Field study on the severity of photovoltaic potential induced degradation

Photovoltaic (PV) systems can be affected by different types of defects, faults, and mismatching conditions. A severe problem in PV systems has arisen in the last couple of years, known as potential-induced degradation (PID). During the early installation stage of the PV system, the PID may not be noticed because it appears over time (months or years). As time passes, it becomes more apparent since the output power may drop dramatically. We studied PV modules over the course of three years that were affected by PID. An electroluminescent and thermal imaging technique helped discover the PID. PID appeared in PV modules after being in different fields for 4–8 months, resulting in a 27–39% drop in power. An anti-PID box was fitted during the second year of the PV operation to recover the PID. Accordingly, it has stabilized the power degradation, but it could not restore the performance of the affected PID modules as compared with healthy/non-PID modules.

Photovoltaic (PV) system reliability and durability have attracted considerable attention in recent years from the PV industry as well as investors, as PV defects and faults have become an essential factor that has been scientifically proven to reduce power generation from PV assets and cause complications. Previous research on PV fault detection and classification has greatly improved our understanding of PV partial shading conditions 1,2 , faulty power converters/inverters 3 , and dead-state battery storage 4 . Although we have gained some insight into the severity of hotspots 5 , cracks 6 , and potential induced degradation (PID) 7,8 , we are still far from understanding how severe the defects are.
Researchers have made significant contributions to the analysis of hotspots and cracks in PV modules and how they contribute to the degraded performance of PV modules 5,6 . Although PID has been explained in a variety of settings, there have been occasionally details on how PID develops outdoors and in the field where there are different environmental conditions. As an example of a PID-affected PV string, what are the actual power losses that are caused by a typical PV string?
There are two different types of PIDs, both of which are attributed to the inverter or power electronic device not having a grounding connection. As a result, it creates either a biasing of + 1000 V or a biasing of −1000 V for the PV string affected. This will lead to the PV modules leaking current from the semiconducting material to the actual framing or glass of the module as a result of the PID 9 . As a consequence, it will result in degrading the modules and resulting in a significant reduction in the amount of output power produced 10 .
There has been some prior research on PV modules that were tested under a controlled environment by applying the existing PID testing procedure: humidity 85%, PV surface temperature at 65 °C, and either a bias voltage of + 1000 V or −1000 V for at least 96 h [9][10][11] . The test can, of course, help us understand how PID affects PV modules. As an example 12 , shows that PID can reduce module output power by more than 30%. However, there is not much information on whether PID results in the same power losses in an outdoor setting. Additionally, it is unknown whether repairing the grounding issues of the affected PID strings will increase output power and improve quality. Through electroluminescence (EL) imaging 9,12,13 , PID can induce shunting, cracks, and breakdown of solar cells.
To identify PV PID, thermal imaging is used via drone inspection 14 . A thermal review of PV systems has become increasingly popular in today's PV market because thermal testing is less expensive than EL testing, which requires that PV strings be disconnected. Even in large PV installations, PID affected PV strings are easily distinguished as they have higher surface temperatures than non-PID strings 15,16 . Temperature increases are caused most often by leakage currents and voltage bias, especially with PID negative biasing, −1000 V.
Additionally, previous experiments on PID of PV systems are usually assessed over short-term testing, 96 h, and the degradation estimation of the affected modules is simply derived by observing the theoretical curves of Results and discussions PID effect on the PV performance ratio. PV installation with 1.2 MW power capacity was tested in Barcelona, Spain. In this project, we investigated how PID develops (changes over time) and how severely they can impact module performance. Therefore, it was necessary to perform a thermal inspection to map the strings affected by PID. As a result, the required data can be extracted using the power converter unit.
The PV site was under construction in 2019 and was operational on 01/08/2019 when the thermal image was taken (Fig. 1a). We took these thermal images during the day when solar radiation was 712 W/m 2 , and the ambient temperature was 35 °C. No anomalies or faults were found. One year later, however, four PV strings were affected by PID, as shown in Fig. 1b. PID is characterized by an increase in PV module surface temperature to nearly 55-60 °C, compared with healthy PV strings which have a surface temperature of almost 37 °C.
