+ All Categories
Home > Documents > THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low...

THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low...

Date post: 06-Sep-2019
Category:
Upload: others
View: 3 times
Download: 0 times
Share this document with a friend
17
International Journal of Innovative Computing, Information and Control ICIC International c 2018 ISSN 1349-4198 Volume 14, Number 4, August 2018 pp. 1389–1405 THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON ARTIFICIAL NEURAL NETWORK Dimas Anton Asfani 1,2,* , Abdillah Fashiha Ilman 1 Nugroho Wisnu Ari Sanjaya 3 , I Made Yulistya Negara 1 , Daniar Fahmi 1 Dian Retno Sawitri 3 , Mochammad Wahyudi 1 and Hadi Lizikri Al-Azmi 1 1 Electrical Engineering Department Institut Teknologi Sepuluh Nopember (ITS) Jalan Raya ITS, Sukolilo, Surabaya 60111, Indonesia * Corresponding author: [email protected] 2 Center of Excellence for Automotive Control and System ITS 3 Electrical Engineering Department Faculty of Engineering Dian Nuswantoro University Jalan Nakula I No. 5-11, Jalan Imam Bonjol No. 207, Semarang 50131, Indonesia Received September 2017; revised February 2018 Abstract. This paper dealt with a dynamic modelling of arc flash phenomenon in low voltage installation system based on artificial neural network (ANN). There were two ANN models employed to this proposed model. The first one is dynamic resistance model and the second one is switch or short circuit contact model. The arc flash energy and the number of filaments are defined as the inputs of these ANN models, whereas the targets are the resistance value for dynamic resistance model and the switch value for switch model. The values used in modelling are obtained from experiment of arc flash initiated by phase to neutral short circuit. This fault location is parallel with the resistive load. The feed-forward back-propagation is selected as an algorithm of ANN. The result shows that the proposed model presented the level of accuracy up to 96.7%. In addition, the simulated model revealed that the lower cable impedance is and the higher load is, the greater current peak is and the shorter duration of arc flash is. Keywords: Parallel arc flash, Phase to neutral fault, Low voltage installation system, Dynamic resistance, Arc flash energy, Feed-forward back-propagation neural network 1. Introduction. The number of fire cases in Indonesia is increased annually. The oc- currence of the fire cases is mostly caused by electrical fault or short circuit which then is followed by arc flash. Based on data of the Indonesian National Disaster Management Agency from August 2011 to April 2015, there are 973 cases of fire with 664 cases of them caused by arc flash [1]. In addition, arc flash has become one of the main causes of fire which occurs in densely populated settlements, shopping areas, and traditional markets. As the national electrification ratio (comparison between households that have electricity supply and not yet) continues to grow per year, the possibility of fire due to electricity will increase in the future. In consequence, an early arc flash detection system, especially in low voltage installation, is highly required as a prevention. Generally, arc flash in low voltage installation system can be initiated by two conductors or cable with different potentials in touch. The presence of damaged (or peeled) insulation of cable is the cause of the mechanism. The main problem of arc flash is its existence that cannot be detected precisely by commercial protection devices in low voltage system DOI: 10.24507/ijicic.14.04.1389 1389
Transcript
Page 1: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

International Journal of InnovativeComputing, Information and Control ICIC International c©2018 ISSN 1349-4198Volume 14, Number 4, August 2018 pp. 1389–1405

THE MODELLING OF LOW VOLTAGE ARC FLASH

BASED ON ARTIFICIAL NEURAL NETWORK

Dimas Anton Asfani1,2,∗, Abdillah Fashiha Ilman1

Nugroho Wisnu Ari Sanjaya3, I Made Yulistya Negara1, Daniar Fahmi1

Dian Retno Sawitri3, Mochammad Wahyudi1 and Hadi Lizikri Al-Azmi1

1Electrical Engineering DepartmentInstitut Teknologi Sepuluh Nopember (ITS)

Jalan Raya ITS, Sukolilo, Surabaya 60111, Indonesia∗Corresponding author: [email protected]

2Center of Excellence for Automotive Control and System ITS3Electrical Engineering Department

