Expert Systems with Applications xxx (2011) xxxxxx

Contents lists available at ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Power quality diagnosis using time frequency analysis and rule based techniques
M. Faisal, A. Mohamed , H. Shareef, A. Hussain
Department of Electrical, Electronic and Systems Engineering, University Kebangsaan Malaysia, 43600 Bangi, Selangor, Malaysia

a r t i c l e

i n f o

a b s t r a c t
Diagnosing a power quality disturbance means identifying the type and cause of the disturbance. Fast diagnosis of power quality disturbances is important so as to assist network operators in performing counter measures and implementing suitable power quality mitigation actions. In this study a novel method for performing power quality diagnosis is presented by using the S-transform and rule based classication techniques. The proposed power quality diagnosis method was evaluated for its functionality in detecting the type of short duration voltage disturbances and identifying the cause of the disturbances which may be due to permanent or non permanent faults. Based on the results, this new method has the potential to be used in the existing real time power quality monitoring system in Malaysia to expedite the diagnosis on the recorded voltage disturbances. 2011 Elsevier Ltd. All rights reserved.

Keywords: Event identication Power quality diagnosis Power quality monitoring Rule-based S-transform

1. Introduction It is widely believed that power quality problems can only be traced to the incoming power from the utility. However, it is not always true because the source of such problems can be traced either to the customer facility or even up to the equipment inside a facility. Power quality problems that originate from the utility often have the greatest impact on a facilitys operation. Typical utility-generated events are such as breaker clearing which can produce voltage sag and arcing contactors which may generate impulses (Yalcinkaya, Bollen, & Crossley, 1998). The power quality problems arise within the customers own installation may be due to lose connections, overloaded circuits and transformers, ground loops and wiring errors. In addition, some automated and computer-based equipments produce harmonics which may propagate into the network and affect other customers. The initial step taken to solve power quality problems is by installing permanent power quality recorders (PQR) in the power supply networks either at the main substations or the point of common coupling (PCC) between the customers and the power utility. A power quality monitoring system (PQMS) as shown in Fig. 1 is usually required to connect all the PQR via a communication network to transmit the real-time measured power quality data to the power utility engineers. The PQMS server immediately sends a summary of the power quality disturbance data through short message system to the utility engineers. However, the existing PQMS does not diagnose power quality disturbances in which the sources and causes of the disturbances are not known. The

Corresponding author. Tel.: +603 89216006.

E-mail address: azah@eng.ukm.my (A. Mohamed). 0957-4174/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.04.047

sources of disturbance may originate from distribution feeder 1 (DF1), transmission feeder 2 (TF2) and other distribution feeders. Usually, the sources and causes of the disturbances are known only after conducting site inspections which is a time consuming process. Many research works focus on automated detection and classication of power quality disturbances but not on diagnosis of disturbances. Some studies on power quality diagnosis consider event identication, stochastic diagnosis, and power quality indexing and cause identication (Azam, Fang, Pattipati, & Rajaiah Karanam, 2004; Il-Yop et al., 2003; Il-Yop, Dong-Jun, Joong-Moon, Seon-Ju, & Seung-Il, 2007; Kezunovic, 2001). For event identication, three functions are considered which include event location, cause identication and solution suggestion. An automated software approach for the analysis of voltage sags, their causes and impacts has been developed using the Fourier and wavelet-transform signal processing techniques and fuzzy expert system (Kezunovic, 2001). Another method for identifying the types and causes of disturbances is by using the model based approach and decision fusion (Azam et al., 2004). A network based power quality diagnostic system (PQDS) that incorporates a graphical user interface of an existing PQMS has also been developed (Il-Yop et al., 2003, 2007). This PQDS has the functions for event detection, power quality trend analysis, event cause identication and event location. However, in this PQDS, the source of voltage sag can only be determined by analyzing the actual power system topology and this presents a disadvantage to the system. In this paper a novel approach to perform automatic power quality diagnosis for identifying the cause of short duration voltage disturbances is presented. The proposed method is developed by using the S-transform and rule based techniques. The excellent

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047

M. Faisal et al. / Expert Systems with Applications xxx (2011) xxxxxx

Send SMS

PQMS Server

Current Input

33 kV

11 kV

PQR
Voltage Input

X A
DF1 DF2

Access using Desktop computer

132 kV 33 kV

X X X X B C D

Source

TF1

DF3

TF2

DF4

X
33 kV 415 Volt

Fig. 1. A power quality monitoring system (PQMS) installed in a distribution network.

