Fault Diagnosis of Smart Grid Distribution System by Using Smart Sensors and Symlet Wavelet Function

Transcription

1 J Electron Test (217) 33: DOI 1.17/s Fault Diagnosis of Smart Grid Distribution System by Using Smart Sensors and Symlet Wavelet Function Mangal Hemant Dhend 1 & Rajan Hari Chile 2 Received: 16 October 216 /Accepted: 16 April 217 /Published online: 29 April 217 # Springer Science+Business Media New York 217 Abstract In today s era of smart grid system scenario, the fault diagnosis is of utmost important task. Present distribution networks change drastically due to expansion and inclusion of large number of distributed generation units into power system at distribution level. To face the challenges of modernized girds, conventional fault diagnosis methodologies require drastic change by making use of advanced infrastructure and technologies. This will be helpful to achieve automation in fault diagnosis tasks, improved power quality, reliability, resilience and self healing property of the power system. This paper proposes the use of smart sensors and advanced communication technology that will be available in future smart grids to carry out automated fault diagnosis tasks using signal processing techniques. Methods of using Standard deviation features of fault transient signal and a fault location factors are proposed. Performance of various scaling levels, features and components of fault transient current signals extracted using the latest non conventional Symlet mother wavelet function are evaluated and compared. The attempt is made to select optimal features and components of fault transient currents to improve the performance of present limited types of available fault locators. The tests are taken on standard model of smart grid distribution system but can be applied for fault Responsible Editor: M. J. Barragan and W. R. Eisenstadt * Mangal Hemant Dhend mangaldhend1@gmail.com 1 2 Rajan Hari Chile rhchile@yahoo.com All India Shri Shivaji Memorial Society s College of Engineering, Pune, India Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded, India diagnosis of any other power equipment. Results show adequate accuracy to extend the use of proposed method for real time applications. Keywords Fault diagnosis. Fault location factor. Feature. Smart sensor. Wavelets transform 1 Introduction In the context of the smart grid (SG), reliable, fast and accurate fault diagnosis (FD) along power transmission and distribution networks is essential. Because of the complexity, including additions of distributed generators (DG) and the changing nature of power distribution networks, there are number of challenges to FD which are more difficult to solve than in transmission systems. Power distribution networks cover wide areas and usually consist of many numbers of nodes, thousands of end user loads, many distribution transformers and short lines with different resistances and inductances. Many of methods like artificial neural network, fuzzy, decision tree and support vector methods are presented. These are knowledge based methods which use characteristics obtained from the time domain signals of current, and voltage. Information of fault location, type of load, settings of protective devices and previously registered faults are also essential. For successful implementation of such a fault location approach, a considerable amount of information of faults in nodes is required, which is difficult in today s fast growing distribution networks. Therefore, some advanced distribution systems have started installing fault locators in the feeder. In order to keep costs down, low-cost test techniques are essential. One strategy is to provide test circuits directly on-chip [16]. Fortunately, using modern technologies for data sensing, recording, signal processing and analysis combined with intelligent algorithms

2 33 J Electron Test (217) 33: or artificial intelligence techniques provides some possible solutions to the automated fault diagnosis and monitoring of the system. Therefore to eliminate the drawback of conventional fault locators, focus on methods using indicating devises with smart sensors is given in this research work. Initially, we had proposed smart monitoring of the SG system in an international conference [8]. At present, some smart distribution monitoring systems make use of intelligent electronic devices, field remote terminal units or fault passage indicators along with global positioning systems for communications, to perform the task of fault location (FL). However, they do not record temporary fault conditions due to the short duration of the voltage collapse. Additionally, for long duration faults, their locations may remain unknown until a permanent fault occurs. With the availability of modern technologies such as optical transducers instead of conventional current transformer and current voltage transformers, future SGs can utilize traveling wave-based methods [19]. Traveling wave-based fault locators require advanced sensors with high sampling rates but these are more