Renewable Energy 66 (2014) 99e110

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Fault detection method for grid-connected photovoltaic plants

W. Chine a, A. Mellit a, b, *, A. Massi Pavan c, S.A. Kalogirou d
a
Faculty of Sciences and Technology, Renewable Energy Laboratory, Jijel University, 18000 Jijel, Algeria
b
Unité de développement des équipements solaires (USES/EPST-CDER), Bousmail, 42000 Tipaza, Algeria
c
Department of Engineering and Architecture, University of Trieste, Via A. Valerio, 6/A, 34127 Trieste, Italy
d
Department of Mechanical Engineering and Materials Science and Engineering, Cyprus University of Technology, P.O. Box 50329, 3603 Limassol, Cyprus

a r t i c l e i n f o a b s t r a c t

Article history: In this work, an automatic fault detection method for grid-connected photovoltaic (GCPV) plants is
Received 22 May 2013 presented. The proposed method generates a diagnostic signal which indicates possible faults occurring
Accepted 29 November 2013 in the GCPV plant. In order to determine the location of the fault, the ratio between DC and AC power is
Available online 22 December 2013
monitored. The software tool developed identifies different types of faults like: fault in a photovoltaic
module, fault in a photovoltaic string, fault in an inverter, and a general fault that may include partial
Keywords:
shading, PV ageing, or MPPT error. In addition to the diagnostic signal, other essential information about
GCPV plant
the system can be displayed each 10 min on the designed tool. The method has been validated using an
Power loss
Fault detection
experimental database of climatic and electrical parameters regarding a 20 kWp GCPV plant installed on
Diagnostic the rooftop of the municipality of Trieste, Italy. The obtained results indicate that the proposed method
can detect and locate correctly different type of faults in both DC and AC sides of the GCPV plant. The
developed software can help users to check possible faults on their systems in real time.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction temperature, and solar irradiance level. In addition, others losses

The number of photovoltaic (PV) systems is increasing rapidly
all over the world. Grid-connected PV plants e with sizes varying
from a few kWp (domestic plants) to several MWp (utility-scale
plants) e represent world-wide the power technology with the
highest rate of growth. This is due to the simplicity of installation,
high reliability, zero fuel costs, very low maintenance costs, and the
lack of noise due to the absence of moving parts [1].
The energy produced by a grid-connected PV (GCPV) plant de-
pends on various factors such as the nominal characteristics of the
components of the PV system, electrical and geometrical configu-
rations, weather conditions of site of installation mainly with
respect to solar radiation availability, the local horizon and the
near-field shading, availability of the plant, failures that may occur
during its operation [2], and other factors which are not very
important. A number of different issues can cause the loss of energy
in the plant. Some losses are related to the PV array, these include
maximum power point tracking error, module parameters disper-
sion (mismatch), wiring losses and ageing [3]. Others are influ-
enced by environmental characteristics such as operating
* Corresponding author. The International Centre for Theoretical Physics (ICTP),
Trieste, Italy. Tel.: þ213 (0)551 998 982.
E-mail addresses: adelmellit2013@gmail.com, amellit@ictp.it (A. Mellit).
are referred to the power conditioning units (DCeDC and DCeAC
converters).
Nowadays, many diagnostic techniques are developed for
possible faults detection in PV systems. Some of these do not
require climate data (solar radiation and module temperature) such
as the earth capacitance measurement (ECM) developed by Taka-
shima et al. [4] that consists of an electrical method for detecting
where a photovoltaic module in a string has been disconnected, the
time-domain reflectometry (TDR) that measures the electrical
characteristic of a transmission line developed by Schirone et al. [5]
which can detect not only the disconnection in the string, but also
the impedance change due to degradation, and a statistical
approach based on the ANOVA (Analysis of Variance) test and non-
parametric KruskaleWallis test that shows a high level of accuracy
and is fast in fault diagnosis [6].
Additionally, a remote monitoring and fault detection method of
small GCPV systems [7] used climate data from satellites observa-
tion that replaces on-site measurements. Then, the expected en-
ergy yield is calculated and compared with the measured one. The
expected system’s energy yields do not have the same accuracy
than yields calculated from real monitored data and values with
root mean square error (RMSE) of about 10% have been reported for
irradiance estimated using these methods [8]. This type of analysis
allows four different types of failures to be distinguished: constant
energy losses (for example due to degradation, soiling, module

0960-1481/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.renene.2013.11.073
100 W. Chine et al. / Renewable Energy 66 (2014) 99e110

Fig. 1. a. The GCPV plant installed on the rooftop of the Trieste local government, data-acquisition system and the designed tool. b. Example of the monitored data (Gi: irradiance, Tc:
cell temperature, output DC and AC power) for a few days (SQ3).