time-frequency resolution characteristic of the S-transform makes it an attractive candidate for the analysis of power quality disturbances under noisy condition and has the ability to detect the single and multiple disturbances correctly. Features extracted by the S-transform are applied to a modular rule based classier for automatic diagnosis of the disturbances as shown in Fig. 2. The process starts with the recording of disturbance data using the on-line PQMS. These data are then processed using the S-transform to extract the features that characterize the disturbances. These features are then applied to three rule based classiers (RBC). The rst RBC detects the type of short duration voltage disturbances which are either voltage sag or swell, while the second RBC diagnoses the cause of sag or swell as either due to permanent or non permanent faults. The third RBC than further classies the non permanent fault as either transient or incipient fault. In this way, the cause of voltage sag or swell can be determined as to whether it is due to a permanent fault, transient fault or incipient fault.

2. Categories of network faults and causes of short duration voltage disturbances Faults that occur in a power network may be categorized as either permanent or non permanent faults as shown in Fig. 3. Faults in a power network generally cause large increase in short circuit current to ow in the network which in turn gives rise to large voltage drops across the impedances of the supply system. These short circuits are caused by external interferences termed as permanent faults. Examples of causes of permanent fault are due to underground cable faults and ashover at medium voltage circuit breakers. Permanent faults that occur in a power network tend to produce short duration voltage disturbances such as voltage sag and swell as specied in IEC/TR 61000-2-8. Another type of network fault is termed as a non permanent fault which occurs at random moments for a nite period of time. The non permanent faults can be further categorized as either transient or incipient faults. A transient fault is dened as a fault that is

START Power Quality Recorders Data (disturbance waveforms) Data acquistion system S-transform Feature extraction technique

1st Rule Based Classifier

Short duration event? NO

YES Categories of faults 2nd Rule Based Classifier

END YES Incipient fault YES 3rd Rule Based Classifier Categories of faults NO Transient fault ? Non permanent fault

Permanent fault ? NO

Fig. 2. Diagnosis of short duration voltage disturbances.

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047

M. Faisal et al. / Expert Systems with Applications xxx (2011) xxxxxx

Network Faults

xt; f p e 2 ei2pft 2p

jf j

t2 f 2

Substituting (2) and (3) into (1), the new S-transform equation is derived as,

Permanent Faults

Non Permanent Faults

jf j Ss; f p 2p

ht e

ts2 2

ei2pft dt

Transient Faults

Incipient Faults

Fig. 3. Categories of network faults.

no longer present if power is disconnected for a short time (White, Transient faults, & IEEE Aerospace Conference, 2004). Typical examples of causes of transient faults are due to momentary tree contact, bird or other animal contact, lightning strike and conductor clash. Incipient faults are on the other hand are considered as non permanent faults caused by partial damage or contamination that progressively weakens the integrity of the network components over time and leads to insulation failure (Weeks & Steiner, 1982). Such faults are intermittent in nature and are considered as self-clearing arcing faults which occur very frequently prior to failing permanently. These events typically last for one half-cycle and extinguish at the rst natural zero crossing of the current. The magnitude of the half-cycle event is primarily dependent on the location of the fault on the feeder. But is also dependent on the point on the voltage waveform where the fault starts. Operational experience suggests that it is benecial to isolate a feeder suspicious of incipient cable faults very early after detecting the rst symptom of an incipient fault. Isolating the feeder can limit the overall energy at the point of fault and it also limits the often-repeated voltage transient seen by the system. However, due to short duration of an incipient fault and the inability to achieve selectivity via time coordination, designing an incipient fault protection function becomes challenging. Incipient faults are predictable and avoidable if the degradation processes are known. Therefore, the detection and isolation of incipient faults in the power supply system is essential for guaranteeing safe, reliable, and efcient operation of customers installations.

