Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis

Abstract

Our world has never faced a far graver threat than it does today, the energy crisis. Low lying rural areas have been identified as mostly suffered areas due to non-availability of power. The energy production from solar photovoltaic system has gained attention worldwide as it a clean and renewable source of energy. The power output of solar panel mainly depends on solar irradiance and temperature. But just like many other power generation sources PV systems does come with its fair share of short comings too that reduces the performance and even causes serious safety concerns. This paper covers major faults like open circuit, short circuit and partial shading. These faults are induced in the PV strings and then output characteristics such as open circuit voltage, short circuit current, current and voltage at maximum power point (MPP) are measured using multi-meter and SP lite 2 sensor. The output characteristics are compared to the healthy strings and based on the difference in current (I) and voltage (v) output at maximum power point (MPP) of each fault induced system, faults are detected. The datasets collected for detection is then analyzed in Minitab software using regression analysis and predictive model is generated for each fault. Based on the amount of solar irradiation, surface temperature, ambient temperature, and voltage, current for each fault is predicted. Machine learning methodologies are then used to validate fault detection and prediction. Random Forest classifier is used for fault classification while linear regression is used for current prediction.

Full text

Journal of Mechanical Engineering Research and Developments
ISSN: 1024-1752
CODEN: JERDFO
Vol. 44, No. 11, pp. 34-49
Published Year 2021
34
Detection and Prediction of Faults in Photovoltaic Solar
Panel Using Regression Analysis
Asfandyar Khalid, Naveed Ullah*, Asim Ahmad Riaz*, Muhammad Zeeshan Zahir, Zuhaib
Ali Khan
Department of Mechanical Engineering, University of Engineering & Technology, Peshawar 25000, Pakistan.
*Corresponding author email: naveedullah@uetpeshawar.edu.pk; engr.asim@uetpeshawar.edu.pk
ABSTRACT
Our world has never faced a far graver threat than it does today, the energy crisis. Low lying rural areas have been
identified as mostly suffered areas due to non-availability of power. The energy production from solar
photovoltaic system has gained attention world-wide as it a clean and renewable source of energy. The power
output of solar panel mainly depends on solar irradiance and temperature. But just like many other power
generation sources PV systems does come with its fair share of short comings too that reduces the performance
and even causes serious safety concerns. This paper covers major faults like open circuit, short circuit and partial
shading. These faults are induced in the PV strings and then output characteristics such as open circuit voltage,
short circuit current, current and voltage at maximum power point (MPP) are measured using multi-meter and SP
lite 2 sensor. The output characteristics are compared to the healthy strings and based on the difference in current
(I) and voltage (v) output at maximum power point (MPP) of each fault induced system, faults are detected. The
datasets collected for detection is then analyzed in Minitab software using regression analysis and predictive
model is generated for each fault. Based on the amount of solar irradiation, surface temperature, ambient
temperature, and voltage, current for each fault is predicted. Machine learning methodologies are then used to
validate fault detection and prediction. Random Forest classifier is used for fault classification while linear
regression is used for current prediction.
KEYWORDS
Short circuit fault, Open circuit fault, Partial shading fault, Detection, Prediction.
INTRODUCTION
One of the most sustainable and economically competitive source is solar energy which is renewable and
environmental friendly[1]. Performance of the solar system is greatly affected by certain factors that results in
system energy loss, photovoltaic voltaic module lifespan or even can cause serious safety issues. There are
different types of faults associated with the solar panel which affect its performance. Faults such as open-circuit,
short-circuit and partial shading has a great impact on the solar panel efficiency and even can be hazardous at
times. Fault detection is necessary to avoid system energy loss and safety breaches. Faults detection methods
include detection using visual method, thermal infrared imaging technique, electrical methods, mathematical
model analysis techniques and artificial intelligence and machine learning algorithms and so on. The first and
foremost step towards fault detection is the visual inspection of the PV panels. Mechanical and electrical deficits
are determined and then further tests are conducted if needed. Some of the visual inspection faults are cracks or
gaps, discoloration, burning spots, delamination or dirt points.
Various inspections guides such as the National Renewable Energy Laboratory (NREL/IEA) are now available as
a checklist to detect various faults in the solar panels [2]. The thermal infrared imaging technique detects and
identifies the fault by the help of an infrared IR camera which measures the surface temperature of the heat caused
by the fault. The IR camera captures the image of the solar panel and based on the difference between the faulty
and normal panel temperatures faults are detected. This method mainly focuses on the detection of hot spots in
PV panel [3]. This method is able to detect faults such as fragmentation, broken grids, black pieces and crack that

