20030111836_2003134957

7/27/2019 20030111836_2003134957

1/8

DETEC TION OF HIGH-IMPEDAN CE ARCING FAULT S INRAD IAL DISTRIBUTION DC SY STEM SMarcel0 C. GonzalezCleveland State UniversityDepartment of Electrical EngineeringCleveland, OH 44115

ABSTRACTHigh voltage, low current arcing faults in DCpower systems have been researched at the NASAGlenn Research Center in order to develop a method fordetecting these hidden faults, in-situ, before damageto cables and components from localized heating canoccur. A simple arc generator was built and high-speedand low-speed monitoring of the voltage and currentwaveforms, respectively, has shown that these highimpedance faults produce a significant increase in high

frequency content in the DC bus voltage and lowfrequency content in the DC system current.Based on these observations, an algorithm wasdeveloped using a high-speed data acquisition systemthat was able to accurately detect high impedancearcing events induced in a single-line system based onthe frequency content of the DC bus voltage or thesystem current. Next, a multi-line, radial distributionsystem was researched to see if the arc location couldbe determined through the voltage information whenmultiple detectors are present in the system. It wasshown that a small, passive LC filter was sufficient toreliably isolate the fault to a single line in a multi-linedistribution system. Of course, no modification isnecessary if only the current information is used tolocate the arc. However, data shows that it might benecessary to monitor both the system current and busvoltage to improve the chances of detecting andlocating high impedance arcing faults.

INTRODUCTIONAs future NASA programs and missions greatlyincrease their required electrical power, the move tohigher and higher distribution voltages becomes anecessity. Programs such as the reusable launch vehicle

(RLV) and more electric aircraft will push electricdistribution in the 500kW to 1MW range. NASAs newPrometheus Program seeks to develop a lOOkW+ spacevehicle before the end of the decade, and commercialcommunication satellites are continuing their upwardtrend in power levels to over 20kW of on-board power.All these trends require significant increases in thedistribution voltages in order for these systems toremain viable.

Robert M. ButtonTechnology Product ManagerNASA Glenn Research CenterCleveland, OH 44 135Current state of the art for space electrical powersystems is the International Space Station. With a totalenergy generation capacity of 75kW and a highlychannelized topology, the nominal distribution voltageis 160VDC. The proposed high power systemsenumerated above will require distribution voltagesstarting at 270VDC on up past 600VDC. Not only willthese high voltage distribution systems pose challengesto electrical component manufacturers, but they alsointroduce a new potential fault that must now beaccounted for- lectrical arcs.As the distribution voltage increases, so too doesthe likelihood of damaging electrical arcs. Currentaerospace electrical distribution systems only protectagainst over current faults using eithermechanical/thermal devices or sometimes solid-stateswitches with complex 12T rip curves. However, theseover current protective features are ineffective atdetecting even high levels of intermittent electrical arcs.It is clear that a new distribution switch with arcingfault detection capabilities is going to be a keyrequirement for these new, high power electricalsystems.There has been much interest in the development

df an arcing fault detection circuit breaker forcommercial aircraft, especially following thecatastrophic loss of SwissAir flight 111 in 1998 whichhas been linked to an electrical fire. However, most ofthe research and development into these arcing faultcircuit breakers has concentrated on 400Hz ACdistribution systems found on commercial aircraft. Formany of the high power space systems on NASAsroadmap, DC power systems are probably more likely.In terms of the published research effort that hasgone into arcing fault detection before and after 1998,most, if not all, have focused on AC (50/60Hz) powerdistribution systems. Of these, most rely on theh a rmon i c content in the current and/or voltagesignature2- for fault detection and others rely onpattern recognition through Neural Networks or NeuralNetwork and Fuzzy Logic combination.* Although it ispossible to detect arcing faults through either thecurrent or voltage signature, most of these researchershave chosen to utilize only the current inf~rmation.~~Since little to no work has been reported on high-impedance arcing faults to return in DC systems, thefirst phase of this investigation focused on evaluating1American [ n s t h t e of Aeronautics and AstronauticsThis is a preprint or reprint of a paper intended for presentation at aconference. Because changes may be made before formalpublication, this is made available with the understanding that it will

nn+kn r b ~ r lI r r r r r r l . . r r r l . . . : I L - . . a A L - ----.--.e-. -8 r L r r . . b L - -

7/27/2019 20030111836_2003134957

2/8

the arcing characteristics, specifically, the frequencycontent of the voltage and current signature, through theFast Fourier Transform (FFT) analysis tool in a simplesystem. It is important to note that this research did notfocus on arcing faults to chassis and low-impedancearcing faults since they can be easily detected withexisting methods, such as a Ground Fault CircuitInterrupter (GFCI). The second phase addressed theissue of arc characterization and location in a radialdistribution system.

