energies

Article
Fault Detection and Power Loss Assessment for Rooftop
Photovoltaics Installed in a University Campus, by Use of
UAV-Based Infrared Thermography
Kyoik Choi and Jangwon Suh *

Department of Energy and Mineral Resources Engineering, Kangwon National University,

Samcheok 25913, Republic of Korea; [email protected]
* Correspondence: [email protected] or [email protected]; Tel.: +82-33-570-6313

Abstract: In contrast to commercial photovoltaic (PV) power plants, PV systems at universities

are not actively monitored for PV module failures, which can result in a loss of power generation.
In this study, we used thermal imaging with drones to detect rooftop PV module failures at a
university campus before comparing reductions in power generation according to the percentage of
module failures in each building. Toward this aim, we adjusted the four factors affecting the power
generation of the four buildings to have the same values (capacities, degradations due to aging,
and the tilts and orientation angles of the PV systems) and calibrated the actual monthly power
generation accordingly. Consequently, we detected three types of faults, namely open short-circuits,
hot spots, and potential-induced degradation. Furthermore, we found that the higher the percentage
of defective modules, the lower the power generation. In particular, the annual power generation of
the building with the highest percentage of defective modules (12%) was reduced by approximately
25,042 kWh (32%) compared to the building with the lowest percentage of defective modules (4%).
The results of this study can contribute to improving awareness of the importance of detecting and
maintaining defective PV modules on university campuses and provide a useful basis for securing
the sustainability of green campuses.

Citation: Choi, K.; Suh, J. Fault

Keywords: module fault; thermal infrared thermography; rooftop photovoltaic; green campus;
Detection and Power Loss sustainability
Assessment for Rooftop
Photovoltaics Installed in a
University Campus, by Use of
UAV-Based Infrared Thermography. 1. Introduction
Energies 2023, 16, 4513. https:// Several types of defects have been found in photovoltaic (PV) modules which reduce
doi.org/10.3390/en16114513
the module power output, preventing modules from reversing to normal behavior and
Academic Editor: Luigi Vesce causing safety issues [1]. The most common panel failures are delamination, loss of
backsheet adhesion, bad junction boxes, broken frames, ethyl vinyl acetate discoloration,
Received: 18 April 2023
cell cracks, snail marks, burn marks, potential-induced degradation (PID), breaks in cell-
Revised: 22 May 2023
and string-interconnect ribbons, defective bypass diodes, micro-arcing in connectors, shunt
Accepted: 2 June 2023
hot spots, broken front glass, and degraded back contact electrodes [2]. Product failures
Published: 4 June 2023
are generally divided into three categories: infant failure, midlife failure, and wear out.
Initial failures generally occur at the beginning of a panel’s operational life, with the most
important failures in the field being junction box failures, glass breakage, cell interconnect
Copyright: © 2023 by the authors. defects, frame looseness, and delamination. Additionally, mid-panel failures show that the
Licensee MDPI, Basel, Switzerland. incidence of defective interconnects within modules is significantly higher, whilst panel
This article is an open access article failures due to glass breakage are also high. The relative failure rates of the junction boxes
distributed under the terms and and cables, cell burn marks, and packaging materials were all relatively high. Aging
conditions of the Creative Commons failures typically occur at the end of a panel’s operational life and determine its maximum
Attribution (CC BY) license (https:// operational life. The main failures of the modules were delamination, separation of cell
creativecommons.org/licenses/by/ pieces owing to cell cracking, and discoloration of the laminate [3,4].
4.0/).

Energies 2023, 16, 4513. https://doi.org/10.3390/en16114513 https://www.mdpi.com/journal/energies

Energies 2023, 16, 4513 2 of 16

To ensure the sustainability of PV power generation facilities, it is necessary to mon-

itor and minimize the loss of power generation due to defective modules. In the case of
commercial PV power plants, continuous monitoring, inspection, and maintenance are
actively carried out since they are operated for maximizing power generation. Conversely,
detection and maintenance of module failures are not sufficient in schools and public
institutions that do not have a commercial purpose for PV power generation systems since
these require significant time and costs [5]. Rooftop PV systems are installed under the
mandatory renewable energy portfolio standard in the Republic of Korea, which requires
that public institutions and local governments obtain a certain percentage of their energy
from new and renewable energy sources (32% or more by 2022) in buildings with an area
of 1000 m2 or more that are newly built, expanded, or renovated [6]. Moreover, there are
many examples of green campuses worldwide that have PV systems (e.g., South Korea [5],
Indonesia [7], Japan [8], India [9], Nepal [10], China [11], Vietnam [12], Pakistan [13],
USA [14], Mexico [15], Brazil [16], Italy [17], and Nigeria [18]). Therefore, it is necessary
to detect defects in PV modules through periodic inspection and monitoring and to pre-
vent power generation degradation through the maintenance of PV systems installed on
university campuses.
Researchers worldwide have conducted studies on the detection of defective PV mod-
ules. The International Energy Agency’s Photovoltaic Power Systems Program summarizes
the types and details of PV module failures/faults described in infrared thermography.
This enables the determination of the type of defect or risk from a thermal image of a PV
module. Unmanned aerial vehicles (UAVs) have often been used to monitor PV plants
at a local scale (<1 km2 ) [19–27]. Several studies have been proposed aiming to automati-
cally derive the line or region of PV modules using thermal imaging drones [19–21] based
on computer vision technology, evaluate high-resolution UAV imagery for overhauling
large-scale PV power plants [22–24], or analyze the suitability of irradiation according
to the tilt angle of PV modules in drone-based PV power plant monitoring [25]. Many
studies have previously reported the use of drones equipped with only thermal imaging
cameras [26–31] or both thermal and optical imaging cameras [32–35] to inspect defects of
PV modules. More recently, research has been reported that adopts artificial intelligence
(AI) technology (e.g., convolutional neural networks, naive Bayes classifiers) in detecting
or monitoring PV module failures. The combined use of drone-based thermal infrared
imaging and AI technology (i.e., machine learning and deep learning) has the advantage of
not only reducing the time and cost of monitoring large-scale PV power plants, but also
detecting defects automatically with high accuracy [36–41]. However, the majority of these
studies were limited to detecting inspections or defects in PV power generation systems,
whilst there remains a lack of studies analyzing the actual power generation degradation
resulting from PV module defects. Considering that the detection of PV module failures is
aimed at reducing power generation losses through maintenance, it is necessary to conduct
a study that can detect PV module failures in addition to analyzing the amount of power
generation degradation these module failures have caused.
The objective of this study is to analyze reductions in power generation for campus
rooftop photovoltaics through module faults detected using UAV-based thermal infrared
images. Toward this aim, we quantitatively analyze the amount of power generation
loss caused by passive maintenance after detecting defects in the rooftop PV systems
of each building on a university campus and evaluate the types and rates of module
defects, using drone-based thermal images to determine the causes. This study differs
from existing research in that it can contribute to improving awareness of the need for the
active maintenance of energy infrastructure to create a sustainable green campuses under
challenging circumstances for university campuses which lack necessary funding.

