© R. Del Prado Santamaría et al., Published by EDP Sciences, 2025

1 Introduction

Regular and efficient operation and maintenance (O&M) procedures are critical for ensuring the longevity and efficiency of PV plants. Among the available diagnostic tools, infrared thermography (IRT) is widely employed due to its non-contact, non-destructive nature. IRT enables the rapid detection of hotspots in PV modules and strings, which may indicate underlying degradation [1–3]. These thermal anomalies can be caused by various fault types, including cell cracks, potential-induced degradation (PID), defective bypass diodes, cell interconnect faults, and external stressors such as soiling—each exhibiting a characteristic thermal signature [4,5].

IRT has gained popularity in utility-scale PV plants because it offers a fast and cost-effective inspection solution, particularly when deployed via drone-mounted thermal cameras. However, the accuracy of IRT-based fault detection is highly dependent on environmental conditions. Optimal performance typically requires a plane-of-array (POA) irradiance above 600 W/m², wind speeds below 28 km/h, and minimal cloud cover or soiling according to current inspection practices, as described in the IEC TS 62446-3 [6]. Such conditions are not consistently present in all regions; for example, in Nordic countries, where IRT inspections are often restricted to summer months.

In addition to weather conditions, the relative position of the thermal camera significantly influences the accuracy of temperature measurements and defect visibility. Previous research has shown that non-optimal imaging angles, particularly in drone-based inspections, can lead to temperature errors of up to 10 °C [7]. Several studies have addressed the challenges of IRT under non-ideal scenarios, such as partial shading [8,9], and have proposed best practices for the inspection setup and procedure.

Moreover, the use of IRT images for classifying defects in PV modules has been widely investigated, with a recent emphasis on machine learning and artificial intelligence. Different approaches, such as Support Vector Machines (SVM), Artificial Neural Networks (ANN), fuzzy logic, and others, have been used for this task [10–13]. These models focus on classifying the modules into defective and non-defective classes by identifying hotspots. However, hotspots can have different shapes and thermal signatures depending on the root cause. Furthermore, these models are often tailored to specific defects and rarely evaluated under suboptimal conditions. In particular, the influence of irradiance and drone altitude on the thermal response of PV modules affected by some defects has not been studied.

This work aims to address this gap by providing an experimental field-based evaluation of IRT images under varying irradiance levels and drone altitudes. We analyzed the thermal behavior of PV modules subjected to different types of degradation, namely, cell cracks, PID, cell interconnect degradation, glass cracks, short-circuited bypass diodes, and soiling. The experiments were conducted across irradiance levels ranging from 200 to 1000 W/m² and drone altitudes from 8 m to 20 m from ground level, equivalent to module spatial resolution of <1 cm/pixel to 3 cm/pixel. The influence of windspeed was not evaluated in this work due to constraints for drone flying safety. Furthermore, theoretical modelling has not been addressed in this study.

The key contribution of this study is to identify the conditions under which IRT remains a reliable diagnostic tool beyond the standard recommendations and to outline its limitations for specific fault types. These findings have practical implications for improving inspection protocols, optimizing drone-based inspection strategies, enhancing the robustness of automated defect detection algorithms, and estimating which failures are happening in the plants, knowing their temperature ranges and signatures.

Chapter 2 will introduce the modules used in this study, starting with a background section on the defects investigated. Chapter 2 also introduces how the module failures were induced, the modules' characterization, deployment, and imaging. In chapter 3, the temperature patterns of each defect type are evaluated on a module and cell level basis. A first evaluation investigates the trends of each defect average module temperature compared to a selection of healthy PV modules for each of the environmental and drone altitude conditions. Secondly, for certain degradation types, a cell-by-cell evaluation has been made, where the module hotspots are evaluated and correlated with the severity of each defect. Finally, in Chapters 4 and 5, the results are further discussed and summarized.

All the data acquired in this study is released as an open dataset containing IRT images of the studied modules at all irradiances and drone altitudes, EL images and I-V curves.

2 Material and methods

2.1 Background

PV modules in the field can be subject to different types of degradation mechanisms that can affect their performance and potentially cause reliability issues and safety risks. The defects considered in this study are cell cracks, PID, degraded cell interconnects, glass cracks and short-circuited bypass diodes. Figure 1 shows EL and visual images of each module defect.

As a benchmark, soiling experiments were added to give context on how the temperature patterns of the rest of the failures compare with soiling in the same type of modules. Soiling in PV plants can be attributed to dust or dirt/soil accumulation, bird droppings, pollution, or other external physical elements that can deposit on the PV modules and shade the solar cells. Regardless of whether the soiling is localized or homogeneous across the modules, it can create hotspots and limit the power production to a large extent. Due to the large impact of soiling on power production, regular cleaning is required, and preventive maintenance with IRT inspections is a valuable method for localizing soiling rapidly. Other studies have investigated the impact of different types of soiling in PV performance [14–16]; in this research, two types of soiling were used as study cases. The named “Soiling 1” consists of punctual soiling mimicking bird drops, and “Soiling 2” recreates homogeneously distributed soiling as it could happen in dryer areas with sand.

Cell cracks can appear in PV modules as a result of mechanical stress, for example during transport and installation, due to extreme weather conditions or through extended thermal cycling and field exposure [17]. These cracks can develop at any point in the module's lifetime; although often considered an early-life failure, they may propagate over time, leading to significant power losses and the formation of hotspots. When hotspots appear, cell cracks pose a safety risk to the installation. Moreover, cracks are not visually detectable, making characterization techniques such as EL or IRT essential for their identification. Cell cracks have been extensively studied in literature. They increase the module's series resistance, which, under illumination and operation at the maximum power point, causes additional losses to dissipate as heat. This thermal behavior forms the basis for using IRT imaging to characterize and detect this defect [18–22].

