You are on page 1of 9

IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO.

1, JANUARY 2018 257

Online Fault Detection and Diagnosis in Photovoltaic


Systems Using Wavelet Packets
B. Pradeep Kumar, Student Member, IEEE, G. Saravana Ilango , Senior Member, IEEE,
M. Jaya Bharata Reddy, Senior Member, IEEE, and Nagamani Chilakapati , Senior Member, IEEE

Abstract—The nonlinear characteristics of a photovoltaic (PV) currents. Unless these faults are cleared by appropriate protec-
array, maximum power point tracking of the PV inverter, presence tion devices, they may cause fire hazards and reduce the power
of blocking diodes, and lower irradiation prevent the conventional output and system efficiency [3], [4].
protection devices to trip under certain faults and lead to reduced
system efficiency and fire hazards. Moreover, the PV character- The conventional protection system for PV arrays consists of
istics under certain partial shading conditions are similar to that overcurrent protection devices (OCPD) and a ground fault pro-
under fault conditions. Hence, it is imperative to detect faults and tection device (GFPD) [5]. However, certain PV array faults are
differentiate faults from the partial shading condition to avoid false not cleared by these protection devices, since the magnitude of
tripping of the system. This paper proposes a simple fault detection fault current is less than the minimum breaking capacity of the
method using the available data of array voltage and current by
means of wavelet packets. The proposed scheme is simulated us- fuse, which is usually 2.1 times of the short-circuit current [6].
ing MATLAB/Simulink and is experimentally tested on a 1.6-kW The factors that mainly influence the magnitude of fault current
4 × 4 PV array to validate its performance. are the presence of blocking diodes, irradiation level, maximum
Index Terms—Faults, partial shading, photovoltaic (PV) system,
power point tracking (MPPT) of the inverter, fault location, po-
wavelet packet transform (WPT). tential difference between the two faulted ends, fault impedance,
and degradation [2]. Furthermore, under certain partial shading
I. INTRODUCTION conditions, the P–V characteristics may be similar to that under
NCREASED concern on environmental pollution and reduc- a fault [7]. Hence, it is imperative to differentiate the partial
I tion in fossil fuels have motivated the search for alternative
energy sources that have a lower footprint on the environment.
shading from the fault condition to avoid the malfunctioning of
the protection system.
Among these renewable alternatives, solar energy is freely avail- Many fault detection methods projected in the past can be
able in unlimited quantities and does not require any moving categorized into three groups namely, performance comparison,
parts for conversion to electrical energy. Furthermore, the total signal processing, and machine learning techniques [8]. The first
installed photovoltaic (PV) capacity has increased due to the group identifies the faults based on the performance compari-
reduction in PV modules’ price and attractive government sub- son of PV systems. Platon et al. [9] presented a fault detection
sidies. According to global status report 2015, the overall PV algorithm (FDA) based on the power loss between the mea-
installed capacity is 227 GW [1]. With such a huge capacity, sured and predicted ac power. For better accuracy, it requires
it is necessary to have a reliable protection system in order to multiple models for various ranges of irradiation. In [10], if the
improve system safety, reliability, and efficiency. According to power loss is more than a threshold value, a reconfiguration
the location of fault or failure in a PV system, it can be classified technique is used to identify and bypass the defective module.
as in [2]: anomalies in the utility grid, failure within power con- However, it requires a massive number of switches for a large
ditioning unit, and faults on PV arrays. The latter of these faults PV system, which makes the protection system uneconomical.
is focused in this paper, i.e., PV array faults, namely, line-to- These comparison-based methods may not be effective due to
ground (LG) faults, line–line (LL) faults, mismatch faults, and the following reasons.
arc faults. Among these, LL and LG faults involve large fault 1) Actual and simulated PV modules may not be the same.
2) False tripping due to similarity between the low percent-
Manuscript received July 14, 2017; revised September 26, 2017; accepted Oc- age mismatch faults, low irradiation, and partial shading
tober 25, 2017. Date of publication December 4, 2017; date of current version conditions.
December 20, 2017. This work was supported by the Ministry of Electronics 3) Model needs to be updated regularly, since PV parameters
and Information Technology, Government of India. This research was performed
at the National Institute of Technology, Tiruchirappalli, India, funded through change with respect to seasonal conditions.
Young Faculty Research Fellowship under the DeitY Visvesvaraya Ph.D. The second group identified and localized the faults based on
scheme. (Corresponding author: G. Saravana Ilango.) the response of the reflected signal (signal-processing-based so-
The authors are with the Department of Electrical and Electronics Engineer-
ing, National Institute of Technology, Tiruchirappalli 620015, India (e-mail: lutions). Time-domain reflectometry, which is an offline method
pradeep301327@gmail.com; gsilango@nitt.edu; jbreddy@nitt.edu; cnmani@ used to detect and locate the open-circuit and short-circuit faults
nitt.edu). and degradation (increase in the series resistance) using the time-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. domain characteristics of the reflected voltage waveform [12],
Digital Object Identifier 10.1109/JPHOTOV.2017.2770159 [13]. Without disconnecting the inverter, using spread spectrum