defect, string defect, etc.), variable energy losses (for example due
to shading [9], grid disconnections, power limitation of the inverter,
MPP tracking failures, temperature, etc.), and losses due to the
presence of snow (blackout).
Other researchers used climate data measured by local sensors
on the plants. This category includes the intelligent based ap-
proaches, as an example the method proposed by Syafaruddin et al.
[10] that used a three layered feed forward neural network, which
allows the identification of the short-circuit location of PV modules
in one string. Another intelligent system for automatic detection of
faults in PV fields based on a TakagieSugenoeKahn Fuzzy Rule-
Based System (TSKFRBS) was described in Ref. [11]. The results
show that the system can recognize more than 90% of fault con-
ditions, even when noisy data are introduced.
Some other techniques are based on learning methods and take
into account the specific condition of the plant under monitoring
[12]. This type of monitoring system simplifies the operation and
maintenance of the PV systems, even though it needs many mea-
surement sensors. Two types of faults were identified by this
method: shading and inverter failure. Firth et al. [13] have devel-
oped a technique that used only few measurement sensors, which
can classify the energy losses in four different categories: sustained
zero efficiency faults, brief zero efficiency faults, shading, and non-
zero efficiency non-shading faults. Moreover, the loss of energy can
be quantified. Nevertheless, this technique does not detect any type
of fault leading to possible loss of energy without any alarm given.
Gokmen et al. [14] developed a simple diagnostic method to
determine the number of open and short-circuited PV modules in a
 W. Chine et al. / Renewable Energy 66 (2014) 99e110 101

Table 1
Characteristics of EC-115 evergreen module [21]. a 5000
4500
Technology Polycrystalline Si

Simulated DC power (W )
4000
Peak power 115 W
3500
Open circuit voltage at STC 21.5 V
Maximum power point voltage at STC 17.3 V 3000
Short circuit current at STC 7.26 A 2500
Maximum power point current at STC 6.65 A
Number of cells 72 2000
Nominal operating cell temperature 44  C 1500
Voltageetemperature coefficient 0.53%/ C
1000
Currentetemperature coefficient 0.049%/ C
Poweretemperature coefficient 0.49%/ C 500
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Measured DC power (W)
string with a small number of sensors. Zhao et al. [15] examined
two types of faults taking into account Over-Current Protection b 4500

Devices (OCPD). The first type of fault is the line to line one that 4000
occurs under low irradiance conditions and the second type is again
the line to line fault but this time occurring in PV arrays where 3500

Simulated AC power (W)

blocking diode have been used.
3000
A fault diagnosis procedure based on the extended correlation
function and the matter element model has been presented by 2500
Chao et al. [16]. The obtained results indicate that the proposed
fault diagnosis method detects the malfunction correctly and 2000
promptly with less memory consumption and the maintenance
1500
staff can confirm the fault types of PV system without system
interruption. Miwa et al. [17] proposed a new method based on the 1000
analysis of the (dI/dV)eV characteristic. The method can detect
the partial shadow phenomenon. 500

Zhao et al. [18] developed a fault detection and classification

0
method based on decision trees (DT). The obtained DT models have 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
shown a high accuracy in fault detection and classification. A fault Measured AC power (W)
detection algorithm acting on the power conditioning system (PCS,
i.e., the inverter) of the PV plant using wavelet transform is pro- Fig. 2. a. Measured versus simulated DC output power. b. Measured versus simulated
posed in Ref. [19]. This method allows the detection of the fault and AC output power.

its location without any additional hardware. This has the disad-
vantage of high cost and re-design problem if the inverter specifi-
AC power ratio, and the ratio between the simulated and measured
cation has been changed. An automatic supervision and fault
DC voltage and current. The fault categories identified by the pro-
detection procedure based on the power losses analysis was pro-
posed technique are:
posed in Ref. [20]. This procedure is able to detect only faults that
occur on the DC side of the PV system. The technique allows the
- faulty modules in a string;
identification of four different types of faults: faulty modules in a
- faulty string;
string, faulty string, false alarm, and a group of different faults such
- faulty inverter;
as partial shadow, ageing, and MPPT error.
- false alarm;
Based on the techniques mentioned above, we can conclude that
- group of faults which include partial shading, ageing of PVs, and
some are not automatic and give only the possible fault types (can
MPPT error.
not identify the exact location of fault), and some others can detect
only the faults occurred on the DC side of the PV system.
Moreover, the technique allows exact location of the failed
In this paper, a simple and fast automatic method is presented. It
string and failed inverter by calculating the DC current ratio of each
allows the detection of faults occurring both in the DC and in the AC
string and the AC power ratio of each inverter. Finally, a software
side of a GCPV system. Initially, the absolute error on the perfor-
tool is designed using MatLab/GUI (graphical user interface) to
mance ratio is calculated and compared with a threshold gener-
display and monitor automatically the possible faults occurring in a
ating a diagnostic signal. Subsequently, three parameters are
PV plant. Essential information about the system is also displayed
calculated and used in order to determine the type of fault: the DCe
and stored each 10 min, which permits users to check their system
in real-time. Therefore, the main contribution of present work is to
Table 2
propose a simple and fast diagnosis method for GCPV systems, to
Mastervolt inverter electrical data [21]. design a user friendly tool using MatLab/GUI and finally to verity
this method using experimental data.
Number of MPPT 4
Output voltage 230 V, 50 HZ
Nominal power 5500 W
European efficiency 94% Table 3
Maximum efficiency 95% Estimated parameters using least square algorithm.
MPPT voltage range 100e380 V
y0 c1 c2 c3 c4
Maximum DC voltage 450 V
Nominal DC current 7.5 A Estimated parameters 1.0733 1.6817 0.0721 1.8440 0.4069
102 W. Chine et al. / Renewable Energy 66 (2014) 99e110