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
35
are not visible with the naked eye [8]. A system was developed based on infrared image analyses that analyses
and recognizes the working status of PV arrays [9]. The time domain reflectometry technique involves the use of
a pulse signal which is injected into the PV module and then by comparing the input signal with the feedback
output signal, faults are identified [4]. Electrical method involves the use of power loss technique in which output
characteristics are measured in real time and compared to the simulated models to detect faults [5]. In
mathematical model analysis technique, a simulated model is analyzed based on the real time data.
The difference between the simulated model output and real time characteristics is then analyzed and faults are
detected [6]. Artificial Intelligence (AI) is an emerging field and involves the use of Machine learning (ML)
methodologies and artificial neural network (ANN) methods to detect faults in PV array systems. A method based
on ANN was developed to detect short circuit faults and partial shading faults of PV array [10, 11]. A method was
also introduced based on fuzzy logics which distinguishes between the short circuit fault and partial shading fault
[12]. A fault detection method in grid connected system was introduced to detect partial shading and degradation
fault [13]. A data driven fault detection approach was proposed for partial shading fault detection approach [14].
A fault detection approach based on comparison between faulty and its accurate model was proposed [15]. A
statistical approach based on EWMA (exponentially weighted moving average) chart was used to detect partial
shading fault and its effect on the output of solar panel [16].
An online algorithm for fault detection in PV module system was proposed based on multi-class support vector
machine, which can differentiate normal condition, line-to-line fault condition and abnormal degradation fault
condition in real time [17]. A peak current mode (PCM) control method was applied with an excessive ramp signal
for fault detection in DC-DC converter connected to PV solar panel [19]. A fault detection method based on power
loss in PV solar panel was introduced to detect three kinds of faults i-e dusting and soiling, shading and material
faults. An automatic supervision system was developed in MATLAB Simulink to analyze the system [20]. In this
paper, open circuit, short circuit and partial shadings faults are detected and their impact on the current (I) and
voltage (v) output is analyzed. Prediction models are then generated using regression analysis for the same data
sets. Prediction and classification of these faults are then also validated using machine learning methodologies.
EXPERIMENTAL SETUP
The proposed faults are evaluated using the real time data with no resistance or load. The data is collected directly
from the solar string and no involvement of inverter or any other device is carried out.
PV MODULE SPECIFICATIONS
Table 1. Rated output power specifications of the solar panels
Nominal Maximum Power
(Pmax)
30 W
Optimum Operating Voltage
(Vmp)
17.4 v
Maximum Operating Current
(Imp)
1.73 A
Open Circuit Voltage
(Voc)
21.7 v
Short Circuit Current
(Isc)
1.92 A
The PV Installation includes 3 PV strings consisting of 3 series-connected PV modules as shown in the Figure 1.
The characteristic parameters such as current (I), voltage (V), surface temperature of the solar panel (Tsur), solar
irradiation (IR), and ambient temperature (Tamb) are measured using the multi-meter (UC33+) and SP Lite 2
Pyranometer and METEON (Figure 2).

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
36
Figure 1. Solar Panel Installations
Figure 2. Temperature measurement using K-type Thermocouple, UC33+Multimeter & Solar Irradiation sensor
FAULT INDUCTION
At first, Series “A” is induced with short circuit fault and series “B” is induced with open circuit fault while series
“C” remains healthy or fault-free. Both current and voltage are measured at the output of each series along with
the solar irradiation and temperature (both ambient and surface). The 3 series connected panels were partial shaded
with different amount of shading on a single panel from each series. First, one panel in series A was shaded 100%
and one panel in Series B was shaded 50% and series C was un-shaded as shown in the Figure 3 (a). Then, on the
next day one panel in series A was 25% shaded, B 40% and C 20% Figure 3 (b).
Figure 3. (a, b) Different amount of shading on Solar Panels
The characteristic parameters are measured from 10:00 am to 4:00 pm. Multi-meter (UC33+) is used to measure
both the current (A) and voltage (V). Solar Irradiation is measured by the intensity meter. Ambient Temperature
and Surface Temperature are measure by the temperature sensor that comes with the UC33+ multi-meter (K type