Cell interconnect disconnections or degradations refer to a physical or electrical break of the tab ribbons between cells in the PV modules. This defect can appear due to thermo-mechanical stress or during manufacturing as a problem with the soldering process. This degradation causes an increase in series resistance of the cells, because of a current mismatch between the cells [23]. As shown in Table 1, this defect can affect up to 15% of the module power production based on the severity of the disconnections, and if severe enough, it could trigger the bypass diodes leading to a larger power loss contribution.

PID is a degradation that appears in PV plants due to large voltage differences between the cells and the grounded module frames. The potential difference can cause ion migration from the module glass into the cells, resulting in increased recombination, shunting, and causing power loss, or in some cases, polarization of the cell's surface, leading to decreased photogeneration [24–26]. PID typically appears as a mid-life failure in PV plants, after some years of field exposure. Past studies show that advanced stages of PID degradation are detectable in IRT inspections as an increase of module temperature around the edges or in a chess pattern similarly to how it appears in EL images [26,27].

Modules with short-circuited bypass diodes pose a significant problem in PV plants. Defective bypass diodes are often reported after a few years of field exposure; however, they have a large impact on power generation, as a short circuit in a PV module bypass diode can lead to power losses ranging from 33% to the full module being bypassed if all diodes fail. This is one of the most common targets of IRT inspections due to their characteristic thermal signal, which is a complete or partial PV module substrings heating with very high temperatures [5]. It should be noted, however, that similar thermal or EL patterns may also arise from other junction box failures that electrically disconnect one or more substrings while the bypass diodes themselves remain intact such as solder joints or interconnection failures. In this work we specifically focus on cases where the bypass diode itself is defective.

Glass cracks may appear in fielded PV modules upon mechanical stress or impact, for example hailstorms, high winds or during installation by improper handling or improper mounting of the frames. Regardless of the crack severity, this failure is a severe safety risk in PV plants, firstly as a compromise of the mechanical stability of the module and potentially leading to isolation failures and inverter ground faults due to moisture ingress through the cracks. Despite cracked modules not showing a large power loss, detecting them timely is crucial as isolation failures pose a safety risk [28,29].

Fig. 1 Visual representation of the defects investigated. a) EL image of a module with cell cracks. b) EL image of a module with degraded cell interconnects. c) EL image of a module with PID. d) EL image of a module with two short-circuited bypass diodes. e) Visual image of a module with glass cracks. f) Visual image of modules with soiling type 1 and 2.

2.2 Experimental setup and methodology

A total of 43 modules has been selected for this study. Table 1 summarizes the types of defects, number of affected modules, estimated power loss ranges, and module technologies. The study includes both monofacial and bifacial Passivated Emitter Rear Cell (PERC) modules with nominal powers of 295 W, 305 W and 540 W. The reported range of power loss corresponds to the difference between the measured maximum power of each module after degradation, obtained from I-V curve flash testing at STC, and the datasheet value provided by the manufacturer.

The defects evaluated in this study are cell cracks, defective cell interconnections, PID, short-circuited bypass diodes, glass cracks, and two types of soiling. All these defect types and external factors have been induced through accelerated stress testing or replicating common failures that modules experience in the field. Thereafter, the modules were installed in different PV strings at DTU's PV plant in Risø.

Cell cracks were introduced by applying controlled mechanical stress on the module's surface, without breaking the glass. Cell interconnect failures were emulated by cutting the ribbons from the modules' back-sheet using a Dremel tool. PID was caused by high voltage stress testing the modules using the foil method [30]. Bypass diodes were shorted by accessing the junction box and soldering a piece of metal to short them. Glass cracks were introduced by puncturing the module glass from the rear side.

Every stress-tested module was characterized before and after degradation by I-V flashing at Standard Test Conditions (STC) (1000 W/m², 25 °C, and AM 1.5, STC) using a class AAA solar simulator; EL images were recorded at 100% I_sc and 10% I_sc bias with a NIKON D3500 CMOS camera. After installing the stress-tested modules in the plant, IRT inspections were carried out. A DJI Mavic M3T drone with a thermal camera of 640 × 480 px and a sensitivity <40 mK was used. The inspections were conducted under four irradiance conditions, monitored by an in-plane reference cell: 200 W/m², 600 W/m², 800 W/m² and >800 W/m². In this paper, >800 W/m² measurements refers to irradiance in the range of 850 to 1000 W/m²; this grouping of irradiances was done due to the inability to obtain consistently high irradiances in the measurement process; therefore, the data has higher variability in the range. The plots that include such measurements in Section 3 are shaded in light grey to indicate that high variability. Note that despite the variability, all measurements for each defect type were recorded in the same day at a stable irradiance. For each irradiance, the drone acquired images at four flight altitudes: 8 m, 10 m, 14 m and 20 m from ground level. It should be noted that drone altitude does not fully describe image quality; rather, the relevant factor is the effective spatial resolution on the PV module surface. With the present setup, the resulting ground sampling distance was <1 cm/pixel at 8 m altitude and approximately 3 cm/pixel at 20 m altitude. According to IEC TS 62446-3, the minimum recommended spatial resolution for PV inspections is 3 cm/pixel. Therefore, the flight altitudes selected in this study cover the practical operating range from very detailed inspections (<1 cm/pixel) to the resolution limit specified by the standard.

During the IRT inspections, the modules were all connected to the rest of the string and operated at maximum power point. Prior to imaging under each irradiance condition, the system was allowed to stabilize for at least 10 min. An example of a string with two modules with defective bypass diodes is shown on Figure 2 under the four altitude conditions and with an irradiance of 800 W/m².

Image processing consisted of module corner detection and perspective correction. For each of the modules, statistical parameters such as the average and maximum module temperature have been calculated. Furthermore, for defects such as cell cracks and degraded cell interconnects, where the degraded cell locations are known from the EL images; the modules' IRT images have been segmented into cells, and the cell mean and maximum temperatures calculated.

Fig. 2 IRT images example of the different inspections carried out. String with two modules showing shorted bypass diode degradation under four drone altitudes (8, 10, 14 and 20 m) corresponding to <1, 1.5, 2 and 3 cm/pixel of spatial resolution. The images correspond to the 800 W/m2 irradiance inspection.