The phase factor in (4) is a phase correction of the denition of the CWT. It eliminates the concept of wavelet analysis by separating the mother wavelet into two parts, the slowly varying envelope (Gaussian function) which localizes in time, and the oscillatory exponential kernel ei2pft which selects the frequency being localized (Pinnegar & Mansinha 2003). It is the time localizing Gaussian that is translated while the oscillatory exponential kernel remains stationary. By not translating the oscillatory exponential kernel, the S-transform localizes the real and the imaginary components of the spectrum independently, localizing the phase spectrum as well as the amplitude spectrum, and is thus directly invertible into the Fourier Transform Spectrum (H). This characteristic of the phase is referred to as absolutely referenced phase information which is given by,

Ss; f ds Hf

The output of the S-transform is an N M matrix called the Smatrix whose rows pertain to the frequency and columns to time. Each element of the S-matrix is complex number and can be used as features to classify the non stationary multiple disturbances. The information in the S-matrix can be plotted as time-frequency contours which can be used to identify the existence of power quality disturbances. 3.1. STMV and STFV Indices Two indices called as S-transform magnitude-time voltage (STMV) and S-transform frequencytime voltage (STFV) are derived from the S-matrices. The STMV indices for all the three phase voltages are given as follows: STMV for the red phase voltage is derived from,

jf j SR V i;j p 2p

1 1

v R te

ts2 2

ei2pft dt

3. S-transform and its application for extracting disturbance features Short duration voltage disturbances can be easily detected by using advanced signal processing technique such as the S-transform which give time frequency representation of a signal (Pinnegar & Mansinha, 2003). The S-transform of a function h(t) can be dened as a continuous wavelet transform (CWT) multiplied by a phase factor ei2pfs and it is given by,

where SR V i;j is the S-matrix for voltage, vR, i = 1 . . . N (number of columns) and j = 1 . . . M (number of rows). The STMV for the red phase voltage is the maximum value for all the columns in the S-matrix which is given by,

V STMVR maxSR V i

Similarly, the STMV for the yellow and the blue phase voltages are,

V STMVY maxSY V i V STMVB maxSB V i

8 9

Ss; f e

i2pf s

W s; d

where W(s, d) is the CWT, s is time, and the dilation factor, d is the inverse of frequency, f. The CWT is a series of correlations of the time series with a function called as wavelet which is dened as,

The STFV indices for all the three phase voltages are next given as follows: STFV for the red phase voltage is derived from,

jf j SR V i;j p 2p

1 1

v R te

ts2 2

ei2pft dt

10

W s; d

htxt s; ddt

where SR V i;j is the S-matrix for voltage, vR, i = 1 . . . N and j = 1 . . . M. The STFV for the red phase voltage is the maximum value for all the rows in the S-matrix and is given by,

where x(t s, d) is the mother wavelet, t and s are both time. The mother wavelet is then given as,

V STFVR maxSR V j

11

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047

M. Faisal et al. / Expert Systems with Applications xxx (2011) xxxxxx

Fig. 4. Examples of the STMV and STFV plots for a non permanent fault.

Fig. 5. Examples of the STMV and STFV plots for a permanent fault.

Similarly, the STFV for the yellow and the blue phase voltages are,

V STFVY maxSY V j V STFVB maxSB V j

12 13

The STMV and STFV are used as input features to the rule based system for identifying the network faults. Examples of STMV and STFV plots for a non permanent fault, permanent fault and incipient fault can be visualized as shown in Figs. 46, respectively. In the rst row of Fig. 4, the per unit (pu) values of the STMV are shown. The STMV plots are almost similar to that of the root mean

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047

M. Faisal et al. / Expert Systems with Applications xxx (2011) xxxxxx

Fig. 6. Examples of the STMV and STFV plots for an incipient fault.

square plots in which the plots show the existence of voltage sags in the red and blue phases. The STFV plots shown in the second row of Fig. 4 are the variations in the system frequency before, during and after a voltage disturbance. Referring to the STFV plots, the frequency oscillations are seen more severe for the phases which experience the voltage disturbances. The main difference between the plots shown in Figs. 4 and 5 is in the oscillations of the system frequency. The permanent fault gives a more severe frequency oscillation and inicted more severe variations in the system frequency during the initiation and ending of the faults as compared to a non permanent fault. The STMV plot in Fig. 6 shows short duration increase in the yellow phase voltage values which signies the existence of high frequency current in one of the cable cores and hence indicates the existence of an incipient fault which is usually initiated by a partial discharge. This high frequency current will partially bridge the gap between phase insulation to ground or phase-to-phase insulation. Normally, when partial discharge is initiated, high frequency transient current pulses will appear and persist for nano-seconds to a milli-second, then disappear and reappear repeatedly (Weeks & Steiner 1982). The event may be detected as a very small change in the current