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
37
thermocouple). The output of healthy series is compared with the faulty series and the difference in current and
voltage is evaluated. Regression analysis is then applied to predict the current of each series. Machine Learning
methodologies are then used for validation.
RESULTS
The different types of faults induced, produced different output characteristics. When an open circuit fault is
induced in the system, short-circuit current decreases significantly. The current loss at the output is equal to the
current of the PV string which is short circuited. The induction of short-circuit fault results in the drop of open-
circuit voltage. When a panel is partial shaded, there is significant decrease in the short-circuit current. The short-
circuit current and open-circuit voltage of the PV solar system, are the fault characteristics in this paper.
Table 2. Short Circuit & Open Circuit Fault Data
S.
No
Time
(am/pm)
Tamb
(˚C)
Tsur
(˚C)
Solar
Irr
(G)
Short Circuit Fault
Open Circuit Fault
Healthy String
Current
(A)
Voltage
(v)
Current
(A)
Voltage
(v)
Current
(A)
Voltage
(v)
1
10:15 AM
34
44
723
0.9
38.8
0.67
38.6
0.93
56.8
2
10:30 AM
37
50
792
0.97
39.2
0.7
39
0.99
57.1
3
10:45 AM
38
52
812
0.98
38.8
0.71
38.9
1.01
57.5
4
11:00 AM
39
54
820
0.99
38.7
0.72
38.5
1.02
57.7
5
11:15 AM
38
53
844
1.01
38.8
0.74
38.4
1.05
57.6
6
11:30 AM
38
51
860
1.03
38.5
0.76
38.3
1.07
57.5
7
11:45 AM
40
55
873
1.06
38.6
0.77
38.5
1.09
57.4
8
12:00 PM
40
57
890
1.08
38.4
0.78
38.3
1.1
57.4
9
12:15 PM
39
55
923
1.09
38.5
0.82
38.4
1.12
57.7
10
12:30 PM
37
52
972
1.19
38.8
0.88
38.7
1.23
58.2
11
12:45 PM
38
57
970
1.18
38.8
0.85
38.7
1.22
58.1
12
1:00 PM
39
55
970
1.16
38.4
0.83
38.2
1.19
57.1
13
1:15 PM
39
54
968
1.15
38.4
0.84
38.4
1.18
57.3
14
1:30 PM
40
60
962
1.16
37.9
0.86
37.7
1.19
56.6
15
1:45 PM
38
55
720
0.89
38
0.67
37.9
0.92
57.1
16
2:00 PM
37
47
580
0.68
39.3
0.50
39.1
0.72
58.5
17
2:15 PM
36
42
373
0.42
38.3
0.34
38.5
0.45
58.2
18
2:30 PM
35
38
263
0.30
38.6
0.28
38.5
0.34
57.6
19
2:45 PM
34
38
187
0.25
38.4
0.22
38.2
0.27
56.6
20
3:00 PM
34
37
137
0.19
37.6
0.16
37.7
0.21
55.8
21
3:15 PM
33
36
122
0.16
37.3
0.14
37.4
0.19
55.5
22
3:30 PM
33
35
113
0.13
37.2
0.11
37.3
0.15
55.1
23
3:45 PM
33
34
94
0.11
37.1
0.10
37.1
0.12
55.0
24
4:00 PM
32
34
75
0.10
37.0
0.08
37.1
0.11
54.6
25
10:15 AM
33
40
673
0.79
39.3
0.65
39.2
0.83
58.8
26
10:30 AM
34
41
720
0.81
39.3
0.67
39.2
0.85
58.8
27
10:45 AM
35
43
743
0.82
39.2
0.67
39.1
0.85
58.7
28
11:00 AM
36
45
770
0.82
39.3
0.68
39.1
0.86
58.8
29
11:15 AM
36
48
803
0.89
39.1
0.73
39
0.92
58.6
30
11:30 AM
36
54
835
0.97
39.0
0.77
38.7
0.99
58.0
31
11:45 AM
35
53
847
0.98
39.1
0.79
38.9
1.02
58.2
32
12:00 PM
36
51
860
1.01
39.0
0.81
39.0
1.05
58.5