2. Study Area
The Samcheok Campus of Kangwon National University was selected as the research
area for this study to analyze the defect detection and power generation degradation of
Energies 2023, 16, 4513 3 of 16

PV modules by using thermal imaging drones. Kangwon National University Samcheok

Campus is located in Samcheok City, Gangwon-do, South Korea, at 37◦ 270 07.900 N latitude
and 129◦ 090 42.800 E longitude.
Originally, six of the 26 buildings on campus were equipped with rooftop PV modules
(silicon-based). However, considering the quantity and quality of the power generation
data for each building monitored here, only four buildings (named building E5, building
E4, building J, and building G) were selected for analysis, as shown in Figure 1. The
installation time (for calculating aging), design capacity, performance degradation, tilt
angle, and azimuthal angle of each facility all differed significantly (Table 1). The PV
facilities in the study area have been in operation for a minimum of one year and a
maximum of 15 years, with design capacities ranging from 50 to 100 kWP . In case of
module performance degradation, they start out with 96–98% performance in the first year
of operation and performance decreases by 0.3–0.8%p per year as they age. The module
performance change through aging is an internal characteristic that is present when the
module is produced by its manufacturer. The performance degradation formula for the
modules installed in each building based on the operating period is shown in Table 1. The
four engineering buildings were composed of four strings with two types of tilt angles. The
remaining three buildings, except for building G, had an azimuth angle of 142◦ because the
PV modules were installed parallel to the building’s rooftop boundary instead of facing
south. In the case of the rooftop PV modules of building G, the installation tilt angle was
14◦ instead of 25◦ (optimal). This is because the effect of the shadow from the front array
installed on the rooftop on the rear array was minimized and the design of the structure
was problematic. Details on the specifications of the PV modules for each building are
summarized in Table 2.
Table 1. Information on the rooftop PV systems at Kangwon National University.

Building Rooftop PV System

Name Operation Type Capacity (kWp ) Degradation (%) * Tilt (◦ ) Azimuth (◦ )
E5 December 2018~present 100.00 (=400 W × 250 EA) 98.5 − 0.5 × age (yr) 23 142
E4 September 2021~present 70.84 (=460 W × 154 EA) 96 − 0.67 × age (yr) 25, 15 142
G January 2020~present Fixed 67.60 (=400 W × 169 EA) 98 − 0.5 × age (yr) 14 180
J Jan 2008~present 50.00 (=173 W × 290 EA) 97 − 0.8 × age (yr) 25 142
* This does not refer to photoelectric efficiency, but rather to performance over age.

Table 2. Specifications of the PV modules installed at Kangwon National University.

Module Building Name

Specification E5 E4 G J
Model LG400N2W-A5 Q.PEAK DUO XL-G9.3 460 LG400N2W-A5 SE-S173
Rated maximum power (Pmax ) [W] 400 460 400 173
Open circuit voltage (Voc ) [V] 49.3 53.66 49.3 43.9
Maximum power voltage (Vmp ) [V] 40.6 45.44 40.6 36.6
Short circuit current (Isc ) [A] 10.47 10.63 10.47 5.13
Maximum power current (Imp ) [A] 9.86 10.12 9.86 4.73
Efficiency [%] 19.3 20.6 19.3 13.9

Since the four buildings were spaced apart and had similar rooftop elevations, the
PV modules were not affected by shadows from the surrounding terrain or buildings.
Furthermore, there are structures such as roof entrances in addition to PV modules on the
rooftops of buildings E5, J, and G. However, in the case of buildings E5 and J, the shadow
effect of the module was insignificant because the structure is located to the north of the
module. In the case of building G, there is a structure on the right, but there is greater
distance from the array on the left (which can affect shadows), so the shadow effect was
minimized. Therefore, it is considered that the shadow effect on the rooftop PV in this
study area is very low. Currently, the rooftop PV facilities are operated and maintained by
the Campus Facilities Support Division.
Energies 2023,16,
Energies2023, 16,4513
x FOR PEER REVIEW 44 of
of 16
17

Figure1.1.Aerial
Figure Aerialphotographs
photographsof
ofthe
therooftop
rooftopPV
PVat
atKangwon
KangwonNational
NationalUniversity,
University,Republic
Republicof
ofKorea
Korea
(image from: http://www.google.com/maps (accessed on 1 June 2023)).
(image from: http://www.google.com/maps (accessed on 1 June 2023)).

3.Table 1. Information on the rooftop PV systems at Kangwon National University.