drawn by the sample under test. A signicant increase either in the PD level or in the developing rate of PD activity can provide an early indication of an incipient fault condition. 3.2. Features for detecting short duration voltage disturbances From the STMV and the STFV indices, several features are extracted and used for detecting the short duration voltage disturbances as depicted in Table 1. 3.3. Features for diagnosing the causes of the short duration voltage disturbances To perform automatic diagnosis of the causes of the voltage disturbances, four features (F1, F2, F3 and F4) are chosen from Table 2 and two new features, F10 and F11 are selected from the STFV index. The overall features shown in Table 2 are used to identify the cause of voltage disturbances which may be due to permanent and non permanent faults. The respective feature values for classifying permanent and non permanent faults are given as depicted in Table 3. The feature values for classifying transient and incipient faults are explained in Table 4. If all these features meet the specic data ranges for incipient fault, then it is considered as an incipient fault.

Table 1 Features selected for detecting short duration voltage disturbances. Feature F1 F2 F3 F4 F5 F6 F7 F8 F9 Index MaxSMT MinSMT Dmax Dmin FR5 FR6 FR7 FR8 FR9 Description Max values of STMV in per unit Min values of STMV in per unit Duration of max values of STMV above 1.10 per unit Duration of min values of STMV below 0.90 per unit Values of frequency resolution from 0.0061 to 0.022 Values of frequency resolution from 0.022 to 0.04 Values of frequency resolution from 0.04 to 0.08 Values of frequency resolution from 0.08 to 0.40 Values of frequency resolution from 0.40 to 0.50

Table 2 Features selected for performing power quality diagnosis. Feature F1 F2 F3 F4 F10 F11 Index MaxSMT MinSMT Dmax Dmin STDSFT DFSFT Description Max values of STMV in per unit Min values of STMV in per unit Duration of max values of STMV above 1.10 per unit Duration of min values of STMV below 0.90 per unit Standard deviation of STFV Difference of max and min of STFV in per unit

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047

M. Faisal et al. / Expert Systems with Applications xxx (2011) xxxxxx

Table 3 Feature values for classifying permanent faults and non permanent faults. Feature F1 Index MaxSMT Description Max values of STMV in per unit Permanent fault STMV(1P) < 0.90 & STMV(2P) > 1.10 STMV(2P) < 0.90 & STMV(1P) > 1.10 STMV(1P) < 0.90 STMV(2P) < 0.90 STMV(3P) < 0.90 Dmax > 10 ms Non-permanent fault STMV(1P) < 0.90 & STMV(2P) > 1.10 STMV(2P) < 0.90 & STMV(1P) > 1.10 0.90 < STMV(2P) < 1.10 & STMV(1P) > 1.10 STMV(1P) < 0.90 STMV(2P) < 0.90 STMV(3P) < 0.90 Dmax > 10 ms

F2

MinSMT

F3

Dmax

F4

Dmin

F10 F11

STDSFT DFSFT

Min values of STMV in per unit Duration of max values of STMV above 1.10 per unit Duration of min values of STMV below 0.90 per unit Std of STFV Difference of max and min of STFV values in per unit

used to classify the non permanent fault as either transient or incipient fault. Thus, three levels of classications are required in diagnosing the short duration voltage disturbances. The overall procedure for implementing power quality diagnosis using the three rule based classiers is as shown in Fig. 1. The rules for the three classiers are developed based on the features dened in Table 1, Table 3 (permanent and non permanent faults) and Table 4 (incipient and transient faults). The details of the rules for the rst, second and third RBC are described as in Tables 57, respectively.

5. Results and discussion The performance of the proposed power quality diagnosis using the S-transform and the rule based classiers was evaluated by testing the RBC with 342 sets of voltage disturbance data obtained from the existing PQMS in Malaysia. The power quality monitoring was performed at the 33 kV network before the rst step down transformer as shown in Fig. 1. The voltage waveforms recorded undergone two transformations before being recorded by the power quality recorder. From the 342 sets of data shown in Table 8, the type and cause of short duration voltage disturbances are known. The causes of the disturbances were validated by referring to the existing database from 2001 to 2008 provided by the national power utility,