3 Results

3.1 Cell cracks

Figure 3 shows an EL image of a selected module with cell cracks and its corresponding IRT images at the four irradiance levels with spatial resolution of <1 cm/pixel. The module presents several cracked cells with different degrees of severity. At 200 W/m² the temperature difference within the module between cracked and healthy cells is very small, between 0.2 and 0.3 °C, however, as the temperature increases the cells on the left start heating up, indicating the cracks location. IRT inspections at 200 W/m² were recorded with ambient temperature of around 10 °C, therefore the maximum module temperature observed in this inspection is considerably lower than the rest of the cases. This is also true for other defects as shown in further sections. In addition, external factors such as wind were not accounted for in this study and may contribute to cooler or spatially non-uniform module temperatures under low irradiance.

At higher irradiances, the temperature differences between cracked and non-cracked cells increase slightly (between 1 and 2 °C). However, these differences remain subtle and are comparable to other sources of non-uniform heating within the module. For instance, apparent hotspots can be observed on the left-hand side of the module in cells without visible cracks in the EL image, highlighting the difficulty of relying on IRT to identify cracks without risking false positives. Therefore, while a marginal increase in mean temperature can sometimes be observed in cracked modules compared to healthy ones, the low contrast (approximately 1 °C) demonstrates that IRT is not a reliable technique for detecting cell cracks. These results emphasize the limitations of IRT for this type of defect.

For investigating the influence of drone altitude in IRT images, all individual cells of the cracked modules were segmented from both the EL and IRT images. Cracked cells have been categorized into low, medium and high severity from the EL images. This categorization is based on thresholding the EL image of the cell with an Otsu adaptive threshold [31]. Cells with cracked area <10% are considered low severity; between 10% and 30% medium severity and >30% high severity. Figure 4 shows an example of three cells from one of the studied modules where each of the cells is categorized. Once the cells are categorized based on severity, the same cells are segmented from the modules' IRT images.

Figure 4 shows the evolution of the mean cell temperature of cracked and non-cracked cells with different drone altitudes sorted by severity, imaged at >800 W/m². All drone altitudes show the same profile as temperature increases with crack severity; however, for cracks under the category “high severity”, the median temperature quartile decreases. A similar finding was pointed out by Dhimish et al. [32].

Although the profiles appear to show a decreasing trend in temperature with increasing crack severity, this cannot be interpreted as a physical effect. Moreover, the statistical distribution is affected by the fact that fewer cracked cells are present overall compared to healthy cells, which amplifies variability in the cracked-cell categories. In practice, the temperature of cells with cracks is within the range of healthy ones (36 °C to 40 °C), make them in practice indistinguishable; this is also true for temperature deltas within the modules, of the range of 1 to 2 °C for the cells with medium crack severity. These findings underline that the small and inconsistent temperature differences observed here highlight a limitation of IRT that without complementary EL imaging, reliable crack diagnosis is highly uncertain.

Fig. 3 Top row: Example of a cell cracked module IRT images at different irradiances. EL image as a reference. Bottom row: average module temperature of all cracked modules vs a selection of healthy ones at different irradiances. Altitude of the inspection 8 m.

Fig. 4 a) EL image of a fielded cracked module. Highlighted cells show an example of a low severity crack with 1.5% cracked area (orange); medium severity with 20.4% cracked area (green) and high severity with 30.6% cracked area (red). b) Boxplot showing the maximum cell temperature trends in drone altitude and cell crack severity compared to reference healthy cells. Irradiance of the inspection >800 W/m2.

3.2 Cell interconnect degradation

Figure 5 shows an example of a module with cell interconnection failures EL image and the resulting IRT images under different irradiances. The bottom plot shows as a boxplot the distribution of the average module temperature of both modules with degraded cell interconnections and healthy modules. This defect shows significantly not only under high irradiances with temperature differences in affected cells of up to 4 °C; furthermore, some cases of severe degradation can also be identified at irradiances of 600 W/m², whose thermal characteristic resembles the pattern of the EL image. It should be noted that only the most severe interconnection failures produce a clear and repeatable thermal signature. Cells with milder degradation, even when clearly visible in the EL image, often remain indistinguishable from healthy cells in IRT, especially at moderate irradiance levels. This reflects the limited thermal sensitivity of IRT to partial or low-resistance disconnections, whose increase in series resistance may not induce sufficient local heating.

Similarly to cell cracks, degraded cell interconnects have been classified into severity classes to assess the impact of drone altitude on the thermal characteristics of the affected cells. Using EL images as reference, the cells were classified into degraded cell interconnects type 1, 2, 3, or 4 based on how many disconnected ribbons are affecting the cell, where type 0 corresponds to healthy cells. Figure 6 shows an example of an EL image of a module affected by degraded cell interconnects, where some of the highlighted cells exemplify the different severity classes. The right plot shows the maximum cell temperature distribution of all categorized degraded cells and the influence of drone altitude on them during the inspection performed at over 800 W/m² POA irradiance. A decrease in cell temperature with increasing flight altitude was observed. This behavior can be attributed to several factors. The effect in this altitude range is a reduction in spatial resolution: at higher flight altitudes, each pixel represents a larger surface area and effectively averages temperature variations within the image. Spatial averaging lowers apparent temperature and reduces image contrast. Small defocusing effects and different camera fields of view may also contribute to the lower apparent temperatures observed at higher altitudes. Since cell interconnection degradation originates form busbars rather than across large cell areas, they are prone to the effect of pixel averaging, leading to a more pronounced reduction in imaged temperature with increasing altitude compared to other defect types.

At low altitude, the trend shows a significant increase in temperature when four interconnects are damaged, with a temperature increase of four to five degrees on average compared to healthy cells. Disconnected cell interconnects of lower severity, one or two degraded interconnects, also present a noticeable temperature difference of around two degrees compared to healthy cells. The interquartile range of maximum cell temperature increases with drone altitude, particularly at 20 m, which reflects again the reduced spatial resolution of IRT images at higher altitudes, increasing measurement variability and reducing consistency in temperature estimation.