a
Climatic data from DAQ Module parameters and plant
(irradiance+module Configuration Electrical data from DAQ
Temperature) (sec 2.1)

Simulation model of
PV module
(sec. 3.1.1)

Real final yield:

Inverterefficiencymodel :
Sec. 3.1.2)

Reference yield: Simulated final yield:

PR_meas=Yf_meas/Yr
PR_sim=Yf_sim/Yr

Yes No

Diagnostic signal=0 Diagnostic signal=1

b Diagnostic signal=1

No Yes

Rp_dc_ac<1

Yes No
-Defect inverter
Rc>1
-Hot inverter

Yes No No Yes

Rv>1 Rv>1
No Yes

Rp_ac j<1

False Faulty No fault in Inverter jis

is
- Partial shadow alarm modules inverterj failed
- Aging
- MPPT error

Faulty
strings

Yes No

Rvi<1

No fault in string i
String i is failed

Fig. 3. a. Flowchart of the fault-diagnosis system of the GCPV plant. b. Flowchart of the fault identification and location procedure.
 W. Chine et al. / Renewable Energy 66 (2014) 99e110 103

Premium). As an example, Fig. 1b shows the measured climatic data

(irradiance and module’s temperature) and the output DC and AC
power of a PV array QS3. The recorded data are from May 1st to May
15th 2009 with a time scale of 10 min. The quality of the measured
data depends mainly to the precision of the data-loggers and sen-
sors used (reference cell, temperature sensors PT100, PT1000, etc.).

3. Methodology

In this section, the modelling of the DC and AC power is presented.

The most significant factors, which involve power losses in both DC
and AC sides of the PV system, are discussed. Finally, the principle of
the fault detection and diagnostic procedure is presented.

3.1. DC and AC output model

3.1.1. 1DC output model

The DC side of the PV system is modelled using the 5-parame-
ters model. The currentevoltage characteristic of a PV module can
be obtained using the single diode model [22,23]:
  !  
VþIRs
V þ IRs
I ¼ IPh  I0 e nsVt
1  (1)
Rsh
Fig. 4. 4 Typical faults in a PV array of 174 modules.
where : Iph is the photo-generated current at Standard Test Condi-
tion (STC), I0 is the dark saturation current at STC, Rs is the module
This paper is organized as follows: the next section provides a series resistance, Rsh is the panel parallel (shunt) resistance, ns is the
brief description of the GCPV plant and the database employed in number of serial cell and Vt is the thermal voltage. This is given by:
this study. Section 3 presents the methodology used. Results and
discussion are presented in Section 4, while Section 5 lists the
conclusions of this work.

2. The GCPV plant and the database employed

2.1. The GCPV plant

The PV system under study is a 20 kWp grid connected PV plant

installed on the rooftop of a building of Trieste local government in
Italy (see Fig. 1a). The system consists by 174 polycrystalline silicon
PV modules with a nominal power of 115 Wp. The specifications of
the modules are reported in Table 1. The photovoltaic modules are
organized in 12 strings and each string is made of 14or 15 series-
connected PV modules. The 12 strings are divided into three
groups of four strings; two strings with 28 PV modules-called
group QS1, QS3, and QS4, and another three groups of two strings
with 30 PV modules-called group QS2, QS5, and QS6. Each group is
connected to the inverter input stage, as reported in Ref. [21]. Each
inverter has a single-phase AC output and is endowed with two
MPPT systems. The inverter used is the Mastervolt QS-6400. Its
electrical characteristics are reported in Table 2.