Cable( L I )(L2)(L3)(L4)(L5)

(L6), (L7)

ARC CHARACTERIZATION

Length (in) Gauge46 1094 1048 10

19.5 1422 1418 +62 10 +110

Acquisition and analysis of the voltage andcurrent signals of interest was done with the graphicalprogramming software LabVIEW 6.0 in conjunctionwith a two-channel, %bit, high-speed digitizer(NATIONAL INSTRUMENTS PCI-5112). Thevoltage signals were first fed through bandpass (BP)filters with corner frequencies at 400 kHz and 30 MHzbefore being sampled at 100 mega-samples per second(MSa/s). The current signals were fed through BPfilters with corner frequencies at 0.16Hz and 9.5 kHzbefore being sampled at 50 kSds. Isolation between theexperimental system and the digitizer input wasprovided by a radio frequency (RF) transformer.Sinde-Line Svstem

The single-line system (SLS), shown in Figure 1below, is comprised of a variable power supply(POWER TEN INC., 15OVDC, 44ADC), an electronicload, IL, in constant-current mode (DYNALOAD,600VDC, 200A, 4000W), and, in parallel with the load,a 10052 planar resistor (OHMITE) in series with anarcing fault mechanism. The 10052 resistor is used tosimulate a high-impedance arcing fault and the arcingfault mechanism was simply a mechanical vise withisolated handles and arcing contacts. The arcingcontacts used for this experiment were carbon weldingelectrodes and the cable lengths and gauges used areshown in Table 1.

Table Contactor

Figure 1: Single-Line Experimental System

Table 1: Cable Length and Gauge

The power supply was set at 15OVDC and thecurrent load was set at either 0.5A or 5A. To inducearcing, the electrodes were manually brought together,by turning the isolated handle, and then slightlyretracted. With the lOOR resistor in series with the arc,the maximum current through the arcing path waslimited to 1.5A.Voltape Signature

The three voltages that were sampled andanalyzed are the arc voltage, Vmc, the load voltage, VL,and the supply voltage, Vs. A snapshot of the arcingvoltage and its corresponding spectral content is shownin Figure 2a and Figure 2b for no-fault and faultcondition with a load of 5A. Note that the magnitude ofthe spectral content of the arc voltage decays withfrequency and frequencies above 30 MHz are presentunder arcing conditions. At the sampling rate of 100MSa/s and 8192 samples per data set, the timedifference between adjacent samples is At, given by

and the spacing between adjacent frequencies, Af, isgiven byf,Af =-=12.207kHz

where f, is the sampling rate and N is the number ofsamples.In a practical application, it is impossible to sensedirectly across the arc voltage, VARC, ince the arcingfault location is unknown. However, the supply outputterminals and the load input terminals provide fixedlocations to sense the arcing fault. Sample plots of thearcing fault signal at the supply and load terminals areshown in Figures 3a,b. Note that the digitizerintroduced a slight DC offset. The supply and loadvoltages were acquired simultaneously but the arcvoltage was acquired separately.

2American Institute of Aeronautics and Astronautics

7/27/2019 20030111836_2003134957

3/8

5 A L o a d , 8 1 9 2 S am p l e s4. 0

3.0

2.0

2 .0p 0.0:E -1 .oWD

4

-2.0-3 o

-4.00 8 1 6 2 4 32 4 0 4 8 5 6 6 4 7 2 80

T i m e (ue)- Ar c Vo l t age-Fau lt - Ar c Vo l t a ge4 0 Fau lt

Figure 2a: Arc Voltage under No-Fault and FaultCondition5 A L o a d , 8 1 9 2 S a m p l e s

1 .o0.90. 80.7

W$ 0.6I-:: .5rnA" 0 4

0.30. 20,I0 .o

e-2

0 . 01 0.1 1 1 0 100F r e q u e n c y (MHz)

-A rc Vo l t age-Fau l t - Ar c Vo l t age-N o Fau l t

Figure 2b: Arc Voltage FFT under No-Fault andFault Condition

5A L o a d , 8 1 9 2 S a m p l es0.8

0.6

0. 4

5 0. 2P o5m

U0-1

-0 .2-0.4-0.6-0.8

0 8 1 6 2 4 3 2 4 0 4 8 56 6 4 7 2 80T i m e (us)

