparatus and Systems, vol. PAS-101, no. 7, pp , July [8] W. L. Weeks and J. P. Steiner, "Instrumentation for te Detection and Location of Incipient Faults on Power Cables" In: IEEE Transactions on Power Apparatus and Systems, vol. PAS-101, no. 7, pp , July [9] N. G. Trin, C. Melançon, C. Vincent, H. Le Hong and G. Larose, "Effect of Capacitor-Switcing Surges on Insulation Aging and Incipient-Fault Breakdown of Distribution Cables and Accessories", In: IEEE Transactions on Power Delivery, vol. PWRD-2, no. 3, pp , July [10] K. K. Kuan and K. Warwick, "Real-time expert system for fault location on ig voltage underground distribution cables", In: IEE Proceedings, vol. 139, no. 3, pp , May [11] M. J. Mousavi and K. L. Butler-Purry, "Detecting Incipient Faults via Numerical Modeling and Statistical Cange Detection", In: IEEE Transactions on Power Delivery, vol. 25, no. 3, pp , July [12] T. S. Sidu and Z. Xu, "Detection of Incipient Faults in Distribution Underground Cables", In: IEEE Transactions on Power Delivery, vol. 25, no. 3, pp , July [13] A. J. Reid, C. Zou, D. M. Hepburn, M. D. Judd, W. H. Siew and P. Witers, "Fault Location and Diagnosis in a Medium Voltage EPR Power Cable" In: IEEE Transactions on Dielectrics and Electrical Insulation, vol. 20, no. 1, pp , February [14] M. Ieda and M. Nawata. "DC Treeing Breakdown Associated wit Space Carge Formation in Polyetylene." IEEE Trans. Electr. Insul., Vol. EI-12 No. 1, pp , February, [15] A. Nicolls and E. Steennis, Water Treeing, State of te Art. Cigré paper 15/21-02, at CIGRÉ Session 1990, Paris, France. [16] W. Su, and S. Boggs. "Effect of Cable Restoration Fluid on Inibiting Water Tree Initiation." IEEE Trans. Power Delivery, vol. 26, No. 1, pp , [17] A. P. Cardillo Magalães, A. C. Siqueira de Lima, M. T. Correia de Barros. Rational Approximation to Full-Wave Modeling of Underground Cables In: IPST 15 International Conference on Power Systems Transients, Cavtat, Croatia, June [Online]. Avaliable: ttp:// [18] A. C. S., Lima and C. Portela. "Closed-form expressions for ground return impedances of overead lines and underground cables." International Journal of Electrical Power & Energy Systems, vol. 38, no. 1, pp: 20-26, [19] A. Ametani. A General Formulation of Impedance and Admittance of Cables, IEEE Trans. Power Apparatus and Systems, Vol. PAS-99, No. 3, pp , June [20] B. Gustavsen, Validation of Frequency-Dependent Transmission Line Models, IEEE Trans. Power Delivery, vol.20, no. 2, pp , Apr [21] C. H. Cien, R. W. G. Bucknall, Analysis of Harmonics in Subsea Power Transmission Cables Used in VSC-HVDC Transmission Systems Operating Under Steady-State Conditions, IEEE Trans. Power Delivery, Vol. 22, No. 4, pp , October 2007.