accurate, and more reliable as compared to phasor-based methods. They also have insensitive high frequency components connected to remote in feed currents from the generator side and so facilitate the estimation of fault location. In an evolution of SGs, smart meters (SM) are essential elements, which have motivated many international institutions and companies to conduct research and development on their use. From these one can build the intelligent electrical network of the future. Therefore we developed a SM and first proposed its extended use for the fault diagnosis task. The hardware scheme and its testing for fault monitoring of a few node systems were published in an international conference [9]. If SMs are strategically placed in nodes or buses of a distribution system, apart from using them only at the consumer end then, they can play a vital role in FD, monitoring, automation and on line pricing. Thus, achieving optimal energy efficiency, security and significant savings of money, many problems can be alleviated to achieve sustainable development of power system services. In this paper, FD is proposed using all these types of SM, fault detectors, or indicators and are referred hereafter as smart sensors (SS). Implementation of advanced communication systems which are being installed currently in various countries can be used for data transferring of this SS from one system end to another. SS with very large satellite would be apt choice for reliable internet communications of faults conditions. It can provide sturdy, critical, fast and dynamic operational requirements of FD and distribution monitoring applications even at remote ends. Paper [27] proposes a fault-location method based on smart feeder meters with voltage sag monitoring capability. But, in this paper it is assumed that voltages are balanced and only magnitudes of voltages are considered. This is very rare case in practical situations. To consider phase angles, synchronization of all smart meters will require increased cost and complexity. In [1], a travelling wave method is used which gives excellent reliability and high accuracy in identifying fault locations. Even if this technique leads to a faster restoration process, it is applied for only high impedance faults and underground distribution systems. In order to improve the efficiency of restoration schemes, many power companies have conducted black start field tests. Research progress of the power system restoration published in last decade is reviewed in [18]. This method increases the effect of inception angle influence of load conditions along with the complexity of the fault locating task. It also requires large amounts of training data and time. Paper [12] proposes a scheme for fault diagnosis in a power distribution network with distributed generators by using relay agents located in the distribution network. The proposed method measures the bus currents at which they are located can detect and locate the faults. The technique used is cumbersome and not applicable for all faults at all bus locations. When differences in relay zones and fault indications are over lapping, no significant fault classification and location is achieved. To overcome these drawbacks, performance of various features extracted from fault transient signal is studied in this work. The method using energy feature is published in [1]. Then the use of mean values of features and FFT was proposed in [11]. In the present paper, use of additional feature of standard deviation is proposed. Comparison of these features with different components of fault transient current is also evaluated. Proposed method has better performance as compared to [1, 11] which is the main contribution of this research work. This work will be also helpful for selecting optimum feature and current component to get good performance in fault diagnosis techniques which use these types of signal processing methods. In addition fault location factors are also calculated and applied. The novelty of this research work is in use of fault locating factors, advanced nonconventional Symlet mother wavelet function, selection of optimal scaling level, feature and components of fault transient currents along with their applicability for smart grid systems. Demonstrated results have proved an increase in the accuracy and robustness of fault diagnosis as compared to the most of similar published methods. The results of this endeavor will be useful to speed the FD process based on signal processing techniques in fault testing of any device or system. The paper is outlined as follows. Section 2 describes the preprocessing stage for current sensing and finding Clark components from a fault transient current. In Section 3, signal processing and feature extraction techniques using wavelet are explained. Section 4 describes the various steps of developing algorithms for fault diagnosis. Section 5 presents experimental set up and Section 6 presents test results. To verify the applicability of the proposed algorithms, performance comparisons