2.2. The database

The monitoring of the photovoltaic plant is made by using two

different data logger (see Fig. 1a); one is dedicated to the climatic
data and the second one to the electrical data recorded for each
group (QS1, QS2, QS3, QS4, QS5, QS6) [21]. The monitored data are:
the irradiance on the array plane (Gi), the module’s temperature
(Tc) and the ambient temperature at array side (Ta), the operating
voltage (Vstr), current (Istr), power (Pstr), grid voltage (Vgrid), the
AC power (Pgrid), and the energy produced both for the DC and for
the AC sides. Two data-loggers were used, the first one (Danfoss
ComLynx Monitor) is used to record the Gi, Tc and Ta, while the Fig. 5. a. PV array output power, measured and simulated results. b. Inverter output
second one to record the electrical data (Mastervolt QS Data Control power, measured and simulated results.
104 W. Chine et al. / Renewable Energy 66 (2014) 99e110

Fig. 6. a. Absolute performance ratio error and diagnostic signal evolution (no fault). b. False alarm observed in the designed software. c. DC current, DC voltage and DCeAC power
ratios evolution. 6d. No fault observed in the designed software.

AKT
Vt ¼
q
where: K is Boltzmann’s constant, q is the charge of the electron,
and A is the diode ideality factor.
These five parameters are determined by solving the transcen-
dental equation (Eq. (1)) using the NewtoneRaphson algorithm
based only on the datasheet of the available parameters [22]. The
algorithm has been developed in MatLab by using the function
“fsolve” available in the Optimization ToolboxÔ. The one-diode
model given in Eq. (1), has been tested using the data collected
during one day by the monitoring system of the GCPV plant
described in Section 2.
Fig. 2a shows a comparison between the measured and the
calculated DC power for February 2, 2009 day. A good agreement
between the two signals has been obtained and the correlation
coefficient is equal to 98%.
3.1.2. AC output model
The efficiency of an inverter converting the DC current and
voltage into the AC current and voltage has been modelled with the
standard approach given by the following equation. The model
describes the efficiency as a function of the input power (Pin) [24].
hDC=AC ¼ y0 þc1 ð1expðc2 Pin ÞÞþc3 ð1expðc4 Pin ÞÞ
where y0, c1, c2, c3, and c4, are five parameters that can be deter-
mined using a non-linear regression algorithm, such as the least
square algorithm, minimizing the following quadratic function:
X
N
(2) SðwÞ ¼ ðPin  hðPin ; wÞÞ2 (4)
i¼1
where: w ¼ (y0, c1, c2, c3, c4) is the vector of the five parameters to be
estimated.
Table 3 reports the values of the estimated five parameters,
using the input power Pin of the PV generator described above.
The model has been verified for several samples; as an
example, Fig. 2b shows a comparison between the AC simu-
lated and measured power. The simulated AC power is very
close to the measured one and the correlation coefficient in
this case is 99%.
3.2. Power losses in DC and AC side of the PV system
3.2.1. Power losses in the DC side (capture losses)
The overall losses occurring in the DC side of the GCPV system
depend on a number of factors. Some of these are related to the PV
module and others to the climatic conditions. The power losses in
the DC side of the photovoltaic plant are defined as “capture losses
(Lc)”. These are related to the operating cell’s temperature, solar
irradiance level, PV shading, and to the sunlight Angle Of Inci-
dence (AOI) [20]. The capture losses are only a part of the DC
losses; other DC losses are related to the efficiency of the
(3) Maximum Power Point Tracker (MPPT) technique employed, to
the mismatch, wiring losses, soiling effect, etc. Nevertheless, the
capture losses are a good indicator of failures occurring on the DC
side of the photovoltaic systems and they are given by the
following equation [25]:
 W. Chine et al. / Renewable Energy 66 (2014) 99e110
Fig. 7. a. PV array output power, measured and simulated results. b. Inverter output
power, measured and simulated results.
Lc ¼ Yr  Ya
where: Yr (h): is the reference yield, and is defined as the total in-
plane solar irradiance Hi (Wh/m2) divided by the array reference
irradiance G0 (1000 W/m2), given as [25]:
Hi
Yr ¼
G0
Ya (h): is the array yield defined as the energy EDC (Wh) pro-
duced by the PV array over a defined period (for example one day,
one month or one year) divided by its nominal power P0 (W). The
array yield is given by Kymakis [25]:
E
Ya ¼ DC
P0
An important part of these losses is associated with the
mismatch effect representing total energy losses between 5% and
11% [26,27].
3.2.2. Power losses in the AC side (system losses)
The losses occurring in the AC side of a PV plant mainly depend
on the inverter, the transformer and on the AC circuits [25]. The
system losses (Ls) are a good indicator of excessive losses and fail-
ures occurring on the AC side of photovoltaic system [27] and can
be written as follows:
105
LS ¼ Ya  Yf (8)
where Yf (h) is the final yield, defined as the annual, monthly or
daily AC energy EAC (kWh) output of the system divided by the
nominal power P0 (kW) of the PV array. It is given by Kymakis [25]:
EAC
Yf ¼ (9)
p0
Another parameter that can give a global idea of the system
behaviour is the performance ratio. This indicates the overall losses
occurring in the PV system and is defined as:
Yf
PR ¼ (10)
Yr
3.3. The diagnostic procedure
The main objective of the fault detection system is to detect and
determine when and where a fault has occurred in the considered
GCPV plant. The system needs a number of sensors to measure the
climatic variables (solar irradiance and module’s temperature) and
electrical variables (DC current and voltage, DC and AC power output).
The measured module’s temperature and solar irradiance are used as
an input to the model of the PV modules so that the PV array output
power is calculated in real time. In a second step, the inverter effi-
ciency model is used in order to calculate the simulated AC power.
The proposed diagnostic procedure is based on the analysis of
the absolute performance ratio error (APRE), which represents the
absolute value of the difference between the measured and the
simulated performance ratio. The accuracy of the developed pro-
cedure depends mainly to the accuracy of the PV and inverter
models employed as well as the accuracy of the sensors and
equipment used. In theory, the APRE should be zero when the
system is in normal operating condition, and should be different
from zero in case of a fault occurrence. In practice the APRE is close
to zero (and not exactly zero) when there’s no fault, to account for
the modelling uncertainty, measurement noise, and the inherent
power losses on the system (such as temperature losses), for this
(5) reason, the APRE has been compared with a threshold.
Several experiments show that the mean daily value of the APRE
is always less than 0.05, hence we can set the threshold equal to a
value higher than 0.05, e.g. 0.07. This means that the signal fault
(diagnostic signal ¼ 1) is given when:
(6) jPRmes  PRsim j > 0:07 (11)
where PRmes and PRsim are the measured and the simulated per-
formance ratios respectively. The fault-diagnosis procedure is
illustrated in the flowchart presented in Fig. 3a.
3.4. Failure identification
(7) In order to determine the kind of fault and where the fault oc-
curs, the two indicators proposed in Ref. [20] have been used. These
are the DC current (Rc) and voltage (Rv) ratios given by the following
expressions:
Ipv_sim
Rc ¼ (12)
Ipv_meas
Vpv_sim
Rv ¼ (13)
Vpv_meas
where: Ipv_sim is the simulated array’s DC current, Ipv_meas is
the measured array’s DC current, Vpv_sim is the simulated
106 W. Chine et al. / Renewable Energy 66 (2014) 99e110
Fig. 8. a. Absolute performance ratio error and diagnostic signal evolution (faulty modules). b. Faulty in PV modules observed in the designed tool. c. DC current, DC voltage, and DC/
AC power ratios evolution.
array’s DC voltage, and Vpv_meas is the measured array’s DC
voltage.
The DCeAC output power ratio (RPDC_AC) given by the following
equation is introduced:
 