The PID occurs in all PV strings due to a ground fault in the inverter, resulting in a −1000 V biasing, with each PV string containing 21 series-connected modules. In this way, the output power of the PV modules can further degrade and current can leak out. As a result, the PV modules will heat up, as shown in Fig. 1, and their output power performance will decrease over time. Note here that in the affected strings (1-4), they are all subjected www.nature.com/scientificreports/ to −1000 V, unlike some other reported PID effects in the literature which their PID is a resultant of + 1000 V biasing 22 . In addition, in 22 , they have also discussed that negative voltage biasing is the worst-case scenario regarding the PID effect on the PV modules, compared with either 0 V or + 1000 V biasing. Each PV string's log of measurements was analyzed and compared with a healthy PV string (string 5, Fig. 1b). Unfortunately, the current-voltage and power-voltage curves of the individual PV strings were not measured during the thermal inspection. The healthy PV string, string 5, was carefully selected as other PV strings had already developed hotspots and other mismatching conditions. We calculated the output performance ratio (PR) for each month. The PR ratio would indicate when the PID developed in the PID-affected PV strings and how much deterioration in output power developed compared with the healthy PV strings.
The results of the monthly calculated PR ratio are presented in Fig. 2. The healthy PV string has no monthly PR ratio below 90% during the entire year of operation. However, the decline of the PR ratio of the other PV strings has been seen at different times. For example, in PV string 1, the monthly PR drop started in November (4-months after field operation); the same observation is valid for PV string 2. The third PV string has a slight decline in the PR starting from December; this suggests that PID has occurred after 5-months of field operation. In contrast, PV string 4 has no significant drop in the PR ratio during the first 7-months of field operation. However, a fast decline in its performance is evident in March, due to PID.
These results demonstrate two critical indicators on the impact of PID: (1) PID can have an instantaneous reduction in the output power of the PV strings. For example, in PV string 2, after 4-months of field operation, the PR has dropped from nearly 90% in November to 77% in December. (2) PID develops in the PV strings at different periods. This implies that the early thermal inspection (day one) is insufficient to detect PID abnormalities.
It is also clear from Fig. 2 that PIDs can accelerate during relatively hot months, May through July. Prior to our examination, there was inadequate information on whether the PID effect could still have an effect on PV modules during these months due to the lower differences between PV surface and ambient temperature. The opposite has been proven by our experiment. In addition, Table 1 shows the PR ratio difference in the first month compared with last month. The PR difference ranges from 27.9 to 39.9% for the PID-affected PV strings, while the PR difference for the healthy PV strings is only 0.3%. These results are consistent with those obtained by laboratory testing 9,12 .
PID examination using electroluminescent imaging. The confirmed PID case in the PV strings was also supported by examining several PV modules under EL testing. For example, the output image of five PV modules from PV string 1 vs in the healthy PV string is shown in Fig. 3. Through EL detection, it is clear that the affected modules by PID, Fig. 3a, has low EL intensity compared with the healthy PV modules, Fig. 3b. In addition, the PID affected modules show shunting (blackout) cells that are likely to degrade the module's performance. However, unlike the healthy PV modules with homogenous EL intensity across the solar cells, these only have some micro cracks. Despite that some solar cells in the healthy PV modules have some cracks or minor Figure 2. Performance ratio of the tested PV strings. The y-axis presents the PR ratio, and the baseline of 90% was selected to indicate low performance; the x-axis gives the month. This chart shows the PR ratio for the PV strings during their first year of operation and the red labelled boxplots for the individual PV strings believed to be the starting month of the PID. A series connection of PV modules in Fig. 3 was used to test the I-V curve under a variety of field conditions. Figure 4a shows the modules under no shading (cloud-free skies), while Fig. 4b shows them under partial shading (some clouds). Last but not least, the modules are evaluated when operating under overcast conditions (Fig. 4c).
The I-V curves were taken for both the healthy and the PID PV modules during each different sky conditions (Fig. 5). In PID affected modules, I-V curves dropped steeply because PID has a significant impact on output voltage. According to Fig. 5a, the open-circuit voltage dropped from 183.6 V for the healthy modules to 134.1 V for the PID modules. PID-affected PV modules do not show a significant drop in short circuit current. In spite of this, the maximum current that can be extracted from the PID affected modules is less than that from healthy modules. As a result, the PID modules will generate less output power due to the voltage and current drop.