3 J Electron Test (217) 33: with the various features and FLF method are given in Section 7. Conclusions are presented in Section 8. 2 Pre Preprocessing Stage 2.1 Sensing Normal and Fault Signal In an actual system, SM or SS installed at various buses can sense the three phase currents along with all other standard electrical quantities. Locations of the sensors can be based on the network configuration, interconnection to the grid, and locations of the DGs. These can be practically located at any node called a sensor node which can be connected with a particular number of nodes that sense transient currents. The sensor node has the ability to control a number of devices. Currents can be sensed before and during the occurrence of a fault with some pre-set sampling frequency. These signals can then be passed to data collection centers for further analysis using intelligent programmes and latest communication infrastructure which will be readily available in smart grid system as mentioned in section 1. This will help to improve the performance of fault locators presently existing in some systems. 2.2 Domain Conversion The Clark transformation allows simplified analysis of three phase currents of balanced, transposed and coupled networks. Quantities measured by SS are converted into three single phase uncoupled networks with the help Eq. (1). This transformation of phase domain quantities into modal domain values reduces the time of analysis [4, 22] I Ia 4 Iα 5 ¼ 4 2 p 1 ffiffiffi 1 5 pffiffiffi 4 Ib 5 ð1þ Iβ 3 3 Ic All six currents including three phase currents Ia, Ib, Ic and three Clark transformed components I, Iα and Iβ are then used to extract features as described in Section 3. 3 Feature Extraction Stage Transient components of fault currents or voltages contain lots of fault information such as some characteristics, high frequency and instant break. Wavelet Techniques (WT), in signal processing have the capability to extract the signal that has sharp changes, tiny discontinuities, and sharp peaks along with ability to identify a small fluctuation occurring in the signal. With the help of wavelet analysis, localization can be achieved in frequency as well as time domain. Which is not the situation with the Fourier transform (FT), as resolution in both domains are fixed. Moreover, FT is only suitable for analysis of signals with slowly varying or periodic stationary characteristics and cannot be used for temporal transient signals. This problem is solved if WT is used, as at elevated frequencies, it uses tiny windows and at short frequencies uses stretched windows. The functions are achieved by dilation of a single or band-pass wavelet. The capabilities of good frequency resolution at low frequencies and Bzoom in^ on singularities make WT extremely popular. Therefore WT is a smart powerful tool in power disturbance and power quality applications. Hence it is specially used for the examination of fast momentary signals which are non periodic containing high frequency signal and localized impulses. Due to all these reasons, WT is preferred and used in proposed technique to identify various faults and analyze their types with location. 3.1 Selection of Mother Wavelet Function Many published work used the Daubechies wavelet function but in this research work due to superior features of the Symlet wavelet, it is selected as mother wavelet for carrying out multi- resolution analysis as mentioned in section 2. It is advanced version of Daubechies and is a reasonably short wavelet in time domain with a good degree of smoothness. It has a bandwidth characteristic almost close to an ideal filter in the frequency domain. Therefore it is possible to allow good separation of various frequencies in reasonable time localization. This mother wavelet function is more symmetrical as compared to Daubechies and has more zooming in capability. 3.2 Selection of Appropriate Coefficient and Scaling Level Fault current signals are decomposed in to 4-levels of scaling by using Symlet discrete WT. Approximate and detail coefficients are designated as ca and cd respectively. They are found out from fault transient currents using usual decomposition techniques. For this work, four levels of scaling of approximate coefficients are designated as ca1, ca2, ca3, ca4 where as detail coefficients are designated as cd1, cd2, cd3 and cd4. All these coefficients are determined and their performance is studied under different fault locations. It is observed that with variations in fault locations and types; variations in detail coefficient cd4 are larger as compared to the approximate coefficient ca4. Figure 1 shows comparisons of variations in values of four different scaling levels of detail and approximate coefficients. Noise is present in some of the samples of the reconstructed detail coefficients and has a very small value which affects the accuracy of estimating fault distance. This can be eliminated by thresholding. It is seen that approximate coefficients have high magnitudes but shows details of variations in low frequency signals.