It is important to note that the cells with Type 4 defective interconnects represent a relatively severe failure case. The analyzed modules have five-busbar cells, and the cells categorized as Type 4 exhibit up to four disconnected interconnects. While such extreme degradation levels are uncommon in the field, it illustrates the upper limit of thermal detectability for this defect type. Less severe interconnection losses (Type 1 to 3) also show measurable, though smaller, temperature increases. These differences are near the detection threshold of IRT and within the temperature range of healthy cells.

Fig. 5 Top row: Example of a module with damaged cell interconnects where the IRT images were taken at different irradiances. EL image as a reference. Bottom row: Average module temperature of all modules with degraded cell interconnects vs a selection of healthy ones at different irradiances. Altitude of the inspection 8 m.

Fig. 6 a) EL image of a fielded module with damaged cell interconnects. Highlighted cells show an example of a cell interconnect failure Type 1 (orange), Type 2 (green), Type 3 (Red) and Type 4 (Purple). b) Boxplot showing the maximum cell temperature trend in drone altitude and cell disconnection severities compared to healthy cells. Irradiance of the inspection >800 W/m2.

3.3 Potential induced degradation

In this study, the PID stress tested modules showed a degradation of around 5% in power loss. Despite the degradation being noticeable in EL images, the degradation is too low to display any thermal anomaly in the IRT images. An example of an EL image before and after stress test of one of the deployed modules is shown on Figure 7. This module shows a general lower EL intensity and darkening on the edges of some cells; this degradation behavior has been reported in commercial PERC modules [33].

Figure 8 shows an example of an IRT image acquired at 8 m altitude and over 800 W/m² where the modules on the bottom are PID affected and the ones on the top row are all healthy. Despite the leftmost module showing some soiling and a hotspot in the bottom left corner, no significant thermal signature was found on the PID affected modules. At 600 W/m², a temperature difference of approximately two degrees was observed in Figure 8b; this cannot be attributed to PID, instead, it is most likely the result of external influences such as wind, transient irradiance changes, or minor calibration uncertainties in the thermal camera. This highlights that environmental and measurement-related factors can introduce temperature variations of this order of magnitude, which complicates the interpretation of IRT images in practice.

The difference in irradiance and altitude showed no influence on these modules' temperature pattern. Past studies showed PID to be detectable under IRT when the affected module was degraded over 40% in power [34]; since the stress tested modules in this study did not reach such level of degradation, no conclusion can be made from the IRT analysis, however, this result highlights the detection limits of IRT for early-stage PID. The absence of a thermal signature despite measurable degradation in EL images shows that early PID may go unnoticed in field IRT inspections. This underlines the importance of complementary inspection methods, particularly EL imaging, and long-term monitoring, as early PID can progress to more severe stages during field exposure. Furthermore, the fact that environmental factors can produce temperature variations of a few degrees reinforces the challenges of diagnosing subtle degradation mechanisms such as early-stage PID, cell cracks, or interconnect damage under realistic field conditions.

Fig. 7 EL image of a PID stress tested module before and after stress test.

Fig. 8 Left: IRT image of a part of two PV strings. The top string is healthy; the bottom string is PID affected. Right: Average module temperature of healthy vs PID affected modules as a function of irradiance. Drone flight altitude of the inspection 8 m.

3.4 Short-circuited bypass diodes

In this study, PV modules were artificially degraded by short-circuiting either one or two bypass diodes per module. Figure 9 (top row) shows an example of a module with one shorted bypass diode and the evolution of the thermal patterns as irradiance increases. Notably, in this particular case, the cells in the middle section of the module exhibit higher temperatures, forming a symmetrical hotspot pattern. The module was bifacial and mounted in portrait orientation, and the rear-side mounting beam likely caused partial shading in the central area, leading to localized heating. The maximum temperature difference reached up to 20 °C at high irradiance (>800 W/m²) and decreased proportionally with irradiance, though the thermal pattern remained visible even at 200 W/m². In addition to the temperature rising with irradiance, some variations in the relative hotspot intensity can be observed. Some individual cells appear hotter at low irradiance (200 W/m²) and relatively cooler at higher levels. This apparent heating does not correspond to a temperature increase of those cells but rather to a relative decrease of neighboring cells as irradiance changes.

Figure 9 (bottom row) presents a monofacial module where two bypass diodes were short-circuited. In this case, the resulting thermal response extends across two-thirds of the module, with multiple hotspot regions visible at all irradiance levels. As irradiance increases, the hotspot temperatures exceed 70 °C, consistent with the presence of current flow through the shorted substrings. Despite this, the thermal pattern remains relatively uniform within each affected substring, showing that the power dissipation is distributed over several adjacent cells. The visibility of these thermal patterns even under low irradiance conditions (200 W/m²) demonstrates that severe bypass-diode faults produce strong and persistent thermal anomalies, which can be clearly distinguished from minor defects. Similarly to the module in the top row, there are some cells with inhomogeneous relative temperature differences. A few cells on the upper part of the module seem to increase in temperature compared to the neighboring cells. Since measurements at each irradiance were conducted on different days, small variations in soiling, illumination uniformity, or rear-side reflections in the case of bifacial modules could contribute to the appearance of hotspots. These observations underline the inherent variability of IRT measurements in outdoor conditions and reinforce the need for careful interpretation of temperature differences.

Modules with short-circuited bypass diodes are also shown to be detectable under high drone altitudes. Figure 10 shows the maximum temperature distributions of all affected modules at the four irradiances and at four different drone altitudes. The relationships between module temperatures remained constant across scenarios. A slight decrease in hotspot temperatures can be observed as drone altitude increases, where the average hotspot at the 800 W/m² inspection decreases from 70 to 65 °C approximately. Similarly to the observed temperature decrease with altitude in the modules with degraded cell interconnects, spatial resolution could affect the recorded temperature by the camera sensor due to pixel averaging or other factors like camera focus.