2156-3381 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
258 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 1, JANUARY 2018

Fig. 1. Block diagram of a typical PV system.

time-domain reflectometry, an LG fault is detected by correlat- strategy in deriving the threshold values. In Section IV, exper-
ing the peaks of the generated and reflected waveforms [14]. imental results are discussed to validate the proposed method.
Both methods require a function generator, and the frequency Section V concludes this paper.
of the transmitted signal can be affected by the presence of long
dc cables, a line frequency transformer, and grounding of the II. DESCRIPTION OF A PHOTOVOLTAIC SYSTEM AND EFFECT OF
system. PHOTOVOLTAIC ARRAY FAULTS
The third group detects the faults by using machine learning
techniques, and these methods use the input data as output PV A. Description of a Photovoltaic System
array voltage (VPV ), PV array current (IPV ), irradiance, and The proposed grid-connected PV system shown in Fig. 1
temperature. Akram and Lotfifard [15] presented a method to consists of a PV array, blocking diodes, bypass diodes, a dc–dc
identify and classify the LL and open-circuit faults, using an converter with MPPT, a centralized inverter, protection devices
intelligent method called probabilistic neural network, in which such as the GFPD and the OCPD, an analog voltage module
a new temperature-dependent PV module model is proposed to (NI 9225), an analog current module (NI 9227), and a personal
validate the method. In [16], a decision-tree-based algorithm is computer (PC). The protection device GFPD is used to protect
proposed to detect and classify the faults. In this, less than 20% the array from the large ground fault current (Ig ) and is placed in
mismatch faults are wrongly classified as the normal condition, between the grounded current-carrying conductor and the sys-
while the method also requires large training data. To overcome tem grounding (shown in Fig. 1). When the current Ig passing
the requirement of a large training data under fault conditions, through the GFPD is more than the threshold value, it will dis-
a graph-based semisupervised learning method is proposed in connect the fault from the PV array. Another protection device
[17]. In another method, the voltage across each module is es- OCPD is used to protect the modules from the large fault cur-
timated by using a three-layer neural network [18] to detect rents. Moreover, to avoid electric shock to persons and to protect
and locate the short-circuit faults. The number of layers and the equipment from ground faults, all the noncurrent-carrying
the complexity in the control law increase with the array size. metal parts such as mounting racks, module frames, and power
These learning methods require fault data, which are costly, and converter chassis are connected to the ground through the equip-
these may not give accurate results, since the training data may ment grounding connector [5] (shown with green-dotted lines
not cover all the possible anomalies under various environmen- in Fig. 1).
tal conditions. In addition, these reference modules may not Further in a PV array, modules are connected in series to build
have the same environmental conditions as that of the working the required output voltage, and such strings are connected in
conditions of the PV array. parallel to obtain the desired output current. VPV and IPV are
Recently, Hariharan et al. [19] have proposed a method to measured at a sampling frequency of 1 KHz by using the analog
detect the PV array faults and partial shading by using the in- voltage module NI 9225 and the analog current module NI
stantaneous measurement of MPPT information (i.e., VPV and 9227, respectively, and their ratings are mentioned in Table I.
IPV ) and the reference module for measurement of irradiation. These two modules are placed in an eight-channel NI CDAQ
However, this method is not suitable for large-size PV plants, 9178 mixed measurement test system, which acts as an interface
where irradiance among the modules is different. between the modules and the PC. Using LabVIEW software,
This paper proposes an online fault detection method based faults are investigated by applying WPTs.
on wavelet packet transforms (WPTs) using only the MPPT Moreover, to excerpt the maximum power from the PV array,
information. The structure of this paper is described as fol- an MPPT (P&O) is used in the dc–dc converter. The extracted
lows. Section II describes a PV system and the effect of faults. maximum power is fed to the ac grid through the Omniksol-
Section III presents an algorithm for the proposed method and a 1.5k-TL inverter with dc input Pm ax = 1.75 kW.
KUMAR et al.: ONLINE FAULT DETECTION AND DIAGNOSIS IN PHOTOVOLTAIC SYSTEMS USING WAVELET PACKETS 259

TABLE I
RATINGS OF THE NI INSTRUMENTS USED FOR MEASURING
THE VOLTAGE AND CURRENT

Sensor Max. sampling Number of DC rating AC rating


frequency channels (rms)
(KHz)

