See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/344629270

Aerial infrared thermography for low-cost and fast fault detection in utility-
scale PV power plants

Article  in  Solar Energy · November 2020

DOI: 10.1016/j.solener.2020.09.066

CITATIONS READS

6 187

3 authors:

Aline Kirsten Vidal de Oliveira Mohammadreza Aghaei

Federal University of Santa Catarina Fraunhofer Institute for Solar Energy Systems ISE
25 PUBLICATIONS   59 CITATIONS    75 PUBLICATIONS   781 CITATIONS   

SEE PROFILE SEE PROFILE

Ricardo Rüther
Federal University of Santa Catarina
211 PUBLICATIONS   3,356 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Centro de Pesquisa e Capacitação em Energia Solar da Universidade Federal de Santa Catarina - Lab. Fotovoltaica UFSC View project

Luminescent Solar Concentrator Photovoltaic (LSC PV) View project

All content following this page was uploaded by Aline Kirsten Vidal de Oliveira on 03 February 2021.

The user has requested enhancement of the downloaded file.

 Solar Energy 211 (2020) 712–724

Contents lists available at ScienceDirect

Solar Energy
journal homepage: www.elsevier.com/locate/solener

Aerial infrared thermography for low-cost and fast fault detection in

utility-scale PV power plants
Aline Kirsten Vidal de Oliveira a, *, Mohammadreza Aghaei b, c, Ricardo Rüther a
a
Universidade Federal de Santa Catarina/UFSC, Caixa Postal 476, Florianópolis, SC, 88056-000, Brazil
b
University of Freiburg, 79085, Freiburg im Breisgau, Germany
c
Eindhoven University of Technology, 5612 AE Eindhoven, Netherlands

A R T I C L E I N F O A B S T R A C T

Keywords: The uptime of utility-scale solar photovoltaic (PV) power plants is of utmost importance for meeting contractual
Large-scale PV systems energy yields and expected capacity factors. Aerial Infrared Thermography (aIRT) is a non-destructive, no-
Operation and maintenance (O&M) downtime, fast and cost-effective method for monitoring large-scale PV power plants and assisting in fault
Aerial inspections
detection. The use of aIRT techniques aims at increasing the quality and service life of PV plants especially in
Infrared thermography
Fault detection
sunny and developing countries such as Brazil, where there is a shortfall of specialised workforce and the costs
for a detailed supervisory system of utility-scale PV power plants are high. This paper presents an analysis of an
aIRT flight campaign over four utility-scale PV plants in the northeast of Brazil. Two types of measurement
equipment have been tested and compared, resulting in more stability and efficiency using a commercially
available solution. This solution was also equipped with a RGB camera that accelerated the inspections, since it
helped to differentiate defects from hot spots caused by soiling and vegetation over the modules, which were
common. Different methods for fault detection were also tested and it was concluded that post-flight image
analysis provides faster results. The most common faults that can happen in the early operation of PV power
plants and how operators should address and prevent them were also discussed. The most common defect
detected during the campaign were disconnected cell substrings. However, disconnected strings had the most
impact on the power plants energy performance.

irradiation) and semi-arid climate in remote sites with difficult access,

including hundreds of kilometres of unpaved roads (Brazilian Ministry
1. Introduction
of Mines and Energy, 2018). In addition, the impact of high-operating
temperatures, extreme over-irradiance events caused by cloud-edge
Globally, renewable energy has contributed to a paradigm change in
and cloud-enhancement effects (Nascimento et al., 2019), and soiling
the electricity market. Mainly European and some North American
on the output of PV power plants is far more extreme in warm and sunny
countries and, in more recent years, China has sped up this growth
countries than in more temperate climates where most of the 600 GW PV
through energy policy. Utility-scale photovoltaics (PV) is now one of the
plants operate worldwide (Jäger-Waldau, 2020; Ruther et al., 2017).
key players in this development, and Brazil has also seen a major
In addition, the lifespan and the durability of PV modules depend on
expansion in installed PV capacity over the last couple of years, with the
the degradation and failure modes, which may be created during PV
first gigawatts of power plants coming online (Agência Nacional de
module production, transportation and installation. Usually, PV mod­
Energia Elétrica (ANEEL), 2020a). As installations increase in number
ules are transported by trucks in ship containers, forcing harsh condi­
and scale, the need for novel methods to ensure the reliability and
tions on a glass product such as a PV module. Not only will the transport
performance of PV systems grows. Approximately 8 GWp of PV plants
of the modules but also their storage and poor handling during instal­
will become operational in the forthcoming years in Brazil (Agência
lation implicate in potential faults and reduction of performance sooner
Nacional de Energia Elétrica (ANEEL), 2020a). Most of them use single-
or later after the beginning of operation. Another critical concern is the
axis tracking technology with bifacial PV modules, and are located in the
quality of these PV modules. Studies have shown that the quality control
northeastern region of the country (see Fig. 1). This region is charac­
methods and qualification tests adopted in current standards are not
terised by high levels of solar radiation (i.e. above 2000 kWh/m2 annual

* Corresponding author.
E-mail address: [email protected] (A. Kirsten Vidal de Oliveira).

https://doi.org/10.1016/j.solener.2020.09.066
Received 15 July 2020; Received in revised form 9 September 2020; Accepted 20 September 2020
Available online 13 October 2020
0038-092X/© 2020 International Solar Energy Society. Published by Elsevier Ltd. All rights reserved.
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

proven to be effective on detecting faults in PV modules. It requires only

Nomenclature a minimum of instrumentation and is carried out under operating con­
ditions, without interrupting the energy production of the PV plant
Abbreviations (Buerhop et al., 2012). Nevertheless, this method is time consuming, not
aIRT Aerial Infrared Thermography cost-effective and not practical for large-scale PV power plants because
CdTe Cadmium Telluride it requires expert manpower, and takes a long time to be performed. For
c-Si Crystalline silicon this purpose, Unmanned Aerial Vehicles (UAVs) have been fitted with
EL Electroluminescence IRT cameras to inspect wide areas quickly, and this technique is known
IRT Infrared Thermography as aerial infrared thermography (aIRT) (Buerhop et al., 2016; Cioaca
O&M Operation and Maintenance et al., 2015).
PB Paraíba The aIRT technique has already been proven to be a simple and
PV Photovoltaic efficient tool for detecting and classifying faults. It has been effectively
RGB Red, Green, Blue applied for the control and commissioning processes of utility-scale PV
RN Rio Grande do Norte power plants (Amstad et al., 2019; Moradi Sizkouhi et al., 2019). It also
UAV Unmanned Aerial Vehicle offers an easy recognition of problems caused by environmental events
such as hailstorms, windstorms, and lightning, as exemplified in Oli­
veira et al, (2018). The aIRT technique can assess a greater number of