Fig. 9 a) IRT images of two modules with a sort-circuited bypass diode at different irradiances imaged at 8 m drone altitude. EL image as a reference. b) Rear side picture of the mouting frame of the PV modules installed in the field. The beam location corresponds to the hoptspot location of the bifacial halfcut modules causing rear side partial shading.

Fig. 10 Distribution of maximum module temperatures for each measured plane of array irradiance at all drone altitudes studied. a) Inspection at 8 m. b) Inspection at 10 m. c) Inspection at 14 m. d) Inspection at 20 m. PV modules with short-circuited bypass diodes (red), compared to healthy modules (blue) of the same type.

3.5 Glass cracks

Previous research shows examples of glass cracks causing hotspots in IRT images due to delamination and corrosion because of moisture ingress. Hotspots can also occur if the glass crack occurred in tandem with a cell crack of the underlying cells. Therefore, low energy cracks and broken glass without long field exposure may not show a hotspot; furthermore, unless the glass is completely shattered, its detection when using visual images is challenging [35].

In this study, we had two modules with both their back and front side cracked. Figure 11 shows an example of a front glass crack in a PV module; this module shows a small cell crack in the point of impact (marked in red) but shows no thermal pattern in the IRT image. The low energy glass cracks observed in the modules from this study did not cause significant hotspots, however, they are prone to expand from field exposure due to extra mechanical stress or thermal cycling. In more severe crack damage such as shattering, the cells could be further mechanically stressed resulting in hotspots. In addition, extended field exposure of glass cracks can potentially lead to moisture ingress in the module leading to corrosion and delamination, which could be visible in IRT, however, becoming a safety issue. The modules from our study were not deployed in the field long enough for such processes to occur.

Fig. 11 Visual image, EL image and IRT image of one of the modules with glass cracks, imaged at 800 W/m2 and 8 m drone altitude (<1 cm/pixel moduel spatial resolution). The red square in the EL image indicates the point of impact on stress test that created a small cell crack.

3.6 Soiling

Figure 12 shows an example of two modules affected by punctual soiling, replicating bird drops and other types of soiling that shade partially or completely one or a few cells in the module. The top module is a bifacial module with 144 half cut cells in butterfly configuration, while the bottom module is a 60 cell monofacial PV module.

The top module has hotspots on two different substrings on its lower side. The temperature of the hotspot increases with irradiance; however, the hottest cell is not the same across conditions. In the image at 800 W/m² the module has two hotspots in the lower half. These hotspots are creating a mirror effect, that causes the opposite string on the upper side to heat up slightly. The hotspot in this case is at around 56 °C, and the substrings above are heating up to 50 °C, a 10 °C difference to the un-soiled module substrings. At 200 W/m², the shaded cell shows a hotspot, but the warmer cell is on the opposite cell substring. In this case, the temperature difference is still significant, with almost a 10 °C increase.

This phenomenon is particular to modules in a butterfly electrical cell layout. When one substring is partially shaded and a hotspot appears due to reverse voltage bias in the affected cells, the bypass diode may activate to protect the module. Since the affected substring and the one above are connected in parallel, when the bypass diode activates, they are forced to operate at the same low voltage. As a result, the non-affected substring will operate near short circuit, which can lead to some of the cells to reverse bias and dissipate power, leading to extra hotspots in the unshaded substring.

The bottom module of Figure 12 also has soiling on two different substrings of the module which is generating two hotspots in cells at 800 W/m² and 600 W/m²; one of the soiling patches does not create a hotspot in the IRT image at 200 W/m². In the case of this module, it was validated with an I-V curve measurement taken at the time the 600 W/m² image was recorded that the hotspots are triggering two of the module's bypass diodes due to the shaded cells operation under reverse bias. This module shows a particular hotspot at 600 W/m² irradiance. The hotspot on the left side of the module that under 800 W/m² shows a temperature of approximately 60 °C increases to over 80 °C. In this case, as soiling was artificially introduced, slight drift could cause larger parts of the cell to be shaded, increasing the reverse bias of the cell and the hotspot temperature.

Figure 13 shows two examples of modules with soiling of type 2, with sand patches distributed across the module surface. This type of soiling attempts to replicate desert-like environments where the PV modules can be covered to a large extent by a layer of sand or dust. The temperature distributions of the analyzed modules are significantly different. Based on the level of sand deposition, if a large patch of sand covers a specific cell, it can generate a hotspot in the same way as it did in the case of Soiling 1; in this case the mirror effect observed previously is prone to happening again. However, if the soiling is homogeneously distributed, it is possible that the module does not experience any extreme hotspot. In cases of extreme soling, power losses of up to 66%, with two bypass diodes activating were observed.

Figure 14 shows the effect of irradiance and drone altitude on the modules affected by soiling. The distribution of the module's hotspots shows a larger dispersion at high irradiance when the soiling is homogeneously distributed (Soiling 2). The modules with Soiling 2 show very large hotspots, reaching 100 °C but if the soiling does not create a hotspot, it increases the module temperature homogeneously. Punctual soiling (Soiling 1) reached over 80 °C hotspots in certain occasions. Drone altitude does not show a significant effect in the identification of soiling; the large hotspots are visible even at drone altitudes of 20 m, and the distributions from Figure 14 show that a clear differentiation can always be made from healthy modules across altitudes and even at low irradiances. However, a trend of temperature decrease in the 800 W/m² inspection can be observed, as the drone's altitude increases. The reduction in spatial resolution, similarly to what shown in previous defects, causes an attenuation effect on the partially shaded cells with hotspots, resulting in hotspots of apparently lower temperature. This apparent temperature decrease is more prominent in the case of Soling 1, as the hotspot area is smaller than in the situation of Soiling 2 (where the whole module tends to heat up), it is more prone to spatial resolution changes and contrast effects.

Fig. 12 Example of two modules affected by punctual soiling (Soiling 1) imaged at different irradiances and at 8 m drone altitude (<1 cm/pixel moduel spatial resolution). Top: Bifacial PV module with buttterfly configuration and 144 half-cut cells. Bottom: Monofacial PV module with 60 6 × 6 inch cells. IV of the module acquired at approximately 600 W/m2.