-Load Voltage-Fault --Supply Voltage-Fault- . -Load Vo l t age-N o Fau l t Supp ly Vo l t a ge4 0 Fau l t

Figure 3a: Load and Su pply Voltage under No-Faultand Fault Condition

0 30

0 25z:2 05I-::0 1 55 0 1 0mm"-=a-1 0 0 5

0 DO0 01

5 A L o a d , 8 1 9 2 S am p l e s

0.1 1 10 1 0 0F r e q u e n c y ("2)

- Load Vo l t age FFT-Fau l t- --Load Voltage FFT -NO Fault

--Supply Voltage FFT-FaultSUpply Voltage FFT-NO Fault

Figure 3b: L oad and Supply Voltage FFT under No-Fault and Fault ConditionAs can be seen from the figures, the voltagesignal generated by the arcing fault has beensignificantly attenuated by the single-line cables. Someinsight into the attenuation of the signal can be obtainedby measuring the frequency response of the system.Using a VENABLE Impedance Measurement System(IMS) for frequencies below 200 kHz and an HP -3577A Network Analyzer for frequencies above 200kHz, a frequency-dependent reference signal was


7/27/2019 20030111836_2003134957

4/8

injected at the arcing terminals and the ratio of thesupply-to-reference signal and load-to-reference signalwas obtained. For the supply voltage, the attenuationwithout the BP filter is over 30 dB below 30 MH z and,for the load, it is over 16 dB for all frequencies (SeeFigure 4). Due to this inherent response of the single-line system, the arcing fault signal detected at the loador supply is mostly composed of high frequency signalscentered around 19 MHz.

Frequenc y Res pons e0

-10-20-30-40Eg -50

W -60-70-80-90

-1 00

.-m

1E-06 I E - 0 5 0 0001 0.001 0.01 0.1 I 10 100Frequency (MHz)

- S u p p l y - L o a d

Figure 4: Supply and Load Gain vs FrequencyResponseEven with this much attenuation, it was stillpossible to reliably detect an induced arcing fault bymonitoring the frequency content around 19 MHz of the

load signal. However, as the cable lengths between thearcing fault mechanism and the sensor locationsincrease, the more difficult it becomes to detect thefaults due to the increased attenuation.Current Sipnature

In contrast to the attenuated high frequencycontent of the DC bus voltage, the arcing fault currentsignature was observed at low frequencies, with most ofthe frequency content below 2 kHz. Figure 5a showstwo sample waveforms of the current, one under no-fault and the other under fault, and Figure 5b showstheir corresponding FFT plots. With the samplingparameters of 50 k S d s and 8192 samples, At =20 psand Af= 6.104 Hz.Taking the same approach as with the voltage, aninduced arcing fault could also be reliably detected bymonitoring the frequency content below 2 H z .

0.5A Load , 8192 Samplesa .4

0 .3

0 .2

0 .1-4E ot5- 0 .1

-0 2

-0 .3

-0.4

1 - C u r r e n t - F a u l t - C u r r e n t - N o F a u n I

Figure 5a: Current under No-Fault and FaultCondition0.5A Load , 8192 S a m p l e s

a .3

0 . 2 5- 0 .24,6

0 1 5LLEt5 0.1V

0 D5

0D D O 1 0 01 0 1 1 10

F r c q u m o y k H r ]10 0

- C u r r e n t F F T -F a u l t - C u r r e n t F F T - N o F a u l t

Figure 5b: Current FFT under No-Fault and FaultConditionARCLOCATION

Having established that it is possible to detect anarcing fault by monitoring the AC content in the DCbus voltage or the DC system current, the next task wasto address the location of the arcing fault in a multi-linesystem, such as a Radial Distribution System (RDS).


7/27/2019 20030111836_2003134957

5/8

Cable(Ll)(L2)(L3)

(L4), (L5)(L6)(L7)

(L81, (L9)(LlO), (L ll )

Radial D istribution Svstem

Length (in) Gauge46 10

73.5 1027.5 1053 12

19.5 1422 1460 12

18 + 62 10+1/0 .

The multi-line system used in this investigation isthe two-channel RDS shown in Figure 6 below with thecable lengths and gauges shown in Table 2. As before,the DC supply was set at 150 V and the arcingimpedance was fixed at 1OOQ. The current load onchannel 1 was set at either 0.5A or 5A and the resistiveload on channel 2 was set at 50Q. The arcing fault wasinduced on channel one and the voltage in each channelwas sensed at the output of the CRYDOM (SOOVDC,10A) solid-state relay (Swl, Sw2) and the return bus;the currents were sensed as shown.