Fig. 1. Brazilian map presenting in red the location of PV power plants that were contracted through regulated and PV-dedicated energy auctions. Based on Agência
Nacional de Energia Elétrica (ANEEL) (2020b). Darker shades of red indicate a multiple PV power plants in the same place.

sufficiently strict to cover the lifetime stress and the evolution of micro- modules in a short time with no system shutdown (trackers only have to
cracks that PV modules experience in working conditions (Ferrara and be in stow mode). The method is even more efficient when it combines
Philipp, 2012). RGB and IRT cameras in the same UAV, allowing a simultaneous
For all these reasons, after PV modules have been installed on their analysis.
mounting structure and during plant commissioning, an effective in­ aIRT is an opportunity for increasing the quality of PV plants and it is
spection method covering a larger area in a short time is crucial. Early especially advantageous in countries such as Brazil, where the lack of
fault identification ensures high uptimes and might prevent outages or specialised workforce and the high costs for a detailed supervisory sys­
expensive repair steps. In recent years, advanced inspection methods tem of the power plants make a fast and cost-effective method for
and techniques have been developed for performance assessment, such monitoring a power plant are even more welcome.
as real-time tracking, I-V curve tracing, infrared thermography (IRT), Many studies showed the potential for the automation of the in­
and electroluminescence (EL). In particular, IRT, as a non-destructive spection process, since the speed in which the inspections are held
method, is one of the key techniques for fault inspections and has generate a great amount of images to be treated and analysed (Li et al.,

713
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

2020; Niazi et al., 2019; Oliveira et al., 2019; Roggi et al., 2020; Siz­ Table 1
kouhi et al., 2020). With the data produced from all the performed in­ AIRT measurement systems features.
spections, it becomes now possible to analyse specific failures occurring System Features System 1 System 2
at various stages of installed PV systems and on the economic effect of
UAV equipment DJI Phantom 3 Professional DJI Matrice 200
these defects on commercial installations, where uptime is paramount. IRT camera MicroCAM 2 FLIR Zenmuse XT2
Additional analysis of these inspections could lead to findings about the UAV Weight 1,280 g 3,800 g
quality of the equipment employed and of the installation process, and System flight autonomy 13 min. 20 min.
the financial return of routine inspections. A similar measurement IRT camera sensitivity less than 60 mK less than 50 mK
Spectral range 7 – 17 µm 7.5 – 13.5 µm
campaign was previously carried out in Italy (Grimaccia et al., 2017), Pixel pitch 17 µm (640x480) 17 µm (640 × 512)
and carried out aIRT inspections for smaller PV power plants that have Frame rate 30 FPS 30 FPS
been operating for at least four years in more temperate climate IRT camera weight 80 g 588 g
conditions. Transmitter BosCam FPV Transmitter Built-in
Antenna Cloverleaf 5.8 GHz Built-in
In this context, the purpose of this paper is to summarise the main
RGB camera None (removed) 12 MP, 1/1.7′′ CMOS
conclusions of a very comprehensive flight campaign carried out in Price US$ 6,000.00 US$ 15,000.00
Brazil over a variety of utility-scale, single-axis tracking PV power plants
of both c-Si and thin film CdTe technologies during their commissioning
phase. It describes common defects found in the different plants dis­ sensitivity, accuracy, temperature range, camera weight and lens type.
playing UAV-captured image data and the association between visible In addition, the camera type and the camera software that are offered
and IRT sensors. It also offers some evidence of typical faults identified are relevant. For the UAV system, the battery flight time, stability of the
and relative failure rates, and correlates results with related research system, maximum altitude, flight duration, maximum payload and full
observations in recent literature. The flight campaign applied different compatibility between instruments are issues that have to be considered,
techniques and tested different sensor technologies in each one of the and which impact on the quality of the flight as well as the costs of the
inspected PV power plants, in order to compare different methods for equipment (Gallardo-Saavedra et al., 2018).
aIRT inspections with the aim of increasing the uptime of these plants. In this experiment, two different measurement systems were used for
The paper is structured as follows. Section 2 summarises the method the aerial inspections, each of them were employed in different test sites,
of aIRT as well as the faults it can detect and the procedures adopted for in order to compare two different types of equipment. Both are described
this study Then, Section 3 describes the test sites where the inspections in Table 1 and were previously tested in order to demonstrate the effi­
were carried out. In Section 4, the results are presented and discussed ciency in detecting all desired faults. System 1 was developed by
and in Section 5 the main findings of the study are summarised. coupling an IRT surveillance camera to a light consumer UAV. The
camera is non-radiometric, which means that it differentiates tempera­
2. Aerial infrared thermography (AIRT) tures with different shades of grey, but it does not provide information
about the temperature at each point. For this retrofit, it was necessary to
As in theory, all PV modules receive the same amount of irradiance in remove the RGB camera of the UAV and to develop a new data trans­
a PV plant, the modules or cells that do not turn photons into electricity mission system, as described in Oliveira et al. (2018). For this reason,
convert them into heat. IRT measures the radiation emitted from the there is no integration between UAV and IRT camera, which creates
surface of any object within the spectrum of infrared wavelengths be­ challenges for inspection, such as not being able to see the camera image
tween 1.4 and 15 μm (Tsanakas and Botsaris, 2013). Energy losses will in the same device that controls the UAV.
therefore be revealed as temperature variations in the IRT images. These System 2 is a high-end, commercially-available professional UAV
thermal differences have been defined and listed in previous research solution that integrates a high quality radiometric IRT camera with an
(Buerhop et al., 2012, 2007; Köntges et al., 2014) and are now stand­ RGB camera in a robust UAV, which offers better image quality, flight
ardised in the international standard IEC TS 62446–3 Version 1.0 (In­ autonomy, stability, and insulation against electromagnetic in­
ternational Electrotechnical Commission and International terferences. The equipment also includes image-processing software that
Electrotechnical Commission (IEC), 2017). facilitates the detection of faults. However, this equipment is much more
Traditionally, IRT assessments are carried out with handheld IRT expensive (almost three times the price) than System 1, what raises the
cameras in the field or on lift platforms to maximise coverage. This question of the necessity of such sophisticated equipment. While the
method relies on human effort and expertise and is very time-consuming equipment cost for these valuable analyses is a negligible fraction of the
and labour-intensive. As a result, the precision of the testing is vulner­ typical costs associated with large-scale PV power plants, these services
able to human error, and it increases the complexity of the inspection are typically offered and carried out by third parties to whom they
method. IRT cameras can be combined with aerial equipment such as represent a considerable cost. Both systems are presented in Fig. 2.
UAVs (drones) to increase cost-effectiveness and allow inspections on
roof-mounted PV systems with limited access or large-scale PV power 2.2. Inspection procedure
plants, in the so-called aIRT. This method increases the speed of the
inspections, but they provide lower images resolutions than the tradi­ In order to optimise flight time, video images are acquired instead of
tional IRT method. (Quater et al., 2014; Tsanakas et al., 2016; Tsanakas still images, also because it facilitates fault detection. It also helps on the
and Botsaris, 2012). The next subsections describe the aIRT techniques, differentiation between faults and measurement artefacts, as the
and the equipment and procedures adopted in this paper. reflection of the sun or other rows of modules move following the UAV
movement.
2.1. aIRT measurement equipment During the flight, different environmental stress factors, such as
weather, air turbulence, and sunlight reflection, can influence the
For the right choice of measurement equipment for aIRT inspections, inspection and, consequently, the quality of the aIRT images. The ideal
many aspects must be considered. For the camera, the resolution is a key conditions for aerial IRT monitoring can be summarised as follows
factor because it determines the maximum flight height, which strongly (Aghaei et al., 2016; Leva et al., 2015; Quater et al., 2014).
influences the time requirements for inspection. This is critical, espe­
cially for large-scale PV power plants, because normally the minimum a. The orientation of the UAV-mounted IRT sensor should be perpen­
irradiance conditions are met only for around six hours per day in the dicular to the PV modules and should be kept constant;
locations of this study. Other important characteristics are thermal

