Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Available online at www.sciencedirect.

com

Solar Energy 94 (2013) 119–127


www.elsevier.com/locate/solener

Automatic fault detection in grid connected PV systems


Santiago Silvestre a,⇑, Aissa Chouder b, Engin Karatepe c
a
MNT Group, Electronic Engineering Department, Universitat Politècnica de Catalunya (UPC) BarcelonaTech, C/Jordi Girona 1-3,
Campus Nord UPC, 08034 Barcelona, Spain
b
Photovoltaic Laboratory, Development Centre of Renewable Energies, BP 62 Route de l’Observatoire, 16340 Bouzareh, Algiers, Algeria
c
Department of Electrical and Electronics Engineering, Ege University, 35100 Bornova, Izmir, Turkey

Received 8 January 2013; received in revised form 2 May 2013; accepted 3 May 2013
Available online 2 June 2013

Communicated by: Associate Editor Nicola Romeo

Abstract

This paper presents a detailed procedure for automatic supervision, fault detection, and diagnosis of possible failure sources leading to
total or partial loss of productivity in grid connected PV systems. The diagnostic procedure is part of the monitoring system allowing, at
the same time, modeling and simulation of the whole system and variables measurements in real time. The fault detection algorithm is
based on the comparison of simulated and measured yields by analyzing the losses present in the system while the identification of the
kind of fault is carried out by analyzing and comparing the amount of errors deviation of both DC current and voltage with respect to a
set of errors thresholds evaluated on the basis of free fault system. The proposed method has been validated in with experimental data in
a grid connected PV system in the Centre de Developpement des Energies Renouvelables (CDER) in Algeria.
Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: PV systems; Fault detection; Diagnostic

1. Introduction ignored by classical overcurrent protection devices which


can lead to the damage of PV modules or to a rise of safety
The cumulative global photovoltaic (PV) capacity has hazards.
been growing exponentially latest years around the world, Therefore, active monitoring and fault detection and
especially due to the installation of grid connected PV sys- diagnosis can provide automatic control of PV systems,
tems (Dincßer, 2011; EPIA, 2012). This growth indicates including the possibility of teleoperation to analyze result-
that PV energy production will have a very important role ing information remotely through internet, resulting in
in the total generation of electricity of the future. But still improved PV system performance and reliability, by opti-
important efforts remain to be done in terms of cost, per- mizing the output of the system to achieve higher energy
formance, and reliability of PV systems. yields. In this work, we present a complete package,
Often monitoring systems are built into inverter and are designed in LabVIEW software, allowing at the same time
mainly designed to connect and disconnect from the utility PV module parameters extraction, dynamic system simula-
during low or high voltage events, prevent islanding and tion and modeling, monitoring of electrical and weather
report on PV status (e.g., current, voltage, power). How- variables, and finally fault detection and diagnosis.
ever, hidden defects or serious output power reduction Most fault detection algorithms reported in the litera-
due to some defects in PV modules can, in most cases, be ture follow the idea of comparing monitored data from
the PV system with model prediction results to identify
⇑ Corresponding author. faults when significant differences are observed between
E-mail address: santiago.silvestre@upc.edu (S. Silvestre). the two sets of data (Yagi et al., 2003; Chao et al., 2008;

0038-092X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.solener.2013.05.001
120 S. Silvestre et al. / Solar Energy 94 (2013) 119–127