Dmin > 10 ms

Dmin > 10 ms

STD > 8 DF > 30

STD < 8 DF < 30

If the data range does not meet the limits of incipient fault, then the fault is categorized as a transient fault. 4. Application of the rule based classiers for implementing power quality diagnosis Three rule based classiers (RBC) have been developed for implementing power quality diagnosis in a structured manner in which the rst level RBC is used to detect the short duration voltage disturbances such as voltage sag or swell, while the second level RBC is used to diagnose the cause of sag or swell as either due to permanent or non permanent faults. The third level RBC is then
Table 4 Feature values for classifying transient and incipient faults. Feature F1 Index MaxSMT Description Max values of STMV in per unit Transient fault STMV(1P) < 0.90 & STMV(2P) > 1.10 STMV(2P) < 0.90 & STMV(1P) > 1.10 STMV(1P) < 0.90 STMV(2P) < 0.90 STMV(3P) < 0.90 Dmax > 10 ms Incipient fault 0.90 < STMV(2P) < 1.10 & STMV(1P) > 1.10

Table 5 The rst level RBC rules for detecting the short duration voltage disturbances. Rules R1 Description If (0.90 < F3 < 1.10), (0.90 < F4 < 1.10), (F5 = 0), (F6 = 0), (F7 = 0), (F8 = 0) and (F9 = 0) then data is only C1 (Signal and only contains sinusoidal waveform) If (F3 < 0.90) & (0.90 < F4 < 1.10), then data is C2 (Voltage sag) If (0.90 < F3 < 1.10) & (F4 > 1.10), then data is C3 (Voltage swell) If (F5 > 0) and (F9 = 0) then data is C4 (Harmonic) If (F6 > 0), (F9 = 0) then data is C5 (Notch) If (F7 > 0), (F9 = 0) then data is C5 (Notch) If (F8 > 0), (F9 = 0) then data is C5 (Notch) If (F5 = 0), (F8 > 0) & (F9 = 0) then data is C6 (Oscillatory Transient) If (F5 = 0) and (F9 > 0) then data is C7 (Impulsive Transient)

R2 R3 R4 R5 R6 R7 R8 R9

Table 6 The second level RBC rules for classifying permanent and non permanent faults. Rules R1 R2 Description If F1 and F3 meet the requirement as in Table 3, F10 > 8 and F11 > 30 then the voltage sag or swell is caused by a permanent fault If F1 and F3 meet the requirement as in Table 3, F10 < 8 and F11 < 30 then the voltage sag or swell is caused by a non permanent fault If F2 and F4 meet the requirement as in Table 3, F10 > 8 and F11 > 30 then the voltage sag or swell is caused by a permanent fault If F2 and F4 meet the requirement as in Table 3, F10 < 8 and F11 < 30 then the voltage sag or swell is caused by a non permanent fault

F2

MinSMT

F3

Dmax

F4

Dmin

F10 F11

STDSFT DFSFT

Min values of STMV in per unit Duration of max values of STMV above 1.10 per unit Duration of min values of STMV below 0.90 per unit Std of STFV Difference of max and min of STFV values in per unit

Not applicable (N/A)

R3 R4

Dmax > 10 ms

Dmin > 10 ms

Dmin > 10 ms

Table 7 The third level RBC rules for classifying incipient and transient faults. Rules R1 Description If F3, F4, F10 and F11 meet the requirement as in Table 4 and F1 = (0.90 < STMV(2P) < 1.10 & STMV(1P) > 1.10) and F2 (No event) then the voltage sag or swell is caused by an incipient fault If F3, F4, F10 and F11 meet the requirement in Table 4 and F1 = (STMV(1P) < 0.90 & STMV(2P) > 1.10, STMV(2P) < 0.90 & STMV(1P) > 1.10) & F2 = (STMV(1P) < 0.90, STMV(2P) < 0.90, STMV(3P) < 0.90) then the voltage sag or swell is caused by a transient fault

STD < 8 DF < 30

STD < 8 DF < 30 R2

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047

M. Faisal et al. / Expert Systems with Applications xxx (2011) xxxxxx Table 8 Description of the power quality data to be diagnosed. Type of short duration voltage disturbance Voltage sag and swell Voltage sag Voltage swell Cause of voltage disturbance Permanent fault Non permanent fault (transient fault) Non permanent fault (incipient fault) Number of data sets 121 111 110

accuracy of 94.1%. Overall, there are 6 errors in diagnosing permanent faults, 7 errors in diagnosing transient faults and 7 errors in diagnosing incipient faults. The errors in the diagnosis of permanent faults are due to low values of the standard deviation of STFV (F10) in which the values for F10 for all the 6 error cases are less than 8. By referring to the rules in Table 6, these faults are considered as non permanent faults. 6. Conclusion