714
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Fig. 2. aIRT measurement systems applied with extra lightweight IRT camera adapted to commercial drone (System 1 (left) in Table 1, and high-end, fully com­
mercial IRT + RGB drone System 2 (right) in Table 1).

Fig. 3. Flowchart illustrating the process of real-time fault analysis of the

power plant (i.e. the detection and analysis of the faults are carried out during
the flight).

b. The flight altitude should be at least five meters to prevent any UAV
self-shading during the inspection;
c. The IRT inspection should be carried out on cloudless, bright, and
dry days. The wind speed should not exceed 4 m/s since higher wind
Fig. 4. Flowchart illustrating the process of post-flight fault analysis of a PV
speeds will create air turbulence and disturb the UAV during the
power plant (i.e. the detection and analysis of faults are carried out after
inspection procedure;
the flight).
d. The optimal period for UAV IRT inspection is about midday, when
the direction of the sun is more perpendicular to the PV modules in
stow mode in single-axis tracking large-scale PV power plants. The For the other three inspections, the route of the UAV was initially
irradiance should be over 600 W/m2 on the plane of the PV array programmed and automatically carried out by the drone software.
under inspection; Furthermore, the fault analysis was evaluated after the flights, by ana­
e. The flight path should be planned before each inspection task and the lysing the images captured. Fig. 4 presents the flowchart that illustrates
route should be well documented. this process. In the image, the grey boxes indicate steps that can be held
outside of the power plant, reducing on-site time.
For all the inspections, the UAV route is a route orthogonal to the In order to compare both measurement systems employed and also
single-axis tracking PV system tracking tables planned in advance, in the two different methods of inspection tested, the duration of the in­
order to optimise and to obtain a shorter flight time and a better view of spections were also registered. The duration of the inspection includes
all PV modules. not only flight time but also image analyses and images captured with
In one of the power plants (Plant 1), the flights were carried out handheld cameras, for verifying faults. With these records, it is possible
manually and the fault analysis was held in real-time. This means that to calculate an estimation of the amount of MW inspected per person per
every time a fault was detected, the UAV was manoeuvred closer to the day, in order to have a metric for comparison between methods ac­
damaged module, for a better visualisation, and a specialist was sent to cording to the amount of manpower required. This does not mean,
the fault in order to check it, making notes and capturing images with however, that a person alone can perform any inspection, since regu­
handheld RGB and IRT cameras. This process is illustrated in Fig. 3. lation on work in power plants requires that any inspection be done at

715
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Table 2
Details of the utility-scale PV power plants inspected in this paper, all located in the sunny and warm Brazilian northeast region. Annual irradiation data is based on
Pereira et al. (2017).
PV Plant Location PV Installed Capacity PV Technology Drone used (from Table 1) Height above sea level Annual irradiation

1 Rio Grande do Norte (RN) 36.7 MW CdTe System 1 50 m 2,150.9 kWh/m2

2 Paraíba (PB) 31.2 MW c-Si System 1 250 m 2,211.9 kWh/m2
3 Paraíba (PB) 31.5 MW c-Si System 2 257 m 2,194.7 kWh/m2
4 Paraíba (PB) 31.5 MW c-Si System 2 257 m 2,194.7 kWh/m2

Fig. 5. Brazilian map with annual averages of the daily total irradiation at latitude tilt, highlighting the two states where the utility-scale PV power plants are located
(RN and PB). The circled areas represent the location of the power plants, coinciding with the highest annual irradiation levels in the country. Based on Pereira
et al. (2017).