Chouder and Silvestre, 2010; Gokmen et al., 2012). This is Measurement of DC voltage and AC voltage is performed
the approach we have used in this work. by a resistive voltage divider and AC transformer in order
Some experiences about remote monitoring and fault to adapt voltage levels to the input of data acquisition
detection in PV systems have been reported in the last years respectively. While the output PV plant, DC current and
(Muselli et al., 1998; Stettler et al., 2006; Drews et al., the output inverter AC current are measured and amplified
2007). These forecasts are based primarily on meteorolog- using hall effect transducers. All the dynamic variables are
ical and satellite observation data or other meteorological gathered in the Agilent 34970A data acquisition. The com-
data bases (Quesada et al., 2011) to evaluate the system munication with a personal computer is achieved by a
output. However, for an accurate evaluation of PV system GPIB bus. In Table 1, it is given the detail of measured
yields in all weather conditions, it is necessary to work with variables as well as information on the precision of the
actual monitored climate data. Moreover, a detailed simu- measurement instruments used in the monitoring process.
lation of the PV system behavior must be performed using The monitoring system is constructed around several
accurate simulation models, experimentally validated, to virtual instruments (VIs), developed in LabVIEW environ-
obtain simulation results with a high degree of accuracy. ment, allowing hardware configuration, data exchanges,
The procedure for diagnostic of grid connected PV sys- parameters extraction of a single PV module and dynamic
tems reported in this work is based in a previous work model simulation of the whole PV system. Furthermore,
about monitoring, modeling, and simulation of PV systems the program handles both measured and simulated data
in real time (Chouder et al., 2013) and on the analysis of in order to offer a convivial environment by displaying all
power losses present in the PV system (Chouder and Silves- variables of concern, plotting in real time measurements
tre, 2010). This diagnostic procedure integrates monitoring, against simulated variable, processing errors, and trigger-
modeling and simulation, and fault diagnosis in a complete ing fault finding process. Detailed reports are generated
package in the same environment and has been validated by creating XLS and HTML files which summarize the
through field testing of a grid connected PV system in Alge- behavior of the system (see Fig. 2).
ria. This diagnostic procedure is able to detect and identify The simulation of the PV system is based on the models
the most likely causes of the major failures of PV systems in developed in previous works that use the five parameters
real time. model for the PV module (Chouder et al., 2013; Chouder
and Silvestre, 2012) and the performance model inverter
2. System description presented by King et al. (2007). The simulation package
developed also in LabVIEW environment has been previ-
The PV system where tests have been carried out is a ously validated experimentally and described by Chouder
9.6 kWp system installed at the roof top of the administra- et al. (2013) and the simulation results include the follow-
tive building of the Centre de Devéloppemnt des Energies ing data: The I(V) and P(V) characteristics as well as the
Renouvelables, Algeria (CDER). coordinates of the maximum power point (MPP) of both,
The PV system includes a total of 90 PV modules (Isofo- solar cell and PV module, DC voltage and current at the
ton 106W-12) and is divided into three arrays linked to the output of the PV array, reference yield (Yr), array yield
main grid via three single phase inverters each one with a (Ya) and final yield (Yf), as well as the system performance
nominal power of 2.5 kW (IG30 Fronius) connected to ratio (PR). Fig. 2 shows the schematic diagram corre-
the national distribution grid. sponding to the monitoring and simulation processes that
allow the extraction of both sets of main parameters of
the PV system and simulated and monitored parameters.
2.1. Monitoring and simulation system description
Finally, Fig. 3 shows the monitored values and simula-
tion results obtained for Ya of the two first weeks of June
Different sensors are setup to measure climate variables:
2012.
Irradiances, in the horizontal plane (GH,p) and in PV mod-
ules plane (GI,p and GI,c), the ambient temperature (Ta), as
3. Proposed method for fault detection and diagnosis
well as electrical variables at the DC and AC side of the PV
plant. Two pyranometers and a reference cell are used to
When PV plants are under operation, they are subject to
monitor incident irradiance in both horizontal and tilted
several inherent power losses. The overall power losses that
plane. Fig. 1 shows the monitoring system and sensors
can appear on the DC and AC outputs of a PV system are
included. The temperature measurement is performed by
enumerated below:
a K type thermocouple where only Ta is considered. Eq.
(1) is used to obtain the cell temperature (Tc) from the irra-
 The incoming radiation energy is lowered by various
diance (G) and the Normal Operating cell Temperature
factors such as the angle of incidence (AOI), dusty mod-
(NOCT) (Alonso Garcıa and Balenzategui, 2004; Malik
ules, and shading objects.
and Damit, 2003).
 At module level, operation above reference temperature
G reduces the efficiency and therefore the output power of
T c ¼ T a þ ðNOCT  20  CÞ ð1Þ
800 the individual PV module.
S. Silvestre et al. / Solar Energy 94 (2013) 119–127 121

Fig. 1. Synoptic of the grid connected system and monitoring system installed at CDER.

Table 1  At PV generator level, mismatch and non-uniform irra-


Measured variables and precision of the instruments used in the diance and temperature distribution over the PV gener-
monitoring process. ator are also sources of power reduction.
Measured Sensors Accuracy  Ohmic losses, maximum power point tracking (MPPT),
variables and DC–AC conversion efficiencies are also cause of
GI,c Reference cell isc = 3.42 A for ±70 mW m2 additional power losses.
1000 W m2
GI,p GH,p Pyranometer CM 11 sensitivity: 5 lV/ ±0.73 W m2
Despite main power reduction causes are known, it is
(W m2)
Ta k type thermocouple ±1 °C very hard to quantify separately the amount of power
VDC Resistive divider ±25 mV reduction; due to an individual effect, more additional data
VAC AC adapter ±0.64 V and specific test bench are required (Firth et al., 2010).
IDC Hall effect ±0.25 mA However, it is established that the overall losses could be
IAC Hall effect ±53 mA
gathered on few power losses indicators as mentioned by

Fig. 2. Schematic diagram of the data acquisition system.