-Chl-F ault -ChZ-Fault -C hl-No Fault Ch2-No Fault

Figure 6: Two-Channel R adial Distribution SystemTable 2: Cable Length and Gauge of RDS

ResultsVoltageFor the RDS under investigation and with Swland Sw2 closed, the input terminals of the BP filters ofchannel 1 and channel 2 are effectively tied to the samepoints. Based on the gain-versus-frequency responseplot in Figure 7, the voltage seen by the BP filter ofchannel 2 will have little to no attenuation compared tothe voltage seen by the BP filter of channel 1 between20 lcHz and 10 MHz. However, as with the SLS, thelower frequencies (below 100 kHz) are highlyattenuated but, in contrast to the SLS, frequencies

above 1 MHz exhibit a maximum attenuation of -30 dBversus -50 dB for the load voltage of the SLS. This isthe reason why most of the arcing fault signature is nowcentered around 4-5 MHz (See Figures 8a,b). Whataccounts for the different attenuation levels is the cableimpedance between the voltage sensors, the load, andsupply in the RDS.

Frequency R e s p onsc0

-10-20-30-40m^-5 0

aV -60-70-8 0-90

-100

I

1E-06 1E-05 0.0001 0.001 0.01 0.1 1 10 100Frequency (MHz)

I- hl RDS- h2-R DS- u p p l y S LS _ Load- S LS IFigure 7: SLS and RD S Gain vs FrequencyResponse

I 5 A Load,8192 Sa mp l e s0. 80. 60. 4

2 .z0 0> -0.2=

-0.4-0.6-0.8

Figure 8a: Ch l and C h2 Voltage under No-Faultand Fault Condition


7/27/2019 20030111836_2003134957

6/8

5 A L o a d , l l S Z Sa m pl es0.30

0.25

2 .20d4Ic 0.15Y

e5 0.10-

0.05

0.000.01 0.I i 10 100

Frequency (MHz)-C hi F FT-F ault -Ch2 FF T-Fault

C hl FF T-No Fault Ch2 FFT-No Fault

Figure 8b: Ch l and Ch2 Voltage FFT under No-Fault and Fault ConditionSince the voltage signature is detected withvirtually equal amplitude and frequency content by bothvoltage sensors, it is not possible to use the voltageinformation for arc location in a RDS as shown inFigure 6 . However, it is possible to isolate the channels,if desired, by inserting an LC filter in each channel, asshown in Figure 9; the voltage and corresponding FFTof channel 1 and channel 2 are shown in Figure 10a andFigure lob. This approach not only enables fault

location through the voltage signature but it alsoreduces the attenuation of the voltage signature at thesensors (compare the first two plots in Figure ll),making it possible to detect arcing faults further awayfrom the DC bus. Unfortunately, as Figure 1 1 shows,the attenuation of the voltage signature, for frequenciesof interest but below 3 MHz, at the sensors increases asthe length of cables (L4) and (L5)ncrease.

Figure 9: Radial Distribution System with LC F ilter

Furthermore, at frequencies above 3 MHz, theparasitic inductance in the system, including theinductance of the cables, comes into play, resulting inan unpredictable frequency response. Yet anotherdrawback at these high frequencies is the significantincrease in transmitted energy above 3 MHz, therebymaking it difficult if not impossible to detect thelocation of an arcing fault through the voltage signatureas the cable length increases beyond 50ft.

5 A L o a d , 8 1 9 2 S am p l e s1 00 80 60 40 2; 0-s -02

-0 4-0 6-0 8-1 0

0 8 16 24 32 40 48 56 6 4 72 80Time (us)

-Chi-Fault, LC -C hZ-Fault, LC.-..Chl-No Fa ult, LC Ch2-N o F ault, LC

Figure loa: C h l and Ch2 Voltage for RD S with LCFilter

0.30

0.25

2 .20w4Ec 0.15Y

eD5 0.1D-s0.05

0.00b5A L o a d , 8 1 9 2 S a m p l es

0.01 0.1 1 10 100Frequency (MHr)

-C hi FF T-Fault, LC---.-Chi FF T-N o Fa ult, LC

-Ch2 FF T-F ault, LCCh2 FFT-No Fault, LC

Figure lob: C hl and Ch2 Voltage FFT for RDS withLC Filter6American Institute of Aeronautics and Astronautics

7/27/2019 20030111836_2003134957

7/8

7/27/2019 20030111836_2003134957

8/8

data is analyzed and an FFT sum is generated. At thesesettings the minimum frequency resolved is only 38 1.47Hz. In Figure 13 the sum of the discrete frequenciespresent between 5 MHz and 7 MHz is plotted for thevoltage and between 381.47 Hz and 1907.35 Hz for thecurrent. Since only 5 discrete frequencies were summedfor the current, the result was multiplied by 50 for easycomparison to the voltage sum.