Methods
Building In this study, we analyzed the defect
Rooftop detection and power generation degradation of PV
PV System
Name Operation modules using
Type a thermal imaging
Capacity (kWp)drone, following the procedure
Degradation (%) * shown
Tilt (°) in Figure
Azimuth2. Firstly,
(°)
thermal images of the PV modules were acquired using a drone thermal imaging camera,
E5 December 2018~present 100.00 (=400 W × 250 EA) 98.5 − 0.5 × age (yr) 23 142
whilst suspected defective modules were detected by considering the relative temperature
E4 September 2021~present 70.84 (=460 W × 154 EA) 96 − 0.67 × age (yr) 25, 15 142
Fixed
difference characteristics and patterns of the modules. In addition, we adjusted the design
G January 2020~present 67.60 (=400 W × 169 EA) 98 − 0.5 × age (yr) 14 180
capacity, the amount of degradation of the modules’ power generation efficiency over
J Jan 2008~present 50.00 (=173 W × 290 EA) 97 − 0.8 × age (yr) 25 142
the operating period, and the tilt and azimuth characteristics of the PV power generation
* This does not refer to photoelectric efficiency, but rather to performance over age.
 3. Methods
In this study, we analyzed the defect detection and power generation degradation of
PV modules using a thermal imaging drone, following the procedure shown in Figure 2.
Firstly, thermal images of the PV modules were acquired using a drone thermal imaging
camera, whilst suspected defective modules were detected by considering the relative
Energies 2023, 16, 4513 temperature difference characteristics and patterns of the modules. In addition, we ad- 5 of 16
justed the design capacity, the amount of degradation of the modules’ power generation
efficiency over the operating period, and the tilt and azimuth characteristics of the PV
power generation
facilities facilities
in the four in the
buildings tofour buildings
be the to be the
same whilst alsosame whilst also
calibrating the calibrating the
monthly power
monthly power generation accordingly. We then analyzed the impacts
generation accordingly. We then analyzed the impacts of defective PV modules on of defective PVthe
modules on the degradation of power generation by comparing the percentage
degradation of power generation by comparing the percentage of defective core modules of defec-
tive core buildings
between modules between
and the buildings and the power
power generation generation
correction values.correction values.

Figure 2.
Figure 2. Flowchart
Flowchart for
for the
the analysis
analysisof
ofdefective
defectivemodules
modulesand
andpower
powergeneration
generationreduction.
reduction.

3.1.
3.1. Detection
Detection of of PV
PV Module
ModuleFailures
Failureswith
withaaThermal
ThermalImaging
ImagingDrone
Drone
In
In this
this study,
study,thetheMAVIC
MAVIC22Enterprise
EnterpriseAdvanced
Advanced(hereafter
(hereafterMAVIC
MAVIC2)2) drone
drone developed
devel-
by
opedDabyJiang Innovation
Da Jiang (DJI)(DJI)
Innovation waswasused to acquire
used to acquirethethe
thermal
thermalcharacteristics
characteristicsof of rooftop
roof-
PV
topmodules.
PV modules. TheTheMAVICMAVIC 2 drone
2 dronehashas
advantages
advantages such
such asas
anan
integrated
integratedhigh-resolution
high-resolu-
thermal imaging
tion thermal sensorsensor
imaging used to quickly
used recognize
to quickly objectsobjects
recognize in the field.
in theItfield.
is alsoItequipped
is also
with a visible light camera and a thermal imaging camera on a stabilized
equipped with a visible light camera and a thermal imaging camera on a stabilized three- three-axis gimbal
to provide visible and infrared images simultaneously (Table
axis gimbal to provide visible and infrared images simultaneously (Table 3). 3).
Standards
Standards from the the International
InternationalElectrotechnical
ElectrotechnicalCommission
Commission(IEC) (IEC) were
were consulted
consulted
for
for the
theacquisition
acquisitionand andanalysis
analysisofofthe
thethermal
thermalimages
imagesofofthethephotovoltaic
photovoltaic modules
modules [42]. IEC
[42].
TS
IEC62446-3:2017
TS 62446-3:2017 (E) defines the the
(E) defines requirements
requirements forfor
thethe
outdoor
outdoor thermal
thermal imaging
imaginginspection
inspec-
of photovoltaic
tion of photovoltaicpower plants
power in operation.
plants in operation.Thermal
Thermal images
imageswith
witha aspatial
spatialresolution
resolution of
3 cm/pix were acquired in accordance with the IEC standards to obtain sufficient samples
for the measured temperatures for all PV cells. Weather conditions such as temperature,
humidity, wind, precipitation, and sunlight were all considered when selecting a drone
flight date; flights were conducted at approximately 10 a.m. and 3 p.m., avoiding times
such as 2 h after sunrise (low temperatures reduce accuracy), midday (strong light causes
reflections), and 2 h before sunset (shadow issues).
For stable and standardized data collection, we used an autoflight function supported
by the Pix4D capture application to fly and capture the drone. The drone-based images
were taken from a height of 30 m above the altitude of each building’s rooftop PV modules
in order to clearly distinguish the temperature differences between cells within a module
and temperature distributions within cells. The surface emissivity was set to a typical
value of 0.9. In addition, the temperature measurement range of each building’s rooftop
PV modules was set automatically, as each type of PV module failure has a different
temperature distribution and pattern.
To detect PV module failures, characterization of the relative temperature distribution
or pattern within the module as seen in the thermal image was analyzed. For this, various
types of PV module failures presented in the photovoltaic failure fact sheet (PVFS) [1] of
the Photovoltaic Power Systems Program of the International Energy Agency (IEA PVPS)
Task 13 were compared and analyzed using thermal images taken in the study area.
Energies 2023, 16, 4513 6 of 16

Table 3. Specifications of the DJI Mavic 2 Enterprise Advanced [43].

Type Features Performance

Dimensions (L × W × H) Folded: 214 × 91 × 84 mm3
Takeoff weight
Unfolded: 322 × 242 × 84 mm3
(Without accessories)
Aircraft 72 kph
Max. speed
(S-mode, without wind)
31 min
Max. flight time
(measured while flying at 25 kph in windless conditions)
Sensor resolution 640 × 512 @30Hz
M2EA Thermal Camera Digital zoom 16×
Accuracy of thermal temperature Measurement: ±2 ◦ C or ±2%, whichever is greater.