NI 9227 50 4 14.1 A 5A
NI 9225 50 4 800 V 230 V

TABLE II
RATINGS OF NEX POWER NH-100 AT 100-W THIN-FILM PV MODULE

S.No Module parameter Module rating

1. Max. Power 100 W Fig. 2. Without blocking diodes, I–V characteristics of the PV array.
2. Open-Circuit Voltage 50.5 V
3. Short Circuit Current 3.3 A
4. Voltage at Max. power 36.5 V which is less than the minimum breaking capacity of the fuse,
5. Current at Max. Power 2.73 A i.e., 6.93 A (= 2.1∗ ISC ). Hence, the fault remains hidden, due
to which fast aging of modules will occur, and also, there is a
disruption in the power supply. To prevent the back-feed cur-
B. Effects of Faults in Photovoltaic Arrays rent and to improve the reliability of the PV system, blocking
diodes are connected in series with each string. However, the
Faults in a PV array are categorized as permanent faults and presence of a blocking diode makes the fault undetectable, even
temporary faults. LL, LG, open-circuit, and mismatch faults are under STC. In addition to blocking diodes, bypass diodes are
the permanent faults, due to which there is a permanent reduc- connected in antiparallel with the PV module to avoid the for-
tion in the output power. Among these, LL and LG faults are mation of hotspot under partial shading conditions. Under the
considered in this paper, since these faults may generate huge partial shading condition, the shaded PV module is bypassed
fault current, which can accelerate the degradation of the PV with the conduction of the bypass diode, and this condition is
modules and also lead to fire hazards. An LL or LG fault oc- similar to that under fault. Hence, to avoid maloperation of the
curs due to an accidental interconnection between two different protection system, an efficient FDA is required.
potential nodes or between one node of the PV array and the
ground, respectively. The LG fault and open-circuit faults are
III. PROPOSED ALGORITHM FOR FAULT DIAGNOSIS
special cases of the LL fault [2], [19], and the severity of an
LL fault depends upon the percentage mismatch. Therefore, the Wavelets techniques are particularly suitable in analyzing the
ability to detect an LL fault implies that the LG or open-circuit transient characteristics of the faults or any other type of dis-
faults can also be detected by the protection system. Moreover, turbance that contains high-frequency components over a short
the partial shading and sudden change in the irradiations are duration of time [11], [20]–[23]. Wavelet analysis is the splitting
temporary faults, because of which there is a degradation in the up of a signal into scaled and shifted versions of the original (or
output power for a short duration of time. The pictorial repre- mother) wavelet. In continuous wavelet transforms, coefficients
sentation of various faults is shown in Fig. 1. are calculated by comparing the input signal with the scaled
A 4 × 4 solar PV array, with a maximum power of 1.6 kW, and shifted versions of wavelets. However, the calculation of all
is considered to investigate the faults in the PV array, and the coefficients is time consuming. To overcome this, the discrete
ratings of the PV module are listed in Table II. F1 in Fig. 1 refers wavelet transform (DWT) is used. In the DWT, low-pass and
to an LL fault with single module mismatch or 25% mismatch high-pass filters are used to discriminate the low-frequency (ap-
(out of four modules, one module is shorted) occurring between proximations) and high-frequency components (details) present
strings 1 and 2, whereas F2, F3, and F4 refer to two-module in the signal, respectively [24]. Furthermore, in each level of
(50%), three-module (75%), and four-module (100% or total the DWT, only approximations are further divided into approx-
string shorted) mismatch faults, respectively. Under standard imations and details, whereas in the WPT, in each level along
test conditions (STC), if an LL fault occurs, there is a signif- with approximations, details are also decomposed into approx-
icant reduction in VPV and a large back-feed current flows, imations and details. Hence, the WPT provides more accurate
which is enough to melt the fuse (OCPD), so that fault may be signal analysis when compared with wavelet transforms. More-
isolated from the healthy system. The magnitude of back-feed over, the Haar wavelet is used as a mother wavelet, and the
current depends on the percentage of mismatch and irradia- coefficients are calculated as sums and differences of the suc-
tion levels. However, under low irradiation (say 300 W/m2 ), an cessive elements [22], [25].
LL fault with two-module mismatch (F2) is created in the ab- Vpv and Ipv are the only input data considered for analyzing
sence of blocking diodes. From the I–V characteristics shown in the faults in the PV array. With these input data, three features,
Fig. 2, at the instant of the fault, the back-feed current is –2.8 A, namely, change in the array voltage, energy of the array voltage,
260 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 1, JANUARY 2018

Fig. 3. Extraction of features for the FDA.