PDC_sim
P simulated ones, given by Eqs (15) and (16), allows the identification
RPDC_AC ¼  P 
DC_meas
(14)
AC_sim
PAC_meas
where PDC_sim is the simulated DC power, PAC_sim is the simulated AC
power, PDC_meas is the measured DC power and PAC_meas is the
measured AC power.
The analysis of the current, voltage, and power ratio allows the
identification of the type of faults. The failure identification method
is as follows:
- If RPDC_AC < 1: it means that the measured AC power is decreased
which indicates that there were possibilities of the inverter
stopping (or malfunctioning), or a wire between the inverter
and the grid line was broken.
- If RPDC_AC > 1: it means that the measured DC power is
decreased which indicates that a fault was occurred on the DC
side of PV system, in this case we can identify four categories of
faults:
a) If Rc > 1 and Rv < 1: It indicates that there is possibility of
string disconnection (or breakdown in string itself).
b) If Rc < 1 and Rv > 1: It indicates that there is possibility of
disconnection of modules in a string (or breakdown in
modules).
c) If Rc > 1 and Rv > 1: It means that a fault was occurred on the
PV array as: partial shadow, ageing, MPPT error.
d) If Rc < 1 and Rv < 1: it means that there are no losses on the
current and voltage; in this case the diagnostic signal state is
considered as a false alarm.
The calculation of the ratio between the measured DC current of
each string and the AC power of each inverter with respect the
of the exact location of the fault string and the fault inverter.
Ipv_measi
Rci ¼ (15)
Ipv_simi
Pac_measj
Rpj ¼ (16)
Pac_simj
where: i and j are the number of strings and inverters in the PV
plant respectively.
The fault identification and location procedure are illustrated by
the flowchart presented in Fig. 3b.
3.5. The automatic fault detection tool
In order to simplify the diagnostic procedure, this is imple-
mented graphically under MatLab/GUI. The graphical user interface
developed can monitor instantaneously the PV plant. Its outputs
are the state of the diagnostic signal, the values of the APRE and the
three indicators DC current, DC voltage, and DCeAC power ratios, as
well as the type and the location of fault. The inputs are the
measured climatic and electrical data which are loaded from the
database, as well as the calculated data.
 W. Chine et al. / Renewable Energy 66 (2014) 99e110 107

a 1.5

a 18000 APRE
Threshold

APRE, threshold and diagnostic signal evolution

16000 Diagnostic signal

1
14000

12000
Output DC power [W]

0.5
10000

8000

6000 0

4000

2000 Simulated output DC power

-0.5
Measured output DC power 100 200 300 400 500 600 700 800
0 Time [min]
0 100 200 300 400 500 600 700 800 900
Time [min] b 2
DC current ratio
b 14000