Fig. 13 Example of two modules affected by large area soiling (Soiling 2) imaged at 800 W/m2 and 8 m drone altitude (<1 cm/pixel moduel spatial resolution). a) Soiling that created hotspots. b) Soiling that did not create hotspots.

Fig. 14 a) Effect of irradiance in maximum temperature distribution of healthy modules and modules with Soiling 1. Altitude of the inspection 8 m. b) Effect of drone altitude in maximum temperature distribution of healthy modules and modules with Soiling 1. Irradiance of the inspection 800 W/m2. c) Effect of irradiance in maximum temperature distribution of healthy modules and modules with Soiling 2. Altitude of the inspection 8 m. d) Effect of drone altitude in maximum temperature distribution of healthy modules and modules with Soiling 2. Irradiance of the inspection 800 W/m2.

4 Discussion

Drone altitude is a relevant parameter in IRT inspections, as it determines the spatial resolution of thermal images and thus the ability to resolve small image features. As drone altitude increases, the spatial resolution of thermal images decreases; each pixel covers a larger surface area of the PV module, resulting in a lower pixel density per cell. This reduction in resolution can compromise the detectability of small or localized defects, potentially hindering reliable fault detection.

Figure 15 illustrates this effect by comparing thermal images of a module with cell cracks and another with cell interconnect failure, captured at 8 m and 20 m drone altitude. A substantial degradation in image detail is observed with increasing altitude, particularly at the cell level. The camera used in this study has a resolution of 640 × 512 pixels with a focal length of 40 mm; since the imaged modules have a surface area between 1.6 and 2.4 m² and one module does not take the full camera field of view, the spatial resolution of the inspection at 8 m altitude corresponds to approximately <1 cm of the module edge per pixel, while the inspection at 20 m corresponds to approximately 3 cm/pixel. Although this reduction in pixel density explains part of the loss in cell-level detail, we the blurrier appearance of the 20 m images could be caused by additional factors. Those could be imperfect focusing of the lens at higher altitude or thermal convection processes that smooth temperature variations. These influences underline that altitude-related effects on image quality are not solely a function of spatial resolution.

As shown in previous sections, the range of drone altitudes investigated in this work does not affect critically the identification of defects.

Severe interconnect failures remained partially detectable at higher altitudes in severe cases, their characteristic elongated thermal patterns became indistinct at 20m. This loss of pattern fidelity complicates root-cause analysis, as the defect-specific thermal signature is no longer discernible.

Large-area defects such as short-circuited bypass diodes and heavy soiling continued to produce prominent and easily identifiable thermal anomalies at high altitudes. These faults typically result in extensive hotspot formation and activation of bypass diodes, yielding strong and characteristic thermal signatures that remain visible even at higher imaging altitudes.

It is important to note that, while the influence of altitude on the absolute temperature signal of defects is not expected to be significant, we included this analysis to provide a systematic comparison. Our results confirm that altitude primarily affects measurement consistency and spatial resolution, which in some cases was linked to a global temperature decrease detected in the modules. Although we did not explore altitudes beyond 20 m, our focus was on the range most representative of detailed inspections (<1 cm/pixel), up to the maximum recommended by the technical specification (3 cm/pixel). Future works should explore beyond this threshold of drone altitudes.

In addition to drone altitude, solar irradiance is one of the most influential parameters affecting IRT reliability. At higher irradiance levels, modules operating at maximum power dissipate more heat at defect locations, enhancing thermal contrasts and improving defect visibility. As recommended by current technical specifications, irradiance levels above 600 W/m² are optimal for accurate IRT inspections. The results from this study confirm that most defect types exhibited increased thermal contrast under high irradiance compared to healthy modules. Conversely, under low irradiance, thermal gradients were minimal, and most defects were difficult or impossible to detect. Notably, large-area faults such as shorted bypass diodes and heavy soiling remained detectable even at low irradiance levels (e.g., 200W/m²), further confirming their robust thermal signature.

Table 2 reports for each defect severity studied in this work a summary of the hotspots identified in the inspection carried out at 8 m altitude and at irradiances of 800 W/m² and over. Max. cell temperature corresponds to the maximum temperature hotspot recorded; ΔT_{max_min} corresponds to the temperature difference between the maximum hotspot to the minimum temperature recorded in that module; and ΔT_{max_healthy} corresponds to the temperature difference between the maximum hotspot to the temperature of an average healthy cell of the module. Note that for the case of cell cracks and defective cell interconnects both ΔT metrics were calculated in the cell level; therefore, the temperature spread is less than in the rest of the degradations.

Under imaging conditions of 8m drone altitude (<1 cm/pixel spatial resolution) and over 800W/m² irradiance, modules with PID, glass cracks, and even high-severity cell cracks exhibited temperature differences of less than 1 °C compared to healthy modules. Specifically, high-severity cell cracks showed a maximum temperature of 38.6 °C versus 38.0 °C for healthy modules, well below the practical detection threshold. Notably, healthy modules show ΔT_{max_min} values of up to 9 °C and ΔT_{max_healthy} of 1.5 °C. These values emphasize that defects causing less than a 1.5 °C increase in temperature will not be distinguished form an average healthy cell; furthermore, the 9 °C ΔT_{max_min} in the full healthy modules indicates that investigating hotspots by comparing maximum and minimum temperature differences can have a significant temperature bias. These temperature differences can occur due to having cooler parts of the module near the edges due to heat convection through the frame; different material emissivity if parts of the module frame remain the in the processed image or reflections in due to camera positioning. Such high temperature differences explain the difficulty of visually identifying defects in IRT images.

Defective cell interconnects exhibited hotspot temperatures ranging from 40.5 °C to 44.2 °C depending on degradation severity. This corresponds to a 2.5 °C to 6.2 °C increase over the healthy module baseline. Importantly, this level of contrast is also more sensitive to reductions in spatial resolution at higher altitudes, especially when the defect affects only a few cells. The same tendency is reflected in both ΔT metrics, where Type 4 disconnections produce the largest spread, confirming their stronger and more localized heating signature.