Fig. 4. Extraction of wavelet packet coefficients V w p c (i) and Z w p c (i). Fig. 5. Flowchart for the proposed method to identify the faults.
TABLE III
and energy for the change in the impedance are extracted (shown THRESHOLD VALUES FOR THE ΔV, E av , AND E Im p
in Fig. 3) to identify the faults and are explained as follows.
1) Change in the array voltage (ΔV): Variable Constant value

ΔV (i) = Vpv (i + 1) − Vpv (i) (1) V1 −19 V


ΔV V2 −11 V
where Vpv (i) and Vpv (i + 1) are the successive samples of V3 −40 V
ε1 31
the array voltage. E av ε2 40
2) Energy of the array voltage (Eav ): Eav is the sum of the ε3 1500
squares of the wavelet packet coefficients of the array ε4 107
voltage (Vwp c ), which is shown in Fig. 4 E Im p ε5 5 × 107


Eav = Vwp
2
c (i) . (2)
Step 1: Get the features ΔV(i), Eav , and EIm p for every t0
3) Energy for the change in the impedance (EI m p ):
seconds (1 s). Set i = 1.
Vpv (i + 1)−Vpv (i) Step 2: If ΔV (i) lies within the limits [V1 , V2 ], then go to
Change in the impedance (Z(i)) = .
Ipv (i + 1)−Ipv (i) step 4; otherwise, go to the next step.
(3) Step 3: If ΔV (i) < V3 , then go to step 5; otherwise, go to
Zwp c (i) is the wavelet packet coefficient for the change in step 8.
the impedance in the node “10” shown in Fig. 4. EIm p is the Step 4: If Eav < ε1 (or) EIm p > ε4 , then go to step 6; other-
sum of the squares of the wavelet packet coefficients Zwp c (i): wise, go to step 7.
 Step 5: If Eav lies within the limits (ε2 , ε3 ) (or) EIm p > ε5 ,
EIm p = Zwp2
c (i) . (4) then go to step 6; otherwise, go to step 7.
Step 6: “Fault occurred.” Go to step 9.
A. Flowchart for the Fault Detection Algorithm Step 7: “Fault has not occurred.” Go to the next step.
Step 8: If i < 1000, increment i by 1 and go to step 2; other-
Features extracted from Fig. 4, namely ΔV, Eav , and EIm p
wise, go to step 1.
are given as input to the FDA (shown in Fig. 5), and their thresh-
Step 9: Disconnect the PV array from the ac grid.
old values are listed in Table III. When a fault or partial shading
occurs in a PV system, instantly VPV decreases (ΔV is negative),
B. Determination of Threshold Values
whereas if irradiation increases, VPV will increase (ΔV is pos-
itive). The main aim is to detect the faults and partial shading; To determine the threshold values, a 4 × 4 PV array shown
therefore, the algorithm will execute only when ΔV is negative. in Fig. 1 is simulated in MATLAB/Simulink under various con-
Furthermore, the step-by-step execution of the algorithm is as ditions such as faults with different percentage mismatch at
follows. discrete irradiance level, faults created at different instants in
KUMAR et al.: ONLINE FAULT DETECTION AND DIAGNOSIS IN PHOTOVOLTAIC SYSTEMS USING WAVELET PACKETS 261

Fig. 6. Variation of ΔV as a function of E av with and without 25% mismatch Fig. 7. At different irradiations, ΔV and E av values with and without fault
fault at different irradiations. having more than 50% mismatch.