122 S. Silvestre et al. / Solar Energy 94 (2013) 119–127

losses by obtaining the deviations between them. An error


parameter, ELc, can be established for the capture losses as
follows:
ELc ¼ jLc meas  Lc sim j ð5Þ
where the subscripts sim and meas indicate simulated and
measured Lc values, respectively.

3.1. Set of the theoretical thresholds for fault detection

The detection of a failure in the PV system operation is


based on the continuous check of the error parameters.
In order to avoid an excessive number of false fault
detections, it is necessary to establish a specific threshold
for ELc to indicate proper operation of the system. For this
purpose, we established thresholds for this parameter. If
Fig. 3. Simulated versus measured array yield in June days. the value does not exceed defined thresholds, then it is con-
sidered that the PV system is working fault free, otherwise
the PV system is considered in faulty operation. These ref-
the international energy Photovoltaic Power Systems (IEA)
erence thresholds have been established after statistical
and described in the standard 61724 of the International
analysis of the evolution of ELc when the PV system is
Electrotechnical Commission (IEC) (1998). At the DC part
working free of faults. Then, the mean value and the stan-
of the PV power plant, the overall power losses are defined
dard deviation are derived for both capture losses error
by the normalized total capture losses Lc. By knowing per-
(ELc), DC current error (Ei) and DC voltage error (Ev),
formance parameters of the PV plant, one can calculate the
respectively, as shown in Table 2.
capture losses by the following expression (Kymakis et al.,
We have identified by trial and error that the PV system
2009):
works in fault free operation when ELc_ref is in between of
Hi Edc the following thresholds:
Lc ¼ Y r ðG; T C Þ  Y a ðG; T C Þ ¼ ðG; T c Þ  ð2Þ
Gref P ref
ELc ref  2rðELc ref Þ 6 ELc 6 ELc ref þ 2rðELc ref Þ ð6Þ
where Yr(G, Tc) and Ya(G, Tc) are the reference and array
This learning procedure can be applied to any grid con-
yields, respectively, at real working irradiance, G, and real
nected PV system in order to obtain its particular thresh-
cell temperature Tc, Hi is the total irradiation in array plane,
olds for a correct fault detection evaluation. Fig. 4 shows
Gref is the reference irradiance at STC (1000 W m2), Edc is
the flow chart of the following procedure to detect mal-
the energy produced by PV array, and Pref is the maximum
function in operational PV system.
power output of PV array.
Moreover, these global losses can be divided into two
types of losses; thermal and miscellaneous capture losses 3.2. Set up of fault diagnosis procedure
denoted as Lct and Lcm, respectively. Thermal losses are
due to the decrease in DC output power when the PV mod- When the ELc parameter exceeds the limit set, indicating
ules are working at temperatures higher than reference the presence of a fault in the PV system is necessary to
temperature (25 °C). Miscellaneous capture losses are determine the most probable cause of this fault. In order
mainly due to shading, ohmic losses, low irradiance losses, to isolate the fault detection and failure type, we define
angle of incidence losses, module failure, string failure, two indicators of the deviation of the measured DC vari-
MPP tracker failure etc. Equations allowing finding ther- ables with respect to the simulated ones. These indicators
mal and miscellaneous capture losses are given by (Haeber- are the current error, Ei, and the voltage error, Ev, given
lin and Beutler, 1995): by the following expressions:

Lct ¼ Y a ðG; 25 CÞ  Y a ðG; T c Þ ð3Þ Ei ¼ jI dc meas  I dc sim j ð7Þ

Lcm ¼ Lc  Lct ð4Þ Ev ¼ jV dc meas  V dc sim j ð8Þ

From the measured weather parameters and electrical vari-


ables, the developed software calculates the instantaneous Table 2
Mean and standard deviation for reference errors.
capture losses, Lc, and then derives thermal and miscella-
neous capture losses. On the other hand, the simulated Standard deviation r Mean value
4
losses are also evaluated using the simulation model from ELc_ref (Wh/Wp/day) 1.55  10 1.8  104
the measured weather variables, G and Tc. This procedure Ei_ref (mA) 108 136
Ev_ref (V) 4.30 4.65
allows the comparison between simulated and measured
S. Silvestre et al. / Solar Energy 94 (2013) 119–127 123