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
38
33
12:15 PM
36
51
860
1.02
39.0
0.81
39.0
1.05
58.5
34
12:30 PM
35
50
860
1.00
38.6
0.81
38.4
1.03
57.3
35
12:45 PM
36
51
854
0.98
38.6
0.80
38.4
1.02
57.2
36
1:00 PM
38
51
846
0.98
38.6
0.79
38.4
1.02
57.7
37
1:15 PM
37
49
813
0.96
38.6
0.77
38.5
0.99
57.6
38
1:30 PM
36
48
778
0.92
38.5
0.74
38.3
0.94
57.5
39
1:45 PM
36
48
744
0.87
38.5
0.72
38.3
0.89
57.4
40
2:00 PM
37
48
720
0.81
38.4
0.70
38.3
0.83
57.3
41
2:15 PM
36
44
433
0.53
38.4
0.37
38.4
0.54
57.3
42
2:30 PM
35
38
234
0.26
38.4
0.24
38.4
0.29
57.2
43
2:45 PM
34
37
227
0.24
38.3
0.24
38.2
0.25
58.3
44
3:00 PM
34
36
206
0.21
38.3
0.22
39.2
0.23
58.5
45
3:30 PM
33
35
78
0.11
36.8
0.10
36.7
0.12
54.3
46
3:45 PM
32
33
46
0.09
36
0.05
36.2
0.10
54
47
4:00 PM
32
33
42
0.08
36
0.05
36.1
0.09
54.2
Figure 4 shows current vs time graph of open circuit fault, short circuit fault and no fault i-e healthy string. The
current of short circuit fault and healthy string has no change. The current drop during an open circuit fault is the
loss of amount of current of the panel that is open circuited. The current and voltage values at maximum power
point (MPP) for the short circuit, open circuit and no fault PV strings are 1.19A, 0.88A and 1.23A, respectively.
Therefore, when there is significant decrease in the current that means an open circuit fault is present in the PV
string.
Figure 4. Current vs Time profile of short circuit fault, open circuit fault and healthy string.
Figure 5 shows the graph between voltage (V) and time (t) of short-circuit, open-circuit and healthy or fault-free
string. As the panels were connected in series, therefore the voltage drop between open circuit and short circuit
remained the same while the voltage for healthy string was maximum. If these panels were connected in parallel
then the voltage for healthy string and open circuit fault would remain almost the same and we would only have
voltage drop for short circuit current. The amount of loss of voltage for short circuit fault would be the voltage
for the panel that was short circuited.
1.19
0.88
1.23
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Current (A)
Time (am/pm)
Current (A) vs Time (am/pm)
Short Circuit Fault
Open Circuit Fault
Healthy String

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
39
Figure 5. Voltage (v) vs Time (t) profile of open circuit, short circuit and healthy string
PARTIAL SHADING FAULT ANALYSIS
Table 3. Partial Shade Data (50% & 100%)
S.
No
Time
(am/pm)
Tamb
(˚C)
Tsur
(˚C)
Solar
Irr
(G)
Partial Shade A
(100%)
Partial shade B
(50%)
Healthy String
Voltage(v)
Current
(A)
Voltage
(v)
Current
(A)
Voltage
(v)
Current
(A)
1
10.15
AM
29
33
623
57.9
0.05
59.4
0.11
59.3
0.76
2
10:30
AM
30
35
691
57.8
0.05
59.3
0.12
59.2
0.83
3
10:45
AM
32
38
733
57.7
0.06
59.2
0.13
59.0
0.88
4
11:00
AM
33
42
787
57.5
0.07
58.3
0.14
58.8
0.95
5
11:15
AM
34
43
806
57.4
0.07
58.2
0.14
58.5
0.98
6
11:30
AM
34
42
827
57.3
0.07
58.0
0.15
58.2
1.01
7
11:45
AM
33
44
823
57.2
0.07
57.9
0.15
58.1
1.03
8
12:00 PM
34
45
875
57.1
0.07
57.8
0.16
57.9
1.07
9
12:15 PM
35
46
878
57.0
0.07
57.7
0.16
58.0
1.07
10
12:30 PM
36
48
882
56.9
0.07
57.7
0.16
58.1
1.07
11
12:45 PM
36
47
851
56.8
0.07
57.6
0.16
58.0
1.05
12
1:00 PM
35
47
829
56.7
0.07
57.6
0.16
57.9
1.02
13
1:15 PM
35
46
822
56.8
0.07
57.7
0.15
58.1
0.99
14
1:30 PM
36
44
813
56.9
0.07
57.8
0.15
58.2
0.96
15
1:45 PM
36
44
811
56.9
0.06
57.8
0.14
58.2
0.93
16
2:00 PM
35
43
802
56.9
0.06
57.8
0.14
58.2
0.93
17
2:15 PM
36
45
792
56.7
0.06
57.6
0.14
58.0
0.92
18
2:30 PM
36
46
788
56.6
0.06
57.6
0.14
57.8
0.91
0
10
20
30
40
50
60
70
Voltage (v)
Time (am/pm)
Voltage (v) vs Time(am/pm)
Short Circuit Fault
Open Circuit Fault
Healthy String