CONCLUSIONThis research has thus far shown that laboratoryinduced arcing faults, under the conditions describedabove, generate a considerable amount of AC voltageacross the arc with frequencies above 30 MHz, but, forthe two systems investigated, most of the voltagesignature is significantly attenuated. Of the twosystems, the radial distribution system exhibitssignificantly less attenuation for frequencies above 100

kHz.Because of this inherent response in the systems,the detectable arcing fault voltage signature was in theMHz range, requiring high-speed data acquisition toimplement an arcing fault detection algorithm.In order to implement arc location in a radialdistribution system via the voltage signature, it wasnecessary to isolate the channels; this was done byinserting a small LC filter in each channel. Thisapproach is fairly successful for relatively short,unshielded cables ((L4),(L5)) of lengths less than Soft.In contrast to the voltage, the current signature isat relatively low frequencies. With this information, anarcing fault can be detected and located with nomodifications to the system and the length of the cablesdoes not affect this information. It just follows then thatthe current signature would be the better choice.However, since the arcing fault signaturemanifests itself at such different frequency ranges, asimple test showed that both signatures are not alwayspresent at the same time. This leads to a preliminaryconclusion that it might be necessary to monitor boththe current and voltage signature but, keeping in mindthe limitations of the voltage signature, less confidencewould have to be placed on the latter.All the results presented thus far were obtainedwith a constant current electronic load and resistiveload. Future work will focus on improving therobustness of the detection algorithm to distinguishbetween transient arcing conditions, such as a relayclosing or opening, or arcing-loads, such as DC brushmotors, and a true arcing fault. At this juncture, all thesignal processing and analysis is done digitally in ahigh-powered data acquisition system. Clearly, this isnot a practical solution, but work has been initiated on astandalone, mixed-signal analog/digital approach.

1.

2.

3.

4.

5 .

6 .

7.

8.

REFERENCESA. F. Sultan, G. W. Swift, and D. J. Fedirchuk,Detection of High Impedance Arcing Faults usingMulti-Layer Perceptron, IEEE Transactions onPower Delivery, Vol. 7, No. 4, October 1992, pp.Dogan Gokhan Ece, Francis M. Wells, and HakanG. Seken, Analysis and Detection of ArcinFaults in Low-Voltage Electrical Power, 7Mediterranean Electrotechnical Conference, Vol. 3,Dogan Gokhan Ece, Behavior of System Voltageduring Arcing Faults and Switching Events,Proceedings of the Mediterranean ElectrotechnicalConference, Vol. 2, 1996, pp. 757 - 760.V. V. Terzija, Z. M. Radojevic, M. B. Djuric, ANew Approach for Arcing Faults Detection andFault Distance Calculation in Spectral Domain,IEEE Proceedings of the Transmission andDistribution Conference, 1996, pp. 573-578.Carl L. Benner and B. Don Russell, PracticalHigh-Impedance Fault Detection on DistributionFeeders, IEEE Transactions on IndustryApplications, Vol. 33, No . 3, MayJJune 1997, pp.K. J. Zoric, M. B. Djuric, and V. V. Terzija,Arcing Faults Detection on Overhead Lines fromthe Voltage Signals, Electric Power & EnergySystems, Vol. 19, No. 5, 1997, pp. 299 -303.Jincheng Li and Jeffrey L. Kohler, New Insightinto the Detection of High-Impedance ArcingFaults on DC Trolley Systems, IEEE Transactionson Industry Applications, Vol. 35, No. 5,September 1999, pp. 1169- 1173.Patricia R. S. Jota and Fabio G. Jota, FuzzyDetection of High Impedance Faults In RadialDistribution Feeders, Electric Power SystemsResearch, Vol. 49, 1999, pp. 169- 174.

1871- 1877.B

1994, pp. 929- 935.

635- 640.


Documents

20030111836_2003134957