the time interval (starting, ending and middle of the interval),


partial shading, and sudden change in the irradiation conditions.
Under STC, the PV array operates at a maximum power point
(MPP) of (144 V, 11.12 A), and the magnitudes of Eav and EIm p
are approximately equal to 35 and 8.3, respectively. At the time
of fault occurrence, depending on the magnitude of ΔV, Eav ,
and EIm p , the threshold values are chosen as follows.
1) V1 , V2 , and ε1 : These limits are determined by simu-
lating an LL fault with 25% mismatch (F1 shown in Fig. 1)
at two extreme irradiation levels: 1) 1000 W/m2 (STC); and
2) 230 W/m2 (low irradiance). Due to this fault, at an ir-
radiance of 1000 W/m2 , VPV drops from 145 to 128 V
(ΔV = 128 V−145 V = −17 V), whereas at 230 W/m2 , ΔV
is −13.2 V (= 105.7 V–118.9 V) and the postfault VM PP oper-
ates at reduced voltage. Based on the voltage drop under extreme
irradiation conditions and with a voltage tolerance of 1% of VPV ,
the voltage limits for the identification of the LL fault with 25% Fig. 8. At different irradiations, variation of E Im p with respect to ΔV with
mismatch are considered as [−19 V, −11 V] (say V1 = −19 V, and without fault.
V2 = −11 V). For certain partial shading and sudden change
in the irradiation conditions, the change in VPV lies within the
limits [V1 , V2 ]. Hence, to differentiate the faults from environ- different irradiation levels, partial shading at more than one
mental disturbances, another feature Eav is used. From Fig. 6, it module with a shaded irradiance of 500, 400, 300, and
is observed that most of the fault data lie within the area bounded 200 W/m2 , and irradiation changes in steps of ±100, ±200,
by the ΔV limits [V1 , V2 ] and below the Eav magnitude of 31 and ±300 W/m2 . In Fig. 7, the entire fault data lie on the left-
(say ε1 ). The undetected faults are marked with circles, since hand side of −40 V (represented with dotted line) and lie within
the magnitude of Eav is approximately equal to the magnitude the Eav limits of [40, 1500] (say [ε2 , ε3 ]).
of Eav under normal conditions. 3) Determination of ε4 and ε5 : Impedance seen by the dc–dc
2) V3 , ε2 , and ε3 : For more than 25% mismatch fault, post- converter changes whenever a fault or irradiation change occurs
fault VM PP is near to prefault VM PP , and the magnitude of Eav in a PV system. Hence, to improve the accuracy in detecting
is increasing, which is contrary to the Eav changing pattern dur- the faults, another feature called energy for the change in the
ing the 25% mismatch fault. Moreover, the magnitude of ΔV impedance (Eim p ) is used. Fig. 8 shows the distribution of sim-
increases with the percentage mismatch. Hence, the lower limit ulated faults and normal condition data at various irradiations.
(V3 ) can be calculated by simulating an LL fault with 50% mis- From the plot, for ΔV < −40, the threshold value for Eim p
match at a low irradiation of 230 W/m2 . At this instant of fault, is chosen as 5 × 107 (say ε5 ), and for ΔV within the limits
the VPV suddenly drops from 120.9 to 79 V (ΔV = −41.9 V), [−19 V, −11 V], the threshold value for EIm p is selected as 107
and hence, the upper limit V3 is chosen as −40 V (say V3 ), be- (say ε4 ), above which it indicates the occurrence of fault in the
low which the fault occurs. Fig. 7 shows the plot between ΔV PV array. The undetected faults in Figs. 6. and 7 are identified
and Eav for various LL faults with more than 25% mismatch at with these threshold values ε4 and ε5 .
262 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 1, JANUARY 2018

TABLE IV
ACCURACY (IN PERCENT) OF THE PROPOSED METHOD

Irradiation (w/m2 ) % mismatch

25% 50% 75% 100%

1000 15/15 15/15 15/15 15/15


800 15/15 15/15 15/15 15/15
600 15/15 15/15 15/15 15/15
230 13/15 15/15 15/15 15/15

Table IV gives the accuracy details of the proposed method


for various mismatch faults at discrete irradiation levels. Hence,
the proposed algorithm exhibits a promising performance in
meeting the challenges.

C. Generalized Procedure to Determine Threshold Values


Step 1: Simulate all the possible percentage mismatch (1, 2, 3, Fig. 9. (a) 4 × 4 PV array. (b) Data acquisition using NI instruments.
. . . N modules) LL faults in the required array configuration (C) Grid-connected inverter.
(say N × M) at extreme irradiation levels (say 1000 and
230 W/m2 ) and at a temperature of 25 °C. the Eav upper limit such that these irradiation changes should
Step 2: At the instant of each fault, calculate the drop in the not be detected as a fault.
array voltage (Vdrop ) and the energy values Eav and Eim p for
a fixed duration of time interval (say 1 s). IV. RESULTS AND DISCUSSION
Step 3: Segregate the faults into two groups, Group A and Group A 4 × 4 PV array is connected in a series–parallel config-
B, based on the postfault operating voltage. uration for experimental verification of the proposed method
Group A: Up to certain percentage mismatch fault, the faulty (shown in Fig. 9). The PV status is indicated with “0” for nor-
string supplies the power to the load, and the corresponding mal operation of the PV array and is indicated with “1” when a
postfault VM PP operates at reduced voltage (say 25% mismatch fault is occurred in the PV array. Array voltage and current are
fault in the 4 × 4 configuration). sampled at a frequency of 1 kHz using NI instruments.
Group B: After certain percentage mismatch fault, the faulty
string is open circuited due to the characteristics of the blocking A. Line–Line Fault With One-Module Mismatch (F1)
diode, and the postfault VM PP operates near to prefault MPP To observe the performance of the proposed method, at the
voltage (say more than 25% mismatch fault in the 4 × 4 con- time of 11:17 hours, an LL fault with one-module mismatch (F1
figuration). shown in Fig. 1) is created at a low irradiation of 360 W/m2
shown in Fig. 10. At this instant, VPV suddenly drops from 141
Step 4: In Group A, for each percentage mismatch fault, the
to 125.2 V, and hence, ΔV = −15.8 V. Since the ΔV value
threshold values for each parameter are calculated as follows:
lies within the limits [−19 V, −11 V], the algorithm checks
ΔV Upper limit = Vdrop at 230 W/m2 , 25 °C + tolerance for the energy condition. Energies EIm p and Eav are calculated
(1% of array voltage). for every 1000 samples per second. At this instant of time, the
ΔV Lower limit = Vdrop at 1000 W/m2 , 25 °C + tolerance magnitude of Eav is 258.5 (< ε1, condition not satisfied), and
(1% of array voltage). the magnitude of EIm p is 5.47 × 108 (> ε4 , condition satisfied).
Eim p Lower limit = Eim p at 230 W/m2 , 25 °C. As ΔV and one of the energy condition (EIm p ) is satisfied, then
Eav Upper limit = Eav at 1000 W/m2 , 25 °C. there is a fault occurred in the PV array at this instant of time
11:17 hours. Other than this point of time, even though VPV is
Step 5: In Group B, for each percentage mismatch fault, the continuously changing, but ΔV is greater than −10 V, which is
threshold values for each parameter are calculated as follows: out of limits. Hence, an LL fault is found to occur at the instant
ΔV Upper limit = Vdrop at 230 W/m2 , 25 °C + tolerance of 11:17 hours, and from this instant, the PV status is changed
(1% of array voltage). to 1.
ΔV Lower limit = Vdrop at 1000 W/m2 , 25 °C + tolerance
(1% of array voltage). B. Line–Line Faults With More Than One-Module Mismatch
Eim p Lower limit = Eim p at 230 W/m2 , 25 °C. (F2, F3, and F4)
Eav Lower limit = Eav at 230 W/m2 , 25 °C. An LL fault with two-module mismatch is created on a cloudy
The Eav upper limit is determined by simulating the various day, during which the incident irradiation, VPV , and IPV are
irradiation changes (say ±100, ±200, and ±300 W/m2 ), and fix recorded from 11:10 to 11:38 hours and are shown in Fig. 11.
KUMAR et al.: ONLINE FAULT DETECTION AND DIAGNOSIS IN PHOTOVOLTAIC SYSTEMS USING WAVELET PACKETS 263