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
40
19
2:45 PM
36
44
698
56.4
0.05
57.5
0.13
57.7
0.86
20
3:00 PM
36
43
671
56.3
0.05
57.6
0.12
57.9
0.83
21
3:15 PM
36
45
622
56.1
0.04
57.3
0.11
57.7
0.75
22
3:30 PM
35
41
511
55.8
0.03
57.1
0.09
57.7
0.61
23
3:45 PM
32
33
147
52.8
0.02
55.2
0.05
55.9
0.25
24
4:00 PM
31
32
113
52.1
0.01
54.6
0.02
55.1
0.11
Figure 6. gives detail about the current output when one panel in series A was 100% shaded, B 50% shaded and
C was left unshaded. The data collected is displayed in Table 3. From Figure 6, we concluded that, the more the
panels are shaded the higher is the current loss in the respective series. The panel which was left unshaded has the
maximum amount of current (I) for the same value of solar irradiance (G). For the maximum value of solar
irradiation from table 4.2 i-e 882 W/m2, the current output of series A is 0.07 (A) and that for series B is 0.16 (A)
while for non-shaded series C the current is 1.07 (A). The difference between the shaded and unshaded current is
significantly large.
Figure 6. Partial Shade Current vs Time Profile, Panel A 100% Shaded, Panel B 50% Shaded & Panel C unshaded
Figure 7. Partial Shade Current vs Time Profile, Panel A 25% shaded, Panel B 40% Shaded & Panel C 20%
shaded
0
0.2
0.4
0.6
0.8
1
1.2
Current (A)
Time (am/pm)
Current (A) vs Time (am/pm)
Partial Shade A (100%)
Partial shade B (50%)
Healthy String
0
0.2
0.4
0.6
0.8
1
1.2
Current (A)
Time (am/pm)
Current (A) vs Time (am/pm)
Partial shade A (25%)
Partial Shade B (40%)
Partial shade C (20%)

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
41
Table 4. Partial Shade Data (20%, 25% & 40%)
S.
No
Time
(am/pm)
Tamb
(˚C)
Tsur
(˚C)
Solar
Irr
(G)
Partial Shade A
(25%)
Partial shade B
(40%)
Partial shade C
(20%)
Voltage
(V)
Current
(A)
Voltage
(V)
Current
(A)
Voltage
(V)
Current
(A)
1
10.15 AM
28
34
754
58.7
0.22
58.6
0.12
58.9
0.25
2
10:30 AM
30
39
788
58.8
0.23
58.6
0.13
59
0.25
3
10:45 AM
32
41
790
58.7
0.23
58.3
0.13
58.9
0.25
4
11:00 AM
34
43
791
58.1
0.23
58.1
0.14
58.5
0.26
5
11:15 AM
34
44
799
58.5
0.23
58.0
0.14
58.3
0.26
6
11:30 AM
34
44
806
57.8
0.23
57.7
0.14
58.1
0.26
7
11:45 AM
34
43
824
57.8
0.23
57.8
0.14
58.2
0.26
8
12:00 PM
34
42
842
57.8
0.24
57.9
0.14
58.3
0.27
9
12:15 PM
33
41
849
57.7
0.24
57.8
0.14
58.4
0.27
10
12:30 PM
32
40
854
58.1
0.24
58.3
0.14
58.7
0.27
11
12:45 PM
32
40
855
58.1
0.24
58.3
0.14
58.7
0.27
12
1:00 PM
33
40
834
58.1
0.22
58.3
0.14
58.6
0.26
13
1:15 PM
33
41
822
58.0
0.21
58.2
0.13
58.4
0.25
14
1:30 PM
32
39
817
57.6
0.21
57.6
0.13
58.1
0.25
15
1:45 PM
32
40
803
57.5
0.2
57.4
0.12
58.0
0.24
16
2:00 PM
32
40
783
57.3
0.2
57.3
0.12
57.7
0.24
17
2:15 PM
33
42
754
57.2
0.19
57.2
0.11
57.5
0.22
18
2:30 PM
33
44
738
57.3
0.18
57.2
0.11
57.6
0.21
19
2:45 PM
34
45
711
57.2
0.17
57.3
0.1
57.7
0.2
20
3:00 PM
34
43
694
57.5
0.16
57.5
0.1
57.8
0.19
21
3:15 PM
33
42
633
57.3
0.15
57.4
0.09
57.5
0.18
22
3:30 PM
33
43
590
57.1
0.13
57.2
0.09
57.6
0.16
23
3:45 PM
32
39
477
57.2
0.11
57.3
0.08
57.8
0.13
24
4:00 PM
32
34
426
57.5
0.09
57.6
0.06
57.9
0.11
Figure 8. Partial Shade Voltage vs Time Profile, Panel A 100% Shaded, Panel B 50% Shaded & Panel C
unshaded
0
10
20
30
40
50
60
70
Voltage (v)
Time (am/pm)
Voltage (v) vs Time (am/pm)
Partial shade B (50%)
Partial Shade A (100%)
Healthy String