3.2. Comparative Analysis of Reduction in Power Generation by Building

At the Kangwon National University Samcheok Campus, the rooftop PV system
monitors power generation from the time of operation. As mentioned above, building
J has been operating for approximately eleven years (January 2012 to December 2022),
while building E4 has approximately one year of data (2022). Therefore, to compare the
power generation degradation in each building due to the failure rate of PV modules, this
study used the monthly power generation from 2022, for which data are available for all
buildings, as a reference value.
To compare the power generation degradation owing to PV module failure, the
weather conditions and system design factors (characteristic values) must be the same for
all other conditions. In the case of the study area, since the area was within 0.3 km2 , the
weather conditions can be assumed to be the same, while characteristic values such as the
capacity of the PV system by building, the module power generation efficiency according
to the operating period, and the tilt and azimuth angles were different (Table 1).
The PV installation capacity was proportional to the power generation. For example,
all else being equal, the 50 kWp facility will generate half as much electricity as the 100 kWp
facility. Considering that silicon-based PV modules generally lose efficiency at a rate of
approximately 0.3–0.8% per year, it is necessary to adjust the power-generation efficiency
of PV modules in buildings with different installation years to be the same. We obtained
the annual degradation rate data for each module by contacting the module manufacturer
in the Republic of Korea. For example, modules installed in building E5 had a first-year
power generation performance of 98.5%, and after four years of operation (as shown in
Table 1) their performance would be 96.5%. The tilt and azimuth angles can both affect
power generation, depending on the angle of incidence of the sun.
Therefore, in this study, the other conditions (influencing factors) were adjusted
equally, as shown in Table 4, to analyze only the power-generation degradation caused by
the failure of the PV modules. For the entire building, the installed capacity of the modules
was set to 50 kWP , the degradation of the power generation efficiency of the module owing
to the time of installation was set to 0%, the tilt angle was set to 25◦ , and the azimuth angle
was set to 142◦ . Thus, the monthly power generation of each building could compensate
for the changes in the values of the four influencing factor characteristics.

Table 4. Property modification of rooftop PV systems for analysis of the reduction in power output.

Building Capacity (kWP ) Degradation by Year (%) * Tilt (◦ ) Azimuth (◦ )

Name Original Revised Original Revised Original Revised Original Revised
E5 100.00 96.50 23 142
E4 70.84 95.30 25, 15 142
50 100 25 142
G 67.60 96.50 14 180
J 50.00 88.20 25 142
* This does not refer to photoelectric efficiency, but rather to performance over age.
Energies 2023, 16, 4513 7 of 16

4. Results & Interpretations

4.1. Detecting PV Panel Failures Using Thermal Imaging Drones
Figure 3 shows the thermal characteristics of each building at the Kangwon National
University Samcheok Campus as detected using the visible and thermal images from the
drone. Faulty modules detected from the temperature distribution and relative temperature
differences of the PV modules were found in all four buildings. In the case of building E5,
building E4, and building G, hot spots (cells with higher temperatures than the surrounding
cells, indicated by white and yellow areas in the thermal image) and PID failures (single
cells with higher temperatures than the surrounding cells and irregular patterns) were
detected, whereas open shorts (certain strings or modules with higher temperatures than
other strings or modules) were detected in building J. In the case of the study area, the
visible-light images did not show any significant features, making it difficult to determine
the fault, whereas the thermal images showed that the temperature of a particular cell
was higher in comparison to that of the surrounding cells, thus indicating that the module
was faulty.
Such defects in PV modules can cause various failures or decreases in system perfor-
mance and power generation. The hot spots on the rooftop PV cells of the fifth and fourth
buildings and building G were caused by cell defects, breakage, or internal connections.
The PVFS reports that hotspot failures could directly cause fires, whilst defects have various
effects on performance. In the case of PIDs, defects can also directly cause fires and are
known to have catastrophic effects on performance. An open short in the rooftop PV cell in
the lab is a case where the module system is not connected, and the fault can have a severe
effect on the performance.
Table 5 shows the number of modules per building, number of faulty modules, and
the ratio of healthy to faulty modules. The highest percentage of defective modules was
12% in building J, followed by 7% in building G, and 4% in buildings E5 and E4. The
management team of the PV facility on the university campus reported that they did not
perform any inspections or maintenance unless power generation had decreased to almost
zero. Considering this situation, it is believed that the high percentage of faulty modules in
building J is due to the fact that it has been operated for the longest period of time (more
than 10 years). However, in the case of building E4, which began operation in September
2021, it was inferred that defects may have occurred early, not long after the start of power
generation.
Defective PV modules can result in reduced power generation. For commercial PV
farms, in which PV power generation is directly tied to revenue, module failure detection
and active maintenance are typically performed. On the other hand, university campuses
strive to create a green campus to comply with government policies and improve their
internal and external image; however, considering that they do not aim to maximize power
generation revenue, they often simply monitor for module failures.

4.2. Comparative Analysis of Power Generation Reduction as a Function of the PV Module Failure
Rate
Table 6 lists the corrected monthly power generation for each building when the four
attribute values were equally adjusted. Building E5, with an installed capacity of 100 kWP ,
generated 11,450 kWh in January. However, if the installed capacity was adjusted to 50 kW,
the corrected January generation was 5725 kWh (50%). The PV modules in building E5 had
an initial power generation efficiency of 98.5% (empirical data for module performance in
the first year of installation provided by the manufacturer), which decreased by 0.5% per
year; after four years, they had a power generation efficiency of 96.5%. Considering that the
modules in the other buildings were of different models and had different power generation
efficiencies, we recalculated the monthly power generation in 2022, assuming a module
efficiency of 100% as of 2022; that is, 100% divided by 96.5% (=100%/96.5%), which was
approximately 1.036, multiplied by the original power generation of 5725 kWh, totaling
5933 kWh. Thus, each building’s monthly generation was multiplied by the capacity
Energies 2023, 16, 4513 8 of 16

Energies 2023, 16, x FOR PEER REVIEW 8 of 17

correction factor and the generation efficiency reduction correction factor to calculate the
corrected generation.

Figure 3. Photos
Figure of the
Photos of defective PV modules
the defective of each building
PV modules of eachtaken using drone-based
building taken usinginfrared
drone-based
thermography.
infrared thermography.
Energies 2023, 16, 4513 9 of 16

Table 5. Number and proportion of defective modules of each building.

No. of Module Operation Period

Building Name Failure Type Proportion (%)
All Defective (Year)
E5 264 12 4 4
Hot spot,
E4 154 7 4 1
PID
G 169 13 7 3
J Open short fault 290 36 12 15

Table 6. Modified monthly power generation of each building considering the calibration of different
capacities, degradations, tilts, and azimuths of PV systems.