Fig. 12. Experimental results for dynamic shading.

Fig. 10. Experimental results for an LL fault with one-module mismatch at At time t = 11:31 hours, an LL fault with two-module mis-
low irradiation level. match is introduced in the PV array (F2 shown in Fig. 1). At the
instant of fault, VPV drops from 142.9 to 0 V, instead of 85% of
open-circuit voltage (Vo c ), which is approximately 85 V (from
the simulation). This is because of limitation in the minimum
operating voltage of the inverter, i.e., 120 V, below which the
inverter fails to operate. Therefore, the inverter will be turned
OFF, and VPV drops to 0 V. After some time, the MPPT and the
inverter start working; then, VPV increases from 0 to 163 V and
settles at 146.4 V, which is called voltage at the MPP (Vm pp )
shown in Fig. 11. At the instant 11:31 hours, the change in the
array voltage (ΔV) is –142.9 V, which is less than ε3 ; then, the
algorithm checks for the energy condition. At this instant, the
magnitude of Eav is 20,650 (< ε3 , condition not satisfied) and
the magnitude of EIm p is 1 × 1010 (> ε5 , condition satisfied).
Hence, an LL fault occurred at this instant, and the PV status is
changed to 1 at 11:31 hours.
For LL faults with three-module (F3) and four-module (F4)
mismatch, the experimental results are similar to that of the two-
module mismatch, since in all these three cases after the fault,
the VPV drops to 0 V because of the inverter failure.

C. Partial Shading
Partial shading is a temporary fault, due to which there is a
transitory reduction in the output power. There are two types of
shadings: static and dynamic shading. Static shading is caused
by the leaves, bird droppings, and dust accumulated on the
glass, whereas dynamic shading occurs due to nearby buildings,
rooftop structures, and chimneys. In dynamic shading, as the
Fig. 11. Experimental results for an LL fault with two-module mismatch.
time progresses, the shading area on the PV modules is chang-
ing, due to which there is a continuous variation in the output
power.
264 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 8, NO. 1, JANUARY 2018