least in pairs. For the comparison between inspection times, it is 3. Experimental SITES
important to notice that the workforce that performed the inspections
were highly skilled and had experience with aIRT in power plants. The The inspections were carried out in four different PV power plants,
real-time approach was only tested using the measurement system 1, totalling around 130 MW and over 600 thousand individual PV modules.
limiting the comparison of productivity of this method with the post- All inspected PV plants follow the trend of most of Brazilian utility-scale
flight analysis approach. PV systems, consisting of N-S single-axis trackers, each one moving some
hundreds of PV modules simultaneously. The inspections were carried
2.3. Image analysis and faults classification out as part of the commissioning phase of the corresponding PV plants.
Table 2 presents some characteristics of the inspected sites. According to
The fault diagnosis occurs by evaluating the PV module’s thermal the PV module technology (i.e. crystalline Si or thin-film CdTe), mea­
pattern, which is uniform for healthy modules, and which is very clearly surement system, and approach, the height of the flight was defined for
impacted by failures evidenced by the IRT image. Examples of detect­ each site, after initial tests.
able failures include: cracks, corrosion, disconnected strings, shading, All sites are located in the countryside of the states of Paraíba (PB)
dirt, broken glass, among others. The classification of detected faults is and Rio Grande do Norte (RN), in the Northeast of Brazil, known for its
performed based on IEC TS 62446–3: 2017 (International Electro­ hot climate and high irradiances. Fig. 5 presents a map of Brazil with
technical Commission and International Electrotechnical Commission annual averages of the daily average of latitude-tilted irradiation,
(IEC), 2017). highlighting the two states in discussion. The annual irradiation aver­
The effect that such faults have on the power plant varies in amount ages in the region exceed 6 kWh/m2 per day (greater than2100 kWh/m2.
of power loss that they cause and in the risks of greater damage, as fire year) and the mean annual temperature is about 26 ◦ C.
hazard for example, which might lead to considerable downtimes. A hot
spot; example, can cause indetectable power loss at first (less than 3%), 4. Results and discussion
but this loss of power can increase over time, as the high temperatures
damage the cell or other elements of the model, such as solder joints 4.1. aIRT measurement system
(Köntges et al., 2014; Ndiaye et al., 2013; Ogbomo et al., 2018)
The data collected are then processed and an actionable report is The inspections carried out with System 2 were faster and easier,
generated. Through precise site planning, it is possible to provide an because of the integrated solution provided by that equipment. The
exact position of the system faults and remediation may be scheduled drone presents a better GPS system, which made the inspections quicker
based on a complete awareness of the site condition. The report is then and safer. Besides, because System 1 consists of an adapted solution, its
sent to stakeholders for correction activities such as connection review transmission and image quality were inferior. System 2 also offered the
or element replacement. convenience of taking aIRT and RGB images simultaneously, unlike the

716
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Table 3
Comparison of both aIRT measurement systems results.
Fault System 1 System 2

Disconnected Substring

Disconnected String

Table 4
Results of the aIRT flight inspections in each utility-scale PV power plant in northeast Brazil.
PV Plant Drone used PV Technology Fault Analysis Inspection Duration Flight Altitude Team Size MW inspected per person per day

Plant 1 System 1 CdTe Real-time 9 days 15 m 4 people 1.02

Plant 2 System 1 c-Si Post-flight 7 days 20 m 2 people 2.23
Plant 3 System 2 c-Si Post-flight 4 days 30 m 2 people 3.94
Plant 4 System 2 c-Si Post-flight 4 days 30 m 2 people 3.94

first equipment that required additional flights with conventional UAVs. thin film CdTe modules, which have a slightly lower efficiency and
Table 3 compares the outcomes of the two systems for the detection of therefore a larger area for the same amount of power.
the most common faults. The poorer image quality and resolution of The duration of the inspection also differs between power plants due
System 1 is clear on the images, on top of the “fisheye” effect of the to the different heights used for the UAV flights. As mentioned before,
camera, which also made the visualization of faults more difficult. The the altitudes were defined according to initial tests, aiming to provide
disconnected substrings faults are highlighted in red in order to facilitate the best results given the employed measurement system and PV module
visualization. technology. The higher image quality of the measurement System 2,
allowed an inspection from a higher altitude, which sped up the in­
4.2. Real-time vs post-flight fault analysis spection process. One further factor that impacted the duration of the
inspection is the presence of an RGB camera in measurement System 2,
Table 4 presents the duration of each inspection. The table also which minimises field time due to the effects of soiling and vegetation
presents an estimation of the amount of MW inspected per person per over the PV modules. For the case of Plant 2, an additional flight with an
day, showing the high productivity obtained in the inspections of plants RGB UAV was performed, to check for such problems, which also takes
3 and 4. extra time not only for the flight but also for the analysis of the RGB
In terms of comparing the different approaches on the PV power images. This additional inspection step can be reduced by performing
plant fault analysis, the analyses carried out after the flight proved to be aIRT right after the cleaning and the vegetation pruning at the power
more efficient. The measurement time in the power plant was longer for plant. Considering that in a day is possible to work for 4 to 6 h due to
the real-time approach and demanded double the workforce (four peo­ environmental conditions, the results of plants 3 and 4 are very similar
ple instead of two) to enable checking and locating the faults during the to other studies as Ulrike Jahn et al (2018). The study affirms that a PV
flight. The real-time approach also complicated the detection of faults plant of 4 MWp requires 5 to 10 h for an aIRT inspection, when all
because of the difficulties of analysing videos in the field, as the screen is required conditions are met.
severely affected by glare, reflections of the sun and dirt from the soil. The results shows that the economy of acquisition costs of System 1
Another reason for the longer time needed for the real-time analysis was does not worth the extra inspection time required using this solution, as
the manual flight. The programmed automatic route provides more the difference is dissolved in the many inspections that the robustness of
stable and quicker flights, but unfortunately do not yet allow real-time System 2 can perform. The cost disparities are much smaller than
analysis, since it is not viable to stop at each fault detected to take workmanship costs that are fixed for every inspection.
notes and take a closer look during automatic route flights. Given these
outcomes, it is suggested that inspections are held with automatic path 4.3. Hot spots caused by soiling and vegetation
planning, prior to the measurements and that the analysis of the footage
is carried out immediately afterwards. There is also the possibility to The most common problems found in the inspections were hot spots
repeat the flight over some areas, in order to check ambiguous results caused by soiling and vegetation resulting in shading of PV cells,
after the first analysis. Plant 1 was also the only power plant to employ affecting some hundreds of modules during the inspections. The faults