The PID causes leakage of the current in the PV modules, resulting in a decrease in the shunt resistance. The shunt resistance is necessary to keep at a high value because it determines the power output of the PV module, especially at low solar radiation conditions. Therefore, a typical reduction in the shunt resistance would accelerate the power reduction as in the case of PID affected modules. Table 2 summarizes the electrical output parameters (compiled from Fig. 5) for the tested PV modules. Due to the reduced current and voltage, the PID modules had a significant drop in power at maximum power point (MPP). PID accelerates the power decrease (−39.69%) due to reduced shunting resistance at lower irradiance conditions (overcast, 408 W/m 2 ). During clear sky and partial shading conditions, respectively, the output power of the PID-affected modules decreases by −37.15 and −36.29%. The I-V curve tracer endeavoured to track approximately the same current at I MPP for the PID compared with the healthy modules at the clear sky and partial shading because the modules are relatively affected by high irradiance (> 650 W/m 2 ). However, at low irradiance  On the contrary, it is incomprehensible to track the identical V MPP due to the PID and the reduction in the shunt resistance. These results explain the differences between healthy and PID modules under different environmental conditions. As a result, the power reduction of the PID affected modules is very high (> 36%), even under high irradiance conditions. Furthermore, under low-irradiance conditions, the output power of the PID-affected PV modules may be further reduced due to reduced shunting resistance.
PID recovery. PV strings affected by PID were equipped with anti-PID boxes between the strings and the inverter. This anti-PID box reverses the voltage of the PV modules, and in effect, polarises all of the PV modules in order to recover the PID of −1000 V of the PV modules [24][25][26] . It was decided to fit the anti-PID box only on the affected PV strings on 06/08/2020 after three days of confirming the PID through the use of thermal images of the PV strings. Figure 6 shows thermal images of PV strings after PID was confirmed (Fig. 6a) and after fitting the anti-PID box for a year (Fig. 6b). During the daytime, both images were taken under relatively the same ambient temperature.  www.nature.com/scientificreports/ As shown in Fig. 6b, the anti-PID box has partly improved the thermal performance of the PV modules. Three different instances were emphasised, where the PV surface temperature has dropped by nearly 9-13 °C; a summary of the results is presented in Table 3.
In spite of the fact that the surface temperature of the modules decreased, the performance ratio of the PV strings did not change significantly because PID reduces the shunt resistance of PV modules permanently. In Table 2. Electrical parameters of the tested PV modules. *The "difference" is calculated as the following:    Fig. 7a, the healthy PV module maintains a superior performance over a period of two years, with a PR ranging from 88 to 96%. PV strings (1, 2, 3 and 4) that were confirmed to have PID had exactly the same PR ratio, even when the anti-PID box was installed, their PR remained consistently below 75% even after the anti-PID box was installed, as shown in Fig. 7b,c.  , and no further reductions (i.e., below 50% PR ratio) have been observed. The results suggest that the anti-PID box has a positive impact on already PID-affected modules, which delays the recycling of the modules when a suitable detection mechanism is in place. Figure 7 illustrates another conclusion about the long-term impact of PID on PV modules. The study shows that PID can severely affect modules, resulting in a 25% to 40% reduction in output power. Sadly, unidentified PIDs can cause rapid power losses, and we see this frequently in large industrial PV assets. Thus, our results highlight the importance of regularly inspecting PV systems to avoid system complications, particularly those related to PID.

V OC (V) I SC (A) V MPP (V) I MPP (A) P MPP (W)
CDF is a widely used statistical analysis function that describes the probability of discovery of an investigated variable 27,28 . To explore the degradation in PR, we plotted the CDF for the affected PID PV strings as shown in Fig. 8. There is a 90% probability that a PV string affected by PID will have a PR degradation between 27.41 and 33.44%. Nevertheless, this assumption is only valid if the PID affected the modules for at least six months, as our previous data shows.
Also shown in Fig. 8 is the standard deviation (StDev) and the mean of the samples. According to the PR, the mean degradation ranges from 22.26 to 28.48%. A low StDev (3 to 4) indicates that the data are clustered around the mean. In summary, the CDF plot shows a degradation in PR for PV strings with affected PIDs. The probability of accelerated power losses in a PV string affected by at least 6-months of PID is approximately 27% to 33% with 90% confidence. Due to the location of our investigated PV system in Spain, Barcelona, these results may also be more accurate during hot summers and cold winters.