Experimentally, at time 09:10 hours, dynamic shading is in- [12] T. Takashima, J. Yamaguchi, and M. Ishida, “Fault detection by signal
troduced by arranging a wooden plank near to the PV array, response in PV module strings,” in Proc. 33rd IEEE Photovoltaic Spec.
Conf., San Diego, CA, USA, 2008, pp. 1–5.
which casts shadow on the PV modules. As the time progresses, [13] L. Schirone, F. P. Califano, U. Moschella, and U. Rocca, “Fault finding
the shadowing part on the module increases and it spreads from in a 1 MW photovoltaic plant by reflectometry,” in Proc. IEEE 1st World
one module to another module, and the corresponding variation Conf. Photovoltaic Energy Convers., Waikoloa, HI, USA, 1994, vol. 1,
pp. 846–849.
of VPV , ΔV, Eav , and EIm p is shown in Fig. 12. From Fig. 12, [14] M. K. Alam, F. Khan, J. Johnson, and J. Flicker, “PV ground-fault de-
even though the ΔV value lies within the limits [−19 V, −11 V], tection using spread spectrum time domain reflectometry (SSTDR),” in
the corresponding Eav (<ε1 ) and EIm p (>ε4 ) conditions are Proc. IEEE Energy Convers. Congr. Expo., Denver, CO, USA, 2013,
pp. 1015–102.
not satisfied. However, at certain time periods, even though the [15] M. N. Akram and S. Lotfifard, “Modeling and health monitoring of dc
ΔV value is small, but the corresponding Eim p value is in the side of photovoltaic array,” IEEE Trans. Sustain. Energy, vol. 6, no. 4,
order of 107 . It is because of small change in the array current pp. 1245–1253, Oct. 2015.
[16] Y. Zhao et al., “Decision tree-based fault detection and classification in
(= I2 − I1 , which is in the order of milliamperes), which re- solar photovoltaic arrays,” in Proc. 27th Annu. IEEE Appl. Power Electron.
sults in large change in the impedance ( = (V2 − V1 )/(I2 − Conf. Expo., Orlando, FL, USA, 2012, pp. 93–99.
I1 )), and the corresponding magnitude of Eim p is very high and [17] Y. Zhao, R. Ball, J. Mosesian, J. F. de Palma, and B. Lehman, “Graph-
based semi-supervised learning for fault detection and classification in
is in the order 107 . It is notable that the proposed algorithm solar photovoltaic arrays,” IEEE Trans. Power Electron., vol. 30, no. 5,
performs well, and a normal (no fault) status is recorded. pp. 2848–2858, May 2015.
[18] Syafaruddin, E. Karatepe, and T. Hiyama, “Controlling of artificial neural
network for fault diagnosis of photovoltaic array,” in Proc. 16th Int. Conf.
V. CONCLUSION Intell. Syst. Appl. Power Syst., Hersonissos, Greece, 2011, pp. 1–6.
[19] R. Hariharan, M. Chakkarapani, G. Saravana Ilango, and C. Nagamani,
This paper has proposed an online method to identify the “A method to detect photovoltaic array faults and partial shading in PV
faults under low irradiation and has also differentiated the partial systems,” IEEE J. Photovoltaics, vol. 6, no. 5, pp. 1278–1285, Sep. 2016.
shading from the faulty conditions based on the WPT. Fault [20] M. J. B. Reddy, D. V. Rajesh, P. Gopakumar, and D. K. Mohanta, “Smart
fault location for smart grid operation using RTUs and computational
diagnosis has been performed by using three variables namely, intelligence techniques,” IEEE Syst. J., vol. 8, no. 4, pp. 1260–1271,
change in the array voltage, energy of the array voltage, and Dec. 2014.
energy for the change in the impedance. This method requires [21] F. H. Magnago and A. Abur, “Fault location using wavelets,” IEEE Trans.
Power Del., vol. 13, no. 4, pp. 1475–1480, Oct. 1998.
only array voltage and array current, which are readily available [22] P. K. Murthy, J. Amarnath, S. Kamakshiah, and B. P. Singh, “Wavelet
at the input of the inverter and hence reduces the cost of the transform approach for detection and location of faults in HVDC system,”
protection system. The proposed method has been validated in Proc. IEEE Region 10 3rd Int. Conf. Ind. Inf. Syst., Kharagpur, India,
2008, pp. 1–6.
with both the simulation and experimental results and shows its [23] M. J. B. Reddy and D. K. Mohanta, “Performance evaluation of an
auspicious achievement in resolving the challenges. adaptive-network-based fuzzy inference system approach for location of
faults on transmission lines using Monte Carlo simulation,” IEEE Trans.
Fuzzy Syst., vol. 16, no. 4, pp. 909–919, Aug. 2008.
REFERENCES [24] M. Misiti, Y. Misiti, G. Oppenheim, and J. Poggi, Wavelet Toolbox. Natick,
MA, USA: MathWorks, 1996.
[1] “Renewables 2016 Global Status Report,” 2016. [Online]. Available: [25] R. Coifman, Y. Meyer, and M. W. Wickerhauser, “Method and apparatus