Faculty of EngineeringDian Nuswantoro University

Jalan Nakula I No. 5-11, Jalan Imam Bonjol No. 207, Semarang 50131, Indonesia

Received September 2017; revised February 2018

Abstract. This paper dealt with a dynamic modelling of arc flash phenomenon in lowvoltage installation system based on artificial neural network (ANN). There were twoANN models employed to this proposed model. The first one is dynamic resistance modeland the second one is switch or short circuit contact model. The arc flash energy and thenumber of filaments are defined as the inputs of these ANN models, whereas the targetsare the resistance value for dynamic resistance model and the switch value for switchmodel. The values used in modelling are obtained from experiment of arc flash initiatedby phase to neutral short circuit. This fault location is parallel with the resistive load.The feed-forward back-propagation is selected as an algorithm of ANN. The result showsthat the proposed model presented the level of accuracy up to 96.7%. In addition, thesimulated model revealed that the lower cable impedance is and the higher load is, thegreater current peak is and the shorter duration of arc flash is.Keywords: Parallel arc flash, Phase to neutral fault, Low voltage installation system,Dynamic resistance, Arc flash energy, Feed-forward back-propagation neural network

1. Introduction. The number of fire cases in Indonesia is increased annually. The oc-currence of the fire cases is mostly caused by electrical fault or short circuit which thenis followed by arc flash. Based on data of the Indonesian National Disaster ManagementAgency from August 2011 to April 2015, there are 973 cases of fire with 664 cases of themcaused by arc flash [1]. In addition, arc flash has become one of the main causes of firewhich occurs in densely populated settlements, shopping areas, and traditional markets.As the national electrification ratio (comparison between households that have electricitysupply and not yet) continues to grow per year, the possibility of fire due to electricitywill increase in the future. In consequence, an early arc flash detection system, especiallyin low voltage installation, is highly required as a prevention.

Generally, arc flash in low voltage installation system can be initiated by two conductorsor cable with different potentials in touch. The presence of damaged (or peeled) insulationof cable is the cause of the mechanism. The main problem of arc flash is its existencethat cannot be detected precisely by commercial protection devices in low voltage system

DOI: 10.24507/ijicic.14.04.1389

1389

Page 2: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1390 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

[2-6], such as miniature circuit breaker (MCB) and fuse. The uniqueness of arc flash inthe form of a high current in short duration can explain this problem. Smoak and Keeth[7] had conducted arc flash experiment on transformer with capacity of 50 kVA and 167kVA and rated voltage of 240 V. The result proved that non-bolted short circuit withvariation of arc length did not result in the opening of circuit breaker.

Other experiments about the characteristic of arc flash showed that the drop voltagewith deformed signal would appear during arc flash [8]. The high current with noise inrange of 10 kHz until 1 GHz was detected in [9]. In addition, the energy of arc flash wasdirectly proportional with current and duration of arc flash [10].

By understanding the characteristics of arc flash, the modelling of this phenomenon intomathematical or simulation model can then be conducted. It is highly necessary sincethe arc flash phenomenon simulated in reality (experimental approach) is accompanied bysome dangers, such as electric shock, arc flash burn, arc flash blast, intense light, soundwave, and projectiles [11].

The general model of momentary arc flash in resistive-inductive system based on methodfor determining angle connection on power was proposed in [4]. An improved model of arcflash based on current analysis was then developed in [12]. A new model that was betterin predicting peak current during arc flash had also been presented in [13]. However, adynamic model of arc flash resistance that is affected by arc flash energy and the number offilaments has not been investigated yet. From this dynamic resistance model, the arc flashcurrent can be obtained and its characteristic can then be understood clearly. Artificialneural network (ANN) is greatly suitable for dynamic modelling. In fact, ANN has beenwidely used and provided satisfactory results in recognizing characteristic, classification,and prediction in some areas [14-21]. ANN is a smart computation method with its abilityto solve complex and non-algorithmic problems [22]. ANN uses the past experience tolearn how to deal with new and unexpected situations. The statistical distribution of datais not required in developing ANN model. No prior knowledge about the relationshipbetween modeled variables is also not required. Moreover, this method has not beenpresented in arc flash modelling at low voltage installation system.