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
42
Table 4 displays data for the panels connected in series with different amount of shading. The amount of shading
applied on series A is 25%, series B 40% and series C 20%. This time one panel in each series was shaded in
order to evaluate the effect on current and voltage output of each shaded series. The graph in Figure 7 illustrates
that the current at maximum power point (MPP) for series A is 0.24 A, for series B is 0.14A and that for series C
is 0.27A. Now, we see that even though the panels were shaded with lesser amount of shading compared to earlier,
the current loss is significantly large.
Figure 9. Partial Shade Voltage vs Time Profile, Panel A 20% shaded, Panel B 25% Shaded & Panel C 40%
shaded
The graphs in Figure 8 and 9 depicts that the voltage remains the same even though the amount of shading applied
in all the series was different. The voltage of the unshaded series and that of shaded series was almost the same.
Fault Detection Flow Chart
0
10
20
30
40
50
60
70
Voltage (v)
Time (am/pm)
Voltage (v) vs Time (am/pm)
Partial shade A (25%)
Partial Shade B (40%)
Partial shade C (20%)

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
43
Table 4. Fault Detection
S.No
Possible Fault
Current (I)
Voltage (V)
01
No Fault
Imf = Imn
Vmf = Vmn
02
Short-circuit Fault
Imf = Imn
Vmf < Vmn
03
Open Circuit Fault
Imf < Imn
Vmf < Vmn
04
Partial Shading
Imf << Imn
Vmf = Vmn
PREDICTION OF FAULTS
The data in Table 2 illustrates the open circuit & short circuit faults (voltage and current) and healthy string
(voltage and current). The data was collected for 2 days for each fault induced in the system. The prediction of
open circuit, short circuit & no-fault current is performed using multiple linear regression in Minitab. Regression
equation is generated for each fault individually and based on the historical data we will be able to predicted the
values of current.
CURRENT PREDICTION USING MULTIPLE-LINEAR REGRESSION IN MINITAB
The current prediction for open-circuit, short-circuit and no fault strings are performed using multiple linear
regression analysis in Minitab software. A regression equation is generated for each fault which then predicts the
current.
Regression Analysis: Current Prediction using Multiple Linear Regression in Minitab
Method
Categorical predictor coding
(1, 0)
Model Summary
S
R-sq
R-sq(adj)
R-sq(pred)
0.0576555
97.52%
97.41%
97.25%
Coefficients
Term
Coef
SE Coef
T-Value
P-Value
VIF
Constant
-0.119
0.424
-0.28
0.779
Ambient Temperature (˚C)
-0.00642
0.00547
-1.17
0.243
6.34
Surface Temperature(˚C)
0.00517
0.00278
1.86
0.065
19.61
Solar Irradiations (W/m²)
0.000985
0.000054
18.25
0.000
12.54
Voltage (V)
0.00456
0.00719
0.63
0.527
176.28
Regression Equation
Fault
nf
Current (A)
=
-0.119 - 0.00642 Ambient Temperature (˚C)
+ 0.00517 Surface Temperature(˚C) + 0.000985 Solar Irradiations (W/m²)
+ 0.00456 Voltage (V)
oc
Current (A)
=
-0.219 - 0.00642 Ambient Temperature (˚C)
+ 0.00517 Surface Temperature(˚C) + 0.000985 Solar Irradiations (W/m²)
+ 0.00456 Voltage (V)
sc
Current (A)
=
-0.062 - 0.00642 Ambient Temperature (˚C)
+ 0.00517 Surface Temperature(˚C) + 0.000985 Solar Irradiations (W/m²)
+ 0.00456 Voltage (V)

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
44
Fits and Diagnostics for Unusual Observations
Obs
Current (A)
Fit
Resid
StdResid
32
0.8500
0.9637
-0.1137
-2.02
R
35
0.8300
0.9446
-0.1146
-2.04
R
53
0.3100
0.1877
0.1223
2.17
R
56
0.2500
0.1179
0.1321
2.35
R
59
0.1900
0.0612
0.1288
2.28
R
62
0.1600
0.0464
0.1136
2.01
R
68
0.1200
0.0071
0.1129
2.00
R
71
0.1100
-0.0052
0.1152
2.05
R
R Large residual
Prediction for Current (A)
Regression Equation
Settings
Variable
Setting
Ambient Temperature (˚C)
38
Surface Temperature(˚C)
52
Solar Irradiations (W/m²)
812
Voltage (V)
38.9
Fault
oc
Prediction
Fit
SE Fit
95% CI
95% PI
0.783168
0.0097973
(0.763791, 0.802546)
(0.667501, 0.898835)