717
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Fig. 6. Examples of vegetation and soiling over PV modules: (a) Soiling pattern image over many c-Si modules, taken with System 2. (b) Bird droppings image over
one c-Si module, taken with System 1. (c) Vegetation image over many c-Si modules taken with System 2. (d) Vegetation image over one CdTe PV module, taken with
System 1.

are not always distinguishable from actual hot spots defects through the effect of such problems detected with the two different measurements
visual aerial images, so all hot spots that are not visible caused by soiling systems and over different PV module technologies. Fig. 6 (a) shows a
or vegetation have to be verified with a ground inspection to complete soiling pattern caused by wind over many PV modules, but which only
the diagnosis. When a hot spot is found in a soiled PV module, the causes hot spots over some of them due to the amount of soiling. The
procedure is to clean the module in order to re-evaluate the thermal soiling pattern consists of white lines on the lower corner of the modules,
pattern to know whether the hot spot was caused by soiling or actual as can be better seen in the detail of the image. Fig. 6 (b) presents a hot
damage. These situations are not accounted for as faults in this analysis spot caused by a bird dropping over a module in an image taken with
because they are not considered damages or defects of the power plant measurement System 1. Because this UAV had its RGB camera removed,
and can be easily fixed. However, they cause power losses and create fire the visual picture (on the right side) was taken with a second UAV.
hazard risks, which can severely affect uptime. Therefore, the images are taken from slightly different positions. This
Fig. 6 presents different examples of soiling and vegetation detected problem slows the association of the two images to compare IRT and
during a flight campaign. On the images, it is possible to compare the RGB images to correctly classify faults. Fig. 6 (c) and (d) present images

718
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Table 5
Most common PV module defects detected in the four utility-scale PV power plants in Brazil.

Number of affected
Number of detected faults Estimated Power Loss (kW)
Disconnected substrings modules

Disconnected substrings

Disconnected substrings
Disconnected strings

Disconnected strings

Disconnected strings
Broken modules

Broken modules

Broken modules
Hot-spots

Hot-spots

Hot-spots
PV Plant

1 0 26 5 1 0 390 5 1 0 44.85 0.01 0.01

2 86 8 0 2 86 168 0 2 9.46 55.44 0 0.01
3 39 2 5 0 39 30 5 0 4.29 9.90 0.03 0
4 34 1 5 0 34 21 5 0 3.74 6.93 0.03 0
Total 159 37 15 3 159 609 15 3 17.49 117.12 0.08 0.02

of modules covered with vegetation. However, due to the difference of • disconnected substrings: disconnection of the cell strings of the
measurement system used and PV module technologies employed, it is module inside the junction box, causing one third of the module to be
clear that the detection is quite different for the two cases. in open circuit;
• disconnected strings: interruption of the connection one full string
4.4. Most common defects detected on the sites with the system, due to failures of connection, fuses or string boxes
problems, causing many modules to be in open circuit;
The most common defects found in the inspections were: • modules with broken glass: modules with the front glass broken due
to shock with rocks or other objects during construction or thermal
stress due to other failures that generate hot spots;

Fig. 7. Examples of disconnected cell substrings in individual PV modules detected during the flight campaign: (a) four defective modules, in the image taken with
measurement System 1, and (b) one defective module, which was taken with measurement System 2. Image (c) shows a disconnected substring IRT image taken with
a hand-held camera.

719
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Fig. 8. Examples of disconnected PV strings: (a) thin film CdTe PV module disconnected string, taken with System 1 and accompanied by an IRT image taken with a
hand-held camera. (b) c-Si disconnected string, taken with System 1. (c) c-Si disconnected string, taken with System 2.

• modules with hot spots defects: points of higher temperature over the From the table we see that despite not being the most common fault,
modules caused by failures such as delamination points, cracks due disconnected strings cause the major loss of power in a power plant,
to mechanical stress, corrosion, soldier joint defects and others. since it affects many modules at once.

Table 5 presents the number of occurrences of each of these faults in 4.4.1. Disconnected substrings
each inspected site as well as the number of modules affected by each The most common defects detected are disconnected cell substrings,
fault (a disconnected string affects 15 to 21 modules, for example). which are usually a manufacturing defect. PV modules are healthy and
These four problems are considered defects in the PV power plant and have an open-circuit voltage according to the product’s datasheet as
are discussed in order of most common occurrence in the next they leave for transport to the PV power plant site, what is confirmed by
subsections. the measurements and the flash report that is carried out before the
The table also estimates the power loss caused by each fault. For the modules leave the manufacturer’s plant. After transport and installation,
calculation of power loss of each fault, the nominal datasheet peak through thermal or mechanical stress, these cells strings became
power of each module was used, considering that disconnected sub­ disconnected in the junction box, causing the bypass diode to take on the
strings cause a loss of one third of the PV module, and disconnected full current of the string. Besides the loss of one-third of the PV module
strings cause the loss of power proportional to the number of modules in peak power, the fault causes unnecessary stress to the bypass-diode.
each string. For the case of hot spots and broken glass PV modules, a loss These results can be compared to a flight campaign in Italy (Grimaccia
rate of 2% of the power of the PV module was applied, assuming that in et al., 2017), in which a similar approach to this work was adopted, only
the commissioning phase of the power plant the impact of the failure is that the experiments were carried out in much smaller PV plants using
minimal. This impact can grow linearly or exponentially over time. PV modules that already had operated for some years, and which were
These power losses estimations do not account for the possible power produced with manufacturing technologies no longer in use. In that
mismatch that the faults brings to the whole system. Disconnected study, disconnected substrings were not as common as in the study
substrings, for example, reduce the string voltage and affect all string carried out in Brazil, which raises the question of the quality of PV
parallel to it, due to voltage mismatch. modules produced under the very demanding price reduction pressure

720
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Fig. 9. Examples of hot pots detected in c-Si PV modules: (a) aIRT image of PV module with broken glass detected with System 2 and its RGB equivalent. (b) aIRT
image of PV module with broken glass detected with System 1.