In this paper, the model of low voltage arc flash using ANN was developed based on thedynamic characteristic of arc flash resistance affected by arc flash energy and the numberof filaments. The model comprised of some variables, such as current, voltage, resistance,power and energy. Back-propagation neural network was chosen as ANN method since itwas a simple architecture of ANN and quite sufficient in showing the relationship betweenthose variables. From the dynamic resistance model of arc flash, the characteristic of arcflash current can be obtained and then be used for detection parameter of arc flash in thefuture.

2. Arc Flash Phenomenon and Artificial Neural Network Method.

2.1. Arc flash in low voltage installation system. Basically, power cable consists oflow resistance to carry current. The cable is covered by insulation to isolate conductorwith other conductors or conductive parts and protect it from surrounding environment[23]. If the insulation of cable is damaged, the current will flow in abnormal path knownas electrical fault. This insulation failure can be caused by stress which is subjected tocable, such as mechanical, electrical, thermal, and environment [24,25].

Electrical fault is also called as short circuit since current flows in a shorter circuit withlower resistance. Based on the circuit of current flowing, electrical fault can generallybe categorized into some types, that are phase to phase, phase to neutral, and phase to

Page 3: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1391

ground. The presence of short circuit, especially phase to neutral, can then lead to arcflash phenomenon [26].

Generally, arc flash is a phenomenon initiated by the two surfaces of conductor which arein contact and create the stepping current between both of them. During the occurrenceof stepping current, the surrounding air will be ionized and result in arc flash [3]. In thiskind of failure event, the magnitude of current is low since it is limited by impedance value[13]. There are two types of arc flash, which are series arc flash and parallel arc flash.Series arc flash occurs when a current-carrying conductor is connected in series with thefaulted line, as illustrated in Figure 1, while parallel arc flash occurs when two conductorswith different polarities are in touch with each other, such as phase and neutral of cable,as depicted on Figure 2. If arc flash occurs when the channel impedance value is highenough, arc flash current does not have sufficient magnitude value or duration to makethe protection system work [3].

Figure 1. Illustration of series arc flash [27]

Figure 2. Illustration of parallel arc flash [27]

The emergence of arc flash will increase temperature significantly at which the shortcircuit occurs. In this case, the temperature associates with thermal energy emitted byarc flash. This thermal energy can be approached by calculation of electrical energy whichis then called as arc flash energy. The number of filaments will affect the severity of arcflash. Besides that, the resistance value in the location of short circuit (called as arcflash resistance) will change dynamically and tend to decreasing during arc flash. Byconsidering the parameters during arc flash, such as the number of filaments, arc flash

Page 4: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1392 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

energy, arc flash resistance, and switch (short circuit contact), a model of low voltage arcflash can be developed appropriately.

2.2. Artificial neural network model for low voltage arc flash. Artificial neuralnetwork (ANN) is a processing unit consisting of some inputs and outputs with archi-tecture as shown in Figure 3 [28,29]. This architecture adopts the structure of humanbrain tissue. ANN compares threshold value with the input of calculation result which isobtained through multiplication with weights and addition with biases. If output valueis greater than the threshold, then the output will be 1; if otherwise, then it will be 0.

Backpropagation is one of ANN methods that is a learning algorithm to reduce theerror value by adjusting the weights and the biases based on the comparison of output anddesired target. Backpropagation is a supervised training method designed for operationof multilayer feed forward artificial neural network.

Backpropagation has three layers in the process of training namely input layer, hiddenlayer, and output layer. This layer is development of a single-layer network which has twolayers; those are input and output layers. With hidden layer, error value is lower thanthe error value in a single-layer network since the hidden layer serves to update weightvalues.

In the proposed modelling of low voltage arc flash, two ANN models were employed asshown in Figure 4. The first one was ANN model for switch, whereas the other one wasANN model for arc flash resistance. Both of models were similar in input, but differentin target. The inputs were arc flash energy and the number of filaments, while the targetfor switch model was switch value (condition of short circuit contact) and the target forresistance model was arc flash resistance value.