A set of boundaries have been defined for these error


parameters following the same procedure used above for
the ELc parameter. So we have obtained values for the
standard deviation, r, of both errors: Ei_ref and Ev_ref.
The values obtained in fault free operation are shown in
Table 2. The diagnostic algorithm supervises the evolution
of the voltage and current errors and evaluates whether
they are not exceeding the established thresholds given by:
Ei ref  2rðEiÞ 6 Ei 6 Ei ref þ 2rðEiÞ ð9Þ

Ev ref  2rðEvÞ 6 Ev 6 Ev ref þ 2rðEvÞ ð10Þ

By managing the actual error for voltage and current is


possible to establish the most probable fault present in the
system when the ELc parameter indicates this situation.
The flowchart given in Fig. 5 shows the diagnostic proce-
dure used to find out the most likely faults which have
caused excessive losses in the photovoltaic power genera-
tion system.

4. Results and discussion


Fig. 4. Flowchart of fault detection procedure.
The diagnostic procedure has been experimentally vali-
dated in a grid connected PV system of 9.6 kWp sited at
where Idc_sim and Idc_meas are the simulated and measured the CDER in Algeria that has been previously described
DC output current, respectively, and Vdc_sim and Vdc_meas in Section 2. The following types of faults have been iden-
are the simulated and measured output DC voltages. tified in the PV system: Inverter disconnection, partial

Fig. 5. Flowchart of the fault diagnosis procedure.


124 S. Silvestre et al. / Solar Energy 94 (2013) 119–127

Fig. 6. Simulated array yield versus measured array yield in case of


normal operation.

Fig. 9. DC voltage error versus reference DC voltage error thresholds in


case of normal operation.

Fig. 7. Capture losses error versus reference capture losses thresholds in


case of normal operation. Fig. 10. Simulated versus measured array yields in case of grid failure.

Fig. 8. DC Current error versus reference current error thresholds in case Fig. 11. Capture losses error versus reference capture losses thresholds in
of normal operation. case of grid failure.
S. Silvestre et al. / Solar Energy 94 (2013) 119–127 125

Fig. 15. Capture losses error versus reference capture losses thresholds in
Fig. 12. DC current error versus reference current errors thresholds in case of shading.
case of grid failure.

Fig. 16. Current error versus reference current errors thresholds in case of
shading.
Fig. 13. DC voltage error versus reference voltage errors thresholds in
case of grid failure.

Fig. 17. DC voltage error versus reference DC voltage errors thresholds in


Fig. 14. Simulated versus measured array yield in case of shading. case of shading.
126 S. Silvestre et al. / Solar Energy 94 (2013) 119–127

Fig. 18. Simulated and measured array yields in case of string failure. Fig. 21. DC voltage error vs reference DC voltage errors thresholds in
case of string failure.

shadowing operation, and disconnection of a string of the


array.
Fig. 6 shows the array yields, Ya, measured and
obtained from simulation when the PV system is working
in fault free operation mode. In this situation, the parame-
ter ELc remains within the established thresholds at all
times as can be seen in Fig. 7. The parameters Ei and Ev
are also within their established boundaries, as Figs. 8
and 9 show, so no alert signals are generated.
When disturbances in the grid induce inverter discon-
nections, deviations between simulated and monitored val-
ues of Ya are clearly observed as shown in Fig. 10. Then,
the parameter ELc is expected to exceed the reference
thresholds limits set indicating a presence of failure or
excessive power losses of the system operation as can be
Fig. 19. Capture losses error versus reference capture losses errors seen in Fig. 11.
thresholds in case of string failure. In this case, three signals of alert are generated; two of
them correspond with the deviations in Ev and Ei going
out of their respective boundaries, and a third one indicat-
ing no AC signal at the inverter output. So, a grid failure is
then identified. This effect is clearly shown in Figs. 12 and
13.
Fig. 14 shows the evolution of Ya, measured and simu-
lated, in case of partial shadowing of the PV array. Again,
the evolution of ELc indicates operation faults for this case
due to the effect of shadowing in Fig. 15.
In this situation, Ev and Ei, Figs. 16 and 17 again exceed
their boundaries generating alert signals, but the inverter is
still injecting energy into the grid. As a result, the system
fault is classified as a fault due to shadowing effects.
Finally, when a string is disconnected from the PV
array, important differences can be observed between sim-
ulated and monitored array yields, as shown in Fig. 18.
This behavior is detected by the ELc parameter as can be
seen in Fig. 19.
In case of faulty string, the parameter Ei exceeds its
Fig. 20. DC current error versus reference DC current errors thresholds in boundaries, Fig. 20, while the parameter Ev remains
case of string failure. between the allowed limits, Fig. 21. This situation allows
S. Silvestre et al. / Solar Energy 94 (2013) 119–127 127