the PV industry has suffered in recent years. under bad environmental conditions, such as clouds and low irradiance.
Fig. 7 presents two examples of detection of disconnected cell sub­ Fig. 8(a) presents also the open string IRT image taken with a hand-held
strings, one detected with each measurement system. System 2 allows a camera. The image shows that the difference of temperature between
clear detection of this defect, and the worse image quality of System 1 disconnected and connected strings is around 4 K, what matches the IEC
results in a more difficult detection. Commonly, as shown in the picture 62446–3, which affirms normal temperature difference for disconnected
taken with System 1, many modules with this defect are found side-by- strings is between 2 and 7 K (International Electrotechnical Commission
side. There are cases with more than four modules in a string with the and International Electrotechnical Commission (IEC), 2017)p.
same defect. This happens because this defect is a manufacturing Disconnected strings also slow down the inspection process, since
problem: since the modules installed close to each other are usually this type of defect disguises other faults, as the only problems that can be
taken from the same pallet, they are more likely to have come from the detected in a disconnected string are short-circuited PV modules or
same PV module production batch, and to be installed and transported substrings. Therefore, the string needs to be reconnected and the ther­
together. Fig. 7(c) shows a hand-held IRT image of a PV module with a mal pattern of the PV modules re-evaluated. The causes for string dis­
disconnected substring. The image shows that the temperature differ­ connections vary from different equipment defects: trackers, inverters,
ence between the affected and the non-affected substrings is around 3 K, and fuses and diodes due to extreme-overirradiance events (Nascimento
value that fits with what is stated in the IEC 62446–3 (differences of 2 to et al., 2019), besides scheduled disconnections for maintenances, or
7 K for disconnected substrings) (International Electrotechnical Com­ power restrictions. Disconnected strings can be detected more easily
mission and International Electrotechnical Commission (IEC), 2017). through the supervisory system when current monitoring is carried out
This defect does not occur in CdTe modules, because thin film PV at a string or stringbox level (depending on topology and PV module
modules do not usually have bypass diodes separating the module in technology). However, with the ever-growing cost pressure on new PV
substrings. power plants, and also because normally the supervisory system is not
fully functional by commissioning phase, the aIRT is still a fast method
4.4.2. Disconnected strings to perform this inspection. For medium-size PV power plants, one can
String disconnection is a failure that causes the highest fraction of raise the question of the viability of having a supervisory system on the
energy losses in a PV power plant energy production, since it affects string level. Depending on the maintenance and equipment costs, a
many modules at once and it is also quite common. Some examples of continuous aerial inspection can provide a better economic balance.
disconnected strings are shown in Fig. 8. The three images were taken in
different PV power plants, which have different string and table con­ 4.4.3. Broken PV modules and other types of hot spots in c-Si PV modules
figurations. In Fig. 8(a) and 8(c) the detection is easier because strings Broken modules were not so commonly found and were mainly
are all wired in the same table, and this provides contrast to the caused by installation events. Other types of hot spots defects were even
disconnected ones. On the other hand, for the power plant shown in rarer since the power plants were connected only for a short-time period.
Fig. 8(b), the inspection should be carried out above a certain height that This is an interesting result in comparison with Grimaccia et al. (2017),
allows a good comparison with other tables, to enable the detection. For where problems detected as hot spots defects were the main findings of
Fig. 8(a) and 8(b), the images were taken with measurement System 1, the flight campaign in Italy. Faults that result in hot spots are most
in two PV power plants using different PV module technologies. For frequently found in PV power plants that have been operating for at least
Fig. 8(c), the image is taken with System 2. It is visible that this type of a few years or have undergone severe weather events. They do not
fault is easily recognised with both systems, even with the lower camera typically induce a substantial loss in PV output at an early level; there­
resolution of System 1. The disconnection can also easily be seen even fore, they are not generally identified by the supervisory systems. They

721
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Fig. 10. Examples of hot spots detected in CdTe modules: (a) PV module with broken front glass. (b) PV module with broken front glass where a whole cell was
damaged (left corner of the module) (c) PV module with a delamination spot. Each image is accompanied by its correspondent hand-held IRT image.

are, however, a possible cause of fire hazards in the power plant or temperature differences, which must be replaced immediately.
electrical shocks. The breakages also usually increase moisture infiltra­
tion and are normally followed by corrosion, discolouration or delami­ 4.4.4. Broken PV modules and other types of hot spots in CdTe PV modules
nation issues (Ndiaye et al., 2013). Fig. 10 presents three images with different types of hot spots defects
Fig. 9 presents examples of modules with hot spots. Both cases shown detected in Plant 1 (CdTe modules). Fig. 10(a) shows an example of a
in the images are c-Si modules with broken front glass. Fig. 9(a) shows module with broken glass that resulted in temperature differences of
an example of the aIRT image and its correspondent RGB image. The more than 30 K above the module temperature, despite IEC 62446–3
detail of the image shows that, in this case, the broken module is not affirming these defects normally causes differences of 0 to 7 K. This type
visible through the visual inspection, even though the problem is clear of failure is easily spotted because it generates a “brighter” hot spot, due
through the IRT image. In these cases, the identification of the broken to the higher temperatures involved. This allows the defect to be
module is performed through ground inspection. There are cases when detected from altitudes higher than 50 m. Fig. 10(b) also presents a
the fault can be identified in the RGB images my using different camera module with broken glass, but in this case, only some cells in the PV
angles. Nevertheless, the RGB image enables the operator to discard the module are hotter than the others and the temperature difference is
hypothesis of vegetation or soiling over the module, assuring that the much smaller (around 4 K), which makes detection more difficult.
hot spot represents a module fault. This combination of RGB and aIRT Fig. 10(c) shows a hot spot that revealed a delamination point in the PV
image reduces on-site time and extra work. Fig. 9(b) presents another module. This fault is much harder to be detected because of the small
example of broken front glass that creates a hot spot with high size and small temperature difference (less than 7 K).