Original Properties (kWh) Revised Properties (kWh)

Month
E5 E4 G J E5 E4 G J
January 11,450 3488 6068 2924 5933 2604 4706 3408
February 13,467 5807 6527 4518 6978 4336 5062 5266
March 13,305 8598 4281 3581 6894 6421 3320 4174
April 15,726 10,658 9726 3694 8148 7959 7543 4305
May 15,221 11,512 10,456 4451 7887 8597 8109 5188
June 13,772 8911 8374 4470 7136 6655 6494 5210
July 13,154 8097 7577 3395 6815 6047 5876 3957
August 11,155 6892 6369 3281 5780 5146 4939 3824
September 12,636 8016 6850 4088 6547 5986 5312 4765
October 10,600 6320 4789 3841 5492 4720 3714 4477
November 9460 6235 4340 2814 4901 4656 3365 3280
December 9913 6139 5006 4079 5136 4585 3883 4754
Sum 149,860 90,673 80,361 45,136 77,648 67,710 62,324 52,606
Modified properties: capacity (50 kWP ), degradation (100%), tilt (25◦ ), azimuth (142◦ ).

As listed in Table 4, the tilt angle of the rooftop PV modules of each building was
adjusted to 25◦ , whilst the azimuth angle was adjusted to 142◦ , and the power generation
was corrected accordingly. Power generation correction by adjusting the tilt angle and
azimuth angle is not as simple as power generation correction by reducing the facility
capacity and degradation by aging. In this study, SAM (system advisor model) software was
used to predict power generation under the original conditions (tilt angle and orientation
angle) and then this was compared with the power generation under the revised conditions
to derive the coefficient of determination (R2 ) [44] and the trend equation (Figure 4) between
two power generation value sets. For example, in the case of the power generation of
building G, the coefficient of determination for two power generation sets under the original
properties (X14–180 ) and revised properties (X25–142 ) was calculated to be 0.9997, indicating
that the accuracies of predictions of the actual value (Y14–180 ) were reliable.
Next, using SAM software, the module tilt angle of building G was modified and
set to 25◦ with an orientation angle of 142◦ to predict the power generation (X25–142 ).
Additionally, the corrected monthly power generation (Y25–142 = 1.0086 × X25–142 − 0.0257)
was calculated by applying the trend equation derived earlier. Thus, power generation
due to tilt and orientation angle adjustments was corrected for all buildings. The intent
is to convey that if the predicted and measured power generation of the PV modules
at a given tilt and orientation angle are highly correlated, it is reasonable to extrapolate
from the predicted value in areas where there are no measured values. However, in the
case of building E4, the inclination angle of each string was divided between 25◦ and 15◦ .
Therefore, the coefficient of determination and trend equation were obtained in the same
manner as above for the string with an inclination angle of 15◦ and calibrated to a level of
power generation at 25◦ .
 tation angle) and then this was compared with the power generation under the revised
conditions to derive the coefficient of determination (R2) [44] and the trend equation (Fig-
ure 4) between two power generation value sets. For example, in the case of the power
generation of building G, the coefficient of determination for two power generation sets
under the original properties (X14–180) and revised properties (X25–142) was calculated to be
Energies 2023, 16, 4513 10 of 16
0.9997, indicating that the accuracies of predictions of the actual value (Y14–180) were relia-
ble.

4. Correlation
Figure 4. Correlationanalysis
analysisbetween
betweenobserved
observed andand predicted
predicted power
power generation
generation using
using SAMSAM soft-
software
ware
for for buildings
buildings G andGE4and (forE4 (for evaluating
evaluating the reliability
the reliability of modified
of modified powerpower generation
generation according
according to the
to the unification
unification of the
of the tilt andtilt and azimuth
azimuth of PV modules).
of PV modules).

The change in the PV power generation of the four buildings according to the adjusted
values of the capacity, degradation, tilt, and orientation angle of the rooftop PV systems
(presented in Table 4) are shown in Figure 5. It was found that in the first step, the installed
capacities of buildings E5, E4, and G were changed to 50 kWp , and the power generation
decreased accordingly. On the other hand, for building J, the original capacity and the
adjusted capacity were identical, so there was no change in power generation. In the second
step, degradation of all buildings’ PV systems was adjusted to 100%. Thus, the power
generation of all buildings was changed. In the last step, power generation of buildings G
and E4 was partially adjusted because only the tilt and orientation angle were adjusted.
Figure 6 shows the percentage of faulty modules for each building in the study area
alongside the calibrated 2022 generation under the same conditions. Building E5, which
had the lowest proportion of faulty modules, generated the most power, whereas building
J, which had the highest proportion of faulty modules, generated the least. That is, the
proportions of faulty modules and power generation were inversely related, and it could
be clearly observed that faulty modules had an impact on power generation.
Under identical conditions for all four rooftop PV systems, the power generation of
buildings E4 and E5 with the same 4% of faulty modules should be the same. However,
the power generation of E5 is higher than that of E4. To interpret the reason for this,
a comparison of the calibrated power generation for buildings E5 and E4 is shown by
in Figure 7, which includes data from the end of 2021 to the end of 2022 when rooftop
PV power values for building E4 began to be monitored. The power generation graphs
followed a similar pattern for approximately 13 months, except for January and February
2022. We concluded that the calibrated generation data were reliable since patterns of
power generation were similar under the same conditions (i.e., capacity, degradation, and
tilt and azimuth angles). The reason for the difference in the actual generation, despite the
same 4% defective panel rate, was attributed to the missing data for the first two months of
2022 for building E4.
 the installed capacities of buildings E5, E4, and G were changed to 50 kWp, and the power
generation decreased accordingly. On the other hand, for building J, the original capacity
and the adjusted capacity were identical, so there was no change in power generation. In
the second step, degradation of all buildings’ PV systems was adjusted to 100%. Thus, the
power generation of all buildings was changed. In the last step, power generation of build-
Energies 2023, 16, 4513 11 of 16
ings G and E4 was partially adjusted because only the tilt and orientation angle were ad-
justed.

Energies 2023, 16, x FOR PEER REVIEW 12 of 17

proportions of faulty modules and power generation were inversely related, and it could
be clearly observed that faulty modules had an impact on power generation.
Changeinin
Figure5.5.Change
Figure power
power generation
generation over
over 2022
2022 forfor each
each PVPV system
system after
after unifying
unifying four
four conditions.
conditions.