Figure 3. Architecture of ANN

Figure 4. Proposed ANN model

Page 5: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1393

3. Experiment of Low Voltage Arc Flash.

3.1. Scheme. This experiment was conducted to collect data of artificial arc flash, whichwere current and voltage. The scheme of experiment was shown in Figure 5, while therealization of experiment could be seen in Figure 6. The experimental equipments con-sisted of 220 V (phase to neutral) source voltage, 6 incandescent lamps (100 W per lamp)as resistive loads, 6 A fuse, NYMHY cables with 20 m length and different diameters suchas 0.75 mm, 1.5 mm, and 2.5 mm, acrylic chamber, and a set of data aquisition, suchas personal computer, National Instrument (NI) PXI, NI – LabView, voltage probe, andcurrent transformer with current probe.

In the first step of experiment, the main cable was connected to the extension cablewith 10 cm length. The end of extension cable was then peeled off its insulation and tied

Figure 5. Scheme of experiment

Figure 6. Realization of experiment

Page 6: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1394 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

Figure 7. Variation of the number of filaments (a) 1, (b) 3, (c) 6, (d) 12,(e) 18, (f) 24

on actuator in order to drive the cable in short circuit condition (phase to neutral). Thismeant that the length of the main cable, which was 20 m, was not reduced during arcflash process. The filaments of extension cable were twisted with variation of the number,such as 12, 18, and 24, as depicted on Figure 7. The extension cable and actuator werelocated in the chamber.

After that, the lamp loads were switched on and NI PXI started to record the waveformof current and voltage. This waveform was then stored and displayed on PC throughLabView program. Later, the actuator was turned on, so that the cable was in shortcircuit condition for a while and created arc flash. The system including the NI PXI andlamp loads was then turned off. This procedure was repeated up to 30 times for eachvariation of number of filaments, so that there would be 540 data for all variations. Table1 showed the types of variations that were tested and the number of experiments.

Table 1. Experimental study case

Cable diameter The number of filaments The number of experiments0.75 mm 1, 3, 6, 12, 18, 24 301.5 mm 1, 3, 6, 12, 18, 24 302.5 mm 1, 3, 6, 12, 18, 24 30

3.2. Data processing. The recorded data were processed in order to obtain the specificcharacteristics of arc flash during experiment with different numbers of filaments, such aswaveform of voltage and current, resistance, power, and energy. The formula to calculatethe power at single phase AC voltage was:

P = V · I · cos θ (1)

where P was power (W), V was voltage (V), I was current (A), and cos θ was powerfactor. As the lamp load was a resistive load, the value of cos θ closed to 1. Energy (E)was the amount of power per time unit which could be expressed by formula:

E =

n∑

0

(P0, P1, P2, . . . , Pn) (2)

The energy value was used as a reference in modelling of short circuit. In this case, thedetermining of energy as reference was conducted by looking at the time span during arcflash. The energy was then used as the input of short circuit model (based on artificialneural network). On the other hand, the resistance value was used as the target of short

Page 7: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1395

circuit model. This resistance was the value of resistance when the cable was connectedshortly (Rarc), excluding the resistance of lamp loads (Rloads). Rarc could be obtained asfollows:

Rt = V /I (3)

Rseries = (Rloads · Rt)/(Rloads − Rt) (4)

Rarc = Rseries − Rcable1 − Rcable2 (5)

The resistance value (Rt) was the resistance total of lamp loads and cable impedan-ce. Equations (4) and (5) were used since Rarc was connected in series with the cableimpedance and connected in parallel with the lamp loads as shown in Figure 7.

4. Modelling of Low Voltage Arc Flash. The experiment of arc flash was transformedinto an electrical circuit as seen in Figure 8. The resistor represented the impedance oflamp loads and cable. The values of resistance were shown in Table 2. The electricalcircuit was then implemented in MATLAB Simulink as depicted on Figure 9.

Figure 8. Arc flash in electrical circuit model

Table 2. Resistance values in electrical circuit model

Resistanceinside (Ω)

Cable impedance (Ω) 100Watt

load (Ω)Rarc(Ω)0.75 mm 1.5 mm 2.5 mm

Neutral Phase Neutral Phase Neutral Phase0.5 0.809 0.805 1.78 1.48 1.4 1.3 484 0.2-50

In this modelling, two models of ANN were employed, which were ANN model forswitch and arc flash resistance. Each of them had two similar inputs, which were the arcflash energy and the number of filaments. The ANN models were trained based on theresult of experiment using cable with 1.5 mm diameter and different numbers of filaments,which was 1 filament and 3 filaments. The targets for switch model and resistance modelwere switch value and arc flash resistance value, respectively. The ANN parameter andconfiguration could be seen in Tables 3 and 4, respectively, while the training parameterwas shown in Table 5. The patterns of input and target which were trained for switchwere depicted on Figures 10 and 11, whereas the patterns of input and target for arc flashresistance were seen in Figures 12 and 13.