Table 9 Results of rst level RBC for detecting short duration voltage disturbances. Type of short duration voltage disturbance Voltage sag and swell Voltage sags Voltage swell Number of data sets 121 111 110 Correct detection 121 111 110 Wrong detection 0 0 0 % Accuracy in detection 100% 100% 100%

Table 10 Results of the second and third level RBCs for diagnosing the cause of voltage disturbance. Type of network fault Permanent fault Non permanent fault (transient fault) Non permanent fault (incipient fault) Number of data sets 121 111 110 Correct diagnosis 115 104 103 Wrong diagnosis 6 7 7 % Accuracy in diagnosis 95.0% 93.7% 93.6%

The new power quality diagnosis method provides diagnostic functions that can give sufcient information for power utility engineers to identify the cause of short duration voltage disturbances. The main advantage offered by the new method is the use of the S-transform to decompose the power line signals into a set of time-frequency components and evaluate the changes in the system frequency in which simple feature extraction can be performed. The features extracted from the S-transform are then used by the three rule based classiers to diagnose the type and causes of the short duration voltage disturbances. The numerical results obtained with actual power quality data recorded in a power distribution system indicated that the new method is effective in diagnosing the type and causes of the disturbances which may be due to permanent faults and non permanent faults categorized as either incipient or transient faults. References
Azam, M. S., Fang, T., Pattipati, K. R., & Rajaiah Karanam, R. (2004). A dependency model-based approach for identifying and evaluating power quality problems. IEEE Transactions on Power Delivery, 19(3), 11541166. Faisal, M. F., & Mohamed, A. (2009). Identication of multiple power quality disturbances in three phase industrial power systems using s-transform and rule based expert system. Journal of Applied Sciences, 9(15), 26882700. IEC/TR 61000-2-8, ELECTROMAGNETIC COMPATIBILITY (EMC) Part 28: ENVIRONMENT Voltage dips and short interruptions on public electric power supply systems with statistical measurement results. International Technical Report IEC 61000-2-8, pp 89. Il-Yop, Chunga., Dong-Jun, Won b., Joong-Moon, Kimc., Seon-Ju, Ahnd., & Seung-Il, Moond. (2007). Development of a network-based power quality diagnosis system. Electric Power Systems Research, 77, 10861094. Il-Yop, Chung., Dong-Jun, Won., Joong-Moon, Kim., Seon-Ju, Ahn., Seung-Il, Moon., Jang-Cheol, Seo., et al. (2003). Development of power quality diagnosis system for power quality improvement. IEEE Power Engineering Society General Meeting, 2, 1217. Kezunovic, M. (2001). Automated analysis of voltage sags, their causes and impacts. IEEE Power Engineering Society Summer Meeting, 2, 11131117. Pinnegar, C. R., & Mansinha, L. (2003). The S-transform with windows of arbitrary and varying shape. Geophysics, 68(1), 381385. Weeks, W. L., & Steiner, J. P. (1982). Instrumentation for the detection and location of incipient faults on power cable. IEEE Transactions on Power Apparatus and Systems, 101(7). White, A.L., 2004. Transient faults and network reliability. In IEEE Aerospace Conference Vol. 1, 613 March 2004, pp 11211128. Yalcinkaya, G., Bollen, M. H. J., & Crossley, P. A. (1998). Characterization of voltage sags in industrial distribution systems. IEEE Transaction on Industry applications., 34(4), 682688. Publisher Item Identier S 0093-999498)04909-3.

Tenaga Nasional Berhad. The data comprises of 121 sets of voltage disturbances caused by permanent faults and 221 sets of voltage disturbances caused by non permanent faults which are either transient or incipient fault. The source of the permanent and non permanent faults originates from either the transmission, distribution or customers own internal networks. By using the 342 sets of data, the RBC are evaluated in which the results for detecting the short duration voltage disturbances are as shown in Table 9. The results for diagnosing the cause of voltage disturbances as to whether the disturbance is caused by a permanent fault, non permanent fault which is either incipient or transient fault are as shown in Table 10. Based on the results in Table 9, the rst level RBC has successfully detected the various short duration voltage disturbances which are voltage sag, voltage swell and combined sag and swell with an accuracy of 100% (Faisal & Mohamed, 2009). The results in Table 10 showed that the new technique successfully diagnosed the causes of short duration voltage disturbances with an average

Please cite this article in press as: Faisal, M., et al. Power quality diagnosis using time frequency analysis and rule based techniques. Expert Systems with Applications (2011), doi:10.1016/j.eswa.2011.04.047