Figure 6 shows the percentage of faulty modules for each building in the study area
alongside the calibrated 2022 generation under the same conditions. Building E5, which
had the lowest proportion of faulty modules, generated the most power, whereas building
J, which had the highest proportion of faulty modules, generated the least. That is, the

Figure Comparisonof
6. Comparison
Figure 6. ofdefective
defectivepanel
panelratios
ratiosand
andpower
powergeneration
generation reductions
reductions forfor each
each building.
building.

Under identical conditions for all four rooftop PV systems, the power generation of
buildings E4 and E5 with the same 4% of faulty modules should be the same. However,
the power generation of E5 is higher than that of E4. To interpret the reason for this, a
comparison of the calibrated power generation for buildings E5 and E4 is shown by in
Figure 7, which includes data from the end of 2021 to the end of 2022 when rooftop PV
power values for building E4 began to be monitored. The power generation graphs fol-
lowed a similar pattern for approximately 13 months, except for January and February
2022. We concluded that the calibrated generation data were reliable since patterns of
Energies 2023,
Energies 16,16,
2023, x FOR
4513 PEER REVIEW 13 of 1716
12 of

Figure 7. Comparison of the monthly power generation of building E5 and building E4.
Figure 7. Comparison of the monthly power generation of building E5 and building E4.
5. Discussion
5. Discussion
Figure 8 shows the four-year calibrated power generation for the buildings with the
Figure
highest 8 shows the
percentage four-year
of failed calibrated
PV modules power generation
(building for the percentage
J) and the lowest buildings with the
(building
highest percentage
E5). From the end of
of failed
2018 toPV modulesthe
mid-2021, (building
analysisJ)showed
and thethat
lowest percentage
the two (building
buildings exhibited
E5). Frompatterns
similar the endofofvariation
2018 to mid-2021,
in power the analysiswith
generation, showed that the two
the differences inbuildings exhib-
power generation
ited similar
being patterns This
negligible. of variation
suggestedin power
that thegeneration, with the
module failure differences
rates of both in power gen-
buildings were
eration
similar.being negligible.
However, after This suggested
the end of 2021,thatthe the
twomodule
buildingsfailure rates some
exhibited of both buildingsin
similarities
were
theirsimilar.
powerHowever,
generationafter the end
patterns, of 2021, the two
although buildings
difference exhibited
in power some similarities
generation was clearly
invisible.
their power generation patterns,
For approximately one yearalthough
(Augustthe 2021difference
to December in power
2022),generation wasin
the difference
Energies 2023, 16, x FOR PEER REVIEW 14 of 1
clearly
powervisible. For approximately
generation between buildingone year
J and(August 2021 to
the building E5December
was 33,292 2022),
kWh.the difference
This suggests
inthat
power generation
a new between
(or additional) building
module J and
failure the
will building
occur by theE5endwasof33,292
2021 inkWh. Thislab.
a shared sug-
gests that a new (or additional) module failure will occur by the end of 2021 in a shared
lab.

Figureof
Figure 8. Comparison 8. monthly
Comparison of monthly
power output power output
between between
building building
E5 and E5 and
building building J (2018–2022).
J (2018–2022).

It is difficult to determine exactly when a PV panel will fail until inspecting it. O
course, if solar irradiation and high temporal resolution monitoring data are available, i
is possible to detect whether the system has failures or defects by analyzing its powe
generation values or electrical characteristics. However, in the case of university cam
Energies 2023, 16, 4513 13 of 16

It is difficult to determine exactly when a PV panel will fail until inspecting it. Of
course, if solar irradiation and high temporal resolution monitoring data are available,
it is possible to detect whether the system has failures or defects by analyzing its power
generation values or electrical characteristics. However, in the case of university campuses,
which do not have commercial power plants, there are no high-resolution monitoring
systems and periodic PV system inspections are not usually performed because of time and
cost issues. Therefore, it is necessary to detect direct and indirect failures and defects of the
PV module to reduce the period in which power generation is left in a degraded state.

6. Conclusions
In this study, we detected defective modules using thermal infrared images of rooftop
PV modules on a university campus acquired using a thermal imaging drone. We sub-
sequently compared and analyzed the level of power generation degradation due to the
percentage of defective modules per building. Hot spots, PID, and open short-circuit faults
that could not be identified through visual analysis were detected in the infrared images.
The percentage of defective modules per building in the study area ranged between 4% and
12%, whilst the amount of power generation degradation increased with the percentage of
suspected defective modules. In addition, a comparative analysis of the changes in power
generation for buildings within the same environment allowed us to estimate when module
failures may occur.
The methodology proposed in this study was insufficient for identifying the exact time
of PV panel failure. To closely analyze the extent of the power generation degradation of
rooftop PV facilities in each building, and its causes, accurate data on the timing of module
failure are required, and further research is still needed. In this regard, solar irradiation
and high-resolution power generation data are necessary for the detailed detection of PV
module failures. In addition, there is a difference between theoretical panel degradation
and actual panel degradation. This is because theoretical degradation values for modules
are needed when predicting power generation for PV systems that have not yet been
installed. Therefore, further research is needed for this also. Moreover, various studies have
reported that AI technology detects defects in PV modules with higher accuracy. Therefore,
it is necessary to apply AI technology to rooftop PV monitoring on university campuses in
the future.
Although no new technology (i.e., machine learning or deep learning) has been de-
veloped herein, the drone-based thermal infrared imagery and power generation data
analysis technique proposed in this study could detect rooftop PV module failures and
provide a proper rationale for maintenance planning. It should be noted that this does not
indicate that defect identification in PV modules installed on university campuses should
be performed with different procedures than for commercial PV plants. Furthermore, this
study can provide a useful basis for minimizing power-generation losses owing to poor
maintenance. On the other hand, university campuses strive to create green campuses
to respond to government policies whilst improving their internal and external image;
however, since they do not aim to maximize power generation revenue, they often simply
monitor panel failures. From this perspective, the results of this study can contribute to
improving awareness of the need for the active maintenance of energy infrastructure to
create sustainable green campuses.