Methods
Photovoltaic system description. A polycrystalline silicon PV module with a peak output power of 220 W was investigated in this work. The modules have the following electrical characteristics: A PV string consists of 21 series-connected PV modules, while a string box consists of four parallel-connected PV strings. The PV modules are single insulated. The PV Modules were installed in the field on a large-scale PV project with a capacity of 1.2 MW in northern Barcelona, Spain. The PV strings are connected to the ICONICA inverter for data monitoring and logging. Data is sampled at 1 sample/10 min. Local authorities have requested that the exact location of the PV system be kept confidential. It is relevant to note that all the pictures, data, and analysis presented in this paper have been approved by the partner company (Al-Marri Ltd).
Although our work only considers the impact of PID on polycrystalline silicon PV modules, some recent research has shown that PID can have the same severe impact on various PV technologies, such as p-type crystalline silicon passivated emitter and rear cell (PERC) solar cells 29 , monofacial PERC solar cells 11 , and bifacial PERC solar cells 30 . Thermal inspection. DJI Mavic 2 Enterprise Advanced drone was used to capture the thermal inspection of the PV system. The maximum tilt angle of this drone is 35 degrees, and it operates wirelessly. The thermal camera used is an uncooled VOx microbolometer, which has thermal sensitivity of 2%. A 35 mm lens and digital zoom up to 32 × are included with the camera.  Electrical parameters field measurements. PV200 PV tester was used to measure the current-voltage curves of the inspected PV modules. PV strings can be traced with up to 15 A short-circuit current and 1000 V open-circuit voltage with this instrument. The I-V curve tracking accuracy is + 0.5%, which means the voltage and current measurements can be within + 0.5%. The I-V were measured directly from the PV strings and downloaded via a USB connection to a personal computer.
Sky imager. STI-440A total sky imager was used to check the cloudiness at PV site. Real-time images of the sky can be viewed via a web browser. Image resolution is 352 by 288 colour, 24-bit, and there is a rotating mirror with a shadow band. The sky imager is situated 30 m away from the PV system, and it can provide accurate cloudiness estimation/images for a 1 km radius.

Anti-PID box
The anti-PID box, PROJOY Electric, was fitted between the affected PID PV strings and the inverter. This box can recover PID in the range 400-1000 V, and it has a load power of 3.75 W. It has 200 kΩ, minimum insulation resistance, and 3.3 mA maximum output current. The Anti-PID box prevents solar panels from losing power because of PID. The negative electrode of the DC string and the ground of the panels are connected in parallel with the inverter, so a high voltage (with a very low amperage) is generated. Therefore, the negative polarity between the negative electrode and ground allows the panel to be repaired at night by freeing it from the accumulated charge during the day.
Performance ratio (PR) estimation. The performance ratio (PR) represents the performance of the PV system compared with the theoretical expected output power. For a typical PV system, a PR ratio above 90% stands very high. For example, in recent work on analysing 1000 polycrystalline silicon-based PV systems across the UK, it was found that the average PR is equal to 92.2% 31 . In this paper, the PR is calculated using Eq. (1) 31 , where P system is the measured dc power of the PV strings in watts, P rated is the rated dc power of the PV string in watts, G poa is the plane-of-array solar radiation in W/m 2 , G R is the reference solar irradiation (1000 W/m 2 ). The variable γ presents the maximum power temperature coefficient of the PV modules at maximum output power (−0.33%/°C, taken from the PV modules' datasheet), T PV is the PV modules surface temperature measured in °C, and T R is the PV modules surface reference temperature at standard test condition, 25 °C.
The relevant environmental parameters such as G poa and T PV are taken from a local weather station positioned near the PV installation. P system is taken from the data logger in the power converter unit connected with the PV strings.
In addition, in this paper, we have plotted the cumulative density function (CDF) for the PV strings affected by PID using Minitab software. The Degradation in the PR ratio is calculated using Eq. (2), where PRratio(%) is the actual PR calculated using Eq. (1) for the PID affected PV strings, and PRratio Healthy (%) is the PR ratio calculated from the healthy PV strings (free PID string).
In Eq. (2), the inclusion of the PR ratio Healthy (%) is to subtract the actual losses of the PR in the healthy strings because this loss is anyways expected in PV modules. Therefore, this would result in an accurate estimation of the PR's exact reduction (degradation) due to the presence of PID in the affected PV strings.

Data availability
The dataset generated and analysed in this study may be available from the corresponding author (G.B.) on reasonable request.