4 332 J Electron Test (217) 33: coefficient (No) units) Scaling level Fig. 1 Variation of approximate and detail coefficients ca cd Coeffcient (No unit) Fault location (Bus No) Fig. 3 Variation in values of detail coefficients cd1 cd2 cd3 cd4 While detail coefficients have low magnitudes but shows clear variations of high frequency signals. Figures 2 and 3 shows variations of detail and approximate coefficients for four different levels of scaling, determined for different fault locations or number of buses in a grid. It is seen that fourth level coefficients ca4 and cd4 shows clear variations of values with respect to the variations in fault locations or number of buses. As per the investigations for different types of features extracted, error with the fourth level detail coefficient cd4 is found minimum as compared to that of the approximate coefficient. Therefore further research work presented here is carried out by using fourth level detail coefficient. 3.3 Feature Selection and Extraction The signal is composed of sinusoidal continuous waves at different frequencies. One of the most imperative issues to take into account for the appropriate fault detection is the signal interpretation. For correct interpretation of the data acquired, a methodology with an accurate comprehension of the different influences of the variables is desired. Different features of wavelet coefficients like peak magnitude, mean value, energy, and entropy are extracted for a disturbance analysis and voltage events in [2, 3, 5 7, 13 15, 17, 2, 21, 23 26, 28, 29]. But majority of these works require different logics for different faults, lot of calculations, an additional classifying technique like artificial neural network or another along in combination with wavelet technique. These methods have drawbacks of multiple fault estimations, large test data requirements, and dependency on system parameters which are minimized in this work. The extended application of various features for fault diagnosis of smart grid system is investigated. To get better accuracy, their performance for selecting Coefficient (Nounit) Fault location ( Bus No) Fig. 2 Variation in values of approximate coefficients ca1 ca2 ca3 ca4 optimum features is considered for further developing the technique. In this research work, six different features of energy, FFT, mean and standard deviation (STD) are extracted from fourth level of detail coefficients. A flowchart for the sequence of steps used in the technique is outlined in Fig. 4 and detail stages carried out for fault diagnosis work are highlighted in Fig. 5. Various features of six current components before and after the fault as mentioned in section 3.3 are extracted under various faults conditions. 4 Fault Diagnosis Stage Data are stored, analyzed and their performance for fault detection, classification and location task is evaluated. Different types of faults can be classified by using features like FFT (or respective feature like STD, depending upon feature method used) of three phase fault transient currents and zero nodal component of current. Whether any phase is involved in fault or not is decided by comparing the feature value, found out by using this phase current with that of found by using the remaining phase currents. If fault occurs on any phase then feature value of faulted phase current becomes higher than that of phases not involving faults. If fault occurs, on the phase A and B then the feature values found out by using these two currents becomes greater than that of found from phase C. Feature values of faulted phases also become higher than those in normal conditions. In this case, the fault can be either AB or ABG. Hence it is required only to ensure whether fault involves ground or not, which can be determined with feature values found using current component I (see Table 1). The threshold values for comparison can be set by storing values under normal conditions. If a fault involves ground then the feature values become higher than that of fault without involving ground or in normal conditions. For location of the fault, respective feature values are determined at different locations and then are compared. At the location of faults, all feature values determined from six current signals become greater than that of other locations. Exact fault diagnosis work is demonstrated in the experimental results and discussions section. The accuracies of using the peak values with different

5 J Electron Test (217) 33: Distribution system Feature Extraction Stage Data acquisition through smart sensors and sampling Preprocessing Stage Get the three phase Decompose the signal in to 4 levels using Symlet Wavelet Fault Diagnosis Stage Identify fault condition Getting nodal components of currents Ia, Ib, Ic, Iα, Iβ and I with Clarks transform Decomposition of six current component signal with Symlet 2 wavelet in four levels of scaling Normalize using Clark s transform Get all four levels of detail and approximate coefficients of three phase and modal normal and fault currents Find faulted phases Classify faults Forming arrays ca4 and cd4 of approximate and detail coefficients of each samples Extract required feature from fourth level coefficient Find fault location features and different current signal varies. Features of FFT and mean used in earlier published work had room for accuracy improvement. So the use of STD feature is proposed for the first time in this work to increase accuracy of the method of [1, 11]. STD is defined as follows- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxi XÞ 2 STD ¼ n 1 STD X Xi Where. Extracting desired feature and Storing values components Investigation and analysis for fault diagnosis Fig. 4 Flowchart for fault diagnosis task ð2þ is the standard deviation is the mean of n numbers of cd4 is the i th coefficient cd4 in group of n samples of cd4 Standard deviation (STD) is found using the fourth level detail coefficient, cd4 using MATLAB software coding. STD shows the dispersion or deviation from its mean value and is a measure of how magnitudes of each sample vary from each other in a group of all samples for a particular sampling frequency. Hence, it is used to give better results as compared to other features like FFT and mean value. So this feature is considered Fig. 5 Different stages of proposed technique for further fault diagnosis analysis. However the accuracy of this method of using different feature values determined only from fault transient signals varies with the type of features selected. Even if the feature values of faulted phases become significantly higher than that of non-faulted phases. The exact percentage may vary depending upon system conditions and fault conditions. Thus, even if accuracy is increased with STD feature still comparison becomes difficult. It is seen that threshold value definition becomes less determined and depends upon the system conditions. To alleviate this drawback the use of fault location factors, FLF is presented in this work, which is our own contribution and is found to give the highest accuracy as compared to many previously presented methods [1, 11, 23, 25]. During normal operating conditions, all six values of features are taken before the occurrence of faults. During the fault conditions these six features are compared with normal Table 1 FFT feature values found out for different faults Fault Type Ia Ib Ic Iα Iβ I ABCG ABC AB AC CG

6 334 J Electron Test (217) 33: values. Fault locating indices or factors are determined for all six currents by using Eq. (3) and are analyzed. FLF ¼ FF FN ¼ STD F STD N Where, FLF FN FF STD F STD N fault locating factor feature value under normal condition feature value in fault conditions STD of cd4 of current during fault condition STD of cd4 of current during normal condition ð3þ Values of FLF are estimated for all buses or nodes in the system by sensing three phase currents before and after the fault conditions. A fault condition is indicated if FLF of any phase current is higher than one. The bus number at which a fault occurs and exact fault distance is located with the highest value of FLF found by using any of six currents. This method of using FLF gives better accuracy and is independent of use of feature current components of the transient signal used. The performance of the different features for different methods is discussed in result and discussion sections. This method can be used for fault diagnosis work of any system or power equipment to improve accuracy and robustness. 5ExperimentalSetUp Proposed technique of fault transient current decomposition and feature extraction as mentioned earlier is implemented by performing experimental work as follows. 5.1 System Used Figure 6 shows a sample 15 bus smart grid distribution system having two machines. The generators were simulated using detailed machine models. Each SS located at buses (number of which can decreased or increased) senses fault signals from the corresponding bus bar located near it. SS transfers fault signals to data centres for further analysis, as mentioned in section Simulation MATLAB software is used to carry out the simulation of the system. By simulating the sample test system for various types of open and short circuit faults on distribution feeders or lines, the three phase transient currents Ia, Ib and Ic and three Clark transformed components of transient current namely Io, Iα and Iβ are determined. Further steps are carried out by 6 developing analytical coding. Sample current signals are taken at a frequency of 1 khz. 5.3 Tests Performed Types of fault class are considered as line to ground (LG), Double line to fault (DLG), all line short circuits (LLL), all line to ground (3LG) total of 11 types of faults. Three LG faults are phase A to ground (AG), phase B to ground (BG), and phase C to ground (CG). Three DLG (ABG, BCG, and CAG), one LLLG (ABCG), three LL (AB, BC, CA) and one LLL (ABC) faults are also considered at different fault locations. These types of faults are simulated on the test model. The fault classification and location is carried out by phase and modal current component by executing comparative analysis. Even though few types of faults like three lines ABC or ABCG occur very rarely in the system they are also considered to demonstrate the wide applicability of the method. While most of the methods presented in literature so far, are applicable only for few types of faults. This is the added advantage of the method presented in this work. 6 Test Results 7 8 DG Fig. 6 Sample 15 bus test system used for experiment 6.1 Faulted Phase Identification Table 1 shows FFT feature values extracted from all six current components of the transient signal. In case of fault CG, it is seen that the feature of current Ic becomes significantly greater than that of phase B and A. In case of fault AB, feature values of current Ia and Ib are about the same but are significantly greater than that of phase C. Feature values of phase A are about the same in fault ABCG, ABC or AB and are also significantly less that of no fault phase. Feature values of any faulted phase become greater than that of non faulted phases.