http://www.ren21.net/wp-content/uploads/2016/06/GSR_2016_Full_Rep for encoding and decoding using wavelet-packets,” U.S. Patent 5 526 299,
ort.pdf Jun. 11, 1996.
[2] Y. Zhao, J. F. de Palma, J. Mosesian, R. Lyons, and B. Lehman, “Line–
line fault analysis and protection challenges in solar photovoltaic arrays,”
IEEE Trans. Ind. Electron., vol. 60, no. 9, pp. 3784–3795, Sep. 2013.
[3] B. Brooks, “The Bakersfield fire: A lesson in ground-fault protection,” B. Pradeep Kumar (S’17) received the B.Tech. de-
Sol. Pro Mag., vol. 4, no. 2, pp. 62–70, Feb./Mar 2011. gree in electrical and electronics engineering from
[4] Duke Energy, “Commercial Rooftop PV Fires Webinar—NC PV DG K.S.R.M. College of Engineering, Kadapa, India, in
Program SEPA Presentation,” Solar ABCs, 2011. [Online]. Available: 2005, and the M.Tech. degree in power electron-
http://www.solarabcs.org/about/publications/meeting_presentations_ ics from B.M.S. College of Engineering, Bangalore,
minutes/2011/12/pdfs/Duke-Webinar-Dec2011.pdf India, in 2008.
[5] U.S. National Electrical Code, “Article 690—Solar Photovoltaic Systems,” He is currently a Full-Time Research Scholar with
2011. the National Institute of Technology, Tiruchirappalli,
[6] Outline of Investigation for Low-Voltage Fuses—Fuses for Photovoltaic India. His research interests include identification
Systems, UL Subject 2579, 2010. of faults, degradation in photovoltaic systems, and
[7] R. Hariharan, M. Chakkarapani, and G. S. Ilango, “Challenges in the power electronics.
detection of line-line faults in PV arrays due to partial shading,” in
Proc. Int. Conf. Energy Efficient Technol. Sustain., Nagercoil, India, 2016,
pp. 23–27.
[8] Y. Zhao, “Fault detection, classification and protection in solar photo- G. Saravana Ilango (SM’17) received the Graduate
voltaic arrays,” Ph.D. dissertation, Dept. Elect. Comput. Eng., Northeast- degree electrical and electronics engineering from the
ern Univ., Boston, MA, USA, Aug. 2015. University of Madras, Chennai, India, in 2000, the
[9] R. Platon, J. Martel, N. Woodruff, and T. Y. Chau, “Online fault detection Master’s degree in power electronics and drives from
in PV systems,” IEEE Trans. Sustain. Energy, vol. 6, no. 4, pp. 1200–1207, Bharathidasan University, Tiruchirappalli, India, in
Oct. 2015. 2001, and the Ph.D. degree in electrical engineer-
[10] X. Lin, Y. Wang, M. Pedram, J. Kim, and N. Chang, “Designing fault- ing from the National Institute of Technology (NIT),
tolerant photovoltaic systems,” IEEE Des. Test, vol. 31, no. 3, pp. 76–84, Tiruchirappalli, in 2009.
Jun. 2014. From 2001 to 2004, he was a Lecturer with the
[11] Z. Yi and A. Etemadi, “Line-to-line fault detection for photovoltaic Noorul Islam College of Engineering, Kumaracoil,
arrays based on multi-resolution signal decomposition and two-stage India. He has been an Assistant Professor with the
support vector machine,” IEEE Trans. Ind. Electron., vol. 64, no. 11, NIT Tiruchirappalli since 2006. His research interests include FACTS con-
pp. 8546–8556, Nov. 2017. trollers, digital controllers, and renewable energy systems.
KUMAR et al.: ONLINE FAULT DETECTION AND DIAGNOSIS IN PHOTOVOLTAIC SYSTEMS USING WAVELET PACKETS 265

M. Jaya Bharata Reddy (SM’13) received the Nagamani Chilakapati (M’10–SM’16) received the
B.E. degree in electrical and electronics engineering M.Tech. degree in power electronics from the In-
from Acharya Nagarjuna University, Guntur, India, in dian Institute of Technology, Kanpur, India, in 1984,
2002, and the M.E. degree in electrical engineering and the Ph.D. degree in electrical engineering from
and Ph.D. degree in power system from Birla Insti- the University of Technology, Sydney, Australia, in
tute of Technology, Ranchi, India, in 2004 and 2008, 2001.
respectively. From 1985 to 1991, she was with the Central
He is currently an Associate Professor with the Power Research Institute, Bangaluru, India. Subse-
Department of Electrical and Electronics Engineer- quently, she joined the Department of Electrical and
ing, National Institute of Technology, Tiruchirappalli, Electronics Engineering, National Institute of Tech-
India. His current research interests include smart nology, Tiruchirappalli, India, where she is currently
grids, substation automation, wide-area protection, digital relaying, and power a Professor. Her research interests include power electronics and drives, renew-
system protection. able energy systems, and FACTS controllers.
Dr. Reddy is an Editorial Board Member of Electrical Power Components
and Systems.

You might also like