Page 8: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1396 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

Figure 9. Arc flash in Simulink model

Table 3. ANN parameter

The initials of parameter ValueNetwork type Feed-forward back-propagationInput data Arc flash energy and number of filamentsTarget data Value of switch and Rarc

Training function TRAINRPAdaptation learning function LEARNGDM

Performance function MSENumber of layers 4

Table 4. ANN layer parameter

Parameter Layer 1 (HL1) Layer 2 (HL2) Layer 3 (HL3) Layer 4 (Output)Total neuron 30 30 30 1Activationfunction

TANSIG TANSIG TANSIG TANSIG

5. Results and Analysis.

5.1. Normal condition. Firstly, the model was simulated in normal condition (withoutshort circuit) for 0.1 s. The waveform of voltage and current was obtained by 5 periodsas depicted on Figure 14. Based on the calculation, the peak voltage during normalcondition was 311.12 V. This value was the result of the multiplication of 220 V and

√2.

The load total of 6 lamps was 600 W. By dividing the load total and the peak voltage,the peak current was calculated by 3.87 A. In Figure 14, the peak voltage and the peakcurrent in normal condition were 310.08 V and 3.84 A, respectively. These values werevery similar to the result of calculation. It proved that the model could represent the lowvoltage installation system with 220 V rated voltage, 50 Hz, and 600 W load.

Page 9: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1397

Table 5. ANN training parameter

Training Parameter ValueshowWindow True

showCommandLine Falseshow 50

epochs 1000time Infgoal 1e-09

min grad 1e-05max fail 1000delta0 0.07delt inc 1.2delt dec 0.5deltamax 50

Figure 10. Pattern of input for switch

Figure 11. Pattern of target for switch

Page 10: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1398 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

Figure 12. Pattern of input for Rarc

Figure 13. Pattern of target for Rarc

5.2. Low voltage arc flash characteristics. The waveform of current and voltagebased on the arc flash experiment was depicted on Figures 15 and 16, respectively. Bothof them were multiplied, so that the graph of power (W ) was obtained as shown in Figure17. The trend of energy based on the graph of power could then be observed as seen inFigure 18. From these results, except the voltage, the occurrence of arc flash was alwayscharacterized by a spike in value. Otherwise, the voltage waveform would be decreasedduring arc flash. The significant decreasing of resistance during arc flash, as seen inFigure 19, could explain the presence of spike in value for current, power, and energy.The resistance pattern was then used as the target for training of the model.

5.3. Model validation. The simulation with impedance value of cable with 1.5 mmdiameter was conducted for 0.1 s. The short circuit condition to create arc flash wasalso given to this simulation. The simulation result compared to experimental result for1 filament and 3 filaments could be seen in Figures 20 and 21, respectively. Of the twofigures, the current waveform in which arc flash occurred was then taken as depicted on

Page 11: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1399

Figure 14. Voltage and current in normal condition based on simulation

Figure 15. Currents during arc flash based on experiment

Figure 22 (for 1 filament) and Figure 23 (for 3 filaments). This would facilitate to obtainthe level of error and accuracy of the proposed model.

The error and accuracy for 1 filament were 5.6% and 94.4%, respectively, while the errorand accuracy for 3 filaments were 5.58% and 94.32%, respectively. These values provedthat the model had successfully imitated the waveform of target in training process. Themodel was then used to predict the arc flash with variation of load and cable impedanceof short circuit.