The largest thermal anomalies were observed for shorted bypass diodes and heavy soiling, with hotspot temperatures ranging from 59.7 °C to over 105.4 °C, differences exceeding 30 °C to 65 °C relative to healthy modules. These defects generate large-area heating that remain visible even under suboptimal irradiance and at high drone altitudes, demonstrating IRT's robustness for detecting such failures. This also reinforces the importance of classifying faults by their thermal footprint, as it strongly influences their detectability under real-world inspection constraints.

Notably, the glass crack cases, exhibited relatively high ΔT_{max_min} values of around 17 °C. This high temperature difference can be observed in Figure 11, as the module edge shows a drastic temperature difference from the rest of the module. The explanation for this observation; similarly to the case of healthy and PID affected modules is heat dissipation through the module's frame or a slightly different material emissivity.

Taken together, these results emphasize that IRT detectability is strongly dependent on both the size and severity of the defect, as well as inspection conditions. While IRT is effective for large or severe faults, it has notable limitations in identifying small-scale or subtle defects, especially when inspections are conducted under low irradiance.

While inspections at low altitude (spatial resolution 1–1.5 cm/pixel) and high irradiance (>800 W/m²) are preferable, these conditions are not always achievable in practice, particularly in colder climates or northern regions where high irradiance events are infrequent and ambient temperatures reduce thermal contrasts. Under such sub-optimal conditions, IRT inspections still provide value, especially for identifying large-area or severe anomalies such as bypass diode failures or heavy soiling, which remain clearly visible even at higher altitudes or low irradiance levels. In contrast, subtle and localized faults such as cell cracks and ribbon disconnections are inherently difficult to detect by IRT, even under optimal imaging conditions, which emphasizes the need for complementary inspection techniques. From an operational perspective, high-altitude flights, despite their reduced diagnostic sensitivity, allow rapid coverage of utility-scale PV plants and would be preferable when the objective is to screen for major faults, whereas low-altitude inspections remain necessary for detailed analysis. Practical constraints such as inspection costs and flight time have not been considered in this study, but they can significantly influence inspection strategies, as frequent low-altitude inspections may be expensive. Optimized flight planning, including hybrid approaches that combine high-altitude scanning for anomaly detection with targeted low-altitude revisits, may help balance accuracy, cost, and efficiency in real-world operations.

Fig. 15 IRT image of two modules with cell interconnection failure (top row) and cell cracks (bottom row) imaged at 8 m and 20 m drone altitude at >800 W/m2. EL image of each module is displayed as a reference.

5 Conclusion

Drone-based infrared thermography is widely employed to assess the operational state of PV power plants. It enables the identification of hotspots in PV modules and strings, which may result from defects, shading, or soiling. While IRT is a valuable diagnostic tool for plant operators, its limitations under non-optimal weather conditions are often overlooked. Current IRT standards recommend that IRT inspections be conducted under irradiance levels above 600W/m², low wind speeds, and minimal cloud cover [6]. However, these ideal conditions are not always achievable all year round in many regions, particularly in northern climates.

Although the thermal signatures of common PV module defects have been previously studied under standard conditions, the behavior of these signatures under sub-optimal weather conditions remains poorly understood. This study investigated the influence of irradiance and drone altitude on the visibility and characteristics of thermal anomalies in IRT images, aiming to assess whether typical PV failures remain detectable under varying inspection scenarios.

An experimental field-based assessment was performed on 43 defective PV modules under controlled field conditions, with irradiance levels ranging from 200W/m² to 1000W/m² and drone altitudes between 8m and 20m (between <1 cm/pixel to 3 cm/pixel in module spatial resolution). The modules exhibited various defect types, including cell cracks, potential-induced degradation, defective cell interconnects, short-circuited bypass diodes, cracked glass, and two forms of soiling. Defects were introduced via stress testing, and modules were operated under realistic conditions in a PV farm environment.

Our findings show that the influence of drone altitude within the tested range is negligible compared to irradiance effects. While increasing altitude naturally reduces spatial resolution, no significant loss of detectability was observed for most defect types. Small defects such as cell cracks and disconnected cell interconnects cannot be identified consistently from healthy cells in practice, regardless of the altitude of the inspection and spatial resolution. In the case of soiling and shorted bypass diodes, the hotspots were large enough to be identified regardless of the drone altitude, although the reduction in spatial resolution showed a decrease in captured module temperatures.

Solar irradiance was found to be a critical parameter for reliable fault detection. While current standards recommend irradiance levels above 600W/m², our results show that certain large-area faults can still be detected at 200W/m². However, small-scale defects become nearly indistinguishable under low irradiance conditions.

Quantitative analysis based on the dataset reveals that a thermal contrast threshold of approximately 5 °C above the healthy module baseline is required for robust detection. Cell cracks and glass cracks, which showed temperature differences below 1 °C, were largely undetectable even under optimal imaging conditions. Interconnect failures exceeded this threshold in severe cases, while shorted bypass diodes and soiling produced the most pronounced hotspots, reaching over 100 °C. In the case of PID and glass cracks, the tested modules did not exhibit strong thermal signatures, and the results are therefore not representative of severe field cases. However, that is also proof that IRT imaging has challenges in diagnosing low severity defects, which have the potential of worsening over extended died exposure, highlighting the importance of alternative and complementary diagnostic tools such as I-V tracing and electroluminescence images.

These results have direct practical implications. For general condition monitoring or the detection of severe, high-impact faults, higher-altitude and low irradiances inspections may suffice, making IRT more viable in regions with limited solar availability.

To further support future efforts in developing and benchmarking automated defect detection algorithms, the IRT image dataset produced in this work has been made publicly available [36]. The dataset provides a comprehensive set of infrared images of the PV modules in this study with their associated EL images and I-V curves. This set of measurements could be useful for developing and benchmarking automated defect detection algorithms. The inclusion of both subtle and severe faults enables the possibility to evaluate the sensitivity of the algorithms to variations in contrast, noise, resolution, and temperature signal.