7 J Electron Test (217) 33: Fault Classification After finding the faulted phases as mentioned above, exact fault classification task is carried out. Whether the fault involves ground or not is decided from feature values of I. From Table 1 it is seen that if the fault is ABC, all FFT feature values of current Ia, Ib and Ic becomes greater than normal values. These values also become approximately same but feature value of I in this case is less than one. In case of fault ABCG feature values of I also become higher than that of the normal condition. In case of a CG fault as shown in Table 1 it is seen that, only phase C feature values become higher than phase A and B. Values of feature of I for fault CG are greater than that of faults involving phase C like AC or ABC. Figure 7 represents variation of FLF values of healthy and faulted phases. Phases during which a fault is occurred are identified from FLF values obtained from three phase currents Ia, Ib and Ic. FLF values in this case will become greater than 1. If phase does not involve a fault then its value remains 1. Thus fault identification and classification can be carried out easily with the FLF method. 6.3 Fault Location The faulted bus is found out by finding bus with highest feature value. Figure 8 shows FFT feature values for AB fault at bus 1, which are highest as compared to that of other buses. In case of FLF method, Fig. 7 indicates that for AB fault, values of FLFs of phase A and B (Ia and Ib) are higher than that of phase C and are also greater than 1. For BG fault value of B phase s FLF is higher than that of phase C and A. In case of three phase faults like ABC, all FLLF s ofthree phases become greater than 1. In case of ABCG fault FLF of I also becomes greater with that of phases A, B, and C. Thus the logic of fault identification, classification and location becomes simple along with easier comparisons as compared to only using feature values extracted from fault transients. FLF (No Units) FLF with Ia FLF with Ib FLF with Ic FLF in Normal condition AB ABC CA AG BG Fault type Fig. 7 Fault location factors of healthy and faulted phases FFT (No units) Number of bus Fig. 8 FFT values of three phases for fault AB at bus 1 7 Performance Comparisons 7.1 Performance of Coefficient Type Selected The average accuracy of the detail level coefficient is higher than that of the approximate coefficient. Figure 9 represents percentage error calculated with fourth level approximate and detail coefficients for faults at different locations. It can be seen that average performance of detail coefficient level is better than that of approximate coefficient. 7.2 Performance of Feature Type In the paper [2, 29] and our published work [1, 11], fault location is done by using either maximum or peak values of features. The logic of using different features is verified and it is found that accuracy is not very dependent on feature type for fault identification and classification. However, for fault location accuracy is greater with STD feature method used in this research work. From Fig. 1,it isseenthatthe FFT feature gives less error as compared to the mean value feature. But the STD feature demonstrates best performance as compared to other methods in [1, 11, 23]. In [23] it is found that the error varies with the fault resistance, fault type, fault location and may be up to 81%. The work of [23] also uses hybrid approach with drawbacks mentioned earlier. 7.3 Performance of FLF Method Table 2 shows FLF values found for faults at bus 1 for various types of faults at different locations. FLF are found by using three features as in the method of using peak value of features. It is seen that FLF values of respective faulted phases are Error (%) Fault location ( Bus No) Fig. 9 Percentage errors with different coefficient Ia Ib Ic ca4 cd4

8 336 J Electron Test (217) 33: Error (%) Bus no Fig. 1 Variation of Errors with different methods STD Mean FFT higher than non faulted phases. This ensures the validity of the method as the easy and accurate fault classification technique. From Fig. 11 it can be seen that FLF for different types of faults at one particular location are same. Therefore, the FLF value gives correct prediction of fault locations and is independent of the feature type. Table 2 represents the fault location with all three features of Mean, FFT and STD. It can be seen that FFT method is more suitable as compared to mean value method. Table 2 Fault bus located with FLF values Fault Type Actual fault bus No Feature Type Type of coef. Fault bus no located by feature of current Ia Ib Ic Iα Iβ I ABCG 1 STD Ca Cd Mean Ca Cd FFT Ca Cd ABG 1 STD Ca Cd Mean Ca Cd FFT Ca Cd CAG 3 STD Ca Cd Mean Ca Cd FFT Ca Cd ABC 3 STD Ca Cd Mean Ca Cd FFT Ca Cd BG 9 STD Ca Cd Mean Ca Cd FFT Ca Cd ABC 9 STD Ca Cd Mean Ca Cd FFT Ca Cd BG 1 STD Ca Cd Mean Ca Cd FFT Ca Cd ABG 1 STD Ca Cd Mean Ca Cd FFT Ca Cd