5.4. Prediction of low voltage arc flash. The model was simulated with lamp loadsof 300 W and 900 W. Moreover, the impedance cable was varied based on the cable’sdiameter, which was 0.75 mm, 1.5 mm, and 2.5 mm as shown in Table 2. The simulationwas also run in short circuit condition. The result of prediction for simulation with 300W load could be seen in Figure 24 (for 1 filament) and Figure 25 (for 3 filaments), while

Page 12: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1400 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

Figure 16. Voltages during arc flash based on experiment

Figure 17. Power during arc flash based on experimental result

Figure 18. Energy during arc flash based on experimental result

Page 13: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1401

Figure 19. Pattern of resistance value during arc flash based on experi-mental result

Figure 20. The comparison of current between the result of experimentand simulation with 1 filament

Figure 21. The comparison of current between the result of experimentand simulation with 3 filaments

Page 14: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1402 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

Figure 22. The comparison of peak current during arc flash between theresult of experiment and simulation with 1 filament

Figure 23. The comparison of peak current during arc flash between theresult of experiment and simulation with 3 filaments

Figure 24. The prediction of arc flash current for 1 filament and 300 W load

Page 15: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1403

Figure 25. The prediction of arc flash current for 3 filaments and 300 W load

Figure 26. The prediction of arc flash current for 1 filament and 900 W load

Figure 27. The prediction of arc flash current for 3 filaments and 900 W load

Page 16: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

1404 D. A. ASFANI, A. F. ILMAN, N. W. A. SANJAYA ET AL.

the result of prediction for simulation with 900 W load was shown in Figure 26 (for 1filament) and Figure 27 (for 3 filaments).

6. Conclusions. By considering the number of filaments, arc flash energy, arc flashresistance, and switch, the proposed method had successfully to model the arc flashin low voltage installation system. The model comprised an artificial neural network(ANN) with simple architecture, which was feed-forward back-propagation configuration,3 hidden layers with 30 nodes per layer, 1 output layer, and Tansig activation function forall layers. Based on the result of comparison between the result of model and experiment,the accuracy of the model was 96.7% (error rate was 3.3%). In addition, based on thegraph of low voltage arc flash prediction result, if the cable impedance was lower and theload was greater, the value of peak current during arc flash would be greater whereas theduration of arc flash was relatively shorter.

REFERENCES

[1] Indonesian National Disaster Management Agency Data’s August 2011-2015.[2] Y. Wu, Z. Song and X. Li, A method for arc fault detection based on the analysis of signal’s

characteristic frequency band with wavelet transform, IEEE the 2nd International Conference onElectric Power Equipment, Matsue, Japan, pp.1-4, 2013.

[3] A. K. Mishra, A. Routray and A. K. Pradhan, Detection of arcing in low voltage distribution systems,IEEE Conference on Industrial and Information Systems, Kharagpur, India, pp.1-3, 2008.

[4] T. Gammon and J. Matthews, The historical evolution of arcing-fault models for low-voltage systems,The 1999 IEEE Industrial and Commercial Power Systems Technical Conference, Sparks, Nevada,1999.

[5] J.-M. Martel, M. Anheuser and F. Berger, A study of arcing fault in the low-voltage electricalinstallation, Proc. of the 56th IEEE Holm Conference on Electrical Contacts, South Calorina, pp.199-209, 2010.

[6] D. G. Ece and F. M. Wells, Analysis and detection of arcing faults in low-voltage electrical powersystems, Proc. of the 7th Mediterranean Electrotechnical Conference, Antalya, Turki, vol.3, pp.929-935, 1994.

[7] A. M. Smoak and A. J. Keeth, An investigation of low voltage arc flash exposure, IEEE IAS ElectricalSafety Workshop, pp.183-185, 2013.

[8] N. K. Medora and A. Kusko, Arcing faults in low and medium voltage electrical systems – Why dothey persist?, 2011 IEEE Symposium on Product Compliance Engineering Proceedings, San Diego,CA, USA, pp.1-6, 2011.

[9] G. Artale, A. Cataliotti, V. Cosentino and G. Privitera, Experimental characterization of series arcfaults in AC and DC electrical circuits, Proc. of IEEE International Instrumentation and Measure-ment Technology Conference (I2MTC), Montevideo, 2014.

[10] E. Carvou, N. B. Jemaa, S. Tian, Z. Belhaja and B. Jusselin, Electrical arc characterization for AC-arc fault applications, Proc. of the 55th IEEE Holm Conference on Electrical Contacts, Vancouver,British Columbia, Canada, 2009.