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
45
Settings
Variable
Setting
Ambient Temperature (˚C)
38
Surface Temperature(˚C)
52
Solar Irradiations (W/m²)
812
Voltage (V)
38.4
Fault
sc
Prediction
Fit
SE Fit
95% CI
95% PI
0.937801
0.0098108
(0.918398, 0.957205)
(0.822130, 1.05347)
Settings
Variable
Setting
Ambient Temperature (˚C)
38
Surface Temperature(˚C)
52
Solar Irradiations (W/m²)
812
Voltage (V)
57.7
Fault
nf
Prediction
Fit
SE Fit
95% CI
95% PI
0.968263
0.0096385
(0.949200, 0.987326)
(0.852648, 1.08388)
The Data collected earlier for open circuit, short circuit and healthy panels is now formatted in such a way that
we had a categorical column as well in which for the same amount of solar irradiation (G), surface temperature
and ambient temperature we had different current and voltage output for the different types of faults induced in
the PV strings. Now, the categorical column contains open circuit (oc), short circuit (sc) and no fault (nf) column
values. Minitab has a feature which converts these values into binary form (0, 1). In our regression analysis, the
categorical predictors contain categorical column of oc, sc and nf.
After analyzing the regression tool, we predicted the value of current for the given amount of solar irradiation,
surface temperature, ambient temperature and fault type as shown in the Figure 3.9.
Figure 10. Current Prediction in Minitab
Feeding in the values of ambient temperature, surface temperature solar irradiation, and voltage and fault type,
current was predicted. For the given values in Figure 3.9, the open circuit current was 0.78A, short circuit current
was 0.94A and no fault current was 0.97A. According to the data collected, for the same value of ambient

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
46
temperature, surface temperature, solar irradiation and voltage, the open circuit current was 0.71A, short circuit
current was 0.98A and no fault current was 1.01A. The result obtained from multiple linear regression in Minitab
is now validated using machine learning methodologies along with the classification of fault as well.
EXPERIMENTAL DATA VALIDATION USING MACHINE LEARNING
The detection and prediction of faults using regression analysis is now validated using machine learning
methodologies. The experimental evaluation of the data sets is analyzed using machine learning in order to show
the effectiveness of the regression tools utilized for detection and prediction of faults in solar photovoltaic system.
In this paper, Machine learning includes the use of Random Forest technique and linear regression using python.
Random forest is a classifier which will classify the type of fault while linear regression will analyze the data set
involving multi-variable to predict the amount of current for each fault.
CURRENT PREDICTION USING LINEAR REGRESSION
The formatted data for analysis of regression in Minitab software is now used for machine learning technique.
For the same amount of surface temperature, ambient temperature and solar radiation, the current and voltage for
different faults are presented. The data is then imported in python and different libraries and classes are also
imported. Random forest classifier and linear regression are used to predict the current for each fault and classify
the type of fault as show in the Figure 11 (a,b,c) & Figure 12 (a,b,c).
The current prediction using Minitab software and machine learning produced identical results. In addition to
current prediction, Machine learning tool also classified the type of fault.
(a)
(b)

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
47
(c)
Figure 11. (a,b,c) Machine learning technique for current prediction
FAULT CLASSIFICATION USING RANDOM FOREST CLASSIFIER
(a)
(b)

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
48
(c)
Figure 12. (a, b, c) Machine Learning Technique for Fault Classification
CONCLUSION
This paper analyzed the effect of faults that occur in the direct current side of solar panel i-e open circuit, short
circuit and partial shading fault based on the real-time data. Photovoltaic systems are exposed to harsh weather
conditions which significantly degrade the panel and affect the system efficiency. Faults such as open circuit,
short circuit and partial shading reduce the output power of the solar system. The detection of these faults is
performed based on the output current (I) and voltage (v) at maximum power point (MPP) as the characteristic
features. The characteristic parameters (current, voltage, solar irradiation, ambient temperature & surface
temperature) are measured using multi-meter and SP2 Lite Meteon sensor.
The output voltage and current from healthy strings are compared to the output current and voltage of short
circuited, open circuited and partial shaded strings. The output of short circuit fault illustrates that whenever there
is a short circuit panel in the solar system, there will be a decrease in the output voltage and current will remain
maximum. When an open circuit fault exists in the solar system, the output voltage will remain maximum while
current will be reduced. The partial shading fault indicates that voltage remains maximum irrespective of the
amount of the shading but there is drastic decrease in the output current when panels are subjected to shading.
The regression analysis is then performed on open circuit, short circuit and healthy string data. The regression
equation generated for each faulty and healthy string is then used to predict the values of current. The prediction
of current is then validated using linear regression in machine learning. In addition to current prediction, fault
classification is also performed using Random Forest classifier.
ACKNOWLEDGEMENT
The authors are thankful to the UET Peshawar for providing necessary technical assistance.
CONFLICT OF INTEREST
The authors declare that they have no conflict of interest.
ETHICAL APPROVAL
All procedures performed in this study involving human participants were in accordance with the ethical standards
of the institutional and/or national research committee.
REFERENCES
[1] Y. Zhao, “Graph-Based Semi-supervised Learning for Fault Detection and Classification in Solar Photovoltaic
Arrays”, IEEE Transactions on Power Electronics, Vol. 30, No. 5, Pp. 2848-2858, 2015.
[2] C.E. Packard, J.H. Wohlgemuth, and S.R. Kurtz, “Development of a Visual Inspection Checklist for
Evaluation of Fielded PV Module Condition, in PV Module Reliability Workshop”, NREL, 2012.
[3] M.K Alam, F.K. Johnson, J. Flicker, “A comprehensive review of catastrophic faults in PV arrays: Types,
detection, and mitigation techniques”, IEEE J. Photovolt, Vol. 15, Pp. 982-997, 2015.