Author Contributions: J.S. designed the research; K.C. gathered and processed data; All the authors
interpreted the data and wrote the paper; J.S. implemented the final work for the paper submission.
All authors have read and agreed to the published version of the manuscript.
Energies 2023, 16, 4513 14 of 16

Funding: This research was supported by (1) “Regional Innovation Strategy (RIS)” through the
National Research Foundation of Korea (NRF) funded by the Ministry of Education (MOE) (2022RIS-
005), (2) the Korea Institute of Energy Technology Evaluation and Planning (KETEP) and the Ministry
of Trade, Industry & Energy (MOTIE) of the Republic of Korea (No. 20224000000080), and (3) Basic
Science Research Program through the National Research Foundation of Korea (NRF) funded by the
Ministry of Education (2019R1I1A3A01062541).
Data Availability Statement: Data sharing not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations

The following abbreviations are used in this manuscript.

PV photovoltaic
PID potential-induced degradation
UAVs unmanned aerial vehicles
DJI Da Jiang Innovation
IEC International Electrotechnical Commission
PVFS photovoltaic failure act sheet
SAM system advisor model

References
1. Köntges, M.; Kurtz, S.; Packard, C.E.; Jahn, U.; Berger, K.A.; Kato, K.; Friesen, T.; Liu, H.; Van Iseghem, M. Review of Failures of
Photovoltaic Modules; Report IEA-PVPS T13-01:2014; International Energy Agency: Paris, France, 2014.
2. Khalil, I.U.; Ul-Haq, A.; Mahmoud, Y.; Jalal, M.; Aamir, M.; Ahsan, M.U.; Mehmood, K. Comparative Analysis of Photovoltaic
Faults and Performance Evaluation of Its Detection Techniques. IEEE Access 2020, 8, 26676–26700. [CrossRef]
3. Köntges, M.; Oreski, G.; Jahn, U.; Herz, M.; Hacke, P.; Weiß, K.A. Assessment of Photovoltaic Module Failures in the Field; Report
IEA-PVPS T13-09:2017; International Energy Agency: Paris, France, 2017.
4. Herz, M.; Friesen, G.; Jahn, U.; Köntges, M.; Lindig, S.; Moser, D. Quantification of Technical Risks in PV Power Systems; Report
IEA-PVPS T13-23:2021; International Energy Agency: Paris, France, 2021.
5. Seo, H.; Suh, J. Optimal layout design according to potential assessment of rooftop photovoltaic system on southeast-facing
buildings for establishing green campus. J. Korean Soc. Miner. Energy Resour. Eng. 2020, 57, 600–608. [CrossRef]
6. Nematollahi, O.; Kim, K.C. A feasibility study of solar energy in South Korea. Renew. Sustain. Energy Rev. 2017, 77, 566–579.
[CrossRef]
7. Thaib, R.; Amudy, M.; Rizal, T.A. Study on Implementation of Rooftop Photovoltaic Power Plant as an Effort to Attain a
Sustainable Campus. Eur. J. Eng. Technol. Res. 2019, 4, 21–25. [CrossRef]
8. Teah, H.S.; Yang, Q.; Onuki, M.; Teah, H.Y. Incorporating External Effects into Project Sustainability Assessments: The Case of a
Green Campus Initiative Based on a Solar PV System. Sustainability 2019, 11, 5786. [CrossRef]
9. Baitule, A.S.; Sudhakar, K. Solar powered green campus: A simulation study. Int. J. Low-Carbon Technol. 2017, 12, 400–410.
[CrossRef]
10. Yadav, B.K.; Rauniyar, P.K.; Sudhakar, K.; Bajracharya, T.R.; Priya, S.S. Sustainable green campus in NEPAL: 3E analysis. Int. J.
Low-Carbon Technol. 2021, 16, 531–542. [CrossRef]
11. Zhu, X.; Lv, Y.; Bi, J.; Jiang, M.; Su, Y.; Du, T. Techno-Economic Analysis of Rooftop Photovoltaic System under Different Scenarios
in China University Campuses. Energies 2023, 16, 3123. [CrossRef]
12. Pham, M.-H.; Phap, V.M.; Trung, N.N.; Son, T.T.; Kien, D.T.; Anh Tho, V.T. A Study on the Impact of Various Meteorological Data
on the Design Performance of Rooftop Solar Power Projects in Vietnam: A Case Study of Electric Power University. Energies 2022,
15, 7149. [CrossRef]
13. Hua, L.-G.; Memon, Q.A.; Shaikh, M.F.; Shaikh, S.A.; Rahimoon, R.A.; Shah, S.H.H.; Qadir, A. Comparative Analysis of Power
Output, Fill Factor, and Efficiency at Fixed and Variable Tilt Angles for Polycrystalline and Monocrystalline Photovoltaic
Panels—The Case of Sukkur IBA University. Energies 2022, 15, 3917. [CrossRef]
14. Choi, Y.; Oh, M.; Kang, J.; Lutzenhiser, L. Plans and Living Practices for the Green Campus of Portland State University.
Sustainability 2017, 9, 252. [CrossRef]
15. Gálvez, D.M.; Kerdan, I.G.; Carmona-Paredes, G. Assessing the Potential of Implementing a Solar-Based Distributed Energy
System for a University Using the Campus Bus Stops. Energies 2022, 15, 3660. [CrossRef]
16. Cordeiro, I.; Bassi, W.; Sauer, I.L. Hosting Capacity Estimate Based on Photovoltaic Distributed Generation Deployment: A Case
Study in a Campus of the University of São Paulo. Energies 2023, 16, 3934. [CrossRef]
Energies 2023, 16, 4513 15 of 16