DC current, DC voltage, and DC AC power ratios

DC voltage ratio
DC AC Power ratio
12000

1.5
10000
Output AC power [W]

8000

6000 1

4000

2000
Simulated output AC power 0.5
100 200 300 400 500 600 700 800
Measured output AC power Time (min)
0
0 100 200 300 400 500 600 700 800 900
Time [min] c
c

Fig. 9. a. PV array output power, measured and simulated results. b. Inverter output
power, measured and simulated results. c. Faulty observed in the designed tool. Fig. 10. a. Absolute performance ratio error and diagnostic signal evolution (faulty
strings). b. Current, voltage, and power ratios evolution. c. DC current ratios of each
string.
108 W. Chine et al. / Renewable Energy 66 (2014) 99e110

a 15000 Simulated output AC power

a 2
Measured output AC power DC current ratio

DC current, DC voltage, and DC AC power ratios

1.8 DC voltage ratio
DC AC Power ratio

10000 1.6
Output AC power [W]

1.4

1.2
5000

0.8
0
100 200 300 400 500 600 700 800
Time [min] 0.6
b 1.5
APRE 0.4
Threshold
100 200 300 400 500 600 700 800
APRE, threshold and diagnostic signal evolution

Diagnostic signal Time (min)

1 b 2
inverter 1
inverter 2
inverter 3
1.5
0.5
AC power ratios

1
0

0.5

-0.5
100 200 300 400 500 600 700 800
Time [min]
0
Fig. 11. a. Inverter output power, measured and simulated results. b. Absolute per-
formance ratio error and diagnostic signal evolution (faulty inverter).

-0.5
4. Experimental validation of the developed system 100 200 300 400 500 600 700 800
Time [min]
In this section, the performance of the developed diagnostic c
method is verified. For this purpose, the acquired data of three
different days (e.g. February 2, 2009, May18, 2009, and May 23, 2009)
have been considered. Four case studies have been examined and
presented. Fig. 4 shows the different faults analysed in this study. The
developed software can automatically display and save the state
(normal/fault) of the GCPV system every 10 min, by making a com-
parison between monitored and calculated electrical current, voltage,
power and the different performance ratios. Monitored and calcu-
lated parameters are also displayed by the software. Moreover, at the
end of the day the user could print and check the system evolution.