9 J Electron Test (217) 33: FLF 5 The STD feature method is giving the most accurate results except for the feature value with current I. Moreover, with FLF, the accuracy of FFT and mean feature is also increased as compared to using peak values of features. The error in [25]is foundtobe3 4% and error in [23] found to be 81%. Whereas error with FLF and STD error is almost % in many cases which is represented by locating the bus correctly. Therefore, use of coefficient cd4, STD feature with FLF method proposed here is found to be most accurate as compared to other methods. Moreover with FLF method, comparison becomes easier as the threshold value is 1 which is independent of the system, fault conditions and also does not require a classifier. 8 Conclusion line 2-3 line 3-4 line AG ABG ABCG ABC Fig. 11 FLF values for different fault at different locations Smart grid distribution system monitors can use advanced technologies with advanced sensors and communication devices for effective fault diagnosis. With the help of smart sensors, feature values can be extracted very accurately using advanced Symlet mother wavelet functions. In this work, it has been demonstrated that that Symlet mother wavelet function gives more robust results and eliminates the drawbacks of multiple locations of faults as seen in existing methods of using conventional mother wavelet functions. With the proposed use of STD deviation feature, improved accuracy is demonstrated as compared to other features. Fault location factors, FLF used in this work is our contribution. This method founds to be giving the highest accuracy as compared to our and many previously published work. The results prove the effectiveness of the proposed method for finding fault type and fault location. It increases the speed, robustness and accuracy of the fault diagnosis process. The developed algorithm can be used for real time applications with additional investigations. The method can be used for fault diagnosis work of any other system or power equipment to improve the accuracy and the robustness of the process. Acknowledgments The authors gratefully acknowledge the management of AISSMS for providing the facilities for executing this research work. Authors are also thankful to the Chief engineer Mr. Satish Karpe and Guruprasad Samsagikar from SCADA DMS center, Bhandup, Mumbai, MS, India for their contributions related to this work. References 1. Bakar AHA, Ali MS, Tan CK, Mokhlis H, Arof H, Illias HA (214) High impedance fault location in 11 kv underground distribution systems using wavelet transforms. Electr Power Energy Syst 55: Borghetti A, Bosetti M, Di Silvestro M, Nucci CA, Paolone M (28) Continuous-wavelet transform for fault location in distribution power networks: definition of mother wavelets inferred from fault originated transients. IEEE Trans Power Syst 23(2): Chen W-H, Liu C-W, Tsai M-S (21) Fast fault section estimation in distribution substations using matrix-based cause-effect networks. IEEE Trans Power Delivery 16(4): Clarke E (1943) Circuit analysis of AC power systems, symmetrical and related components, vol 1. Wiley, New York 5. Costa FB, Silva KM, Dantas KMC, Souza BA, Brito NSD (27) A wavelet-based algorithm for disturbances detection using oscillographic data. Proc. Int. Conf. Power Systems Transients 6. Costa FB, Souza BA, Brito NSD (28) A wavelet-based algorithm to analyze oscillographic data with single and multiple disturbances. Proc. IEEE Power and Energy Society General Meeting: Conversion and Delivery of Electrical Energy in the 21st Century, pp de Apraiz M, Barros J, Diego RI (21) A laboratory for time frequency analysis of power quality disturbances using wavelets. Proc. 14th IEEE International Conf. on Harmonics and Quality of Power, pp Dhend MH, Chile RH (215) Smart distribution monitoring system for smart grid. Int J Disc 47(22): Dhend M, Chile R (215) Innovative scheme for smart grid distribution system. Conf. Proc. IEEE PEL Society s Int. Future Energy Electronics Conf, Taiwan 1. Dhend MH, Chile RH (216) Efficient fault diagnosis in smart grid using non conventional mother wavelet function. Proc. IEEE International Conf. Asia Pacific Power and Energy, Xian, China 11. Dhend MH, Chile RH (216) Fault diagnosis methodology in smart grid with distributed energy generation. IEEE international conf, Birmingham 12. El-Zonkoly AM (211) Fault diagnosis in distribution networks with distributed generation. Electr Power Syst Res 81: He H, Shen X, Staryk JA (29) Power quality disturbances analysis based on EDMRA method. Electr Power Energy Syst 31: Hu G-S, Zhu F-F, Ren Z (28) Power quality disturbance identification using wavelet packet energy entropy and weighted support vector