[11] J. F. Maida, Electrical Awareness, Arc Flash Overview and Qualifications, Maida Engineering, 2009.[12] T. L. Gammon, Improved Arcing-Fault Current Models for Low-Voltage Power Systems (<1 KV),

Ph.D. Thesis, Georgia Institute of Technology, 1999.[13] T. Gammon and J. Matthews, Arcing-fault models for low-voltage power systems, Conference Record

of the 2000 IEEE Industrial and Commercial Power Systems Technical Conference, Clearwater, FL,USA, pp.119-126, 2000.

[14] E. Ozgan and R. Demirci, Neural networks-based modeling of traffic accidents in interurban ruralhighways, Journal of Applied Sciences, vol.8, pp.146-151, 2008.

[15] V. Himanen, J. Ratio, P. Nijkamp and A. Reggiani, Neural Network in Transport Applications,Ashgate Pub. Co., USA, 1998.

[16] H. Chaoui, C. C. Ibe-Ekeocha and H. Gualous, Aging prediction and state of charge estimation of aLiFePO4batteryusing input time-delayed neural networks, Electric Power Systems Research, vol.146,pp.189-197, 2017.

Page 17: THE MODELLING OF LOW VOLTAGE ARC FLASH BASED ON … · Artificial neural network model for low voltage arc flash. Artificial neural Artificial neural network (ANN) is a processing

THE MODELLING OF LOW VOLTAGE ARC FLASH 1405

[17] A. S. Guedes, S. M. Silva, B. J. C. Filho and C. A. Conceicao, Evaluation of electrical insulationin three-phase induction motorsand classification of failures using neural networks, Electric PowerSystems Research, vol.140, pp.263-273, 2016.

[18] J. M. Diaz-Chacon, C. Hernandez and M. A. Arjona, Finite element and neural network approachfor positioning a magnetic shunt on the tank wall of a transformer, IET Electric Power Applications,vol.10, pp.827-833, 2016.

[19] A. A. Mas’ud, B. G. Stewart and S. G. McMeekin, An investigative study into the sensitivityof different partial discharge ϕ-q-n pattern resolution sizes on statistical neural network patternclassification, Measurement, vol.92, pp.497-507, 2016.

[20] G. H. Bazan, P. R. Scalassara, W. Endo, A. Goedtel, W. F. Godoy and R. H. C. Palacios, Stator faultanalysis of three-phase induction motors usinginformation measures and artificial neural networks,Electric Power Systems Research, vol.143, pp.347-356, 2017.

[21] E. Celik, H. Gor, N. Ozturk and E. Kurt, Application of artificial neural network to estimate powergeneration and efficiency of a new axial flux permanent magnet synchronous generator, InternationalJournal of Hydrogen Energy XXX, pp.1-8, 2017.

[22] N. B. Karayiannis and A. N. Venetsanopoulos, ANNs: Learning Algorithms, Performance Evalua-tion, and Applications, Kluwer Academic Publisher, Boston, 1993.

[23] G. F. Moore and BICC Cables Ltd. (eds.), Electric Cables Handbook, 3rd Edition, Oxford, Malden,Mass, Blackwell Science, 1997.

[24] G. C. Stone, E. A. Boulter, I. Culbert and H. Dhirani, Electrical Insulation for Rotating Machines:Design, Evaluation, Aging, Testing and Repair, IEEE Press, John Wiley, 2004.

[25] Electric Machinery Committee, IEEE Recommended Practice for Testing Insulation Resistance ofElectric Machinery, IEEE Press, 2013.

[26] Z. Cao, X. Liu, Z. Zhang, Q. Wang, C. Shi and C. Li, Simple analysis of the measurement methodsof arc fault, International Conference on Intelligent Systems Design and Engineering Applications(ISDEA), Hunan, China, pp.914-917, 2014.

[27] J. J. Shea, Comparing 240 Vrms to 120 Vrms series arcing faults in residential wire, The 54th IEEEHolm Conference on Electrical Contacts, Orlando, FL, USA, pp.218-224, 2008.

[28] M. H. Purnomo and A. Kurniawan, Supervised Neural Networks and Its Application, Graha Ilmu,Yogyakarta, Indonesia, 2006 (in Indonesian).

[29] J. J. Siang, Artificial Neural Network and Its Programming Using MATLAB, Andi, Yogyakarta,Indonesia, 2005 (in Indonesian).


Recommended