4.1. Case 1: no fault

The simulated DC and AC power is closed to the measured one,

as shown in Fig. 5a and b. The APRE curve is lower than the pre- Fig. 12. a. Current, voltage, and power ratios evolution. b. AC power ratio for each
defined threshold as shown in Fig. 6a. According to this analysis; inverted. c. Faulty inverter observed.
the fault diagnosis system provides a diagnostic signal in low state,
indicating that no faults are detected in both DC and AC side of the possibility of breakdown or malfunctioning of modules, strings, or
GCPV plant (see Fig. 6a) except one false alarm observed in the inverters. In the case of no fault, the designed tool reported in
figure. More details about this fault are clearly observed in the Fig. 6d, indicates clearly that the diagnostic signal is set to 0 and the
designed tool shown in Fig. 6b, although the diagnostic signal is set rest of information shown in the tool are coherent.
to 1 the rest information current, voltage and power indicate that
there is no error. Furthermore, to verify this graphically, the three 4.2. Case 2: faulty modules
indicators DC current, DC voltage, and DCeAC power ratios shown
in Fig. 6c confirm the state of the diagnostic signal as they have In order to test the ability of the method to detect the problem of
small variation about the unity, and this means that there is no faulty modules, five modules in one string have been disconnected.
 W. Chine et al. / Renewable Energy 66 (2014) 99e110
As a consequence, the voltage of the string falls to a wrong value
and this fact impacts both the DC and the AC power. In this case, the
diagnosed result is as follows:
In Fig. 7a and b, the unexpected DC and AC power is shown with
respect to the simulated ones. Absolute performance ratio error and
diagnostic signal evolution are depicted in Fig. 8a, as a conse-
quence, the APRE exceeds the assigned threshold, and the diag-
nostic signal is one. More details are reported in the designed tool,
Fig. 8b. In addition, Fig. 8c shows the evolution of the current,
voltage, and power ratios of the GCPV plant. Analysing the plots
depicted in both Fig. 8b and c, some information regarding the type
of faults can be obtained; the measured output voltage reduction
increases the DC voltage ratio, while the DCeAC power and DC
current ratios remain unchanged. This fact is related to the
disconnection of modules in a string (or breakdown in modules).
4.3. Case 3: faulty strings
In this case, the disconnection of four strings connected to the
first inverter has been simulated. The result is a reduction in the
current and power of the GCPV plant as can be seen in Fig. 9a and b.
More information about this fault are reported in the designed tool,
see Fig. 9c, where it is clearly shown that the measured current and
power are lower than those simulated.
As shown in Fig. 10a, the APRE exceeds the threshold so that the
diagnostic signal goes one and the fault is detected. Fig. 10b depicts
the current, voltage and power ratios evolution of the system, the
fault type can be detected by analysing both Figs. 10b and 9c. It can
be observed, that the measured current reduction increases the DC
current ratio, while the DC voltage and the DCeAC power ratios
remain unchanged; this means that there was a possibility of
strings disconnection (or breakdown in the same string).
The location of the four strings disconnected can be determined
by analysing the ratio of the measured output current of each string
with respect to the simulated one, as shown in Fig. 10c.
4.4. Case 4: faulty inverter
To confirm whether a failure which happens in the inverter
could be detected by this method, an inverter disconnection fault is
generated on the GCPV plant. Fig. 11a shows the reduction at the
measured AC power compared with the simulated one. As a result,
the APRE level given in Fig. 11b is higher than the expected
threshold, and the diagnostic signal in this case equals to one.
To isolate the type of fault, we analyse the DCeAC power ratio
shown in Fig. 12a, which has a value lower than one, and in addition
the value of the DC current and voltage ratios is near to one. This
means that there are no losses on the DC output power of the array,
because each inverter is endowed with four Maximum Power Point
Tracking (MPPT) systems. This information confirms that there is
no possibility that a wire between modules/strings and inverter
was broken or a breakdown occurs in strings and/or modules.
Therefore, inverter stopping or (malfunctioning) is the most likely
source of power loss.
We can also identify exactly the location of the failed inverter by
analysing the measured AC output power of each inverter with
respect to the simulated one as given in Fig. 12b. More details about
this fault are reported in the designed tool presented in Fig. 12c.
5. Conclusions
A fault detection method for a grid-connected photovoltaic
(GCPV) plant based on the comparison of the APRE with a threshold
is proposed and verified experimentally. The method generates a
diagnostic signal as an indicator of normality/abnormality operation
109
in the GCPV plant. In order to detect the failure occurred in the
inverter, we have defined an indicator named DCeAC power ratio,
for identifying clearly the precise location of faulty string and faulty
inverter, by calculating the DC current and AC power ratios of each
string and each inverter respectively. This procedure has been tested
using monitored data of a 20 kWp GCPV plant. Four case studies have
been presented: the normal operation with false signal, faulty
modules, faulty strings and faulty inverter operation. Results
confirm the feasibility of the method for fault diagnosis of GCPV
plants with good accuracy.
An automatic system is developed graphically under MatLab/
GUI to display and monitor instantaneously the PV plant, and
therefore, the designed software has the potential for becoming a
useful, widespread tool for GCPV design engineers, O&M personnel,
as well as scientists.