machines. J Exp Syst Appl 35: Huang S-J (22) Application of immune-based optimization method for fault-section estimation in a distribution system. IEEE Trans Power Delivery 17(3): Li Y, Bielby S, Chowdhury A, Roberts GW (216) A jitter injection signal generation and extraction system for embedded test of highspeed data I/O. J Electron Test 32(4): Liu Y, Schulz N (22) Knowledge-based system for distribution system outage locating using comprehensive information. IEEE Trans Power Syst 17(2): Liu Y, Fan R, Terzija V (216) Power system restoration: a literature review from 26 to 216. Springer J Mod Power Syst Clean Energy 4(3): Livani H, Evrenosoglu CY (214) A machine learning and waveletbased fault location method for hybrid transmission lines. IEEE Trans on Smart Grid 5(1): Moravej Z, Abdoos AA, Pazoki M (21) Detection and classification of power quality disturbances using wavelet transform and support vector machines. Electr Power Comp Syst 38: Panigrahi BK, Pandi VR (29) Optimal feature selection for classification of power quality disturbances using wavelet packet-based fuzzy k-nearest neighbor algorithm. IET Gener Transm Distrib 3:296 36

10 338 J Electron Test (217) 33: Perera N, Rajapakse AD, Buchholzer TE (28) Isolation of faults in distribution networks with distributed generators. IEEE Trans on Power Delivery 23(4): Meisam Pourahmadi, Ali Akbar Safavi (211) Path characteristic frequency- based fault locating in radial distribution systems using wavelets and neural networks. IEEE Trans Power Delivery 26(2): Sadeh J, Bakhshizadeh E, Kazemzadeh R (213) New fault location algorithm for radial distribution systems using modal analysis. Electrical Power and Energy Systems 45: Salim RH, de Oliveira KRC, Filomena AD, Resener M, Bretas AS (28) Hybrid fault diagnosis scheme implementation for power distribution systems automation. IEEE Trans Power Delivery 23(4): Thukaram D, Khincha H, Vijaynarasimha H (25) Artificial neural network and support vector machine approach for locating faults in radial distribution systems. IEEE Trans Power Delivery 2(2): Trindade FCL (214) Fault location in distribution systems based on smart feeder meters. IEEE Tsransactions on Power Delivery 29(1): Uyar M, Yildirim S, Gencoglu MT (28) An effective waveletbased feature extraction method for classification of power quality disturbance signals. Electr Power Syst Res 78: Zhang Xiaoli, Zeng Xiangjun, Lei Li, S. S. Choi, Wang Yuanyuan (27) Fault location using wavelet energy spectrum analysis of traveling waves. Int conf. Power Engineering Conference (IPEC), pp 1 6 Mangal Hemant Dhend received her Diploma in electrical engineering from Bombay Technical Board in 1987 from Government College of Enginnering, Karad, and BE Electrical Degree from Government College of Engineering, Karad, Shivaji University, Kolhapur in 199. She has also done Advance Diploma in Computer System, Software, System Analysis and Applications from Bombay Technical Board in 1992, and ME Power System Degree from Government College of engineering Pune from University of Pune in Currently she is pursuing her doctoral research work under the guidance of Dr. R. H. Chile, Professor of Instrumentation Department at Shri Guru Gobvind Singh College of Engineering, Nanded, Swami Ramanand Tirth Marathwada University, Nanded. She has worked as a RESEARCH AND DEVELOPMENT ENGINEER in Kirlosakar Pneumatic Ltd, Pune, and since last 25 years has been working as an ASSISTANT PROFESSOR in Electrical Engineering Department of AISSMS College of Engineering, Pune. She was a Head of Department for more than 12 years. Recently she has received distinguished women engineer award in March 217 and best paper award in Springers International Conference in December 216. Her research topic is BSome investigations of distribution system monitoring and fault location techniques in smart grid system^.she is an author of four books and has published more than 45 papers in International or national journals and conferences. Dr. Rajan Hari Chile is a PROFESSOR in Department of Instrumentation Engineering of SGGS Institute of Engineering and Technology, Nanded. He obtained his B.E. (Instrumentation) and M.E.(Instrumentation) degrees in 1987 and 1992, respectively, from SGGS College of Engineering and Technology, Nanded. He has done Ph.D. from Department of Electrical Engineering, Roorkee University, Roorkee, in Nov He has 28 years of Teaching Experience of UG and 23 years of PG teaching. His area of specialization is Process Control and Instrumentation, Instrumentation System Design, Control Engineering, and Adaptive Control and its Applications to Process Industries. He has to his credits total 1+ research papers, 25 papers in International Journals, 2 in National Journals and more than 55 in International and National Conferences.