Further works could also focus on modelling the thermal response of modules of different architectures under varied irradiance conditions to further understand the generation of hotspots and their apparent non-homogeneity across irradiance levels.

Funding

This project has received funding from the European Union under grant agreement no. 101146377 as part of the SOLARIS project − Solar Operational Lifecycle and Asset Reliability Intelligence System.

Conflicts of interest

The authors declare no conflict of interest.

Data availability statement

This article provides the IRT, EL images pairs and IV curves of the defective modules in the study as associated data. The dataset can be found at [36]. Licensed under a CCBY 4.0 license.

Author contribution statement

Conceptualization, R.S., M.D., G.B. and S.S.; Methodology, R.S., R.F., T.K., G.B. and S.S.; Software, R.S., R.F. and T.K.; Validation, R.S., G.B. and S.S.; Formal Analysis, R.S., R.F. and T.K.; Investigation, R.S.; Resources, R.S.; Data Curation, R.S., R.F. and T.K.; Writing − Original Draft Preparation, R.S.; Writing − Review & Editing, R.S., G.B., S.S., A.M., M.D. and T.K.; Visualization, R.S.; Supervision, S.S.; Project Administration, P.P. and S.S.; Funding Acquisition, P.P.

References

All Tables

All Figures

	Fig. 1 Visual representation of the defects investigated. a) EL image of a module with cell cracks. b) EL image of a module with degraded cell interconnects. c) EL image of a module with PID. d) EL image of a module with two short-circuited bypass diodes. e) Visual image of a module with glass cracks. f) Visual image of modules with soiling type 1 and 2.
In the text

	Fig. 2 IRT images example of the different inspections carried out. String with two modules showing shorted bypass diode degradation under four drone altitudes (8, 10, 14 and 20 m) corresponding to <1, 1.5, 2 and 3 cm/pixel of spatial resolution. The images correspond to the 800 W/m2 irradiance inspection.
In the text

	Fig. 3 Top row: Example of a cell cracked module IRT images at different irradiances. EL image as a reference. Bottom row: average module temperature of all cracked modules vs a selection of healthy ones at different irradiances. Altitude of the inspection 8 m.
In the text

Fig. 4 a) EL image of a fielded cracked module. Highlighted cells show an example of a low severity crack with 1.5% cracked area (orange); medium severity with 20.4% cracked area (green) and high severity with 30.6% cracked area (red). b) Boxplot showing the maximum cell temperature trends in drone altitude and cell crack severity compared to reference healthy cells. Irradiance of the inspection >800 W/m2.

In the text

	Fig. 5 Top row: Example of a module with damaged cell interconnects where the IRT images were taken at different irradiances. EL image as a reference. Bottom row: Average module temperature of all modules with degraded cell interconnects vs a selection of healthy ones at different irradiances. Altitude of the inspection 8 m.
In the text

	Fig. 6 a) EL image of a fielded module with damaged cell interconnects. Highlighted cells show an example of a cell interconnect failure Type 1 (orange), Type 2 (green), Type 3 (Red) and Type 4 (Purple). b) Boxplot showing the maximum cell temperature trend in drone altitude and cell disconnection severities compared to healthy cells. Irradiance of the inspection >800 W/m2.
In the text

	Fig. 7 EL image of a PID stress tested module before and after stress test.
In the text

	Fig. 8 Left: IRT image of a part of two PV strings. The top string is healthy; the bottom string is PID affected. Right: Average module temperature of healthy vs PID affected modules as a function of irradiance. Drone flight altitude of the inspection 8 m.
In the text

	Fig. 9 a) IRT images of two modules with a sort-circuited bypass diode at different irradiances imaged at 8 m drone altitude. EL image as a reference. b) Rear side picture of the mouting frame of the PV modules installed in the field. The beam location corresponds to the hoptspot location of the bifacial halfcut modules causing rear side partial shading.
In the text

	Fig. 10 Distribution of maximum module temperatures for each measured plane of array irradiance at all drone altitudes studied. a) Inspection at 8 m. b) Inspection at 10 m. c) Inspection at 14 m. d) Inspection at 20 m. PV modules with short-circuited bypass diodes (red), compared to healthy modules (blue) of the same type.
In the text

	Fig. 11 Visual image, EL image and IRT image of one of the modules with glass cracks, imaged at 800 W/m2 and 8 m drone altitude (<1 cm/pixel moduel spatial resolution). The red square in the EL image indicates the point of impact on stress test that created a small cell crack.
In the text

	Fig. 12 Example of two modules affected by punctual soiling (Soiling 1) imaged at different irradiances and at 8 m drone altitude (<1 cm/pixel moduel spatial resolution). Top: Bifacial PV module with buttterfly configuration and 144 half-cut cells. Bottom: Monofacial PV module with 60 6 × 6 inch cells. IV of the module acquired at approximately 600 W/m2.
In the text

	Fig. 13 Example of two modules affected by large area soiling (Soiling 2) imaged at 800 W/m2 and 8 m drone altitude (<1 cm/pixel moduel spatial resolution). a) Soiling that created hotspots. b) Soiling that did not create hotspots.
In the text

Fig. 14 a) Effect of irradiance in maximum temperature distribution of healthy modules and modules with Soiling 1. Altitude of the inspection 8 m. b) Effect of drone altitude in maximum temperature distribution of healthy modules and modules with Soiling 1. Irradiance of the inspection 800 W/m2. c) Effect of irradiance in maximum temperature distribution of healthy modules and modules with Soiling 2. Altitude of the inspection 8 m. d) Effect of drone altitude in maximum temperature distribution of healthy modules and modules with Soiling 2. Irradiance of the inspection 800 W/m2.

In the text

	Fig. 15 IRT image of two modules with cell interconnection failure (top row) and cell cracks (bottom row) imaged at 8 m and 20 m drone altitude at >800 W/m2. EL image of each module is displayed as a reference.
In the text