In the future, we aim to implement experimentally the devel-
oped automatic procedure into low cost microcontroller, with LCD
display and flash alarm to notify users about the state of their
systems in real time.
Acknowledgement
Authors wish to thank the municipality of Trieste and Mr. F.
Romeo for providing the experimental data of climate and electrical
signals. Particular thanks also to Mr. L. Farneti for his efforts in
making possible the remote access of the electrical and climate data
at the University of Trieste. Also, the second author would like to
thank the International Centre for Theoretical Physics (ICTP),
Trieste (Italy) for providing the means and the computers facilities
for performing the present work.
References
[1] Pacas JM, Molina MG, dos Santos Jr EC. Design of a robust and efficient power
electronic interface for the grid integration of solar photovoltaic generation
systems. Int J Hydrogen Energy 2012;37:10076e82.
[2] Almonacid F, Rus C, Pérez-Higueras P, Hontoria L. Calculation of the energy
provided by a PV generator. Comparative study: conventional methods vs.
artificial neural networks. Energy 2011;36:375e84.
[3] Nordmann T, Jahn U, Nasse W. Performance of PV systems under real con-
ditions. In: European workshop on life cycle analysis and recycling of solar
modules, the waste challenge, Brussels 2004.
[4] Takashima T, Yamaguchi J, Otani K, Oozeki T, Kato K, Ishida M. Experimental
studies of fault location in PV module strings. Sol Energy Mater Sol Cells
2009;93:1079e82.
[5] Schirone L, Schirone L, Califano FP, Pastena M. Fault detection in a photo-
voltaic plant by time domain reflectometry. Prog Photovolt Res Appl 1994;2:
35e44.
[6] Vergura S, Acciani G, Amoruso V, Patrono G. Inferential statistics for moni-
toring and fault forecasting of PV plants. In: Proceedings of the IEEE inter-
national symposium, industrial electronics, July 2008, Cambridge, UK 2008.
pp. 2414e9.
[7] Drews A, de Keizer AC, Beyer HG, Lorenz E, Betcke J, van Sark WGJHM, et al.
Monitoring and remote failure detection of grid-connected PV systems based
on satellite observations. Sol Energy 2007;81:548e64.
[8] Muselli M, Notton G, Canaletti JL, Louche A. Utilization of Meteosat satellite
derived radiation data for integration of autonomous photovoltaic solar en-
ergy systems in remote areas. Energy Convers Manag 1998;39:1e19.
[9] Drif M, Mellit A, Aguilera J, Pérez PJ. A comprehensive method for estimating
energy losses due to shading of GC-BIPV systems using monitoring data. Sol
Energy 2012;86:2397e404.
[10] Syafaruddin S, Karatepe E, Hiyama T. Controlling of artificial neural network
for fault diagnosis of photovoltaic array. In: Proceedings of the 16th inter-
national conference on intelligent system application to power systems
(ISAP), September 2011, Greece. p. 1e6.
[11] Ducange P, Fazzolari M, Lazzerini B, Marcelloni F. An intelligent system for
detecting faults in photovoltaic fields. In: Proceedings of the 11th interna-
tional conference on intelligent systems design and applications IEEE,
November 2011, Cordoba, Spain. p. 1341e6.
[12] Yagi Y, Kishi H, Hagihara R, Tanaka T, Kozuma S, Ishida T, et al. Diagnostic
technology and an expert system for photovoltaic systems using the learning
method. Sol Energy Mater Sol Cells 2003;75:655e63.
[13] Firth SK, Lomas KJ, Rees SJ. A simple model of PV system performance and its
use in fault detection. Sol Energy 2010;84:624e35.
110 W. Chine et al. / Renewable Energy 66 (2014) 99e110
[14] Gokmen N, Karatepe E, Celik B, Silvestre S. Simple diagnostic approach for
determining of faulted PV modules in string based PV arrays. Sol Energy
2012;86:3364e77.
[15] Zhao Y, Lehman B, DePalma JF, Mosesian J, Lyons R. Challenges to overcurrent
protection devices under line-line faults in solar photovoltaic arrays. In: IEEE
energy conversion congress and exposition (ECCE), Sept. 2011, Phoenix, Ari-
zona 2011. pp. 20e7.
[16] Chao KH, Ho SH, Wang MH. Modelling and fault diagnosis of a photovoltaic
system. Electr Power Syst Res 2008;78:97e105.
[17] Miwa M, Yamanaka S, Kawamura H, Ohno H, Kawamura H. Diagnosis of a
power output lowering of PV array with a (-dI/dV)-V characteristic. In: Pro-
ceeding of IEEE 4th world conference on photovoltaic energy conversion
2006. pp. 2442e5.
[18] Zhao Y, Yang L, Lehman B, DePalma JF, Mosesian J, Lyons R. Decision-based
fault detection and classification in solar photovoltaic arrays. In: Twenty-
seventh annual IEEE applied power electronics conference and exposition,
5-9 Feb. 2012. p. 93e9.
[19] Il-Song K. Fault detection algorithm of the photovoltaic system using wavelet
transform. In: Proceedings of the IEEE India international conference on po-
wer electronics, January 2010, New Delhi. p. 1e6.
[20] Chouder A, Silvestre S. Automatic supervision and fault detection of PV systems
based on power losses analysis. Energy Convers Manag 2010;51:1929e37.
[21] Mellit A, Pavan AM. Performance prediction of 20 kWp grid connected
photovoltaic plant at Trieste (Italy) using artificial neural network. Energy
Convers Manag 2010;51:2431e41.
[22] Sera D, Teodorescu R, Rodriguez P. PV panel model based on datasheet values.
In: Proceedings of the IEEE international symposium on industrial electronics
2007. pp. 2392e6.
[23] Stein JS, Cameron CP, Bourne B, Kimber A, Posbic J, Jester T. A standardized
approach to PV system performance model validation. In: Photovoltaic spe-
cialists conference (PVSC), 2010 35th IEEE 2010. pp. 1079e84.
[24] So JH, Jung YS, Yu GJ, Choi JY, Choi JH. Performance results and analysis of
3 kW grid-connected PV systems. Renew Energy 2007;32:1858e72.
[25] Kymakis E, Kalykakis S, Papazoglou TM. Performance analysis of a grid connected
photovoltaic park on the island of Crete. Energy Convers Manag 2009;50:433e8.
[26] Chouder A, Silvestre S. Analysis model of mismatch power losses in PV Sys-
tems. J Sol Energy Eng 2009;131:024504e9.
[27] Spertino F, Akilimali JS. Are manufacturing IeV mismatch and reverse cur-
rents key factors in large photovoltaic arrays? IEEE Trans Ind Electron
2009;56:4520e31.