722
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

5. Conclusions Amstad, D., Oliveira, A.K.V. de, Häberle, A., Rüther, R., 2019. Fault Inspection of CIGS
PV Plant using Aerial Infrared Thermography. In: 36th European Photovoltaic Solar
Energy Conference and Exhibition (EU PVSEC). Marseille, France, pp. 1569–1574.
In order to maximise uptime and increase capacity factors, this paper https://doi.org/10.4229/EUPVSEC20192019-5CV.4.6.
offers an analysis of the most common defects found during the Brazilian Ministry of Mines and Energy, 2018. Photovoltaic projects on energy auctions -
commissioning phase of four different utility-scale PV power plants in Characteristics of PV projects on auctions from 2013 to 2018. [WWW Document].
URL http://www.epe.gov.br/sites-en/publicacoes-dados-abertos/publicacoes/
Brazil, analysing PV module faults of two different technologies: c-Si and PublicacoesArquivos/publicacao-184/EPE_DEE_NT_091_2018_r0 (English).pdf
CdTe. Two different measurement equipment have been tested and (accessed 5.30.19).
compared, and as expected, the results show that System 2 provided a Buerhop, C., Jahn, U., Hoyer, U., Lercher, B., Wittmann, S., 2007. Abschlussbericht
Machbarkeitsstudie Überprüfung der Qualität von Photovoltaik- Modulen mittels
faster and easier experience. This is due to the integrated solution pro­ Infrarot-Aufnahmen.
vided by the equipment, the better quality of its camera and GPS and the Buerhop, C., Pickel, T., Dalsass, M., Scheuerpflug, H., Camus, C., Brabec, C.J., 2016. AIR-
existence of a RGB camera on the UAV, what allows a quicker differ­ PV-check: A quality inspection of PV-power plants without operation interruption.
In: 2016 IEEE 43rd Photovoltaic Specialists Conference (PVSC). Portland, OR, USA,
entiation between faults and soiling and vegetation. Because of its ad­ pp. 1677–1681. https://doi.org/10.1109/PVSC.2016.7749909.
vantages, System 2 is more expensive than System 1. However, results Buerhop, C., Schlegel, D., Niess, M., Vodermayer, C., Weißmann, R., Brabec, C.J., 2012.
showed that this cost is compensated by reducing the inspection time, Reliability of IR-imaging of PV-plants under operating conditions. Sol. Energy Mater.
Sol. Cells 107, 154–164. https://doi.org/10.1016/j.solmat.2012.07.011.
and therefore, reducing workmanship costs. Cioaca, C., Pop, S., Boscoianu, E.C., Boscoianu, M., 2015. Aerial Infrared Thermography:
In terms of comparing real-time with post-flight fault analyses, the A scalable procedure for photovoltaics inspections based on efficiency and
analyses carried out after or in between flights proved more efficient and flexibility. Appl. Mech. Mater. 772, 546–551. https://doi.org/10.4028/www.
scientific.net/AMM.772.546.
required less time on-site. The duration of the inspection was also
Ferrara, C., Philipp, D., 2012. Why do PV modules fail? Energy Procedia 15, 379–387.
influenced by PV module technology and quality of the camera of the https://doi.org/10.1016/j.egypro.2012.02.046.
measurement system employed. Gallardo-Saavedra, S., Hernández-Callejo, L., Duque-Perez, O., 2018. Technological
The most common problems found during the aerial inspections were review of the instrumentation used in aerial thermographic inspection of
photovoltaic plants. Renew. Sustain. Energy Rev. 93, 566–579. https://doi.org/
hot spots due to soiling and vegetation, which caused shading of PV 10.1016/j.rser.2018.05.027.
cells. They are not considered faults but can affect the outcomes of the Grimaccia, F., Leva, S., Dolara, A., Aghaei, M., 2017. Survey on PV modules’ common
aIRT and the total PV plant downtime. They can be distinguished from faults after an O&M flight extensive campaign over different plants in Italy. IEEE J.
Photovoltaics 7, 810–816. https://doi.org/10.1109/JPHOTOV.2017.2674977.
actual damage of the PV cells through RGB images. International Electrotechnical Commission, International Electrotechnical Commission
Among the actual faults of the PV systems detected during the flight (IEC), 2017. IEC TS 62446-3 - Photovoltaic (PV) systems - Requirements for testing,
campaign, the most common, in number of occurrences, were discon­ documentation and maintenance - Part 3: Photovoltaic modules and plants - Outdoor
infrared thermography.
nected cell substrings, followed by disconnected strings, hot spots de­ Jäger-Waldau, A., 2020. Snapshot of photovoltaics-February 2020. Energies 13. https://
fects and modules with broken glass. However, when comparing the doi.org/10.3390/en13040930.
amount of lost power caused by each fault, the highest downtimes are Köntges, M., Kurtz, S., Packard, C.E., Jahn, U., Berger, K., Kato, K., Friesen, T., Liu, H.,
Van Iseghem, M., 2014. Review of Failures of Photovoltaic Modules. IEA-
caused by disconnected strings, followed by disconnected substrings and Photovoltaic Power Systems Programme.
hot spots, with broken front glass in last. Leva, S., Aghaei, M., Grimaccia, F., 2015. PV power plant inspection by UAS : Correlation
The results of these flight campaigns are extremely important between altitude and detection of defects on PV modules. In: Environment and
Electrical Engineering (EEEIC), 2015 IEEE 15th International Conference On.
because they show the most common faults that can happen in the early
https://doi.org/10.1109/EEEIC.2015.7165466.
operation of PV power plants and from that, operators can decide how to Li, X., Li, W., Yang, Q., Yan, W., Zomaya, A.Y., Member, S., Yang, Q., Member, S.,
address and prevent them, maximising the uptime of utility-scale PV Yan, W., Zomaya, A.Y., 2020. An Unmanned Inspection System for Multiple Defects
power plants. Detection in Photovoltaic Plants. IEEE J. Photovoltaics 10, 568–576. https://doi.
org/10.1109/JPHOTOV.2019.2955183.
Moradi Sizkouhi, A.M., Majid Esmailifar, S., Aghaei, M., Vidal De Oliveira, A.K., Ruther,
R., 2019. Autonomous Path Planning by Unmanned Aerial Vehicle (UAV) for Precise
Declaration of Competing Interest Monitoring of Large-Scale PV plants. In: 2019 IEEE 46th Photovoltaic Specialists
Conference (PVSC). Chicago, IL, USA, pp. 1398–1402. https://doi.org/10.1109/
The authors declare that they have no known competing financial PVSC40753.2019.8980862.
Nascimento, L.R.do., de Souza Viana, T., Campos, R.A., Rüther, R., 2019. Extreme solar
interests or personal relationships that could have appeared to influence overirradiance events: Occurrence and impacts on utility-scale photovoltaic power
the work reported in this paper. plants in Brazil. Sol. Energy 186, 370–381. https://doi.org/10.1016/j.
solener.2019.05.008.
Ndiaye, A., Charki, A., Kobi, A., Kébé, C.M.F., Ndiaye, P.A., Sambou, V., 2013.
Acknowledgments Degradations of silicon photovoltaic modules: A literature review. Sol. Energy 96,
140–151. https://doi.org/10.1016/j.solener.2013.07.005.
The authors acknowledge with thanks ENGIE Brasil Energia and the Niazi, K.A.K., Akhtar, W., Khan, H.A., Yang, Y., Athar, S., 2019. Hotspot diagnosis for
solar photovoltaic modules using a Naive Bayes classifier. Sol. Energy 190, 34–43.
ANEEL R&D Program for the financial support under contract number https://doi.org/10.1016/j.solener.2019.07.063.
PD-0403-0042/2016, and the Alexander von Humboldt Foundation for a Oliveira, A.K.V. de, Aghaei, M., Madukanya, U.E., Nascimento, L., Ruther, R., 2018.
Research Group Linkage Project travel grant that allowed Dr. Aghaei to Aerial Infrared Thermography of a Utility-Scale PV Plant After a Meteorological
Tsunami in Brazil. In: 2018 IEEE 7th World Conference on Photovoltaic Energy
spend time at the Fotovoltaica/UFSC Solar Energy Laboratory in Flo­ Conversion (WCPEC). IEEE, Waikoloa Village, HI, USA, pp. 684–689. https://doi.
rianópolis-Brazil. A. K. V. Oliveira wishes to acknowledge the members org/10.1109/PVSC.2018.8548019.
of Fotovoltaica/UFSC that assisted her during the inspections of the PV Oliveira, A.K.V. de, Aghaei, M., Rüther, R., 2019. Automatic Fault Detection of
Photovoltaic Arrays by Convolutional Neural Networks During Aerial Infrared
plants.
Thermography. In: 36th European Photovoltaic Solar Energy Conference and
Exhibition (EU PVSEC). Marseille, France, pp. 1302–1307. https://doi.org/10.4229/
References EUPVSEC20192019-5BO.6.4.
Ogbomo, Osarumen O., Amalu, Emeka H., Ekere, N.N. Ekerea, Olagbegi, P.O., 2018.
Effect of operating temperature on degradation of solder joints in crystalline silicon
Agência Nacional de Energia Elétrica (ANEEL), 2020a. Sistema de Informações de
photovoltaic modules for improved reliability in hot climates. Solar Energy 170,
Geração (SIGA) [WWW Document]. accessed 7.11.20. https://www.aneel.gov.
682–693. https://doi.org/10.1016/j.solener.2018.06.007 (in press).
br/siga.
Pereira, E.B., Martins, F.R., Gonçalves, A.R., Costa, R.S., de Lima, F.J.L., Rüther, R., de
Agência Nacional de Energia Elétrica (ANEEL), 2020b. Results of Generation Auction in
Abreu, S.L., Tiepolo, G.M., Pereira, S.V., de Souza, J.G., 2017. Atlas Brasileiro de
the Regulated Environment [WWW Document]. accessed 6.2.20. https://app.
Energia Solar, 2a, Edição. ed. INPE, São José dos Campos.
powerbi.com/view?r=eyJrIjoiZTZiNDhjNjctZTQ2NC00YzFmLTgxYTUtZmY5Y
Quater, P.B., Grimaccia, F., Leva, S., Mussetta, M., Aghaei, M., 2014. Light Unmanned
jEzNmI3MjdkIiwidCI6IjQwZDZmOWI4LWVjYTctNDZhMi05MmQ0LWVhNGU5Y
Aerial Vehicles (UAVs) for cooperative inspection of PV plants. IEEE J. Photovoltaics
zAxNzBlMSIsImMiOjR9.
4, 1107–1113. https://doi.org/10.1109/JPHOTOV.2014.2323714.
Aghaei, M., Leva, S., Grimaccia, F., Milano, P., 2016. PV Power Plant Inspection by
Image Mosaicing Techniques For IR Real-Time Images. IEEE 43rd Photovolt. Spec.
Conf. 3100–3105 https://doi.org/10.1109/PVSC.2016.7750236.