classifying the fault as a string disconnected in the PV Dincßer, F., 2011. The analysis on photovoltaic electricity generation
array. status, potential and policies of the leading countries in solar energy.
Renewable and Sustainable Energy Reviews 15, 713–720.
Drews, A., de Keizer, A.C., Beyer, H.G., Lorenz, E., Betcke, J., Van Sark,
5. Conclusion W.G.J.H.M., et al., 2007. Monitoring and remote failure detection of
grid-connected PV systems based on satellite observations. Solar
This work presents a detailed procedure for automatic Energy 81, 548–564.
supervision and fault diagnosis in grid connected PV sys- EPIA, 2012. European Photovoltaic Industry Association, Report: Global
Market outlook for PV until 2016, Brussels, Belgium, May 2012.
tems. A complete monitoring and simulation package is Firth, S.K., Lomas, K.J., Rees, D.J., 2010. A simple model of PV system
provided in the same environment, allowing parameters performance and its use in fault detection. Solar Energy 84 (4), 624–
extraction of the base module, dynamic simulation model 635.
of the entire operating PV system and measurement of both Gokmen, N., Karatepe, E., Celik, B., Silvestre, S., 2012. Simple diagnostic
electrical and weather variables. For fault detection proce- approach for determining of faulted PV modules in string based PV
arrays. Solar Energy 86, 3364–3377.
dure, we established reference thresholds based on the error Haeberlin, H., Beutler, Ch., 1995. Normalized representation of energy
between simulated and measured capture losses in case of and power for analysis performance and on-line error detection in pv-
free fault system operation. These thresholds serve as com- systems. In: Proceedings of the 13th EUPV Conference on Photovol-
parison boundaries with the actual capture losses error. taic Solar Energy Conversion, Nice, France, October 1995, pp. 934–
For diagnosis procedure, we established reference thresh- 937.
King, D.L., Gonzalez, S., Galbraith, G.M., Boyson, W.E., 2007.
olds of both current and voltage in order to state whether Performance model for grid-connected photovoltaic inverters. Sandia
the fault is due to current reduction or voltage reduction. Report: SAND2007-5036.
Based on this, the most likely fault can be determined. Kymakis, Emmanuel, Kalykakis, S., Papazoglou, Thales M., 2009.
Performance analysis of a grid connected photovoltaic park on the
References island of Crete. Energy Conversion and Management 50, 433–438.
Malik, A.Q., Damit, Salmi.Jan.Bin.Haji., 2003. Outdoor testing of single
crystal silicon solar cells. Renewable Energy 28, 1433–1445.
Alonso Garcıa, M.C., Balenzategui, J.L., 2004. Estimation of photovoltaic
Muselli, M., Notton, G., Canaletti, J.L., Louche, A., 1998. Utilization of
module yearly temperature and performance based on nominal
Meteosat satellite derived radiation data for integration of autono-
operation cell temperature calculations. Renewable Energy 29, 1997–
mous photovoltaic solar energy systems in remote areas. Energy
2010.
Conversion & Management 39, 1–19.
Chao, K.-H., Hob, S.-H., Wang, M.-H., 2008. Modeling and fault
Quesada, B., Sánchez, C., Cañada, J., Royo, R., Payá, J., 2011.
diagnosis of a photovoltaic system. Electric Power Systems Research
Experimental results and simulation with TRNSYS of a 7.2 kWp
78, 97–105.
grid-connected photovoltaic system. Applied Energy 88, 1772–1783.
Chouder, A., Silvestre, S., 2010. Automatic supervision and fault detection
Stettler, S., Toggweiler, P., Remund, J., 2006. SPYCE: satellite photovol-
of PV systems based on power losses analysis. Energy Conversion and
taic yield control and evaluation. In: Proceedings of the 21st European
Management 51, 1929–1937.
Photovoltaic Solar Energy Conference. Dresden, Germany, September
Chouder, A., Silvestre, S., 2012. Modeling and simulation of a grid
2006, pp. 2613–2616.
connected PV system based on the evaluation of main PV module
Yagi, Y., Kishi, H., Hagihara, R., Tanaka, T., Kozuma, S., Ishida, T.,
parameters. Simulation Modeling Practice and Theory 20 (1), 46–58.
et al., 2003. Diagnostic technology and an expert system for photo-
Chouder, A., Silvestre, S., Taghezouit, B., Karatepe, E., 2013. Monitor-
voltaic systems using the learning method. Solar Energy Material &
ing, modeling and simulation of PV systems using LabVIEW. Solar
Solar Cells 75, 655–663.
Energy 91, 337–349.

You might also like