17. Di Somma, M.; Buonanno, A.; Caliano, M.; Graditi, G.; Piazza, G.; Bracco, S.; Delfino, F. Stochastic Operation Optimization of the
Smart Savona Campus as an Integrated Local Energy Community Considering Energy Costs and Carbon Emissions. Energies
2022, 15, 8418. [CrossRef]
18. Gbadamosi, S.L.; Ogunje, F.S.; Wara, S.T.; Nwulu, N.I. Techno-Economic Evaluation of a Hybrid Energy System for an Educational
Institution: A Case Study. Energies 2022, 15, 5606. [CrossRef]
19. Roggi, G.; Niccolai, A.; Grimaccia, F.; Lovera, M. A Computer Vision Line-Tracking Algorithm for Automatic UAV Photovoltaic
Plants Monitoring Applications. Energies 2020, 13, 838. [CrossRef]
20. Kim, D.; Youn, J.; Kim, C. Automatic Photovoltaic Panel Area Extraction from UAV Thermal Infrared Images. J. Korean Soc. Surv.
Geod. Photogramm. Cartogr. 2016, 34, 559–568. [CrossRef]
21. Díaz, J.J.V.; Vlaminck, M.; Lefkaditis, D.; Vargas, S.A.O.; Luong, H. Solar panel detection within complex backgrounds using
thermal images acquired by Uavs. Sensors 2020, 20, 6219. [CrossRef]
22. Jung, S.; Jo, Y.; Kim, Y.J. Aerial surveillance with low-altitude long-endurance tethered multirotor UAVs using photovoltaic power
management system. Energies 2019, 12, 1323. [CrossRef]
23. Grimaccia, F.; Aghaei, M.; Mussetta, M.; Leva, S.; Quater, P.B. Planning for PV plant performance monitoring by means of
unmanned aerial systems (UAS). Int. J. Energy Environ. Eng. 2015, 6, 47–54. [CrossRef]
24. Moudrý, V.; Beková, A.; Lagner, O. Evaluation of a high resolution UAV imagery model for rooftop solar irradiation estimates.
Remote Sens. Lett. 2019, 10, 1077–1085. [CrossRef]
25. Kim, K.; Choi, Y.; Shim, K.; Jeon, H.; Commerford, J.; Matson, E.T. Analyzing the range of angles of a solar panel to detect
defective cells, using a UAV. In Proceedings of the Third IEEE International Conference on Robotic Computing (IRC), Naples,
Italy, 25–27 February 2019; pp. 471–476. [CrossRef]
26. Lee, G.S.; Lee, J.J. The Detection of Heat Emission to Solar Cell Using UAV-Based Thermal Infrared Sensor. J. Korean Soc. Geospatial
Inf. Syst. 2017, 25, 71–78. [CrossRef]
27. Henry, C.; Poudel, S.; Lee, S.W.; Jeong, H. Automatic detection system of deteriorated pv modules using drone with thermal
camera. Appl. Sci. 2020, 10, 3802. [CrossRef]
28. Lee, D.H.; Park, J.H. Developing inspection methodology of solar energy plants by thermal infrared sensor on board unmanned
aerial vehicles. Energies 2019, 12, 2928–2941. [CrossRef]
29. Grimaccia, F.; Leva, S.; Niccolai, A. PV plant digital mapping for modules’ defects detection by unmanned aerial vehicles. IET
Renew. Power Gener. 2017, 11, 1221–1228. [CrossRef]
30. Cubukcu, M.; Akanalci, A. Real-time inspection and determination methods of faults on photovoltaic power systems by thermal
imaging in Turkey. Renew. Energy 2020, 147, 1231–1238. [CrossRef]
31. Lee, G.-S. Test of fault detection to solar-light module using UAV based thermal infrared camera. J. Korean Assoc. Geogr. Inf. Stud.
2016, 19, 106–117. [CrossRef]
32. Quater, P.B.; Grimaccia, F.; Leva, S.; Mussetta, M.; Aghaei, M. Light unmanned aerial vehicles (UAVs) for cooperative inspection
of PV plants. IEEE J. Photovolt. 2014, 4, 1107–1113. [CrossRef]
33. Zefri, Y.; Elkettani, A.; Sebari, I.; Lamallam, S.A. Thermal Infrared and Visual Inspection of Photovoltaic Installations by Uav
Photogrammetry—Application Case: Morocco. Drones 2018, 2, 41. [CrossRef]
34. Liao, K.C.; Lu, J.H. Using UAV to detect solar module fault conditions of a solar power farm with IR and visual image analysis.
Appl. Sci. 2021, 11, 1835. [CrossRef]
35. Li, X.; Yang, Q.; Chen, Z.; Luo, X.; Yan, W. Visible defects detection based on UAV-based inspection in large-scale photovoltaic
systems. IET Renew. Power Gener. 2017, 11, 1234–1244. [CrossRef]
36. Akram, M.W.; Li, G.; Jin, Y.; Chen, X.; Zhu, C.; Ahmad, A. Automatic detection of photovoltaic module defects in infrared images
with isolated and develop-model transfer deep learning. Sol. Energy 2020, 198, 175–186. [CrossRef]
37. Wang, X.; Yang, W.; Qin, B.; Wei, K.; Ma, Y.; Zhang, D. Intelligent monitoring of photovoltaic panels based on infrared detection.
Energy Rep. 2022, 8, 5005–5015. [CrossRef]
38. Tsanakas, J.A.; Ha, L.; Buerhop, C. Faults and infrared thermographic diagnosis in operating c-Si photovoltaic modules: A review
of research and future challenges. Renew. Sustain. Energy Rev. 2016, 62, 695–709. [CrossRef]
39. Ali, M.U.; Khan, H.F.; Masud, M.; Kallu, K.D.; Zafar, A. A machine learning framework to identify the hotspot in photovoltaic
module using infrared thermography. Sol. Energy 2020, 208, 643–651. [CrossRef]
40. Niazi, K.A.K.; Akhtar, W.; Khan, H.A.; Yang, Y.; Athar, S. Hotspot diagnosis for solar photovoltaic modules using a Naive Bayes
classifier. Sol. Energy 2019, 190, 34–43. [CrossRef]
41. Starzyński, J.; Zawadzki, P.; Harańczyk, D. Machine Learning in Solar Plants Inspection Automation. Energies 2022, 15, 5966.
[CrossRef]
42. IEC (International Electrotechnical Commission). Available online: https://webstore.iec.ch/publication/28628/ (accessed on
27 February 2023).
Energies 2023, 16, 4513 16 of 16

43. DJI Mavic 2 Enterprise Advanced. Available online: https://www.dji.com/mavic-2-enterprise-advanced/ (accessed on

27 February 2023).
44. Di Bucchianico, A. Coefficient of Determination (R2 ). In Encyclopedia of Statistics in Quality and Reliability; Wiley: Hoboken, NJ,
USA, 2007.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.