723
A. Kirsten Vidal de Oliveira et al. Solar Energy 211 (2020) 712–724

Roggi, G., Niccolai, A., Grimaccia, F., Lovera, M., 2020. A computer vision line-tracking Tsanakas, J.A., Botsaris, P.N., 2012. An infrared thermographic approach as a hot-spot
algorithm for automatic UAV photovoltaic plants monitoring applications. Energies detection tool for photovoltaic modules using image histogram and line profile
13. https://doi.org/10.3390/en13040838. analysis. Int. J. Cond. Monit. 2, 22–30. https://doi.org/10.1784/
Ruther, R., Nascimento, L.R.do., Campos, R.A., 2017. Performance assessment issues in 204764212800028842.
utility-scale photovoltaics in warm and sunny climates. Renew. Energy Environ. Tsanakas, J.A., Ha, L., Buerhop, C., 2016. Faults and infrared thermographic diagnosis in
Sustain. 2, 35. https://doi.org/10.1051/rees/2017035. operating c-Si photovoltaic modules: A review of research and future challenges.
Sizkouhi, A.M., Aghaei, M., Esmailifar, S.M., Mohammadi, M.R., 2020. Automatic Renew. Sustain. Energy Rev. 62, 695–709. https://doi.org/10.1016/j.
Boundary Extraction of Large-Scale Photovoltaic Plants using a Fully Convolutional rser.2016.04.079.
Network on Aerial Imagery. IEEE J. Photovoltaics. https://doi.org/10.1109/ Jahn, Ulrike, Herz, M., Köntges, M., Parlevliet, D., Paggi, M., Tsanakas, I., Stein, J.S.,
JPHOTOV.2020.2992339. Berger, K.A., Ranta, S., French, R.H., Richter, M., Tanahashi, T., 2018. Review on
Tsanakas, I., Botsaris, P.N., 2013. On the detection of hot spots in operating photovoltaic Infrared and Electroluminescence Imaging for PV Field Applications. IEA PVPS.
arrays through thermal image analysis. Mater. Eval. 71, 457–465.

724

View publication stats