Detection and Prediction of Faults in Photovoltaic Solar Panel Using Regression Analysis
49
[4] T. Takashima, J. Yamaguchi, M. Ishida, “Fault Detection by Signal Response in PV Module Strings IEEE
Specialist Conference”, San Diego, CA, USA, Pp. 11-16, 2008.
[5] S. Silvestre, A. chouder, E. Karatepe, “Automatic Fault Detection in Grid Connected PV Systems”, Pp. 119-
127, 2013.
[6] A. Drews, A.C. De Keizer, H.G. Beyer, E. Lorenz, J. Betcke, W.G. Van Sark, W. Heydenreich, E. Wiemken,
S. Stettler, P. Toggweiler, S. Bofinger, “Monitoring and Remote Failure Detection of Grid-Connected PV
systems Based on Satellite Observation”, SOL Energy, Vol. 81, Pp. 548-567 2007.
[7] Y. Yadav, S. Umashankar, D. Vijayakumar, “Real Time Simulation of Solar Photovoltaic Module”, 2013.
[8] B. Nian, Z.F.L. Wang, X. Cao, “Automatic detection of defects in solar modules: Image processing in
detecting. 6th international conference on wireless communication networking and mobile computing
(WiCOM)”, chengdu, China, 2010.
[9] W. Peizhen, “Fault Analysis of Photovoltaic array Based on Infrared Image”, Pp. 197-201, 2010.
[10] W. Chine, A.M.V. Lughi, A. Malek, G. Sulligor, “A novel fault diagnosis technique for photovoltaic systems
based on artificial neural networks”, Renewable Energy, Pp. 501-512, 2016.
[11] M. Dhimish, V.H.B. Mehrdadi, M. Dales, and P. Mather, “Photovoltaic fault detection algorithm based on
theoretical curves modelling and fuzzy classification”, 2017.
[12] H. Mekki, A.M.H. Salhi, “Artificial neural network based modelling and fault detection of partial shaded
photovoltaic modules”, Pp. 1-13, 2016.
[13] S. Silvestre, M.A.d.S., A. chouder, D. Duasech, E. Karatepe, “New procedure for fault detection i grid
connected PV systems based on the evaluation of current and voltage indicators”, Pp. 241-249, 2014.
[14] S. Fadhel, “PV shading fault detection and classification based on I-V curve using principal component
analysis: Application to isolated PV system”, Solar Energy, Vol. 179, Pp. 1-10, 2019.
[15] M.H. Ali, “Real Time Fault Detection in Photovoltaic Systems”, Energy Procedia, Vol. 111, Pp. 914-923,
2017.
[16] E. Garoudja, “Statistical fault detection in photovoltaic systems”, Solar Energy, Vol. 150, Pp. 485-499, 2017.
[17] L. Wang, “Online Fault Diagnosis of Photovoltaic Modules Based on Multi-Class SVM”, Vol. 17, Pp. 978-
984, 2017.
[18] J. Mark, and G.B. Albers, “Comparative Evaluation of DC Fault-Mitigation Techniques in Large PV
Systems”, IEEE Journal Of Photovoltaics, Vol. 5, 2015.
[19] N. Kashyap, “Fault Tolerant Peak Current Mode Controlled Flyback Converter For Solar PV Modules”,
IEEE International Conference on Power and Embedded Drive Control (ICPEDC), 2017.
[20] D. Singh, “Fault Prediction and Analysis Techniques of Solar Cells and PV Modules”, International Journal
of Engineering Sciences & Research Technology, 2018.